STATISTICS THROUGH

APPLICATIONS
This page intentionally left blank
STATISTICS THROUGH
APPLICATIONS
SECOND EDITION

DAREN S. STARNES
THE LAWRENCEVILLE SCHOOL

DANIEL S. YATES
STATISTICS CONSULTANT

DAVID S. MOORE
PURDUE UNIVERSITY

W. H. FREEMAN AND COMPANY / NEW YORK

Executive Publisher: Craig Bleyer
Publisher: Ruth Baruth
Executive Editor: Ann Heath
Senior Developmental Editor: Shona Burke
Executive Marketing Manager: Cindi Weiss
Marketing Manager: Nicole Sheppard Klophaus
Media Editor: Laura Capuano
Associate Editor: Katrina Wilhelm
Editorial Assistant: Catriona Kaplan
Project Editor: Jane O’Neill
Text and Cover Designer: Blake Logan
Photo Editor: Cecilia Varas
Photo Researcher: Elyse Rieder
Senior Illustration Coordinator: Bill Page
Illustrations: Network Graphics
Production Coordinator: Paul Rohloff
Composition: Aptara
Printing and Binding: RR Donnelley

TI-84TM and TI-NspireTM screen shots are used with permission of the publisher: ©1996, Texas Instruments
Incorporated. TI-84TM and TI-NspireTM Graphic Calculators are registered trademarks of Texas Instruments
Incorporated. Minitab is a registered trademark of Minitab, Inc. Microsoft© and Windows© are registered
trademarks of the Microsoft Corporation in the United States and other countries. Excel screen shots are
reprinted with permission from the Microsoft Corporation. Fathom Dynamic Statistics is a trademark
of KCP Technologies.

Library of Congress Control Number: 2009936387

ISBN-13: 9-781-4292-1974-7
ISBN-10: 1-4292-1974-2

© 2011, 2005 by W. H. Freeman and Company

All rights reserved

Printed in the United States of America

First Printing

No part of this book may be reproduced by any mechanical, photographic, or electronic process, or in the
form of a phonographic recording, nor may it be stored in a retrieval system, transmitted, or otherwise
copied for public or private use, without the written permission of the publisher.

W. H. Freeman and Company

41 Madison Avenue
New York, NY 10010
Houndmills, Basingstoke RG21 6XS, England
www.whfreeman.com
 CONTENTS

About the Authors vii CHAPTER 4 DESCRIBING

Preface ix RELATIONSHIPS 141
4.1 Scatterplots and
CHAPTER 1 STATISTICS: THE ART Correlation 142
AND SCIENCE OF DATA 1 4.2 Regression and Prediction 168
1.1 Where Do Data Come From? 2 Chapter 4 Review 194
1.2 Drawing Conclusions from Chapter 4 Review Exercises 195
Data 18
Chapter 1 Review 28
Chapter 1 Review Exercises 29 Part B Producing Data 199

CHAPTER 5 SAMPLING AND

Part A Analyzing Data 33 SURVEYS 201
5.1 Samples, Good and
CHAPTER 2 DESCRIBING Bad 202
DISTRIBUTIONS OF DATA 35 5.2 What Do Samples
2.1 Displaying Distributions Tell Us? 219
with Graphs 36 5.3 Sample Surveys in the
2.2 Describing Distributions Real World 236
with Numbers 59 Chapter 5 Review 252
2.3 Use and Misuse of Statistics 82 Chapter 5 Review Exercises 253
Chapter 2 Review 94
Chapter 2 Review Exercises 96
CHAPTER 6 DESIGNING
EXPERIMENTS 257
CHAPTER 3 MODELING 6.1 Experiments, Good
DISTRIBUTIONS OF DATA 101 and Bad 258
3.1 Measuring Location in a 6.2 Experiments in the
Distribution 102 Real World 273
3.2 Normal Distributions 116 6.3 Data Ethics 292
Chapter 3 Review 136 Chapter 6 Review 305
Chapter 3 Review Exercises 137 Chapter 6 Review Exercises 306
vi Contents

Chapter 9 Review 464

Part C Chance 309
Chapter 9 Review Exercises 466
CHAPTER 7 PROBABILITY: WHAT
ARE THE CHANCES? 311 CHAPTER 10 INFERENCE IN
7.1 Randomness, Probability, PRACTICE 469
and Simulation 312 10.1 Chi-Square Tests: Goodness of Fit
7.2 Probability Rules 329 and Two-Way Tables 470
7.3 Conditional Probability 10.2 Inference about a Population
and Independence 348 Mean 493
Chapter 7 Review 364 Chapter 10 Review 516
Chapter 7 Review Exercises 365 Chapter 10 Review Exercises 517
Notes and Data Sources N-1
CHAPTER 8 PROBABILITY Solutions to Odd-Numbered Exercises S-1
MODELS 369 Index I-1
8.1 Probability Distributions 370 Tables T-1
8.2 Counting and Probability 387 Table A Standard Normal
8.3 Binomial Distributions 399 Probabilities T-1
Chapter 8 Review 413 Table B Random Digits T-2
Chapter 8 Review Exercises 414 Table C t Distribution Critical Values T-3
Table D Critical Values of z* T-4
Table E Chi-Square Critical Values T-4
Part D Inference 417 Additional Topics on CD-ROM and at
www.whfreeman.com/sta2e
CHAPTER 9 INTRODUCTION
S.1 Measuring
TO INFERENCE 419
S.2 The Consumer Price Index and
9.1 What Is a Confidence
Government Statistics
Interval? 420
9.2 What Is a Significance Test? 438
9.3 Use and Abuse of Statistical
Inference 453
 ABOUT THE AUTHORS

DAREN S. STARNES holds the endowed Master Teacher chair in Mathematics at The
Lawrenceville School near Princeton, New Jersey. He earned his MA in Mathematics
from the University of Michigan. In 1997, he received a GTE GIFT Grant to integrate AP
Statistics and AP Environmental Science. He was named a Tandy Technology Scholar
in 1999. Daren has led numerous one-day and weeklong AP Statistics institutes for new
and experienced AP teachers, and he has been a reader, table leader, and question leader
for the AP Statistics exam. In 2001–2002, he served as coeditor of the Technology Tips
column in the NCTM journal The Mathematics Teacher. Since 2004, Daren has served
on the ASA/NCTM Joint Committee on the Curriculum in Statistics and Probability
(which he chaired in 2009). While on the committee, he edited the Guidelines for
Assessment and Instruction in Statistics Education (GAISE) Pre-K–12 Report and
coauthored (with Roxy Peck) Making Sense of Statistical Studies, a capstone module in
statistical thinking for high school students. Daren is also coauthor of the popular text
The Practice of Statistics, second and third editions.

DANIEL S. YATES has taught AP Statistics in the Electronic Classroom (a distance

learning facility affiliated with Henrico County Public Schools in Richmond, Virginia).
Before he was a high school teacher, Dan was on the mathematics faculty at Virginia
Tech and Randolph-Macon College. He has a PhD in Mathematics Education from
Florida State University. He has served as President of the Greater Richmond Council
of Teachers of Mathematics and the Virginia Council of Teachers of Mathematics.
Named a Tandy Technology Scholar in 1997, Dan is a 2000 recipient of the College
Board/Siemens Foundation Advanced Placement Teaching Award. Although recently
retired from classroom teaching, he stays in step with trends in teaching by frequently
conducting College Board workshops for new and experienced AP Statistics teachers
and by monitoring and participating in the AP Statistics electronic discussion group.
Dan is coauthor of The Practice of Statistics, first, second, and third editions.

DAVID S. MOORE is Shanti S. Gupta Distinguished Professor of Statistics at Purdue

University and 1998 President of the American Statistical Association. David is an elected
fellow of the American Statistical Association and of the Institute of Mathematical
Statistics and an elected member of the International Statistical Institute. He has served
as program director for statistics and probability at the National Science Foundation.
David has devoted his attention to the teaching of statistics. He was the content developer
for the Annenberg/Corporation for Public Broadcasting college-level telecourse Against
All Odds: Inside Statistics and for the series of video modules Statistics: Decisions
Through Data, intended to aid the teaching of statistics in schools. He is the author of
influential articles on statistics education and of several leading textbooks, including
Introduction to the Practice of Statistics (written with George P. McCabe and Bruce
Craig), The Basic Practice of Statistics, Statistics: Concepts and Controversies, The
Practice of Business Statistics, The Practice of Statistics in the Life Sciences (written
with Brigitte Baldi), and most recently Essential Statistics.
vii
This page intentionally left blank
 PREFACE
Statistics Through Applications, second edition (STA 2e), is investigate, discuss, and make use of statistical ideas and
designed to support a first course in statistics that empha- methods. Examples and exercises have been carefully
sizes statistical thinking. The focus of this text is on statistical selected to pique students’ interest and curiosity. Although
ideas and reasoning and on their relevance to such fields as the text has a fairly relaxed style and demands only an
medicine, education, environmental science, business, algebra background, its emphasis on ideas and reasoning
psychology, sports, politics, and entertainment. asks more of the student than many texts that emphasize
In its publication Principles and Standards for procedures over concepts.
School Mathematics, the National Council of Teachers Students learn to think about data by working with data.
of Mathematics (NCTM) identifies “Data Analysis and Consequently, we have included many elementary graphical
Probability” as one of five content standards. Here is the and numerical techniques in STA 2e. We have not, however,
precise statement of this NCTM standard: allowed techniques to dominate concepts. Our intention is to
invite discussion and even argument about statistical ideas
Instructional programs from prekindergarten rather than to focus exclusively on computation (though
through grade 12 should enable all students to— some computation remains essential). The coverage in
• formulate questions that can be addressed with STA 2e is considerably broader than that of some traditional
data and collect, organize, and display relevant statistics texts, as the table of contents reveals.
data to answer them;
• select and use appropriate statistical methods to THE STRUCTURE OF STA 2e
analyze data; STA 2e contains 10 chapters organized in four parts:
• develop and evaluate inferences and predictions analyzing data, producing data, chance, and inference. Each
that are based on data; chapter consists of two or three sections, each devoted to a
• understand and apply basic concepts of probability.1 single coherent set of ideas. Sections are further divided into
small subsections of material that can be fully addressed in
More recently, the American Statistical Association (ASA) a standard 45- or 50-minute high school period. Exercises
published its Guidelines for Assessment and Instruction appear at the end of each subsection, section, and chapter.
in Statistics Education (GAISE) Pre-K–12 Report to Summaries at the end of each section and chapter help
elaborate on the NCTM Data Analysis and Probability students distill what they have learned.
standard. The joint curriculum committee of the ASA
and the Mathematical Association of America (MAA)
recommends that any first course in statistics “emphasize THE SECOND EDITION
the elements of statistical thinking” and feature “more In revising Statistics Through Applications, we have
data and concepts, fewer recipes and derivations.”2 STA 2e attempted to build on the elements that made the first
incorporates the recommendations made by the NCTM, edition successful:
ASA, and MAA by taking a conceptual and verbal approach • Clear explanations of statistical ideas and
rather than a methods-oriented one. terminology are presented.
Statistics Through Applications is ideally suited for • Hands-on Activities that allow students to explore
a non-AP-level introduction to statistics for high school statistical concepts begin each section.
students. It may be used effectively in either a one- or a
• Abundant examples from a variety of fields
two-semester course.
illustrate important statistical ideas.
• Varied exercises (with descriptive titles) reflect the
THE NATURE OF STA 2e usefulness of statistics in many different subject areas.
Statistics Through Applications is written to be read by
students. It is somewhat informal, with thought-provoking • Numerous cartoons enliven the pages.
stories (“nuggets”) in the margins and cartoons and photo- • More than 30 short marginal nuggets offer
graphs interspersed throughout. Activities, Applications, interesting and often amusing anecdotes about
and Data Explorations give students an opportunity to statistics and its impact.
ix
x Preface

• Applications at the end of each section require • Many of the Activities that open each section have
students to put what they have learned to use in a been revised to engage students in “doing statistics,”
real-world setting. and a number of the Applications have been updated
• A short Summary reviews the big ideas at the end to ask students to critique and interpret results from
of each section. Each chapter concludes with a published statistical studies.
Chapter Review that lists the skills students should • New Data Exploration boxes put students in the
have acquired by that point and Chapter Review role of data detectives as they use graphical and
Exercises to check understanding. numerical tools to investigate variation in data.
• The four-step statistical problem-solving process,
introduced on page 19 and first used in Example 1.10,
WHAT’S NEW IN THE SECOND EDITION? provides a framework for helping students to think
Combining our classroom experience, input from users of
through the process of asking and answering good
the first edition, and a careful review of NCTM, ASA, and
statistical questions. This four-step process is used
MAA recommendations, we have made some important
as a learning tool to aid in interpreting and analyzing
changes in content and features. Here is a brief summary.
data in selected examples and exercises.
Content Changes • Calculator Corners provide step-by-step instructions
In response to users’ overwhelming preference to get with screen shots for performing important statistical
students involved in doing data analysis more quickly, we functions on the TI-84 and the TI-Nspire.
have reorganized the first six chapters. In addition, we • Solutions to odd-numbered exercises in the back
have expanded the coverage of probability concepts and of the book now include figures that are part of the
techniques in Chapters 7 and 8. answers.
• The revised Chapter 1 gives students an overview • Expanded use of technology—Java applets,
of the statistical problem-solving process—from computer software, and Web links—helps enrich AP
PLET

defining a research question to producing, student learning. Look for the new Applet icon and
analyzing, and drawing conclusions from data. Web icon. www

• Chapters 2, 3, and 4 provide a foundation in • Output from Fathom, Minitab, and Excel is
exploratory data analysis—graphical and numerical integrated in examples and exercises throughout the
methods for describing one- and two-variable data. textbook.
• The details of sampling, surveys, and experiments
can now be found in Chapters 5 and 6. SUPPLEMENTS
• We have significantly expanded the coverage of A full range of supplements is available to help teachers
probability in Chapters 7 and 8 to include conditional and students use STA 2e.
probability, basic counting techniques (permutations
Teacher’s Resource Binder (TRB),
and combinations), and binomial distributions.
1-4292-5111-5
• In Chapter 10, we added the chi-square test for
This indispensable resource compiled by Daren Starnes,
goodness of fit and inference for means using the
Dan Yates, and Michael Legacy (The Greenhill School)
t distributions.
gives teachers a multitude of useful tools for planning
• Former Section 1.2, “Measuring,” and Section 5.2, and managing their courses, assessing student progress,
“The Consumer Price Index and Government and enhancing day-to-day instruction. The TRB offers the
Statistics,” have been removed from the textbook. following:
They are still available on a CD and at the STA 2e
• Two sample quizzes for every section of every chapter
Web site, www.whfreeman.com/sta2e.
• Two sample chapter tests for every chapter
Features • Extensive teaching suggestions, including additional
• Four-color design and photographs add interest examples for in-class use, for each chapter
value and visual appeal. • Additional Activities and Applications
• Each chapter begins with a brief opening story that • New worksheets that focus on acquisition of
gives an overview of issues that will be addressed in vocabulary and basic skills, many written by Sally
the pages ahead. Miller of T. C. Williams High School
 Preface xi

• Suggestions for short-term and long-term projects, Resources include the following:
with sample grading rubrics • Online quizzes with instant feedback to help
• Expanded technology tips for using the TI-84, students master material and prepare for testing
TI-Nspire, Fathom software, and Java applets • Interactive statistical applets that allow students to
• An extensive list of references to books, journal manipulate data and see the corresponding results
articles, video series, data sources, interesting graphically
statistics-related Web sites, and more • Calculator program files and data sets
• Key tables and formulas summary sheet
Instructor’s Solutions Manual (ISM),
• All tables from the text in .pdf format for quick,
1-4292-5110-7
easy reference
Written by Sarah Streett and Duane Hinders, the ISM
has been thoroughly revised and updated to reflect the • Optional content (S.1, “Measuring,” and S.2, “The
content changes made to the second edition. It contains Consumer Price Index and Government Statistics”)
fully worked out solutions to every problem in the book. • Statistics on the Web: links to useful sites and
Where appropriate, solutions now follow the four-step resources, such as the Census Bureau, Gallup Poll,
statistical problem-solving process outlined in the text. Nielsen, etc.

Instructor’s Resource CD, 1-4292-5107-7 Interactive eBook (Access code or

The CD contains the following: online purchase required)
• Microsoft Word files of the contents of the Teacher’s This online resource integrates a complete version of the
Resource Binder text with its media resources for ease of use. Students
may quickly search the text and personalize the eBook
• The Teacher’s Solutions Manual in Adobe .pdf
with notes, highlights, and bookmarks just as they would
format
with the print version.
• All text figures in an exportable presentation format,
• Allows students and teachers to bookmark pages,
JPEG for Windows users and PICT for Macintosh
take notes, and highlight content
users
• Provides answers to the odd-numbered problems
• PowerPoint Slides that may be used directly or
with a click of the “Short Answer” button, allowing
customized to fit your needs
students to check their answers immediately without
• Applets that allow students to manipulate data and having to flip to the back of the book
see the corresponding results graphically
• Applets are linked directly to the supporting
• Data sets from the text in a variety of formats activity via hot-linked icons that encourage
• Sections 1.2 and 5.2 from the first edition in .pdf students to explore and experience hands-on
format simulations
• Multiple-choice questions from the Test Bank as • Data sets and calculator programs are accessed
Microsoft Word files to allow teachers to add, edit, easily when they are needed
and resequence test items • Definitions to key terms are easily found through
the online glossary
Printed Test Bank, 1-4292-5108-5 • EESEE Case Studies developed by The Ohio
Tim Brown (The Lawrenceville School) has updated the
State University Statistics Department provide
printed Multiple-Choice Test Bank to reflect the changes
additional support to teach students to apply their
to the second edition and has added 150 new problems,
statistical skills by exploring actual case studies,
bringing the total number to 1000.
using real data, and answering questions about
Student Companion Web Site the study
(www.whfreeman.com/sta2e) ACKNOWLEDGMENTS
The STA 2/e Student Companion Web Site offers a variety To Craig Bleyer and Ruth Baruth, I offer sincere gratitude
of resources to help students succeed in their high school for their unwavering support of STA as it continues to evolve.
statistics course. This site is open to all students; no My heartfelt thanks go to Shona Burke for shepherding the
registration is required. manuscript through development, and especially for her
xii Preface

timely responses to my occasionally frantic emails and Cheryl Hughes, Providence School, FL
phone calls. Jennifer Battista did yeoman’s work coordinat- Jean-Marie Johnson, Shrewsbury High School, MA
ing the external reviews of the manuscript and consolidating
Melissa Johnson, Robinson Secondary School, VA
disparate opinions into coherent recommendations. Thanks
to Ann Heath, who came in late in the game as our “closer” John Thomas Matsik, West Mecklenburg High
and pitched a shutout with her careful attention to the design School, NC
of the new edition and the supplements and saw us through Sally C. Miller, T. C. Williams High School, VA
production. Pamela Bruton did her usual superb copyediting, Jim Mydloski, Monroe High School, MI
leaving no stone unturned and no grammar rule violated.
Marcia Prill, Madison High School, NJ
Jane O’Neill capably fulfilled her role as project editor, and
Blake Logan worked tirelessly to help us develop and refine Sue Ravalese, Concord-Carlisle Regional High
the design and cover. School, MA
All of us involved in the project appreciate the thoughtful Gerry Reynolds, Wayne Valley High School, NJ
comments and suggestions made by our colleagues who Linda Ruppenthal, Hancock High School, MD
reviewed draft chapters of the STA 2e manuscript:
Colleen A. Russell, Warwick Valley High School, NY
Matthew J. Adam, McHenry High School—West Dianne Saugier, Granada High School, CA
Campus, IL Tom Shafer, Leslie High School, MI
Peter R. Apple, Badger High School, WI Holly K. Shrader, Edison High School, OH
Barbara Aylworth, Alexander High School, GA Allison Stanley, McHenry High School—East
Virginia Blake, Lincoln-Sudbury Regional High Campus, IL
School, MA Matt Townsley, Solon High School, IA
Denise Bogues, Hartland High School, MI Natalie Tucker, Concord High School, NC
Kevin Brenner, Cloquet Senior High School, MN Doug Tyson, Central York High School, PA
Timothy Brown, Lawrenceville School, NJ Melissa Warfield, Boulder High School, CO
Katie Bruck, Secondary Technical Education Lisa Weiss, Lincoln-Sudbury Regional
Program, MN High School, MA
Robert Bunio, Anoka High School, MN David White, Ponderosa High School, CA
Julie A. Burke, North Springs Charter School, GA Robert Wickwire, Simi Valley High School, CA
Joani Byrum-Wright, Louisville Male High Jodi N. Wilson, West Carrollton High School, OH
School, KY Jeannette Wilt, Brandywine High School, DE
Lyn Davies, Denver School of the Arts, CO Christine Wittenberg, Warwick Valley High School, NY
Diane DeSantis, Algonquin Regional
High School, MA Special thanks go to Tim Brown (The Lawrenceville
School) and Michael Legacy (The Greenhill School) for
Allan DeToma, Pinewood Preparatory School, SC
their detailed reviews of the chapters that involved major
Melissa E. DosSantos, Bay Shore High School, NY content changes in the new edition and for their efforts
Erika Draiss, Celina High School, OH in revising the Test Bank and Teacher’s Resource Binder,
Karen Eggebrecht, Charlotte Catholic High respectively. I also owe a debt of gratitude to Eric Zagorski
School, NC from The Lawrenceville School, who proved his mettle as
one of my AP Statistics students and as a student reviewer for
Vickie Fosdick, David W. Butler High School, NC
STA 2e. Eric offered some remarkably insightful feedback
Terry C. French, Lake Braddock Secondary School, VA at several critical stages in the writing process. Finally,
Eva Galdo, Robinson Secondary School, VA I would like to thank Sarah Streett for her thorough and
Adam Goudge, Wheat Ridge High School, CO careful work in preparing the Instructor’s Solutions Manual
and Ann Cannon (Cornell College) for her fastidiousness in
Steve Haldeman, Hempfield High School, PA
checking the book and solutions for accuracy.
Katherine Harris, Patrick Henry High School, VA As the saying goes, behind every good man, there is a
Karen Huff, Spencer–Van Etten High School, NY great woman. For those of you who have had the pleasure
 Preface xiii

of meeting my wife and soul mate, Judy, you know that the necessity. With the publication of STA in 2005, Dan’s
adage is true. I cannot possibly thank her enough for the vision became reality. His pioneering efforts have touched
sacrifices she has made to ensure that STA 2e got done well the lives of thousands of students and teachers.
and on time. She inspires me every day. As I started work on the second edition, Dan provided
invaluable advice on several potential changes. He read
every manuscript chapter, offering meaningful sugges-
tions that enhanced the quality of the finished product.
Dan’s influence can be seen throughout the pages of
Statistics Through Applications was Dan Yates’s STA 2e, from his conviction that students learn statistics
brainchild. He articulated a clear vision for STA as a text best by doing statistics (hence all the Activities, Applications,
that would show students the applicability of statistics in and Data Explorations) to the many cartoons pulled from
their daily lives. Following his successful launch of The his extensive collection.
Practice of Statistics as the first book written specifically To my mentor, colleague, and friend, Dan Yates, I
for high school AP Statistics, Dan persuaded the good dedicate STA 2e as a tribute to his innumerable contri-
folks at W. H. Freeman and Company that a textbook for butions to statistics education.
non-AP high school statistics students was an absolute Daren S. Starnes
This page intentionally left blank
 CHAPTER 1

Statistics: The Art

and Science of Data
1.1 Where Do Data Come From?
1.2 Drawing Conclusions from Data

Alamy
Improving health with the Wii?
In the past, video game players spent hours sitting idly in front of a television
or computer screen. Critics argued that excessive video game play was a major
contributor to the increase in obesity. With the growing popularity of the
Nintendo Wii, however, the days of passive video gaming may be over. More
importantly, the Wii may actually lead to better health for some individuals.
Some doctors have begun using the Wii to help stroke victims regain
movement in their arms and legs. Patients who find traditional exercise too
boring or too difficult are motivated to play the motion-intensive Wii games.
“Wii therapy” has been used successfully with recovering military patients, too.
The Wii has also caught on among senior citizens, who enjoy the social and
competitive nature of games like bowling, tennis, and golf.
Research has even been conducted on whether playing Wii games regularly
can help prevent childhood obesity. A study at Liverpool’s John Moores
University compared teenagers’ energy use when playing with a traditional video
game console and when using the Wii console. The 12 participants in the study
increased their energy expenditure by an average of 60% over resting values
when playing the traditional video game. When playing with the Wii, the teens
increased their energy expenditure by an average of 156% over resting values.
However, as Professor Tim Cable, the lead researcher for this study points out,
“Parents should encourage other physical activities and outdoor pursuits in
order for their children to lead well-balanced lives.”1

1
2 CHAPTER 1 Statistics: The Art and Science of Data

1.1 Where Do Data Come From?

ACTIVITY 1.1
The “real facts”
Materials: Computer with Internet access • Termites eat through wood 2 times faster when
Do you drink Snapple? If so, then you have probably listening to rock music.
noticed a “Real Fact” printed on each bottle cap. • 27% of Americans have spent at least one night
Here is one we found: Licking a stamp burns 10 in jail. (This one is actually an incorrect choice in
calories. How did they determine this? Does this a previous Real Facts game at the Snapple Web
“fact” seem reasonable? In this Activity, you will site. Do you think the correct percent is higher
start to think carefully about how the data were or lower? Why?)
produced.
2. On each Snapple bottle cap, you are
www
1. Consider the following “Real Facts.” For each encouraged to “Get all the Real Facts at
one, (a) discuss whether the result makes sense, and snapple.com.” Go to the Snapple Web site
(b) suggest how you might find data to support or (www.snapple.com) and locate the Real Facts page.
oppose the “fact.”
3. Find two “Real Facts” that are different from
• The average American walks 18,000 steps a
the ones above and that involve a numerical result.
day.
Record each fact. Then tell how you think the
• August has the highest percent of births.
data were produced and whether the result makes
• The average person spends about 2 years on the
sense.
phone in a lifetime.

You can hardly go a day without meeting data and statistical studies. Have you ever
wondered how the numbers were produced? Consider these examples.
• In the year 2007, 12.5% of drivers involved in fatal crashes were 15
to 20 years old, according to the National Highway Traffic Safety
Administration.2
• According to researchers in the Netherlands, people who are 30 or
more pounds overweight could lose up to 7 years from their life
expectancy.3
• The Gallup Poll reports (November 2007) that 51% of U.S. adults engage in
“vigorous exercise” at least once a week.
• A major medical study concluded that taking aspirin regularly reduces the
risk of a heart attack.
Where do these data come from? Why can we trust them? Or maybe we can’t
trust them. You can observe a lot just by watching. But you can’t discover just by
watching that 51% of adults exercise vigorously or that aspirin reduces heart attack
risk. Good data are the product of intelligent human effort. Bad data result from
laziness or lack of understanding, or even the desire to mislead others. “Where do
the data come from?’’ is the first question you should ask when someone throws a
number at you.
 1. 1 Where Do Data Come From? 3

Data beat personal experiences

Suppose you want data on a question that interests you. Do cell phones cause can-
cer? What percent of people recycle? Does listening to music while studying im-
prove student learning? Why did pollsters incorrectly predict the results of a recent
election? Can a new drug help people quit smoking? How will you get relevant data
to help answer your question?
It is tempting to base conclusions on your own experiences. You probably
know many people who use cell phones. Do they have cancer? And polls aren’t
always right, are they? If one of your friends swears that listening to music helps
him concentrate better, you might be persuaded to try it. But our own experiences
may not be typical. In fact, the incidents that stick in our memory are often unusual.
We are much safer relying on carefully produced data.

Example 1.1 Do cell phones cause brain cancer?

In August 2000, Dr. Chris Newman appeared as a guest on CNN’s Larry King Live.
Dr. Newman had developed brain cancer. He was also a frequent cell phone user.
Dr. Newman’s physician suggested that the brain tumor may have been caused
by cell phone use. So Dr. Newman decided to sue the cell phone maker, Motor-
ola, and the phone company that provided his service, Verizon. As people heard
Dr. Newman’s sad story, they began to worry about whether their own cell phone
use might lead to cancer.
Since 2000, several statistical studies have investigated the link between
cell phone use and brain cancer. One of the largest was conducted by the Dan-
ish Cancer Society. Over 400,000 Denmark residents who regularly used cell
phones were included in the study. Researchers compared the brain cancer rate
for the cell phone users with the rate in the general population. The result: no
difference.4 In fact, most studies have produced similar conclusions. In spite of
the evidence, many people are still convinced that Dr. Newman’s experience is
typical.

Talking about data: individuals and variables

Statistics is the art and science of dealing with data. Good judgment, good math,
and even good taste make good statistics. A big part of good judgment is deciding
what to measure in order to produce data that help answer your questions. Measure-
ments are made on individuals and organized in variables.

Individuals and variables

Individuals are the objects described by a set of data. Individuals may

be people, animals, or things.

A variable is any characteristic of an individual. A variable can take

different values for different individuals.
4 CHAPTER 1 Statistics: The Art and Science of Data

For example, here is a small part of a data set from the Cyber Stat Corporation:

A B C D E F
1 Name Job Type Age Gender Race Salary
2 Cedillo, Jose Technical 27 Male White 52,300
3 Chambers, Tonia Management 42 Female Black 112,800
4 Childers, Amanda Clerical 39 Female White 27,500
5 Chen, Huabang Technical 51 Male Asian 83,600
6
7
8
Ready NUM

The individuals are company employees. In addition to each employee’s name,

there are five variables. The first says what type of job the employee holds. The sec-
ond variable gives the age, the third records gender, and the fourth reports race.
The fifth variable records employee salaries.
Statistics deals with numbers, but not all variables are numerical. Of the five
variables in the company’s data set, only age and salary have numbers as values. It
makes sense to talk about the average age or the average salary of the company’s
employees. However, it wouldn’t make sense to ask about the average job type,
average race, or average gender. For nonnumerical variables like these, we can use
counts or percents to summarize the data. We might give the percent of employees
who work in management, for example, or the number who are Asian.
When we examine data, it is helpful to distinguish these two types of variables.

Categorical and quantitative variables

A categorical variable places an individual into one of several groups

or categories.

A quantitative variable takes numerical values for which arithmetic

operations such as adding and averaging make sense.

For the data from Cyber Stat Corporation, age and salary are quantitative vari-
ables. Job type, gender, and race are categorical variables.
Deciding what variables to measure and how to measure them isn’t easy. Bad
judgment in choosing variables can result in data that cost lots of time and money
to produce but that don’t tell us much. Here is an example that shows some chal-
lenges in deciding what and how to measure.

Example 1.2 Who recycles?

Do wealthier people recycle more than poorer people? Researchers spent lots of time
and money weighing the stuff put out for recycling in two neighborhoods in a Califor-
nia city, call them Upper Crust and Lower Mid. The individuals here are households,
 1. 1 Where Do Data Come From? 5

because trash and recycling pickup are done for residences. The vari-
able measured was the weight in pounds of the curbside recycling basket
each week.
The Upper Crust households contributed more pounds per week on
average than those in Lower Mid. Can we say that the rich are more seri-
ous about recycling? No. Someone noticed that Upper Crust recycling
baskets contained lots of heavy glass wine bottles. In Lower Mid, they put
out lots of light plastic soda bottles and light metal beer and soda cans.
Weight tells us little about commitment to recycling.5
Jupiterimages

Exercises 1.1 Making the grade Here are a few lines from a statistics teacher’s grade book:
Name S Homeroom Gr Calculator # Test 1
Hsu, Danny M Blair Sr. B319 81
Iris, Francine F Kingsley Sr. B298 92
Ruiz, Ricardo M Alfonso Jr. B304 87

(a) What individuals does this data set describe?

(b) For each individual, what variables are given? Identify each variable as
categorical or quantitative.
1.2 Hotel choices A high school’s debate team is planning to go to Salt Lake City
for a three-day competition. Team members will be responsible for sharing the
cost of staying in a hotel for two nights. The competition’s sponsor provides a list
of available hotels, along with some information about each hotel. Here is a chart
that summarizes the hotel options.

Exercise Internet Distance to Room Room rate

Hotel Pool room? ($/day) Restaurants competition service? ($/day)
Comfort Inn Out Y 0 1 8.2 mi Y 149
Fairfield Inn & Suites In Y 0 1 8.3 mi N 119
Baymont Inn & Suites Out Y 0 1 3.7 mi Y 60
Chase Suite Hotel Out N 15 0 1.5 mi N 139
Courtyard In Y 0 1 0.2 mi Dinner 114
Hilton In Y 10 2 0.1 mi Y 156
Marriott In Y 9.95 2 0.0 mi Y 145

(a) What individuals does this data set describe?

(b) Clearly identify each of the variables. Which are quantitative?
(c) If you were a debate team member, which hotel would you recommend? Why?
1.3 Who recycles more? In Example 1.2, weight is not a good measure of which of
the two neighborhoods recycles more. What variables would you measure instead?
1.4 Choosing a college Popular magazines rank colleges and universities on their
“academic quality.” Describe four variables that you would like to see measured for
6 CHAPTER 1 Statistics: The Art and Science of Data

each college if you were choosing where to apply. Identify each as categorical or
quantitative. Give reasons for your choices.
1.5 More “Real Facts” According to the Real Facts page at www.snapple.com,
www
Americans on average eat 18 acres of pizza every day.
(a) Tell how you think the data were produced.
(b) Discuss whether this result makes sense.
1.6 Chart toppers Visit the Billboard Music Web site at www.billboard.com.
www
(a) What are the top five songs on the Billboard Hot 100 chart?
(b) How do you think they determined these rankings?

Observational studies
Sometimes all you can do is watch. To learn how chimpanzees in the wild behave,
watch. To study how a teacher and young children interact in a schoolroom, watch.
It helps if the watcher knows what to look for. The chimpanzee expert may be inter-
ested in how males and females interact, in whether some chimps in the troop are
dominant, in whether the chimps hunt and eat meat. In fact, chimps were thought to
be vegetarians until Jane Goodall watched them carefully in Gombe National Park,
Tanzania. Now it is clear that meat is a natural part of the chimpanzee diet.
At first, the observer may not know what to record. Eventually patterns seem
to emerge. Then we can decide what variables we want to measure. How often do
chimpanzees hunt? Alone or in groups? How large are hunting groups? Males alone,
or both males and females? How much of the diet is meat? Observation that is orga-
nized and measures clearly defined variables is more convincing than just watching.
A carefully designed observational study is far better than educated guesswork.

Observational study

An observational study observes individuals and measures variables

of interest but does not attempt to influence the responses. The pur-
pose of an observational study is to describe some group or situation.

Here is an example of highly organized observation.

Example 1.3 Pain and suffering

An Associated Press news article begins, “people hurt in traffic accidents actually
recover more quickly when they cannot collect money for their pain and suffering,
researchers say in a new study.”6 The Canadian province of Saskatchewan changed
its insurance laws. The old system allowed lawsuits for “pain and suffering.” The new
no-fault system paid for medical costs and lost work but not for subjective suffer-
ing. The study looked at insurance claims filed between July 1 of the year before the
change and December 31 of the year after the change. Under the new system there
were not just fewer claims of whiplash neck injuries but faster recovery with less
pain for the people who filed claims. This is a comparative observational study.
 1. 1 Where Do Data Come From? 7

It is possible that the effect of the change in the insurance system is mixed up
with some other difference between the two time periods. The study gives reasonably
convincing evidence that people report less pain when they can’t collect money for it.
But the study’s design prevents us from concluding that the change in insurance law
causes a decrease in pain.

Sampling
You don’t have to drink the whole pitcher of lemonade to know that it’s too sweet. A
few sips from a well-mixed pitcher will tell you. That’s the idea of sampling: to gain
information about the whole by examining only a part. Observational studies that
use sampling are called sample surveys. They attempt to measure characteristics of
some group of individuals (the population) by studying only some of its members
(the sample). The individuals in the sample are selected not because they are of
special interest but because they represent the larger group. Here’s the vocabulary
we use to discuss sampling.

Populations and samples

The population in a statistical study is the entire group of individuals

about which we want information.

A sample is a part of the population from which we actually collect

information, which is then used to draw conclusions about the
whole.

Notice that the population is the group we want to study. If we want informa-
tion about all U.S. high school students, that is our population even if students at
only one high school are available for sampling. To make sense of any sample re-
sult, you must know what population the sample represents. Did a preelection poll,
for example, ask the opinions of all adults? Citizens only? Registered voters only?
Democrats only? The sample consists of the people we actually have information
about. If the poll can’t contact some of the people it selected, those people aren’t in
the sample.
The distinction between population and sample is basic to statistics. The fol-
lowing examples illustrate this distinction and also introduce some major uses of
sampling. These brief descriptions also indicate the variables measured for each in-
dividual in the sample.

Example 1.4 Public opinion polls

Polls such as those conducted by Gallup and many news organizations ask people’s
opinions on a variety of issues. The variables measured are responses to questions
about public issues. Though most noticed at election time, these polls are conducted
on a regular basis throughout the year. For a typical opinion poll:
8 CHAPTER 1 Statistics: The Art and Science of Data

Population: U.S. residents 18 years of age and over. Noncitizens and even illegal
immigrants are included.
Sample: Between 1000 and 1500 people interviewed by telephone.
www For more information about Gallup polls, visit www.gallup.com.

Example 1.5 The Current Population Survey

Government economic and social data come from large sample surveys of a nation’s
individuals, households, or businesses. The monthly Current Population Survey
(CPS) is the most important government sample survey in the United States. Many
of the variables recorded by the CPS concern the employment or unemployment of
everyone over 16 years old in a household. The government’s monthly unemploy-
ment rate comes from the CPS. The CPS also records many other economic and
social variables. For the CPS:
Population: The more than 110 million U.S. households. Notice that the in-
dividuals are households rather than people or families. A household consists of
all people who share the same living quarters, regardless of how they are related to
each other.
Sample: About 50,000 households interviewed each month.
www For more information about the Current Population Survey, visit www.bls.gov/cps/.

Example 1.6 TV ratings

Market research is designed to discover what consumers want and what products
they use. One example of market research is the television-rating service of Nielsen
Media Research. The Nielsen ratings influence how much advertisers will pay to
sponsor a program and whether or not the program stays on the air. For the Nielsen
national TV ratings:
Population: The over 110 million U.S. households that have a television set.
Sample: About 25,000 households that agree to use a “people meter’’ to record
the TV viewing of all people in the household.
The variables recorded include the number of people in the household and
their ages and gender, whether the TV set is in use at each time period, and, if so,
what program is being watched and who is watching it.
For more information on Nielsen ratings, visit www.nielsenmedia.com and
www
search for “television rankings.”

Example 1.7 The General Social Survey

Social science research makes heavy use of sampling. The General Social Survey
(GSS), carried out every second year by the National Opinion Research Cen-
ter at the University of Chicago, is the most important social science sample
survey. The variables cover the participant’s personal and family background,
 1. 1 Where Do Data Come From? 9

experiences and habits, and attitudes and opinions on subjects from abortion
to war.
Population: Adults (aged 18 and over) living in households in the United
States. The population does not include adults in institutions such as prisons and
college dormitories. It also does not include persons who cannot be interviewed
in English.
Sample: About 3000 adults interviewed in person in their homes.
www For more information on the General Social Survey, visit www.norc.org.

We reserve the term “sample survey” for studies that use an organized plan to
choose a sample that represents some specific population, like those in Examples
1.4 to 1.7. By our definition, the population in a sample survey can consist of people,
animals, or things. Some people use the terms “survey” or “sample survey” to refer
only to studies in which people are asked one or more questions, like the opinion poll
of Example 1.4. We’ll avoid this more restrictive terminology.
Most statistical studies use samples in the broad sense. Consider the people who
had traffic accidents in the year following the change in insurance laws from Exam-
ple 1.3 (page 6). They are supposed to represent all accident victims (not just those
from that particular year) following the change in insurance laws. Expert judgment
says they are probably typical, but we can’t be sure. A sample survey doesn’t rely on
judgment: it starts with an entire population and chooses a sample to represent it.
Chapter 5 discusses the art and science of sample surveys.

Exercises 1.7 Populations and samples, I For each of the following sampling situations,
identify the population and the sample as exactly as possible.
(a) A furniture maker buys hardwood in large batches. The supplier is supposed
to dry the wood before shipping (wood that is not dry won’t hold its size and
shape). The furniture maker chooses five pieces of wood from each batch and
tests their moisture content. If any piece exceeds 12% moisture content, the entire
batch is sent back.
(b) An insurance company wants to monitor the quality of its procedures for
handling loss claims from its auto insurance policyholders. Each month the
company selects a sample from all auto insurance claims filed that month to
examine the accuracy and promptness with which they were handled.
1.8 Populations and samples, II For each of the following sampling situations,
identify the population and the sample as exactly as possible.
(a) A business school researcher wants to know what factors affect the survival
and success of small businesses. She selects a sample of 150 restaurants from those
listed in the phone book for a large city.
(b) Your local television station wonders if its viewers would rather watch a
local college basketball team play or an NBA game scheduled at the same time.
It announces that it will show the NBA game and receives 89 calls asking that it
show the local game instead.
10 CHAPTER 1 Statistics: The Art and Science of Data

1.9 Bad apples A truckload of apples arrives at an apple juice production

plant. The plant’s quality control team selects three large buckets of apples from
various locations within the truck. These apples are inspected carefully. Based
on inspection results, the entire truckload is either accepted or rejected by
the plant. Identify the population, sample, individuals, and variable(s) in this
setting.
1.10 J. K. Rowling’s words Different types of writing
can sometimes be distinguished by the lengths of the
words used. A student interested in this fact wants to
study the lengths of words used by J. K. Rowling in her
Harry Potter books. She opens a Harry Potter book at
random and records the lengths of the first 50 words on
the page.
(a) What is the population in this study? What is the
sample?
AP Photo/Bizuayehu Tesfaye (b) What variable does the student measure?
1.11 Nielsen ratings Find a current copy of the Nielsen television ratings. (Many
local newspapers report these ratings weekly.)
(a) What individuals are being measured?
(b) What variables are recorded? For each quantitative variable, identify the units
of measurement.
(c) How were the data produced?
1.12 Gallup Poll Go to the Gallup Poll Web site, www.gallup.com. Locate the
www
results of a sample survey on a topic that interests you.
(a) What questions did the survey include?
(b) Who were the individuals in the sample?
(c) Summarize one or two of the important results from the survey. Be
sure to cite source information for your article, including the title, date,
and Web URL.

Census
A sample survey looks at only a part of the population. Why not look at the entire
population? A census tries to do this.

Census

A census is a sample survey that attempts to include the entire population

in the sample.

The U.S. Constitution requires a census of the American population every

10 years. A census of so large a population is expensive and takes a long time.
 1. 1 Where Do Data Come From? 11

Even the federal government, which can afford a census, uses samples such as
the Current Population Survey to produce timely data on employment and many
other variables. If the government asked every adult in the country about his or
her employment, this month’s unemployment rate wouldn’t be available until
next year!
So time and money favor samples over a census. Samples can have other ad-
vantages as well. If you are testing fireworks or fuses, the individuals in the sample
are destroyed. Moreover, a sample can produce more accurate data than a census. A
careful sample of an inventory of spare parts will almost certainly give more accu-
rate results than asking the employees to count all 500,000 parts in the warehouse.
Bored people don’t count accurately.
Until 2000, even the every-10-years census included a sample survey.
Is a census old- A census “long form’’ that asked many more questions than the basic cen-
fashioned? sus form was sent to a sample of one-sixth of all households. Starting with
The United States has taken the 2010 census, the long form will be replaced by the American Community
a census every 10 years Survey (ACS).
since 1790.Technology
marches on, however, and
replacements for a national
census look promising.
Example 1.8 Census undercounts and overcounts
Denmark has no census, and
France plans to eliminate A census can only attempt to sample the entire population. The Census Bureau
its census. Denmark has a estimates that the 1990 census missed 1.6% of the American population. These
national register of all its res- missing persons included an estimated 4.4% of the black population, largely in
idents, who carry identifica- inner cities.7 In 2000, the census overcounted the U.S. population by about 1.3
tion cards and change their
register entry whenever they million. Millions of people who live in two places—like college students—were
move. France will replace counted twice. About 1.8% of blacks and 0.7% of Hispanics were not counted
its census by a large sample at all. A census is not foolproof, even with the resources of the government
survey that rotates among behind it.
the nation’s regions.The U.S.
Census Bureau has a similar
idea: the American Community
Survey has already started Why take a census at all? The government needs block-by-block population
and will replace the census figures to create election districts with equal populations. The main function of the
“long form.”
U.S. census is to provide this local information.
What kind of information can we get from the U.S. census? You’ll find out in
the following Data Exploration.

DATA EXPLORATION
Sampling from the Census
We used Fathom software to take a random sample data for the first 10 people selected in the sample.
of 100 individuals from the 2000 U.S. census. For (The full data set is available on the book’s Web site,
each individual, the values of several variables were www.whfreeman.com/sta2e.) www
recorded. The table on the following page displays
DATA EXPLORATION (continued)
12 CHAPTER 1 Statistics: The Art and Science of Data

DATA EXPLORATION (continued)

Sex Age Race Siblings Marital status Education

Male 52 White 0 Married_ spouse present 1 to 3 years of college
Male 50 White 0 Married_ spouse present 1 to 3 years of college
Female 8 White 3 Never married/single (N/A) Grade 1_ 2_ 3_ or 4
Female 18 Black/Negro 2 Never married/single (N/A) Grade 11
Female 7 White 1 Never married/single (N/A) None or preschool
Female 60 White 0 Married_ spouse present Grade 12
Female 11 White 1 Never married/single (N/A) Grade 5_ 6_ 7_ or 8
Male 33 White 0 Never married/single (N/A) 1 to 3 years of college
Female 45 White 0 Never married/single (N/A) 1 to 3 years of college
Male 79 White 0 Married_ spouse present Grade 12

1. Describe the population and the sample as

exactly as possible.
2. What variables were recorded? Identify each as
categorical or quantitative.
In Chapter 2, you will learn about graphical and
numerical tools for analyzing data. Here is a brief
preview of what you’ll see. (You may find that you
have seen, or even used, some of these methods in
previous math courses.)
Let’s begin with a categorical variable, like
sex. We have constructed a bar graph (often
called a bar chart) and a summary table that
display information about this variable for our
sample.
Sex Count
3. What percent of the individuals in the sample are
male? Male 47
Female 53
4. What percent of the population is male? Justify
Total 100
your answer.

Now let’s look at the quantitative variable age. 6. About what percent of the sample is 65 or
We have constructed a dotplot to display the ages of older?
the individuals in our sample. 7. About what percent of the population is 65 or
5. How old is the youngest person in the sample? older? Justify your answer.
The oldest person?
 1. 1 Where Do Data Come From? 13

“Now eat that banana. The nice statistician is watching us”

Experiments
Our goal in choosing a sample is a picture of the population, disturbed as little as
possible by the act of gathering information. All observational studies share the prin-
ciple “observe but don’t disturb.’’ When Jane Goodall first began observing chimpan-
zees in Tanzania, she set up a feeding station where the chimps could eat bananas.
She later said that was a mistake, because it might have changed the apes’ behavior.
In experiments, on the other hand, we want to change behavior. In doing an
experiment, we don’t just observe individuals or ask them questions. We actively
impose some treatment in order to observe the response. Experiments can answer
questions such as “Does aspirin reduce the chance of a heart attack?” and “Do a
majority of high school students prefer Pepsi to Coke when they taste both without
knowing which they are drinking?’’

Experiments

An experiment deliberately imposes some treatment on individuals in

order to observe their responses. The purpose of an experiment is to
study whether the treatment causes a change in the response.

Example 1.9 Curing a cold

It has been claimed that large doses of vitamin C will prevent colds. An experiment
to test this claim was performed in Toronto during the winter months. About 500
volunteer subjects were assigned “at random” to each of two groups. Group 1 received
1 gram per day of vitamin C and 4 grams per day at the first sign of a cold. (This is a
large amount of vitamin C; the recommended daily allowance of this vitamin for adults
is only 60 milligrams, or 60/1000 of a gram.) Group 2 served as a control group and
received a placebo pill—identical in appearance to the vitamin C capsules but with no
active ingredient. Both groups were regularly checked for illness during the winter.
Some of the subjects dropped out of the experiment for various reasons, but 818
completed at least two months. Groups 1 and 2 were very similar in age, occupation,
14 CHAPTER 1 Statistics: The Art and Science of Data

smoking habits, and other background variables. At the end of the winter, 26% of the
subjects in Group 1 had not had a cold, compared with 18% in Group 2. Thus, vita-
min C did appear to prevent colds better than the placebo, but not much better.8

The vitamin C example illustrates the big advantage of experiments over

observational studies: in principle, experiments can give good evidence for cause and
effect. If we design the experiment properly, we start with two very similar groups of
subjects. The individual subjects differ from each other in age, occupation, smoking
habits, and other respects. But the two groups resemble each other when we look
at those variables for all subjects in each group. During the experiment, the sub-
jects’ lives differ, but there is only one systematic difference between the two groups:
whether they take vitamin C or a placebo. So we should be able to say whether tak-
ing vitamin C reduces the likelihood of getting a cold.
The fact that experiments can give good evidence that a treatment causes a response
is one of the big ideas of statistics. A big idea needs a big caution: statistical conclusions
hold “on the average’’ for groups of individuals. They don’t tell us much about one in-
dividual. On the average, the subjects taking vitamin C had fewer colds than those who
were taking a placebo. That says vitamin C was somewhat effective. It doesn’t say every-
one who takes vitamin C will be healthy. And a big idea may also raise big questions: if
we think vitamin C will prevent colds, is it ethical to offer it to some and not to others?
Chapter 6 explains how to design good experiments and looks at ethical issues.

Reprinted with special permission of King Features Syndicate

Exercises 1.13 Tasty muffins Before a new variety of frozen muffin is put on the market, it
is subjected to extensive taste testing. People are asked to taste the new muffin and
a competing brand and to say which they prefer. Is this an observational study or
an experiment? Explain your answer.
1.14 Cell phones and cancer One study of cell phones and the risk of brain cancer
looked at a group of 469 people who have brain cancer. The investigators matched
each cancer patient with a person of the same sex, age, and race who did not have
brain cancer, then asked about use of cell phones.9 Result: “Our data suggest that use
of hand-held cellular telephones is not associated with risk of brain cancer.”
(a) Is this an observational study or an experiment? Why?
(b) What individuals are measured, and what variables are recorded?
 1. 1 Where Do Data Come From? 15

1.15 Choose your study purpose Give an example of a question about high
school students, their behavior, or their opinions that would best be answered by
(a) an observational study.
(b) an experiment.
1.16 Child care In 2001, researchers announced that “children who spend most
of their time in child care are three times as likely to exhibit behavioral problems
in kindergarten as those who are cared for primarily by their mothers.”10
(a) Was this likely an observational study or an experiment? Why?
(b) Can we conclude from this study that child care causes behavior problems?
Why or why not?
1.17 Teenage drivers Go to the National Highway Traffic Safety Administration
www Web site: www.nhtsa.dot.gov. Locate a traffic safety fact about young drivers that
includes a supporting chart or graph. If possible, print the image.
(a) Summarize your safety fact with a few well-written sentences in your own
words. Be sure to describe what the chart or graph tells you. Cite the title, date,
and Web URL of the article from which this information comes.
(b) How did the NHTSA obtain these data? Justify your answer.
1.18 Whiter teeth In an effort to make their teeth whiter, some people use
whitening strips or bleaching gels. Are new “light-activated bleaching techniques”
any more effective? Researchers carried out a study to help answer this question.
They gathered a small group of dental patients who were willing to participate.
On one side of patients’ mouths, the researchers used bleaching gel alone. On the
other side, they used light plus bleaching gel. The results? The patients’ teeth got
whiter on both sides of their mouths; there was no real difference in whiteness
between the two sides.
(a) Was this study an experiment or an observational study? Explain.
(b) Why did the researchers apply both possible treatments to each patient, rather
than using bleaching gel on only half of the patients and light plus bleaching gel
on the other half?

ws
Ne
ws
Ne

APPLICATION 1.1 B]
RO
g
<
Se
a

Making sense of the census

www
The U.S. Census Bureau Web site contains 2. Use the “Population Clocks” section of the site to
a vast amount of data about the U.S. popula- answer these questions:
tion. In this Application, you will take a closer look • What are the current U.S. and world populations,
at where these data come from. according to the Population Clocks?
1. Go to the Census Bureau’s site at • How is the U.S. population figure obtained?
www.census.gov. • When did the world population reach 6 billion?
APPLICATION 1.1 (continued)
16 CHAPTER 1 Statistics: The Art and Science of Data

APPLICATION 1.1 (continued)

U.S. Census Bureau/www.census.gov

3. Click on the link for the most recent census to 5. Click on the American Community Survey
answer these questions: (ACS) link.
• How large was the U.S. population? • Why does the Census Bureau conduct surveys,
• What percent of the U.S. population was male? such as the ACS?
How many people were aged 15 to 19? • How many people does the ACS survey, and
• What percent of the U.S. population was how are they selected?
Hispanic? How does this compare with the 6. Click on “Subjects A to Z.” For the most recent
Hispanic population in the previous census? census:
• How did the Census Bureau obtain their • Computer use and ownership: What percent
data? of homes had a computer? What percent had
4. Go to the “QuickFacts” section of the site. For Internet access? Where did these data come
the most recent census: from?
• What was your state’s population? • Educational attainment: How do males and
• For your state, what was the average travel time females compare in terms of bachelor’s, master’s,
to work? How do you think this value was and higher degrees earned? Where do these data
obtained? come from?

Section 1.1 Summary

Any statistical study records data about some individuals (people, animals, or
things) by giving the value of one or more variables for each individual. Quantitative
variables, such as age and income, take numerical values. Categorical variables,
such as occupation and gender, do not. Be sure the variables in a study really do tell
you what you want to know.
The most important fact about any statistical study is how the data were
produced. Observational studies try to gather information without disturbing
the scene they are observing. Sample surveys are an important kind of obser-
vational study. A sample survey chooses a sample from a specific population
 1. 1 Where Do Data Come From? 17

and uses the sample to get information about the entire population. A census
attempts to measure every individual in a population. Experiments actually
do something to individuals in order to see how they respond. The goal of an
experiment is usually to learn whether some treatment actually causes a certain
response.

Section 1.1 1.19 Public housing To study the effect of living in public housing on family
Exercises stability in low-income households, researchers obtain a list of all applicants for
public housing in Chicago last year. Some applicants were accepted, while
others were turned down by the housing authority. The researchers interview
both groups and compare them. Is this an experiment or an observational study?
Explain your answer.
1.20 Baking bread A flour company wants to know what fraction of Minneapo-
lis households bake some or all of their own bread. A sample of 500 residential
addresses is taken, and interviewers are sent to those addresses. All of the inter-
views take place during regular working hours on weekdays.
(a) What population is the flour company interested in? What is the sample?
(b) Do you think this sample will provide accurate information to the flour com-
pany? Why or why not?
1.21 Fit to lead? Is there a relationship between physical fitness and leadership
ability? To answer this question, researchers recruit 100 students who are willing
to take part in an exercise program. The volunteers are divided into a low-fitness
group and a high-fitness group on the basis of a physical examination. All stu-
dents then take a test designed to measure leadership, and the results for the two
groups are compared.
(a) Is this an experiment or an observational study? Explain your answer.
(b) What population does it appear that the investigators were interested in? What
variables did they measure?
(c) Explain why we might prefer a sample survey to using students who volunteer
for a fitness program.
1.22 Cool cars You and your friends want to find out which student at your school
has the “coolest” car.
(a) What variables would you record? Which are quantitative?
(b) What are the individuals in your data set?
(c) What method would you use to produce your data? Justify your choice.
1.23 Angry bees My grandmother once told me that the color red makes bees
angry. Here’s a method I’ve designed to test her claim. I’ll select half of my
students (by drawing names from a hat) to wear red clothes and the other half
to wear white clothes. Then I’ll turn a bunch of bees loose in our classroom and
record how many times each student is stung.
(a) Is this an observational study or an experiment? Why?
(b) What variables are recorded?
(c) If students wearing red clothes are stung much more often than students wear-
ing white, can we conclude that the color red causes bees to sting more? Why or
why not?
(d) Comment on any flaws you see with my methods.
18 CHAPTER 1 Statistics: The Art and Science of Data

1.24 Fast food Many people eat fast food as a regular part of their diet. Is fast
food unhealthy? The best answer may be “It depends on what you eat.” Here are
some nutritional data about four popular fast-food burgers:

Burger Restaurant Calories Fat Cholesterol Sodium

Quarter Pounder
with cheese McDonald’s 430 30 95 1310
Classic Single with
everything Wendy’s 410 19 70 890
Whopper with
cheese Burger King 706 43 113 1164
Cheeseburger In-N-Out 480 27 60 1000

(a) What individuals and variables are recorded in this data set?
(b) In what units is each quantitative variable measured?
(c) Are fast-food chicken sandwiches healthier than burgers? Collect some data
for yourself. (Most fast-food restaurants will provide nutritional information on
request. You can also search the Web.)

1. 2 Drawing Conclusions from Data

ACTIVITY 1.2
See no evil, hear no evil?
Materials: Index cards (two per student, prepared 3. Flip a coin to determine which task you will
by your teacher); clock, watch, or stopwatch perform first: heads = Task A first; tails = Task B first.
to measure 30 seconds; coin (one per pair of
4. Flip the coin again to decide which of the two
students)
index cards will be used for the first task.
Confucius said, “I hear and I forget. I see and I Task A: Study the pairs of numbers on one of your
remember. I do and I understand.” Do people index cards for 30 seconds. Then turn the card over.
really remember what they see better than what Recite the alphabet aloud (A, B, C, etc.). Then tell your
they hear?11 In this Activity, you will perform an partner what you think the numbers on the card are.
experiment to try to find out. Task B: Your partner will read the pairs of numbers
1. Your teacher will give each student two index on one of your index cards aloud three times slowly.
cards with 10 pairs of numbers on them. Do not Next, you will recite the alphabet aloud (A, B, C,
look at the numbers until it is time for you to do the etc.). Then you will tell your partner what you think
experiment. the numbers on the card are.

2. With your partner, decide who will perform the 5. Perform both tasks in the order specified by the
experiment first and who will be the timer. coin toss. Your partner will record how many pairs
of numbers you recalled correctly for each task.
The experiment consists of two tasks; let’s call them
Task A and Task B. 6. Switch roles, and repeat Steps 3 to 5.
 1. 2 Drawing Conclusions from Data 19

7. Your teacher will make a chart on the board 9. Did students in your class remember numbers
like the one shown below. Record the data for each better when they saw them or when they heard
partner on a separate row of the chart. them? Give appropriate evidence to support your
answer.
Student First “See” “Hear”
number task? correct correct Difference 10. Based on the results of this experiment, can
1 A 5 3 2 we conclude that people in general remember
2 B 4 5 1 better when they see than when they hear? Why
or why not?
8. As a class, make a dotplot (see page 12) of the
difference values on the board.

The statistical problem-solving process

Activity 1.2 outlines the steps in the statistical problem-solving process.12 We
begin with a question of interest. Then, we produce data using an observational
study or experiment. Next, we analyze the data using graphs and numerical
summaries. Finally, we interpret the results of the analysis and draw conclusions
about our original question.

Statistical problem-solving process

I. Ask a question of interest A statistics question involves some char-

acteristic that varies from individual to individual.

II. Produce data The methods of choice are observational studies

and experiments.

III. Analyze data Graphs and numerical summaries are the tools for
describing patterns in the data, as well as any deviations from those
patterns.

IV. Interpret results The results of the data analysis should help
answer the question of interest.

You have already seen some examples of the kinds of questions that statistics
can help answer. Here are a few more. How common is texting while driving?
Can yoga help dogs live longer? How well do SAT Writing scores predict the
GPAs of college freshmen? If a casino’s roulette wheel comes up red five times
in a row, is the casino cheating? Will a low-fat diet or a low-carbohydrate diet
reduce weight more effectively in the long run? Do angry people have more
heart disease? How much time per day does the average teen spend on social-
networking sites? Do employers discriminate against job candidates based on
their names?
20 CHAPTER 1 Statistics: The Art and Science of Data

What do all of these questions have in common? Variation. Individuals vary.

Chance outcomes—like spins of a roulette wheel or tosses of a coin—vary. Data
vary. Some would say that statistics is really the study of variation. The following
example illustrates how the statistical problem-solving process can be used to make
sense of variation.

Example 1.10 Who has tattoos?

I. Ask a question of interest What percent of U.S. adults have one or more tattoos?
II. Produce data The Harris Poll conducted an online survey of 2302 adults dur-
ing January 2008. According to the published report, “Respondents for this survey
were selected from among those who have agreed to participate in Harris Interac-
tive surveys.”13
III. Analyze data The pie chart below summarizes the responses from those
who were surveyed: 14% said they had one or more tattoos. Did responses differ
by gender? The accompanying bar graph shows that males and females responded
similarly (13% of females said Yes, compared with 15% of males).

Do you have a tattoo? Who has tattoos?

16
14
Percent with tattoos

Yes
14% 12
10
8
6
4
No 2
86%
0
Female Male
Gender

IV. Interpret results In this sample of adults, 14% said they had one or more tat-
toos. Of course, a different sample of 2302 adults might yield different results. That’s
the idea of sampling variability. If this sample is truly representative of the adult
population, then we would expect about 14% of the population to say that they had
at least one tattoo if asked.
www
You can find reports about various sample surveys—like this one on tattoos—at the
Harris Poll’s Web site: www.harrisinteractive.com/harris_poll/.

Interesting questions can come from almost anywhere—something you

read in the newspaper, see while surfing the Internet, or hear on television or
the radio. As one of the authors learned, interesting questions can even emerge
during a family lunch. In this case, Mr. Starnes’s mother mentioned that several
people she knew had begun filling their car tires with nitrogen instead of
 1. 2 Drawing Conclusions from Data 21

compressed air. Her friends were convinced that nitrogen-filled tires would lead
to a smoother ride, steadier air pressure, and better gas mileage. After lunch,
Mr. Starnes took to the Internet to research this issue. To learn what he discovered,
keep reading.

Example 1.11 Nitrogen in tires—a lot of hot air?

Most automobile tires are inflated with compressed air, which consists of about 78%
nitrogen. Aircraft tires are filled with nitrogen, which is safer than air in case of fire.
I. Ask a question of interest Does filling automobile tires with nitrogen improve
safety, performance, or both?
II. Produce data Consumers Union designed a study to test whether nitrogen-
filled tires would maintain pressure better than air-filled tires. They obtained pairs
of tires from each of several brands and then filled one tire in each pair with air
Courtesy Judy Starnes and one with nitrogen. All tires were inflated to a pressure of 30 pounds per square
inch and then placed outside for a year. At the end of the year, Consumers Union
measured the pressure in each tire.14
III. Analyze data The amount of pressure lost (in pounds per square inch) dur-
ing the year for the air-filled and nitrogen-filled tires of each brand is shown in the
following table.

Brand Air Nitrogen Brand Air Nitrogen

BF Goodrich Traction T/A HR 7.6 7.2 Pirelli P6 Four Seasons 4.4 4.2
Bridgestone HP50 (Sears) 3.8 2.5 Sumitomo HTR H4 1.4 2.1
Bridgestone Potenza EL400 2.1 1.0 Yokohama Avid H4S 4.3 3.0
Bridgestone Potenza G009 3.7 1.6 BF Goodrich Traction T/A V 5.5 3.4
Bridgestone Potenza RE950 4.7 1.5 Bridgestone Potenza RE950 4.1 2.8
Continental Premier Contact H 4.9 3.1 Continental ContiExtreme
Cooper Lifeliner Touring SLE 5.2 3.5 Contact 5.0 3.4
Dayton Daytona HR 3.4 3.2 Continental ContiProContact 4.8 3.3
Falken Ziex ZE-512 4.1 3.3 Cooper Lifeliner Touring SLE 3.2 2.5
Fuzion Hrl 2.7 2.2 General Exclaim UHP 6.8 2.7
General Exclaim 3.1 3.4 Hankook Ventus V4 H105 3.1 1.4
Goodyear Assurance Tripletred 3.8 3.2 Michelin Energy MXV4 Plus 2.5 1.5
Hankook Optimo H418 3.0 0.9 Michelin Pilot Exalto A/S 6.6 2.2
Kumho Solus KH16 6.2 3.4 Michelin Pilot HX MXM4 2.2 2.0
Michelin Energy MXV4 Plus 2.0 1.8 Pirelli P6 Four Seasons 2.5 2.7
Michelin Pilot XGT H4 1.1 0.7 Sumitomo HTR 4.4 3.7

A brief inspection of the data on pressure loss reveals a lot of variation. What are
some possible sources of this variation? Some tires were filled with compressed air, while
others were filled with nitrogen. So part of the variation in pressure loss is due to what
the tires were filled with. Some brands of tires might maintain pressure better than
others. Thus, some of the variation we see is due to differences among the brands of
tires that were used in this study. Even two tires of the same brand are not completely
identical. As a result, some of the variation in the data may be due to differences in
22 CHAPTER 1 Statistics: The Art and Science of Data

individual tires of the same brand. How can we separate the effects of these sources of
variation?
We’ll continue the data analysis process shortly.

Exercises 1.25 A misleading graph When we entered the results of the tattoo survey of
Example 1.10 (page 20) into Excel, the program produced the default bar graph that
you see below. What’s wrong with this picture?

Who has tattoos?

16
15
Percent 15
14
14
13
13
12
Female Male
Gender

1.26 Tattoos across the United States An Excel bar graph displaying more results
from the tattoo survey of Example 1.10 (page 20) is shown below. Write a couple
of sentences describing what the graph tells you.

Adults with tattoos by region

25
20
Percent

15
10
5
0
East Midwest South West
Region

1.27 Variation in tire pressure In Example 1.11, we discussed three possible

sources of variation in tire pressure loss. Give specific evidence to show that each
of these three types of variation is present in the experiment.
1.28 Nitrogen and gas mileage Can filling car tires with nitrogen instead of
compressed air increase gas mileage? Describe how you would design a study
to try to answer this question. Be as specific as you can about the details of
your study.
1.29 Teens and tattoos What percent of teens have tattoos? Describe how you
would design a study to try to answer this question. Be as specific as you can
about the details of your study.
 1. 2 Drawing Conclusions from Data 23

1.30 Who talks more—women or men? According to Louann Brizendine,

author of The Female Brain, women say nearly three times as many words per
day as men. Skeptical researchers devised a study to test this claim. They used
electronic devices to record the talking patterns of 396 university students
from Texas, Arizona, and Mexico. The device was programmed to record 30
seconds of sound every 12.5 minutes without the carrier’s knowledge. What
were the results?
According to a published report of the study in Scientific American, “Men
showed a slightly wider variability in words uttered. . . . But in the end, the sexes
came out just about even in the daily averages: women at 16,215 words and men
at 15,669.”15
(a) Did researchers conduct an observational study or an experiment? Explain.
(b) Why was it important for the recording device to operate without the knowledge
of the carrier?
(c) To what population can the results of this study be generalized? Justify your
answer.

Statistics in action: from tattoos to tires

The air versus nitrogen study of Example 1.11 is an experiment, because researchers
deliberately imposed treatments—filling with air or filling with nitrogen—on the in-
dividuals (tires) in the study. A common goal of experiments is to draw conclusions
about cause and effect. Consumers Union hoped that their experiment would provide
convincing evidence about whether filling tires with nitrogen rather than air causes a
reduction in pressure loss. Did they achieve their goal? Let’s examine the data more
closely before we try to answer this question.

DATA EXPLORATION
Nitrogen in Tires
The nitrogen in tires data set from
www
Example 1.11 is available at the book’s Web
site, www.whfreeman.com/sta2e. If you have a
TI-Nspire graphing calculator, you can download
the file tires.tns from your teacher or the book’s
Web site.
1. The accompanying TI-Nspire screen shot shows
a comparative dotplot of the pressure loss data.
What does this graph tell you about pressure loss
for air-filled versus nitrogen-filled tires?

DATA EXPLORATION (continued)

24 CHAPTER 1 Statistics: The Art and Science of Data

DATA EXPLORATION (continued)

2. Quick calculations reveal that the average A dotplot of the differences in pressure loss is
pressure loss for the air-filled tires was 3.9 pounds shown in the TI-Nspire screen shot below.
per square inch (psi), compared with 2.7 psi for the
nitrogen-filled tires. That seems like a pretty large
difference. Can we conclude that this difference was
due to what the tires were filled with? Or is there
another possible explanation?
3. Did you realize that our examination of the
tire pressure data in Questions 1 and 2 ignored
the fact that the data values are paired by brand
of tire? Since the data are paired, it makes sense
to examine the differences in pressure loss for
the air-filled and nitrogen-filled tires in each
pair. For example, for the BF Goodrich Traction
T/A HR tires, the air-filled tire lost 7.6 psi in a
year. The nitrogen-filled tire lost 7.2 psi over that Are most of the difference values positive or
same period. The difference in pressure loss (air  negative? The average difference in pressure loss is
nitrogen) for this brand is 7.6  7.2  0.4 psi. 1.2 psi. What do these results suggest?

What have we learned so far from the tire experiment? On average, a nitrogen-
filled tire of a given brand lost 1.2 psi less pressure than an air-filled tire of the same
brand over a one-year period. For 28 of the 31 pairs of tires tested, the nitrogen-filled
tire lost less pressure than the air-filled tire. Are we safe concluding that these ob-
served differences in tire pressure loss are due to the different effects of filling with air
and filling with nitrogen? Or could such results have happened “just by chance”?

Example 1.12 Nitrogen in tires—a lot of hot air? (continued)

IV. Interpret results In the tire pressure experiment, the air-filled tire lost more
pressure than the nitrogen-filled tire for 28 of the 31 tire pairs. There are two possi-
ble explanations for this result: (1) filling tires with nitrogen rather than air caused
a reduction in pressure loss, or (2) filling tires with nitrogen rather than air did not
reduce pressure loss, but just by chance, the tire that was going to lose more pres-
sure anyway happened to get filled with air in 28 of the 31 pairs.
Imagine flipping a coin 31 times to simulate assigning air and nitrogen to all 31
pairs of tires. There is virtually no chance of selecting the tire that will end up losing
more pressure in 28 of those pairs to fill with air just by the luck of the coin toss.
As a result, explanation (2) is not plausible. Based on the results of this experiment,
we conclude that there is convincing evidence that filling tires with nitrogen rather
than air caused a reduction in pressure loss.

So there you have it: a brief introduction to how statistics can be used to draw
conclusions from data. Whether it’s an observational study about who has tattoos or
 1. 2 Drawing Conclusions from Data 25

an experiment about filling tires with nitrogen instead of air, the statistical problem-
solving process can help us make sense of the world around us. In the chapters that
follow, you’ll learn more about the last three steps in this process—producing data,
analyzing data, and interpreting results.
Statistics Through Applications presents the important ideas of statistics in four
parts:
Part A: Analyzing Data (Chapters 2 to 4) concerns strategies for exploring,
organizing, and describing data using graphs and numerical summaries. You can
learn to look at data intelligently even with very simple tools, as you have seen in
this chapter.
Part B: Producing Data (Chapters 5 and 6) describes methods for data produc-
tion that can give clear answers to puzzling questions. As you learned in this chapter,
where the data come from is really important. Basic concepts about how to select
samples and design experiments are the most fundamental principles in statistics.
Part C: Chance (Chapters 7 and 8) gives us the language we need to describe
variation, risk, and probability. Because variation is present in any statistical prob-
lem, we need a deeper understanding of the role that chance plays in contributing
to that variation.
Part D: Inference (Chapters 9 and 10) moves beyond the data in hand to draw con-
clusions about some wider universe. Done properly, inference will allow us to estimate
some characteristic of a population based on results from a sample. Other inference
methods help us answer the question “Could this result have happened just by chance?”

APPLICATION 1.2
What is “normal“ body temperature?
If a thermometer under your tongue reads higher QUESTIONS
than 98.6°F, do you have a fever? Maybe not. 1. Explain why this is an observational study and
People vary in their “normal” temperature. Your not an experiment.
own temperature also varies—it is lower around 2. To what population does the conclusion about
6 a.m. and higher around 6 p.m. normal body temperature apply? Justify your answer.
Researchers designed an observational study to
3. A dotplot of Shoemaker’s temperature data
investigate the long-accepted value of 98.6°F, which
is shown below. We have added a vertical line at
was established in 1878 by German doctor Carl
98.6°F for reference. What does this graph tell you
Wunderlich. In the study, 148 healthy men and
about “normal” body temperature?
women aged 18 to 40 had their temperatures taken
orally several times over a three-day period. A total of
700 temperature readings were produced.16
When researchers published their results, they
did not provide the original data. This is unfortu-
nately a common practice. By working backward
from graphs and numerical information given in 96 97 98 99 100 101
the article, however, Allen Shoemaker from Calvin Temp.
College produced a data set with many of the same 4. How do male and female temperatures compare?
characteristics as the original temperature readings.17 The dotplot on the following page shows body
APPLICATION 1.2 (continued)
26 CHAPTER 1 Statistics: The Art and Science of Data

APPLICATION 1.2 (continued)

temperature readings by gender. Write a few The average of the 700 temperature readings
sentences comparing normal body temperatures for was 98.2°F. For the men in the study, the average
males and females in the study. temperature reading was 98.1°F. For the women
in the study, the average temperature reading was
98.4°F. Only 8% of the temperatures recorded were
Male

equal to 98.6 degrees.

Researchers used a statistical test to determine
Gender

that an average temperature of 98.2°F in the sample

of 700 measurements was extremely unlikely if
Female

the true average temperature in the population

was 98.6°F. They concluded that Carl Wunderlich’s
96 97 98 99 100 101
value of 98.6°F is not a reasonable value for “nor-
Temp. mal” body temperature.
5. List and describe as many possible sources of
variation in this study as you can.

Section 1.2 Summary

Statistics is the study of variation—among individuals, groups, and measurements.
The statistical problem-solving process can help us make sense of the variation we
see. The process begins with some question of interest about the world around us—a
question that involves variation. To answer the question, we need some relevant data.
The most common ways of obtaining data are through an observational study or an
experiment. With data in hand, we proceed to analysis. Graphs and numerical sum-
maries help us make sense of observed variation in the data. Finally, we interpret the
results of our analysis and answer the original question of interest. You’ll learn more
about each of these aspects of the statistical problem-solving process in later chapters.

Section 1.2 1.31 Meat eaters According to a How often Americans eat meat
Exercises USA Today Snapshot, about 3% of 6%
Once a
Americans never eat meat. A pie chart 8% week or 3%
Twice a less Never
presented in the Snapshot shows that week
49% eat meat daily, 34% eat meat at 49%
34%
least three times a week, 8% eat meat 3-plus Daily
twice a week, and 6% eat meat once per times a week

week or less.
(a) What would be a reasonable ques-
tion of interest in this case?
(b) Do you think these results were
obtained using an observational study or an experiment? Explain.
(c) Who are the individuals in this study? What variable was measured? Is the
variable categorical or quantitative?
 1. 2 Drawing Conclusions from Data 27

1.32 TV and obesity According to the National Institute on Media and the
Family, a preschoolers’ risk of obesity jumps 6% for every hour of television
watched per day. The risk increases by 31% if the TV is in their bedroom.18
(a) What would be a reasonable question of interest in this case?
(b) Do you think these results were obtained using an observational study or an
experiment? Why?
(c) The actual study that produced these results involved 2761 low-income adults
in New York with children aged 1 to 4 years. Who are the individuals in this
study? What variable(s) were measured?
1.33 Throw in the towel! Which paper towel brand absorbs best? Samantha,
a 12-year-old, carried out a science project to compare the absorbency of four
brands of paper towels—Bounty, Scott, Sparkle, and Western Family. She started
with four individual sheets of each brand. One at a time, Samantha placed a sheet
of paper towel in a container with 200 milliliters of water. After 15 seconds, she
removed the paper towel from the water and squeezed out as much water as
she could into a graduated cylinder. Samantha then determined the amount of
water that had been absorbed.19 Here are the data she recorded:

Bounty 33 31 49 32
Scott 42 34 60 36
Sparkle 53 36 47 33
Western Family 47 26 41 31

(a) Identify two sources of variation in this study. Give specific evidence to show
that each of these types of variation is present.
(b) Write a brief report describing Samantha’s study. Follow the four-step statistical
problem-solving process.
1.34 Healthy cereal? Researchers collected data on 77 brands of cereal at a local
supermarket.20 For each brand, the values of several variables were recorded,
including sugar (grams per serving), calories per serving, and the shelf in the
store on which the cereal was located (1  bottom, 2  middle, 3  top). Here
is a dotplot of the data on sugar content and shelf:
3
Shelf
2
1

0 2 4 6 8 10 12 14 16
Sugars
28 CHAPTER 1 Statistics: The Art and Science of Data

Critics claim that supermarkets tend to put sugary kids’ cereal on lower shelves,
where the kids can see it. Do the data from this study support this claim? Follow
the four-step statistical problem-solving process.
1.35 Creative artwork Does competing for a prize improve children’s artistic
creativity? A psychologist carried out a study with about 100 third-graders
that was designed to help answer this question. In the study, the children were
divided into two groups and instructed to make a “silly” collage using materi-
als that were provided. Before they started, the children in one group were
told that their collages would be judged by experts and that the winners would
receive prizes. The children in the other group were told that they would share
their collages in an art party. In fact, expert judges rated the creativity of all the
collages.21
(a) Was this an observational study or an experiment? Justify your answer.
(b) The judges were not aware of which students’ collages came from which
group. Why is this important?
(c) Describe two possible sources of variation in this study.
(d) Children in the group who believed they were competing for prizes
received much lower average creativity ratings than children in the art
party group. Can we conclude that the difference in creativity scores was
caused by the difference in what the two groups of children were told? Why
or why not?
1.36 Find your own study Locate a published report about an experiment or an
observational study.
(a) Print the report if possible. Be sure to record all relevant source information.
(b) Describe carefully how the study reflects the four-step statistical problem-solving
process.

CHAPTER 1 REVIEW
The first and most important question to ask about any statistical study is “Where did
the data come from?” The distinction between observational and experimental data is a
key part of the answer. Good statistics starts with good designs for producing data. Then,
we measure the characteristics of interest to obtain numbers we can work with. With
data in hand, we turn to analysis. Graphs and numerical summaries help us explore
variation in the data. Finally, we interpret the results of our data analysis in light of
the original question of interest. Ask a question, produce data, analyze data, interpret
results—these are the four steps in the statistical problem-solving process.

Here is a review list of the most important skills you should have developed
from your study of this chapter.
A. DATA PRODUCTION
1. Recognize the individuals and variables in a statistical study.
2. Distinguish between categorical and quantitative variables.
3. Identify the population and the sample in a statistical study.
4. Distinguish observational studies from experiments.
5. Determine which method of data production is most appropriate for answering
a given question of interest.
 Chapter 1 Review Exercises 29

B. DRAWING CONCLUSIONS FROM DATA

1. Identify the four steps in the statistical problem-solving process.
2. Describe possible sources of variation in a statistical study.
3. Interpret the results of a statistical study from basic graphical displays.
4. Recognize limitations in the kinds of conclusions that can be drawn from a statis-
tical study based on how the data were produced.

CHAPTER 1 REVIEW EXERCISES

1.37 Multi-airline flights A passenger who takes a trip involving two or more
airlines pays the first carrier, which then owes the other car rier a portion of the
ticket cost. It is too expensive for the airlines to calculate exactly how much they
owe each other. Instead, a sample of about 12% of tickets sold is examined and
accounts are settled on that basis.
(a) What is the population that the airlines want
information about? What is the sample?
(b) Will the sample results definitely, probably, prob-
ably not, or definitely not yield the correct amount of
Sam Forencich/NBAE via Getty Images

money for each airline? Explain.

1.38 NBA All-Stars In the National Basketball
Association (NBA), fans vote to decide which play-
ers get to play in the NBA All-Star game. Kevin
Garnett of the Boston Celtics led all players, with
2,399,148 votes for the 2008 All-Star game. The
table below provides data about the top five vote-
getters that year:22

Points per
Player Team Votes game Height Position

Kevin Garnett BOS 2,399,148 18.8 83 F

LeBron James CLE 2,108,831 30.0 80 F
Dwight Howard ORL 2,066,991 20.7 83 C
Kobe Bryant LAL 2,004,940 28.3 78 G
Carmelo Anthony DEN 1,723,701 25.7 80 F

(a) What individuals are measured?

(b) What variables are recorded? Which are quantitative? In what units is each quanti-
tative variable recorded?
(c) Which variable or variables do you think most influence the number of votes
received by an individual player?
1.39 Cow-a-bunga! According to scientists, a typical cow burps about every 40 seconds.
When cows burp, they release methane gas that is built up while digesting their food.
Some estimates suggest that livestock produce over 25% of methane emissions each year.
Researchers in Germany have developed a pill that is designed to reduce the amount of
methane that cows produce.
30 CHAPTER 1 Statistics: The Art and Science of Data

(a) What would be a reasonable question of interest in this case?

(b) Describe how you would design a study to answer the question of interest from (a).
Be as specific as you can about the details of your design.
1.40 Noisy snowmobiles In spite of a study that predicted harmful environmental effects,
the government lifted a ban on snowmobiles in two national parks in November 2002.
The study found that air pollutants would be far higher with snowmobiles than with
the snowcoaches that were currently being used. Noise from snowmobiles would be
audible over half the time in the parks.
(a) Was this an observational study or an experiment? Why?
(b) What individuals were measured, and what variables were
recorded?
Money for kids
(c) Why do you think the government decided to
allow snowmobiles, in spite of the study results?
1.41 How much allowance? According to a USA 39%
Today Snapshot, about 24% of children aged 5 to $11–$20
12 receive more than $20 a month in allowance.
A pie chart presented in the Snapshot shows that 23%
13% receive $5 or less, 23% get $6 to $10, and $6–$10
39% receive $11 to $20.
(a) What would be a reasonable question of 24%
$21 or more
interest in this case? 13%
(b) Do you think these results were obtained $5 or less
using an observational study or an experiment?
Why?
(c) Who are the individuals in this study? What variable
was measured?
(d) Think: why don’t the four percents given add up to 100%?
1.42 Creative writing Do external rewards—things like money, praise, fame,
and grades—promote creativity? Not according to the study in Exercise 1.35
(page 28). Children who competed for prizes generally produced less creative
collages than children who simply shared their artwork. Will similar results
hold for college students? The psychologist of Exercise 1.35 designed another
study—this time involving 47 experienced creative writers who were college
students. Students were divided into two groups using a chance process (like
drawing names from a hat). The students in one group were given a list of state-
ments about external reasons for writing, like public recognition, making money,
or pleasing their parents. Students in the other group were given a list of state-
ments about internal reasons for writing, such as expressing yourself and enjoy-
ing working with words. Both groups were then instructed to write a poem about
laughter. Each student’s poem was rated separately by 12 different poets using a
creativity scale.23
(a) Was this an observational study or an experiment? Justify your answer.
(b) Why were students divided into the two groups using a chance process?
(c) None of the poets was aware of which students were in which group. Why is this
important?
1.43 More creative writing Refer to Exercise 1.42. The 12 poets’ ratings of each
student’s poem were averaged to obtain an overall creativity score. Parallel dotplots
of the two groups’ creativity scores are shown on the facing page.
 Chapter 1 Review Exercises 31

Creativity

Reward
E
I
0 5 10 15 20 25 30
Avg. rating

What do you conclude? Follow the four-step problem-solving process.

1.44 Find another study Locate a published report about a sample survey.
(a) Print the report if possible. Be sure to record all relevant source information.
(b) Describe carefully how the study reflects the four-step statistical problem-solving
process.
This page intentionally left blank
 PA RT A

ANALYZING DATA

Words alone don’t

tell a story. A writer
organizes words into
sentences and organizes
the sentences into a
story. If the words are
badly organized, the
story isn’t clear. Data
also need organizing if
they are to tell a clear
story. Too many words
can make a subject "Tonight we’re going to let the statistics speak for themselves."
confusing.Vast amounts
of data are even harder to digest—we often need a brief summary to highlight
essential facts. How to organize, summarize, and present data are our topics in
this part of the book.
As the old saying goes, a picture is worth a thousand words. So we start
by looking at graphical methods for presenting data. Then we add appropriate
numerical summaries—like counts or percents for categorical variables, and
the mean and standard deviation for quantitative variables. The final step is to
interpret what the graphs and numbers tell us about the data.
Organizing and summarizing a large body of facts opens the door to
distortions, both unintentional and deliberate. This is also the case when
the facts take the form of numbers rather than words. We’ll point out some
common pitfalls to avoid when presenting data graphically and numerically.
Mostly, though, we’ll show you how to present data clearly and accurately.

33
This page intentionally left blank
 CHAPTER 2

Describing
Distributions
of Data

Corbis Premium RF/Alamy

2.1 Displaying Distributions with Graphs
2.2 Describing Distributions with Numbers
2.3 Use and Misuse of Statistics

Does education pay?

People with more education earn more, on average, than people with less
education. How much more? A simple way to answer this question is to
compare median incomes, the amount such that half of a group of people earn
more and half earn less. Here are the median incomes of adults aged 25 and
older with four different levels of education:

High school Some college, Bachelor’s Advanced

graduate only no degree degree degree

$24,800 $30,000 $45,000 $62,000

These numbers come from 113,320 people interviewed by the Current Population
Survey (CPS) in March 2008.1 Every March, the CPS asks detailed questions about
income. Boiling down many incomes into four numbers helps us get a quick picture
of the relationship between education and income. The median college graduate,
for example, earns about twice as much as the median high school graduate.
We know that incomes vary a lot. In fact, the highest income reported
by the 33,943 people in the CPS sample with only a high school diploma was
$692,270. The medians compare the centers of the four income distributions.
Can we also describe the spreads with just a few numbers? The largest and
smallest incomes in a group of people don’t tell us much. Instead, let’s give the
range covered by the middle half of the incomes in each group. Here they are:

High school Some college, Bachelor’s Advanced

graduate only no degree degree degree

$11,000 to $40,000 $14,645 to $48,000 $23,545 to $72,050 $37,550 to $100,000 35

36 CHAPTER 2 Describing Distributions of Data

The message of the 113,320 incomes is now clear: going to college but not
getting a degree doesn’t raise income much. People with bachelor’s degrees,
however, earn quite a bit more, and an advanced degree gives income another
healthy boost. But there’s a lot of variation among individuals—some very rich
people never went to college. Finally, remember that this observational study
says nothing about cause and effect. People who get advanced education are
often smart, ambitious, and well-off to start with, so they might earn more even
without their degrees.

2.1 Displaying Distributions with Graphs

ACTIVITY 2.1A
Did you know?
How much do you know about your classmates? any numerical information that you think helps
Without looking, what is the most common eye summarize the data.
color in your class? Without asking, how many 3. Share your team’s analysis with the rest
siblings does a “typical” classmate have? In this of the class. What is the most common eye
Activity, you’ll begin by collecting some data to help color?
you answer these questions. Then, you’ll work with
a few teammates to analyze the data. 4. Now work with your teammates to produce a
graph that displays the siblings data effectively. Add
1. Have each member of the class record his or her any numerical information that you think helps
eye color and number of siblings on the board or summarize the data.
overhead projector.
5. Share your team’s analysis with the rest of the
2. Work with your teammates to produce a graph class. How many siblings does a “typical” student
that displays the eye color data effectively. Add have?

Bar graphs and pie charts

How well educated are young adults in the United States? Table 2.1 presents some
data for people aged 25 to 34 years. This table illustrates some good practices for
data tables. It is clearly labeled so that we can see the subject of the data at once. The
table title gives the general subject of the data and the date because these data will
change over time. Labels within the table identify the variables and state the units
in which they are measured. Notice, for example, that the counts are in thousands.
The source of the data appears at the bottom of the table. Note that these data come
from our old friend, the Current Population Survey.
Table 2.1 starts with the counts of young adults with each level of education.
Rates (percents or proportions) are often clearer than counts—it is more helpful
 2. 1 Displaying Distributions with Graphs 37

Table 2.1 Level of Education of People 25 to 34 Years

Old in the United States, 2007

Number of persons
Level of education (thousands) Percent

Less than high school 5,126 12.9

High school graduate 11,408 28.6
Some college 10,961 27.5
Bachelor’s degree 9,076 22.8
Advanced degree 3,299 8.3
Total 39,868 100

Source: U.S. Census Bureau, Current Population Survey, 2007.

to hear that 12.9% of young adults did not finish high school than to hear that
there are 5,126,000 such people. The percents also appear in Table 2.1. The last two
columns of the table present the distribution of the variable “level of education”
in two forms. Each column gives information about what values the variable takes
and how often it takes each value.

Distribution of a variable

The distribution of a variable tells us what values the variable takes

and how often it takes these values.

Did you check Table 2.1 for consistency? The total number of people should be

5,126  11,408  10,961  9,076  3,299  39,870 (thousands)

The table gives the total as 39,868. What happened? Each entry is rounded to the near-
est thousand. The rounded entries don’t quite add to the total, which is rounded sepa-
rately. Such roundoff errors will be with us from now on as we do more arithmetic.
The distribution in Table 2.1 is quite simple because the categorical variable
“level of education” has only five possible values. To picture this distribution, we
might use a bar graph.

Example 2.1 Education level in the United States:

a bar graph
Figure 2.1 (on the next page) shows a bar graph of the data. The height of each
bar shows the percent of young adults who have the level of education that is
marked at the bar’s base. The bar graph makes it clear that there are more people
38 CHAPTER 2 Describing Distributions of Data

with only a high school diploma than there are people who also have some college—
the “HS grad.” bar is taller.
35
30
25

Percent
20
15
10
5
0
Not HS HS grad. Some Bachelor's Advanced
grad. college degree degree
Education level
Figure 2.1 Bar graph of the distribution of level of education for persons aged 25 to 34
in the United States in 2007.

We can also display the distribution of education level with a pie chart.

Example 2.2 Education level in the United States:

a pie chart
Figure 2.2 is a pie chart of the level of education of young adults. Pie charts show
how a whole is divided into parts. Because it is hard to see this from the wedges in
the pie chart, we added labels with the actual percents.

Advanced
degree
8.3%
Not
HS grad.
12.9%
Bachelor’s
degree
22.8% HS grad.
28.6%

Some college
27.5%

Figure 2.2 Pie chart of the distribution of education level among persons aged 25 to 34
in the United States in 2007.

To make a pie chart, first draw a circle. The circle represents the whole—in this
case, all people in the United States aged 25 to 34 years. Wedges within the circle
represent the parts. The angle spanned by each wedge is proportional to the size of
that part. For example, 22.8% of young adults have a bachelor’s degree but not an
 2. 1 Displaying Distributions with Graphs 39

advanced degree. Because there are 360 degrees in a circle, the “bachelor’s degree”
wedge spans an angle of 0.228  360  82 degrees.
You can calculate the angles for the other wedges in the same way and then use
a protractor to help you draw an accurate pie chart. Our advice is to let technology
draw the graph for you.

Pie charts force us to see that the parts do make a whole. But because angles
are harder to compare than lengths, a pie chart is not the best way to compare the
sizes of the various parts of the whole. A bar graph is easier to draw than a pie chart
unless a computer is doing the drawing for you.
When we think about graphs, it is helpful to distinguish between quantitative
variables, like height in centimeters and SAT scores, and categorical variables, such
as sex, eye color, and level of education. Pie charts and bar graphs are useful for
displaying distributions of categorical variables. Although both pie charts and bar
graphs can show the distribution (either counts or percents) of a categorical vari-
able such as level of education, bar graphs have other uses, too.

Example 2.3 Hang up and drive!

The Harris Poll conducted an online survey of 2085 U.S. adults about cell phone
use while driving.2 Over half (56%) of those who responded said that it is danger-
ous or very dangerous for someone to use a cell phone while driving. One survey
question asked, “How often do you talk on a cell phone while you are driving?”
Respondents could choose from the options “Always,” “Sometimes,” or “Never.”
Figure 2.3 shows the percent who chose “Always” or “Sometimes” by region of the
United States in which they live.

90
Percent using cell phones

80
70
while driving

60
50
40
30
20
10
0
East Midwest South West
U.S. region
Figure 2.3 Bar graph displaying the percent of adults from each region of the United
States who said they sometimes or always talk on a cell phone while driving.

We cannot replace Figure 2.3 by a pie chart, because it compares four separate
quantities, not the parts of some whole. A pie chart can compare only parts of a
whole. Bar graphs can compare quantities that are not parts of a whole.
40 CHAPTER 2 Describing Distributions of Data

Exercises 2.1 Marital status In the Statistical Abstract of the United States we find these
data on the marital status of adult American women as of 2007:
Marital status Count (thousands)
Never married 25,262
Married 65,128
Widowed 11,208
Divorced 13,210
Total 114,807

(a) How many women were not married in 2007?

(b) Make a bar graph to show the distribution of marital status.
(c) Would it also be correct to use a pie chart? If so, make a pie chart for these data.
2.2 Consistency? Refer to the previous exercise. What is the sum of the counts for the
four marital status categories? Why is this sum not equal to the total given in the table?
2.3 College freshmen A survey of college freshmen asked what field they planned
to study. The results: 25.2% arts and humanities, 19.3% business, 7.1% education,
16.6% engineering and science, 7.8% professional, and 15.3% social science.3
(a) What percent plan to study fields other than those listed?
(b) Make a graph that compares the percents of college freshmen planning to
study various fields.
2.4 Girls excel Is it true that girls perform better than boys in the study of
languages and so-called soft sciences? Here are several Advanced Placement
subjects and the percent of examinations taken by female candidates in 2007:
English Language/Composition, 63%; French Language, 70%; Spanish Language,
64%; and Psychology, 65%.4
(a) Explain clearly why we cannot use a pie chart to display these data, even if we
know the percent of exams taken by girls for every subject.
(b) Make a bar graph of the data. Order the bars from tallest to shortest; this will
make comparisons easier.
(c) Do these data answer the question about whether girls perform better in these
subject areas? Why or why not?
2.5 Cell phones and driving The Harris Poll survey of Example 2.3 (page 39)
also provided results about cell phone use and driving by age group. Here is a
table that summarizes the percent of each age group who said they “Always” or
“Sometimes” use cell phones while driving
Age group

Gen. Y Gen. X Baby boomers Matures

86% 79% 76% 48%

(a) Would it be appropriate to use a pie chart to display these data? If so, do it. If
not, explain why not.
(b) Make a bar graph of the data. Describe what you see.
 2. 1 Displaying Distributions with Graphs 41

2.6 Lottery sales States sell lots of lottery tickets. Table 2.2 shows where the
money comes from. Make a bar graph that shows the distribution of lottery sales
by type of game. Is it also proper to make a pie chart of these data?

Table 2.2 State Lottery Sales by Type of Game, 2007

Game Sales (millions of dollars)

Instant games 29,736

Three-digit games 5,586
Four-digit games 3,499
Lotto 10,014
Other games 3,579
Total 52,414

Source: 2009 Statistical Abstract of the United States.

Displaying quantitative variables: dotplots

ACTIVITY 2.1B
How good is your balance?
Materials: Stopwatch, plus open space with a 2. The doer should get in position—standing on one
desk, table, or wall nearby for each pair of foot with eyes closed—and then say, “Go.” The timer
students. starts the stopwatch on “Go” and stops it when the
doer hops, opens his or her eyes, or touches anything.
Can you stand on one foot without hopping or
Record the doer’s time to the nearest second.
touching anything? Try it. (You might need to
practice!) In this Activity, you and a partner will 3. Switch roles and repeat Step 2.
take turns trying to balance on one foot—with your 4. While you are collecting data, your teacher will
eyes closed! put a blank data table on the board or overhead
1. With your partner, decide who will perform the projector. When both team members have finished,
activity first (the “doer”) and who will be the timer. record your individual times in the data table.
Your teacher may give you some practice time 5. Once all teams have finished, make a dotplot of
before data collection begins. the class data. Then describe what you see.

In Chapter 1, we used a dotplot to display:

• ages of a random sample of 100 people from the U.S. census
• pressure loss values in an experiment involving 31 brands of tires
• temperature readings of 148 healthy adults in an observational study about
normal body temperature
• sugar content of 77 brands of cereal
42 CHAPTER 2 Describing Distributions of Data

What do all of these variables—age, pressure loss, temperature, and sugar content—
have in common? They’re all quantitative variables. A dotplot is the simplest graph
for displaying the distribution of a quantitative variable.

Example 2.4 Are you driving a gas guzzler?

The Environmental Protection Agency (EPA) is in charge of how fuel economy
ratings for cars are determined and reported (think of those large window stick-
ers on a new car). For years, consumers complained that their actual gas mileages
were noticeably lower than the values reported by the EPA. It seems that the EPA’s
tests—all of which are done on computerized devices to ensure consistency—did
not consider things like outdoor temperature, use of the air conditioner, or realis-
tic acceleration and braking by drivers. In 2008, the EPA changed the method for
measuring a vehicle’s fuel economy to try to give more accurate estimates.
Table 2.3 displays the EPA estimates of highway gas mileage for 24 model year
2009 midsize cars. Let’s construct a dotplot that shows the distribution of EPA
highway mileage ratings.

Table 2.3 Highway Gas Mileages for Model Year 2009

Midsize Cars

Model mpg Model mpg

Acura RL 22 Lexus GS 350 26

Audi A6 Quattro 23 Lincoln MKZ 28
Bentley Arnage 14 Mazda 6 29
BMW 528I 28 Mercedes-Benz E350 24
Buick Lacrosse 28 Mercury Milan 29
Cadillac CTS 25 Mitsubishi Galant 27
Chevrolet Malibu 33 Nissan Maxima 26
Chrysler Sebring 30 Rolls-Royce Phantom 18
Dodge Avenger 30 Saturn Aura 33
Hyundai Elantra 33 Toyota Camry 31
Jaguar XF 25 Volkswagen Passat 29
Kia Optima 32 Volvo S80 25

Source: Fuel Economy Guide 2009, from the U.S. Environmental Protection Agency’s Web
site at www.fueleconomy.gov.

Step 1: Label your axis and title your graph. Draw a horizontal line and label it with
the variable, highway gas mileage (mpg). Title your graph.
Step 2: Scale and number the axis based on the values of the variable. For these data,
the smallest value is 14 and the largest is 33.
Step 3: Mark a dot above the number on the horizontal axis corresponding to each
data value.
 2. 1 Displaying Distributions with Graphs 43

Figure 2.4 displays the dotplot.

Figure 2.4 Minitab dotplot of EPA highway gas mileage ratings for 24 model year 2009
midsize cars.

Current information on vehicle fuel economy ratings is available at www.epa.gov/

www
fueleconomy.

Making a statistical graph is not an end in itself. The purpose of the graph is to
help us understand the data. After you (or your computer) make a graph, always
ask, “What do I see?” Here is a general strategy for looking at graphs.

Pattern and deviations

In any graph of data, look for an overall pattern and also for striking
deviations from that pattern.

In the dotplot of Figure 2.4, we can see three distinct clusters of values:
cars that get around 25 mpg, cars that get about 28–30 mpg, and cars that get
around 33 mpg. We also see two cars with unusually low gas mileage ratings—
the Bentley Arnage and the Rolls-Royce Phantom. These cars are potential
outliers.

Outliers

An outlier in any graph of data is an individual observation that falls

outside the overall pattern of the graph.

Whether an observation is an outlier is to some extent a matter of judgment.

Both the Arnage and the Phantom are clearly separated from the main body of
observations, which seems to put them “outside the overall pattern.” In Section 2.2,
we’ll establish a numerical procedure for identifying outliers. For now, once you
have spotted potential outliers, look for an explanation. Many outliers are due to
mistakes, such as typing 4.0 as 40. Other outliers point to the special nature of some
observations. Explaining outliers usually requires some background information.
Does it surprise you that the Bentley and Rolls-Royce midsize cars have much lower
gas mileages than the other models?
44 CHAPTER 2 Describing Distributions of Data

Displaying quantitative variables: stemplots

Sometimes the values of a variable are too spread out for us to make a reasonable
dotplot by hand. In these cases, we can consider another simple graphical display—
a stemplot.

Example 2.5 Where do older folks live?

Table 2.4 presents the percent of residents aged 65 years and over in each of the 50
states. Let’s make a stemplot to display this distribution.

Table 2.4 Percent of Residents Aged 65 and Over in the States, 2007

State Percent State Percent

Alabama 13.5 Montana 14.0

Alaska 7.0 Nebraska 13.3
Arizona 12.9 Nevada 11.1
Arkansas 14.0 New Hampshire 12.6
California 11.0 New Jersey 13.1
Colorado 10.1 New Mexico 12.7
Connecticut 13.5 New York 13.2
Delaware 13.6 North Carolina 12.2
Florida 17.0 North Dakota 14.6
Georgia 9.9 Ohio 13.5
Hawaii 14.3 Oklahoma 13.3
Idaho 11.7 Oregon 13.1
Illinois 12.1 Pennsylvania 15.2
Indiana 12.5 Rhode Island 13.9
Iowa 14.7 South Carolina 13.0
Kansas 13.0 South Dakota 14.3
Kentucky 13.0 Tennessee 12.9
Louisiana 12.2 Texas 10.0
Maine 14.8 Utah 8.9
Maryland 11.8 Vermont 13.6
Massachusetts 13.3 Virginia 11.8
Michigan 12.7 Washington 11.7
Minnesota 12.2 West Virginia 15.5
Mississippi 12.5 Wisconsin 13.1
Missouri 13.4 Wyoming 12.2

Source: 2008 Statistical Abstract of the United States.

 2. 1 Displaying Distributions with Graphs 45

Step 1(a): Separate each observation into a stem, consisting of all but the final (right-
most) digit, and a leaf, the final digit. Stems may have as many digits as needed, but
each leaf contains only a single digit. For the “65 and over” percents in Table 2.4, the
whole-number part of the observation is the stem and the final digit (tenths) is the
leaf. The Alabama entry, 13.5, has stem 13 and leaf 5.
Step 1(b): Write the stems in a vertical column with the smallest at the top, and
draw a vertical line at the right of this column. For these data, the stems run from
7 to 17.
Step 2: Write each leaf in the row to the right of its stem. For Alabama, we put the leaf
5 on the 13 stem. For Alaska, we add a 0 leaf to the 7 stem. We continue in this way
for the other 48 states in alphabetical order.
Step 3: Sort the leaves in increasing order as they move out from the stem.

Figure 2.5 shows the steps in making a stemplot for the data in Table 2.4.

7 7 0 7 0
8 8 9 8 9
9 9 9 9 9
10 10 10 10 01
11 11 078187 11 017788
12 12 915272567292 12 122225567799
13 13 55600343125319061 13 00011123334555 669
14 14 0378063 14 0033678
15 15 25 15 25
16 16 16
17 17 0 17 0
Step 1: Stems Step 2: Add leaves Step 3: Order leaves

Figure 2.5 Making a stemplot of the data in Table 2.4. Whole percents form the stems,
and tenths of a percent form the leaves.

To describe the distribution in Figure 2.5, it is easiest to begin with deviations

from the overall pattern. Which states stand out as unusual? Florida has 17% of
its residents over age 65, which is 1.5% higher than for any other state. There’s a
gap in the stemplot that calls our attention to Florida. On the other end of the
stemplot, only 7% of Alaska’s residents are 65 or older. That’s 1.9% lower than for
any other state. So Alaska and Florida are clear outliers. Is Utah, with 8.9% of its
population over 65, an outlier? It’s not as obvious from the stemplot that Utah
is an outlier, but it is. As we’ll see in Section 2.2, so is Georgia at 9.9%. It is not
surprising that Florida, with its many retired people, has many residents over 65
and that Alaska, the northern frontier, has few. How would you explain Utah and
Georgia?
46 CHAPTER 2 Describing Distributions of Data

To see the overall pattern of a distribution, ignore any outliers. Here is a simple
way to organize your thinking.

Overall pattern of a distribution

To describe the overall pattern of a distribution:

• Give the center and the spread.

• See if the distribution has a simple shape that you can describe in a
few words.

We will learn how to describe center and spread numerically in Section 2.2. For
now, we can describe the center of a distribution by its midpoint, the value with half
the observations taking smaller values and half taking larger values. We can describe
the spread of a distribution by giving the smallest and largest values, ignoring any
outliers.

Example 2.6 Where do older folks live? (continued)

Look again at the stemplot in Figure 2.5. Shape: The distribution has a single peak.
It is roughly symmetric—that is, the pattern is similar on both sides of the peak.
Center: The midpoint of the distribution is close to the single peak, at about 13%.
Spread: The spread is about 10% to 15.5% if we ignore the outliers.

Exercises 2.7 Gooooaaal! The number of goals scored by each team in the first round of
the California high school soccer playoffs is shown below.

5 0 1 0 7 2 1 0 4 0 3 0 2 0
3 1 5 0 3 0 1 0 1 0 2 0 3 1

(a) Make a dotplot of the data.

(b) Describe the distribution of the variable “number of goals scored.” Be sure to
discuss the overall pattern (shape, center, and spread) and any deviations from that
pattern.
2.8 Poverty in the states, I Table 2.5 shows the percent of people living below the
hoto

poverty line in the 26 states east of the Mississippi.

iStockp

(a) Make a stemplot of these data.

(b) Describe the distribution of the variable “percent living in poverty.” Be sure
to discuss the overall pattern (shape, center, and spread) and any deviations from
that pattern.
 2. 1 Displaying Distributions with Graphs 47

Table 2.5 Percent of State Residents Living in Poverty, 2007

State Percent State Percent State Percent

Alabama 16.9 Maryland 8.3 Pennsylvania 11.6

Connecticut 7.9 Massachusetts 9.9 Rhode Island 12.0
Delaware 10.5 Michigan 14.0 South Carolina 15.0
Florida 12.1 Mississippi 20.6 Tennessee 15.9
Georgia 14.3 New Hampshire 7.1 Vermont 10.1
Illinois 11.9 New Jersey 8.6 Virginia 9.9
Indiana 12.3 New York 13.7 West Virginia 16.9
Kentucky 17.3 North Carolina 14.3 Wisconsin 10.8
Maine 12.0 Ohio 13.1

Source: U.S. Census Bureau, Current Population Survey, Annual Social and Economic Supplements.

2.9 Driving in town In Example 2.4, we examined data on highway gas

mileages of model year 2009 midsize cars. The Minitab dotplot below
shows the EPA estimates of city gas mileages for these same 24 car models.
Describe the overall pattern of the distribution and any deviations from that
pattern.

2.10 Fuel efficiency Refer to the previous exercise. The Minitab dotplot below
shows the difference (Highway—City) in EPA mileage ratings for each of the 24
car models from Example 2.4. What does the graph tell us about fuel economy in
the city versus on the highway for these car models? Be specific.
48 CHAPTER 2 Describing Distributions of Data

11 44 2.11 Where do the young live? Figure 2.6 is a stemplot of the percent of residents
11 66778
aged 25 to 34 in each of the 50 states. As in Figure 2.5 (page 45) for older residents, the
stems are whole percents and the leaves are tenths of a percent. This time, each stem
12 0134
has been split in two, with values having leaves 0 through 4 placed on one stem and
12 666778888 values ending in 5 through 9 placed on another stem.
13 000000111144 4
(a) Utah has the highest percent of residents aged 25 to 34. What is that percent?
13 7788999 Why does Utah have an unusually high percent of residents in this age group?
14 0044
(b) Describe the shape, center, and spread of the distribution, ignoring Utah.
14 567
(c) Is the distribution for young adults more or less spread out than the distribution
15 11
in Figure 2.5 for older adults? Justify your answer.
15
2.12 Watch that caffeine! The U.S. Food and Drug Administration limits the
16 0
amount of caffeine in a 12-ounce can of carbonated beverage to 72 milligrams.
Figure 2.6 Stemplot of That translates to a maximum of 48 milligrams of caffeine per 8-ounce serving.
the percent of each state’s Data on the caffeine content of popular soft drinks is provided in Table 2.6.
residents who are 25 to 34 How does the caffeine content of these drinks compare to the USFDA’s limit?
years old.

Table 2.6 Caffeine Content (in Milligrams) for an 8-Ounce Serving of

Popular Soft Drinks

Caffeine (mg Caffeine (mg

Brand per 8-oz serving) Brand per 8-oz serving)

A&W Cream Soda 20 IBC Cherry Cola 16

Barq’s Root Beer 15 Kick 38
Cherry Coca-Cola 23 KMX 36
Cherry RC Cola 29 Mello Yello 35
Coca-Cola Classic 23 Mountain Dew 37
Diet A&W Cream Soda 15 Mr. Pibb 27
Diet Cherry Coca-Cola 23 Nehi Wild Red Soda 33
Diet Coke 31 Pepsi One 37
Diet Dr. Pepper 28 Pepsi-Cola 25
Diet Mello Yello 35 RC Edge 47
Diet Mountain Dew 37 Red Flash 27
Diet Mr. Pibb 27 Royal Crown Cola 29
Diet Pepsi-Cola 24 Ruby Red Squirt 26
Diet Ruby Red Squirt 26 Sun Drop Cherry 43
Diet Sun Drop 47 Sun Drop Regular 43
Diet Sunkist Orange Soda 28 Sunkist Orange Soda 28
Diet Wild Cherry Pepsi 24 Surge 35
Dr. Nehi 28 TAB 31
Dr. Pepper 28 Wild Cherry Pepsi 25

Source: National Soft Drink Association.

 2. 1 Displaying Distributions with Graphs 49

(a) You could construct a dotplot of the caffeine content data, but a stemplot
might be preferable. Explain why.
(b) Construct a stemplot of the data using the first digit as the stem and the
second digit as the leaf. What problem do you see with this display?
(c) Figure 2.7 shows a stemplot with “split stems” for the caffeine content data. This time,
values having leaves 0 through 4 are placed on one stem, while values ending in 5 through
9 are placed on another stem. Describe the shape, center, and spread of the distribution.

1 5 5 6
2 0 3 3 3 4 4
2 5 5 6 6 7 7 7 8 8 8 8 8 9 9 Key: 2 /8 m eans t he s oft dr i nk
3 1 1 3 c ont ai ns 28 m g of c affei ne per
3 5 5 5 6 7 7 7 8 8-ounc e s er vi ng.

4 3 3
4 7 7
Figure 2.7 Stemplot showing the caffeine content (in milligrams per 8-ounce serving)
of various soft drinks.

Displaying quantitative variables: histograms

ACTIVITY 2.1C
The nutritional value of cereals
PLET
AP What makes a “healthy” breakfast cereal? 2. Click on the Stemplot tab to view a stemplot
One without much sugar? A cereal of the sodium data. Try the “Split stems” option.
with lots of vitamins? You probably Which graph displays the data more effectively?
didn’t think about how much salt the cereal 3. Now click on the Histogram tab. Let’s explore this
contains. In this Activity, you will examine data new type of graph. The variable of interest, sodium
on the sodium content (in milligrams) for 77 content, is on the horizontal axis. The applet divides
brands of breakfast cereal. In the process, you the values on the horizontal axis into “classes” of equal
will meet another kind of graph for displaying width. The vertical axis uses a frequency scale that
the distribution of a quantitative variable—a indicates how many values fall in each class interval.
histogram.
Your teacher will give you a data set called 4. You can adjust the width of the classes by clicking
CEREALS that provides nutritional information and dragging with your mouse. Try this for yourself.
for the 77 brands of breakfast cereal.5 For each As you do this, do the features of the histogram
brand of cereal, we show the number of calories per change your impression of what the data say? Explain.
serving, milligrams of sodium, grams of sugar, and 5. Once you have found an interval width that shows
grams of fat. the distribution well, describe the shape, center, and
1. Launch the One-Variable Statistical Calculator applet spread of the distribution. Are there any outliers?
at the book’s Web site, www.whfreeman.com/sta2e. In 6. Write a few sentences describing what you have
the Data Sets tab, choose Cereals. Click on the Data tab learned about the sodium content of these breakfast
to look at the values of sodium content. cereals.
50 CHAPTER 2 Describing Distributions of Data

We can use a dotplot or a stemplot to display the distribution of quantitative variables

that have relatively few values. What about variables such as the SAT scores of students
admitted to a college or the income of families? These variables, like the sodium con-
tent of cereals in Activity 2.1C, take so many values that a graph of the distribution is
clearer if nearby values are grouped together. The most common graph of the distri-
bution of a quantitative variable is a histogram.

Example 2.7 College costs

Tonya is a student at Grand Valley State University in her home state of Michigan.
Grand Valley State charges in-state students $8196 in tuition and fees. Tonya wonders
how Grand Valley State’s tuition compares with the amount charged by other Michigan
colleges and universities.
There are 90 colleges and universities in Michigan. Their tuition and fees for the
2008–2009 school year run from $1873 at Oakland Community College to $30,823
at Kalamazoo College.6 It isn’t easy to compare Grand Valley State with a list of 89
other colleges. So let’s make a graph. The histogram in Figure 2.8 shows that many
colleges charge less than $3000, but the distribution also extends out to the right. At
the upper extreme, 8 colleges charge more than $24,000. Grand Valley State is one
of 19 colleges in the $6000 to $9000 group.

25 24

20 19

15
Frequency

10 9
Figure 2.8 Minitab 7 7 7
5
histogram of the tuition 5 4 4
3
and fees charged by 90 1
Michigan colleges and uni- 0
versities in the 2008–2009 0 6000 12000 18000 24000 30000
Tuition
academic year.

The distribution of tuition and fees at Michigan colleges has a distinct shape.
There is a strong peak at the lowest cost class. And even though most colleges
charge less than $9000, there is a long right “tail” extending up to $33,000. We call
a distribution with a long tail on one side skewed. The center is roughly $8500 (half
the colleges charge less than this). The spread is large, from less than $3000 to more
than $30,000. There are no outliers—the colleges with the highest tuition just con-
tinue the long right tail that is part of the overall pattern.

When you describe a distribution, concentrate on the main features. Look for
major peaks, not for minor ups and downs in the bars of the histogram like those in
Figure 2.8. Look for clear outliers, not just for the smallest and largest observations.
Look for rough symmetry or clear skewness.
 2. 1 Displaying Distributions with Graphs 51

Symmetric and skewed distributions

A distribution is symmetric if the right and left sides of the graph are
approximately mirror images of each other.

A distribution is skewed to the right if the right side of the graph

(containing the half of the observations with larger values) extends
much farther out than the left side. It is skewed to the left if the left
side of the graph extends much farther out than the right side.

In mathematics, symmetry means that the two sides of a figure like a histogram
are exact mirror images of each other. Data are almost never exactly symmetric, so
we were willing to call the stemplot in Figure 2.5 (page 45) roughly symmetric as
an overall description. The tuition distribution in Figure 2.8, on the other hand, is
clearly skewed to the right.

Example 2.8 How to make a histogram

Table 2.4 (page 44) presents the percent of residents aged 65 years and over in each
of the 50 states. We made a stemplot of the data in Example 2.5. To make a histo-
gram of this distribution, proceed as follows.
Step 1: Divide the range of the data into classes of equal width. The data in Table 2.4
range from 7.0 to 17.0, so we choose as our classes

7.0  percent over 65  8.0

8.0  percent over 65  9.0
(
17.0  percent over 65  18.0

Be sure to specify the classes precisely so that each individual falls into exactly one
class. A state with 7.9% of its residents aged 65 or older would fall into the first class,
but 8.0% falls into the second.
Step 2: Count the number of individuals in each class. Here are the counts:
Class Count Class Count Class Count
7.0 to 7.9 1 11.0 to 11.9 6 15.0 to 15.9 2
8.0 to 8.9 1 12.0 to 12.9 12 16.0 to 16.9 0
9.0 to 9.9 1 13.0 to 13.9 17 17.0 to 17.9 1
10.0 to 10.9 2 14.0 to 14.9 7

Step 3: Draw the histogram. Put the variable whose distribution you are displaying
on the horizontal axis. That’s “percent of residents aged 65 and over” in this exam-
ple. The scale runs from 6 to 19 because that range spans the classes we chose. The
vertical axis contains the scale of counts. Each bar represents a class. The base of
the bar covers the class, and the bar height is the class count. There is no horizontal
space between the bars unless a class is empty, so that its bar has height zero. Figure 2.9
(on the next page) is our histogram.
52 CHAPTER 2 Describing Distributions of Data

20
18 17
16
14

Number of states
12
12
10
8 7
6
6
4
2 2
2 1 1 1 1
0
8 10 12 14 16 18
Percent of residents aged 65 and over
Figure 2.9 Histogram of the percent of residents aged 65 and older in the 50 states.

Just as with bar graphs, our eyes respond to the area of the bars in a histogram.
Be sure that the classes for a histogram have equal widths. There is no one right
choice of the classes. Too few classes will give a “skyscraper” histogram, with all val-
ues in a few classes with tall bars. Too many classes will produce a “pancake” graph,
with most classes having one or no observations. Neither choice will give a good
picture of the shape of the distribution. You must use your judgment in choos-
ing classes to display the shape. Technology will usually choose classes for you.
The computer’s or calculator’s choice is usually a good starting point, but you can
change the settings if you want, as you’ll see in the following Calculator Corner.

CALCULATOR CORNER Entering data and making histograms

The TI-84 and TI-Nspire will make histograms for you. Here’s how to
do it.

TI-84
1. Enter the data from Table 2.4 (page 44).
• Press A and choose 1:Edit…
• Type the values into list L1.
2. Set up a histogram in the statistics plots menu.
• Press yo (STAT PLOT)
• Press b or À to go into Plot 1.
• Adjust your settings as shown.
 2. 1 Displaying Distributions with Graphs 53

3. Get a “quick graph” using ZoomStat.

• Press q and choose 9:ZoomStat.
• Press r to look at the class intervals chosen by the calculator.

4. Set the window to match the class intervals chosen in Example 2.8.
• Press p.
• Enter the values shown.

5. Graph the histogram.

• Press s. Then press r.
• Compare with Figure 2.9.

TI-Nspire
1. Enter the data from Table 2.4 (Page 44). First, open a new document with a
Lists & Spreadsheets page.
• Press c and choose 6:New Document.
• Press b ; choose 3:Add Lists & Spreadsheet.
• In the name box at the top of column A, type “pctolder.”
Second, type the values into list pctolder.

2. Make a “quick graph” of the data.

• Press b ; choose 3:Data and 5:Quick Graph. You should see a
dotplot of the data.
• Press b ; choose 1:Plot Type and 3:Histogram to change the graph
to a histogram.
3. Change the horizontal scale to match the class intervals chosen in Example 2.8.
• Press b ; choose 5:Window/Zoom and 1:Window Settings.
• Change Xmin to 6, Xmax to 19, and Ymax to 20, and click OK.
• Press b ; choose 2:Plot Properties and 2:Bin Settings.
• Change Width to 1, Alignment to 6, and click OK.
54 CHAPTER 2 Describing Distributions of Data

Exercises 2.13 Minority students in engineering Figure 2.10 is a histogram of the number
of minority students (black, Hispanic, Native American) who earned doctorate
degrees in engineering from each of 115 universities over a 5-year period.7 Briefly
describe the shape, center, and spread of this distribution.

50

40

Number of universities
30

20

10

0
0 10 20 30 40 50 60
Number of doctorates earned by minorities

Figure 2.10 The distribution of number of engineering doctorates earned by minority

students at 115 universities over a 5-year period.

2.14 Lightning flashes Figure 2.11 comes from a study of lightning storms in
Colorado.8 It shows the distribution of the hour of the day during which the first
lightning flash for that day occurred. Describe the shape, center, and spread of
this distribution. Are there any outliers?

25
Count of first lightning flashes

20

15

10

0
7 A.M. 8 A.M. 9 A.M. 10 A.M.11 A.M. Noon 1 P.M. 2 P.M. 3 P.M. 4 P.M. 5 P.M.
Time of day

Figure 2.11 Histogram of time of day at which the day’s first lightning flash occurred.
 2. 1 Displaying Distributions with Graphs 55

2.15 Poverty in the states, II Refer to Exercise 2.8 (page 46). Joe made the
histogram shown below from the poverty data in Table 2.5. Write a sentence or
two clearly explaining to Joe why his histogram is not legitimate.

Number of states
6

4.2 6.4 7.5 8.5 10.0 13.0 16.2

Percent living below poverty line

2.16 Poverty in the states, III Refer to Exercise 2.8 (page 46). Make a histogram
for the data in Table 2.5. Use seven classes of width 2, beginning with 7–9. If you
made a stemplot of these data in Exercise 2.8, how does your histogram compare?
2.17 Yankee money Table 2.7 gives the salaries of the players on the New York
www
Yankees baseball team for the 2008 season.
(a) Make a histogram of these data.

Table 2.7 Salaries of the New York Yankees, 2008

Alex Rodriguez $28,000,000 Morgan Ensberg $1,750,000

Jason Giambi $23,428,571 Andrew Brackman $1,184,788
Derek Jeter $21,600,000 Wilson Betemit $1,165,000
Bobby Abreu $16,000,000 Brian Bruney $725,000
Andy Petite $16,000,000 Billy Traber $500,000
Mariano Rivera $15,000,000 Melky Cabrera $461,200
Jorge Posada $13,100,000 Phil Hughes $406,350
Johnny Damon $13,000,000 Shelley Duncan $398,300
Hideki Matsui $13,000,000 Sean Henn $397,448
Mike Mussina $11,071,029 Ian Kennedy $394,275
Carl Pavano $11,000,000 Jeffrey Karstens $393,300
Kyle Farnsworth $5,916,666 John Albaladejo $393,225
Chien-Ming Wang $4,000,000 Ross Ohlendorf $391,425
LaTroy Hawkins $3,750,000 Joba Chamberlain $390,000
Robinson Cano $3,000,000 Humberto Sanchez $390,000
Jose Molina $1,875,000

Source: USA Today Salary Database at http://content.usatoday.com/sports/baseball/salaries/default.aspx.

56 CHAPTER 2 Describing Distributions of Data

(b) Is the distribution of Yankees’ salaries roughly symmetrical, skewed to the left,
or skewed to the right? Explain briefly.
(c) What is the spread of the distribution?
PLET
AP 2.18 Activity 2.1C follow-up Use the CEREALS data from Activity 2.1C
and the One Variable Statistical Calculator applet on the book’s Web site
(www.whfreeman.com/sta2e) to investigate the calories, sugar, or fat content
of the 77 brands of cereal. Make a histogram of the variable you chose. Write a
brief report describing what you learned.

ws
Ne
ws
Ne

APPLICATION 2.1 B]
RO
g
<
Se
a

Mercury in tuna
Do you like to eat tuna? Many people do. Unfortu- by Defenders of Wildlife set
nately, some of the tuna that people eat may contain out to answer this ques-
high levels of mercury. (Yes, mercury, like the stuff tion. Defenders collected
found in glass thermometers—yuck!) Exposure to a sample of 164 cans of
mercury can be especially hazardous for pregnant tuna from stores across
women and small children. How much mercury the United States. They
is safe to consume? That depends on your body sent the selected cans to
weight. According to the Environmental Protec- a laboratory that is
tion Agency (EPA), a 180-pound man would be often used by the
safe eating one 6-ounce can of tuna per week if EPA for mercury
the mercury content was less than 0.31 parts per testing. The table
million (ppm). For a 45-pound child to be safe, the below shows data
same can of tuna would have to contain less than 0.08 on the brand,

s
Pic
ppm of mercury. The Food and Drug Administra- country of origin,

re
atu
/Fe
tion will take action (such as removing the product type of tuna (light or ners
Ron Sum
from store shelves) if the mercury concentration in albacore), and mercury
a 6-ounce can of tuna is 1.00 ppm or higher. content (ppm) for 8 of the 164 cans.9 You can get
What is the typical mercury concentration the entire set of data from your teacher or from the
in cans of tuna sold in stores? A study conducted book’s Web site, www.whfreeman.com/sta2e.

Can Brand Country Type Mercury (ppm)

11 Dave’s Ahi Tuna United States Light 0.790
12 Van Triunfo Ecuador Light 0.760
13 Whole Foods Thailand Albacore 0.730
14 Tuna Real Ecuador Light 0.720
15 Starkist United States Albacore 0.710
16 Tuna Real Ecuador Light 0.700
17 Tuna Real Ecuador Light 0.660
18 Tuny Mexico Light 0.640

QUESTIONS
1. What is the population of interest? What is the 2. List the variables that were measured. Identify
sample? each as categorical or quantitative.
 2. 1 Displaying Distributions with Graphs 57

3. The Fathom bar graph below displays 4. The Fathom histogram below displays the
information about the country of origin for the mercury concentration in the sampled cans. Write
whole sample. About what percent of the cans in the a few sentences describing this distribution.
sample originated in the United States? Is this value
a good estimate for the percent of all cans of tuna
that originate in the United States? Why or why not?

We’ll return to the tuna data a little later, so keep

your work handy.

Section 2.1 Summary

To see what data say, start with graphs. The choice of graph depends on the type of
data. The distribution of a variable tells us what values the variable takes and how
often it takes each value. To display the distribution of a categorical variable, use a pie
chart or a bar graph. Pie charts always show the parts of some whole, but bar graphs
can compare any set of numbers measured in the same units. To display the distri-
bution of a quantitative variable, use a dotplot, stemplot, or histogram. We usually
favor dotplots or stemplots when we have a small number of observations. For large
amounts of data, histograms are usually better.
When you look at a graph, look for an overall pattern and for deviations from that
pattern, such as outliers. We can describe the overall pattern of a dotplot, stemplot, or
histogram by giving its shape, center, and spread. Some distributions have simple shapes
such as symmetric or skewed, but others are too irregular to describe by a simple shape.

Section 2.1 2.19 Feeling sleepy? Students in a college statistics class responded to a survey
Exercises designed by their teacher. One of the survey questions was “How much sleep did
you get last night?” Here are the data (in hours):

9 6 8 6 8 8 6 6.5 6 7 9 4 3 4
5 6 11 6 3 6 6 10 7 8 4.5 9 7 7

(a) Make a dotplot to display the data.

(b) Describe the overall pattern of the distribution and any deviations from that pattern.
2.20 Bad habits According to the National Household Survey on Drug Abuse,
31.8% of adolescents aged 12 to 17 years used alcohol in 2007, 12.5% used mari-
juana, 1.5% used cocaine, and 15.7% used cigarettes. Explain why it is not correct
to display these data in a pie chart.
2.21 Low birth weights Figure 2.12 (on the next page) shows the distribution of
percent of low birth weights for the 26 states east of the Mississippi.10 Notice that the
58 CHAPTER 2 Describing Distributions of Data

vertical scale in Figure 2.12(a) is the number of states in each class. In Figure 2.12(b),
we have calculated and displayed the percent of states in each low-birth-weight class.

(a) (b)
12 50
Number of states

Percent of states
10 40
8
30
6
20
4
2 10
0 0
6 7 8 9 10 11 6 7 8 9 10 11
Percent low birth weight Percent low birth weight
Figure 2.12 (a) The distribution of infants born with low birth weight in the 26 states east of the Mississippi. (b) The
heights of the bars are the same, but the vertical scale has been changed to show the percent of states in each class.

(a) Describe the shape, center, and spread of the distribution.

(b) When might it be preferable to use percents rather than counts on the vertical
axis of a histogram?
(c) Mississippi had the highest percent of low birth weight, 10.1. Why do you
think this state has such a high percent of low-birth-weight infants?
2.22 Skewed left
(a) Sketch a histogram for a distribution that is skewed to the left.
(b) Suppose that you and your friends emptied your pockets of coins and recorded
the year marked on each coin. The distribution of dates would be skewed to the
left. Explain why.
Exercises 2.23 to 2.26 refer to the following setting. CensusAtSchool is an
international project that collects data about primary and secondary school
students using surveys. As of early 2009, students in Australia, Canada, New
Zealand, South Africa, and the United Kingdom have participated in the
project. Data from CensusAtSchool surveys are available from the project’s
Web site, www.censusatschool.com. We used the “Random Data Selector”
30 to choose a simple random sample of 50 Canadian stu-
Frequency of method of

dents who completed the survey in 2007–2008. The data

25
set CASCAN0708 can be found on the book’s Web site,
communication

20 www.whfreeman.com/sta2e.
15 www
2.23 Let’s chat The bar graph on the left displays data
10
on students’ responses to the question “Which of these
methods do you most often use to communicate with your
5 friends?”
(a) Would it be appropriate to make a pie chart for these
In person

Text messaging
Cell phone

MySpace,
Internet chat
or MSN

Facebook, blogs

Telephone
(landline)

data? If so, do it. If not, explain why not.

(b) Summarize what the bar graph tells you about students’
communication preferences.
www
2.24 Travel time The dotplot on the facing page displays
data on students’ responses to the question “How long
Method of communication does it usually take you to travel to school?”
 2. 2 Describing Distributions with Numbers 59

(a) Make a well-labeled histogram of the data.

(b) Describe the shape, center, and spread of the distribution. Are there any
outliers?

0 20 40 60 80 100
Travel time (minutes)

2.25 Who’s left-handed? Students were asked, “Are you right-handed, left-
www
handed or ambidextrous?” The responses of the 50 randomly selected
Canadian students are shown below (R  right-handed; L  left-handed;
A  ambidextrous).

R R R R R R R R R R R L R R
R R R R R R R R R R R R R A
R R R R A R R L R R R R L A
R R R R R R R R

(a) Make an appropriate graph to display these data.

(b) Over 10,000 Canadian high school students took the CensusAtSchool survey
in 2007–2008. What percent of this population would you estimate is left-handed?
Justify your answer.
2.26 How tall are you? Here are the heights (in centimeters) of the 50 randomly
www
selected Canadian students who participated in CensusAtSchool in 2007–2008.

166.5 170 178 163 150.5 169 173 169 171 166 190 183 178 161
171 170 191 168.5 178.5 173 175 160.5 166 164 163 174 160 174
182 167 166 170 170 181 171.5 160 178 157 165 187 168 157.5
145.5 156 182 168.5 177 162.5 160.5 185.5

Make a stemplot of these data. Describe the shape, center, and spread of the distri-
bution. Are there any outliers?

2 .2 Describing Distributions with Numbers

ACTIVITY 2.2A
Do you know your geography?
How well do you know the U.S. state capitals? The geographic knowledge by estimating the population
locations of major rivers and mountain chains? of a capital city.11
The populations of states and cities? In this 1. Your teacher will give each student a slip of paper
Activity, you’ll have a chance to demonstrate your with two questions about the population of a U.S.
ACTIVITY 2.2A (continued)
60 CHAPTER 2 Describing Distributions of Data

ACTIVITY 2.2A (continued)

capital city. Answer the questions to the best of your not included in either of these halves.) Use the
ability. procedure from Step 3 to find the median of the
lower half of the estimates. This value is called the
2. Your teacher will form two groups. Students
first quartile. (Why do you think we give it that
in each group should line up in order of their
name?)
estimates, from smallest to largest.
5. Repeat the process from Step 3 again to find
3. Find the median of your population estimates the median of the upper half of the estimates. The
as follows. Have the student at each end of the resulting value is called the third quartile.
line begin counting off with “One.” The next
student in line at each end should call out “Two.” 6. Record the following five values for the estimates
Continue until you locate the middle person in in your line: the minimum, first quartile, median,
line (if the last person is the only one calling out third quartile, and maximum. This is known as the
a number) or the middle two people in line. In five-number summary.
the latter case, average the estimates of the two 7. Your teacher will provide the capital city’s actual
middle students. population. How close was your estimate?
4. The median from Step 3 divides the line 8. Which group did better at estimating the city’s
into two halves. (Note that the median itself is population? How can you tell?

Measuring center: the median

A natural way to describe the center of a distribution is to use the “middle value”
in a histogram or stemplot. That is, find the number such that half the observa-
tions are smaller and the other half are larger. This is the median of the distribu-
tion. We will call the median M for short. Although the idea of the median as the
midpoint of a distribution is simple, we need a precise rule for putting the idea
into practice.

Example 2.9 Finding the median

To find the median of the numbers

8 4 9 1 3

arrange them in increasing order:

1 3 4 8 9

The boldface 4 is the center observation, because there are 2 observations to its
left and 2 to its right. When the number of observations n is odd, there is always
one observation in the center of the ordered list. This is the median, M  4.
If n is even, there is no one middle observation. But there is a middle pair, and
we take the median to be the mean of this middle pair, the point halfway between
them. So the median of

8 4 1 9 1 5
 2. 2 Describing Distributions with Numbers 61

is found by arranging these numbers in increasing order,

1 1 4 5 8 9
and averaging the middle pair,

415
M5 5 4.5
2

Here is our rule for finding medians.

The median M

The median M is the midpoint of a distribution, the number such that

half the observations are smaller and the other half are larger. To find
the median of a distribution:

1. Arrange all the observations in order of size, from smallest to largest.

2. If the number of observations n is odd, the median M is the center
observation in the ordered list.
3. If the number of observations n is even, the median M is the average
of the two center observations in the ordered list.

Example 2.10 How many text messages?

In a September 28, 2008, article titled “Letting Our Fingers Do the Talking,” the
New York Times reported that Americans now send more text messages than they
make phone calls. According to a study by Nielsen Mobile, “Teenagers ages 13 to 17
are by far the most prolific texters, sending or receiving 1,742 messages a month.”
Mr. Brown, a high school statistics teacher, was skeptical of the texting result stated
in the article. So he collected data from his first-period statistics class on the number
of text messages they had sent and received in the past 24 hours. Here are the data:

0 7 1 29 25 8 5 1 25 98 9 0 26 8 118 72 0 92 52 14 3 3 44 5 42

Let’s find the median number of text messages sent and received in the past
24 hours. We begin by sorting the values from lowest to highest:

0 0 0 1 1 3 3 5 5 7 8 8 9 14 25 25 26 29 42 44 52 72 92 98 118

Since there are 25 values, the median is the bold 9 in the center of the ordered list.

Measuring spread with quartiles

In Example 2.10, the median number of text messages in the past 24 hours for
Mr. Brown’s students was 9. That information is helpful but incomplete. Did most
62 CHAPTER 2 Describing Distributions of Data

students have close to 9 text messages in the previous 24 hours, or were the data
values very spread out?
The simplest useful description of a distribution consists of both a measure of
center and a measure of spread. If we choose the median (the midpoint) to describe
center, the quartiles give us a natural way to measure spread. Again, the idea is
clear: find the points one-quarter and three-quarters up the ordered list of observa-
tions. Here’s an example.

Example 2.11 How many text messages? (continued)

The first half of the ordered list of number of text messages sent and received in the
past 24 hours is

0 0 0 1 1 3 3 5 5 7 8 8

There are 12 data values less than the median, and we have put the two middle
numbers in boldface type. The first quartile, Q1, is the average of the middle two
numbers, which is 3. The 12 data values greater than the median are

14 25 25 26 29 42 44 52 72 92 98 118

The third quartile, Q3, is the average of the two middle numbers, also in bold.

42 1 44
Q3 5 5 43
2

We can measure the spread of the values around the median by calculating the
distance between Q1 and Q3:

Q3 2 Q1 5 43 2 3 5 40

This is known as the interquartile range (IQR for short).

Note: If there is an odd number of observations less than or greater than the
median, then the middle number of that half of the data is Q1 or Q3. Remember
that in finding the quartiles, you never include the median, whether it is a data
value or not.

The quartiles get their name because, with the median, they divide the
observations into quarters—one-quarter of the observations lie below the first
quartile, half the observations lie below the median, and three-quarters lie
below the third quartile. The interquartile range (IQR) measures the range of
the middle 50% of the data. That’s the idea. Again, we need a rule to make the
idea precise. The rule for calculating the quartiles and the IQR uses the rule
for the median.
 2. 2 Describing Distributions with Numbers 63

The quartiles Q1 and Q3; the interquartile range (IQR)

To calculate the quartiles:

1. Arrange the observations in increasing order and locate the median

M in the ordered list of observations.
2. The first quartile Q1 is the median of the observations whose posi-
tion in the ordered list is to the left of the location of the overall
median.
3. The third quartile Q3 is the median of the observations whose posi-
tion in the ordered list is to the right of the location of the overall
median.

The interquartile range (IQR) is defined as

IQR 5 Q3 2 Q1

The five-number summary and boxplots

The smallest and largest observations tell us little about the distribution as a whole,
but they give information about the tails of the distribution that is missing if we
know only the median and the quartiles. To get a quick summary of both center
and spread, combine all five numbers.

The five-number summary

The five-number summary of a distribution consists of the smallest

observation, the first quartile, the median, the third quartile, and the
largest observation, written in order from smallest to largest. In sym-
bols, the five-number summary is

Minimum Q1 M Q3 Maximum

These five numbers offer a reasonably complete description of center and spread.
We can use them to make a new type of graphical display, a boxplot.

Example 2.12 How many text messages? A boxplot

The five-number summary for the text message data of Examples 2.10 and 2.11 is

0 3 9 43 118

Figure 2.13 (on the next page) shows a boxplot of the distribution.
64 CHAPTER 2 Describing Distributions of Data

0 20 40 60 80 100 120
Number of text messages
sent/received in last 24 hours
Figure 2.13 A boxplot of the text message data for Mr. Brown’s students from
Examples 2.10 and 2.11.

Here are the main properties of boxplots.

Boxplot

A boxplot is a graph of the five-number summary:

• A central box is drawn from the first quartile (Q1) to the third
quartile (Q3).
• A line in the box marks the median.
• Lines extend from the box out to the smallest and largest observa-
tions that are not outliers.

We will adopt the convention of identifying outliers as isolated points in the boxplot.
In Example 2.12, the student who sent and received 118 text messages in the past
24 hours seems like a clear outlier. What about the student with 98 text messages? This
observation is not considered an outlier, because it isn’t “far enough” from the other val-
ues in the data set. How far is “far enough?” Here’s the rule that’s most commonly used.

Identifying outliers

Any observation that is more than 1.5 times the interquartile range
(IQR) above Q3 or below Q1 is considered an outlier.

For the text message data of Example 2.12, IQR  43  3  40 and 1.5 1 IQR 2 5
1.5 1 40 2 5 60. Any observation that is above
Q3 1 1.5 1 IQR 2 5 43 1 60 5 103
or below
Q1 2 1.5 1 IQR 2 5 3 2 60 5 257
 2. 2 Describing Distributions with Numbers 65

would be identified as an outlier. So the student with 98 text messages is not an

outlier, but the one with 118 text messages is.

Exercises 2.27 Median income You read that the median income of U.S. households in
2007 was $50,233. Explain in plain words what “the median income” is.
2.28 Acing the first test Here are the scores of Mrs. Liao’s students on their first
statistics test:

93 93 87.5 91 94.5 72 96 95 93.5 93.5 73 82 45 88 80

86 85.5 87.5 81 78 86 89 92 91 98 85 82.5 88 94.5 43

(a) Find the median. Explain what this number means in this setting.
(b) Find the quartiles and the IQR. Interpret the meaning of the IQR in this
setting.
(c) Are there any outliers? Justify your answer with appropriate calculations.
(d) Make a boxplot of the test score data.
2.29 Phone calls, I After hearing about the text message study of
Example 2.10, Mrs. Krebs asked the students in her statistics class how many
phone calls they had made or received in the past 24 hours. Here are their
responses:

3 0 12 10 8 4 2 7 10 11 6 7 0 35 12 3 10 2 7 9 15 5 14 15 6

(a) Find the median. Explain what this number means in this setting.
(b) Find the quartiles and the IQR. Interpret the meaning of the IQR in this
setting.
(c) Determine whether there are any outliers. Show
your work.
2.30 Phone calls, II Refer to the previous exercise. Make
a boxplot of the phone call data from Mrs. Krebs’s class.
How would you describe the distribution of the variable
“number of phone calls made or received in the past
24 hours”?
2.31 Texting or calling, I After hearing about the text
message study of Example 2.10, Mr. Williams collects
data from each student in his math class on the number
of cell phone texts and calls sent or received in the
past 24 hours. Figure 2.14 is a boxplot of the difference
20 0 20 40 60 80 100 120
(texts  calls) in the number of text messages and calls
Difference (texts – calls)
for each student.
Figure 2.14 Boxplot showing the difference
(a) Estimate the five-number summary for these data from
(texts – calls) in the number of text messages and
calls sent and received by the students in Mr. Williams’s
the boxplot.
math class. (b) Interpret the value of Q1, M, and Q3 in this setting.
66 CHAPTER 2 Describing Distributions of Data

2.32 Texting or calling, II Refer to the previous exercise.

(a) Do Mr. Williams’s students seem to prefer texting or talking on the phone?
Give appropriate evidence to support your answer.
(b) Can we draw any conclusion about the preferences of all students in the school
based on the data from Mr. Williams’s math class? Why or why not?

Measuring center: the mean

The five-number summary is not the most common numerical description of a dis-
tribution. That distinction belongs to the combination of the mean to measure cen-
ter and the standard deviation to measure spread. The mean is familiar—it is the
ordinary average of the observations. We’ll meet the standard deviation shortly.

The mean x–

The mean, denoted by x (pronounced “x-bar”), of a set of observa-

tions is their average. To find the mean of n observations, add the
values and divide by n:
sum of the observations
x5
n

Let’s consider a simple example involving the mean.

Example 2.13 Do you have pets?

Nine elementary school children were asked how many pets they had. Here are their
responses, arranged from lowest to highest:12

1 3 4 4 4 5 7 8 9

Since there are 9 observations, the median is the middle number: 4 pets. What’s the
mean number of pets for this group of children? It’s

sum of observations 11314141415171819

x5 5 55
n 9

But what does that number tell us? Here’s one way to look at it: if every child in the
group had the same number of pets, each would have 5 pets. In other words, the
Jonathan Brizendine/iStockphoto
mean is the “fair share” value.

As we saw in Example 2.13, the mean of a set of data is more than just the nu-
merical average. The mean tells us how large each observation in the data set would
be if the total were split equally among all the observations. In the language of young
children, the mean is the “fair share” value. The mean of a distribution also has a
physical interpretation, as the following Activity shows.
 2. 2 Describing Distributions with Numbers 67

ACTIVITY 2.2B
What does the mean mean?
Materials: One pad of Post-it Notes observation from 5 to 3. Move two other sticky
notes to the right so that the mean becomes
In this Activity, you’ll use the pet data from
5 again.
Example 2.13 to explore an interesting property of
the mean. 5. Start with all nine sticky notes above 5 on the
1. Draw a number line on the board. Label the dotplot. Change the value of one observation to 9,
horizontal axis from 0 to 9, with the tick marks the maximum number of pets among the children.
spaced out by at least the width of a Post-it Note. Now move another sticky note to keep the mean
at 5. Change the value of one of the 5s to a 3.
2. Make a sticky-note dotplot with all nine values Move another sticky note from the stack above 5
equal to 5. Mark the location of the mean. to make the mean 5 again. Finally, change three
3. Change the value of one observation from 5 to 9 of the 5s to 4s. Move one of the remaining sticky
by moving a sticky note. Move one other sticky note notes from above 5 to return the mean of the
so that the mean of the distribution becomes 5 again. distribution to 5.
4. Return the two sticky notes to their original 6. Do you see why the mean is sometimes called the
location at 5. Now change the value of one “balance point” of a distribution?

The mean and median give two different ways to describe the center of a dis-
tribution. The median is right in the middle of the ordered data. But the median
ignores the size of the values at either end of the distribution. As a result, we say
that the median is resistant. The mean, on the other hand, incorporates every
value in the data set. As a result, extreme values such as outliers can have a large
effect on the mean.

Example 2.14 The median is resistant; the mean is not

In Example 2.10 (page 61), we examined data on the number of text messages sent
and received in one day by Mr. Brown’s statistics class. The median number of texts
was 9. What about the mean? If we add the number of text messages and divide by
25, we get
0 1 7 1 1 1 29 1 c1 42 687
x5 5 5 27.48
25 25

That’s a lot more than 9. Notice that the 12 values below the median are much less
spread out than the 12 values above the median. Since the mean is the balance point
of the distribution, it is pulled far to the right by the several large values above the
median.

In Example 2.14, the mean is pulled far above the median by extreme skewness
and an outlier. Because of this, the mean is the preferred measure of center only when
68 CHAPTER 2 Describing Distributions of Data

a distribution is roughly symmetric with no outliers. When a distribution is skewed

PPLET
or has outliers, the median does a much better job of summarizing its center.
A
You can compare the behavior of the mean and median by using the Mean and
Median applet at the book’s Web site, www.whfreeman.com/sta2e.

Measuring spread: the standard deviation

When we use the median to measure the center of a distribution, the interquartile
range (IQR) is our corresponding measure of spread. If we summarize the center
of a distribution with the mean, then the standard deviation is used to describe the
spread of observations around the mean. The idea of the standard deviation is to
give the average distance of observations from the mean.

Example 2.15 Do you have pets? (continued)

In Example 2.13, we found the mean number of pets for the group of children to
be x 5 5. Let’s look at where the observations in the data set are relative to the
mean. The table below shows the deviation (value  mean) of each value in the
data set from the mean. Add up the deviations from the mean. Did you get 0? You
should, because the mean is the balance point of the distribution. Since the sum of
the deviations from the mean will be 0 for any set of data, we need another way to
calculate spread around the mean.

Value: 1 3 4 4 4 5 7 8 9
Value  mean: 1  5 35 45 45 45 55 75 85 95
Deviation: 4 2 1 1 1 0 2 3 4

How can we fix the problem of the positive and negative deviations canceling
each other out? We could take the absolute value of each deviation. Or we could
square the deviations. For mathematical reasons beyond the scope of this book, stat-
isticians choose to square rather than to use absolute values.
The table below shows the pet data and the deviations from the mean once
again. We have added a row at the bottom that shows the square of each deviation.
Add up the squared deviations. Did you get 52?

Value: 1 3 4 4 4 5 7 8 9
Deviation: 4 2 1 1 1 0 2 3 4
Squared deviation: 16 4 1 1 1 0 4 9 16

Now we compute the average squared deviation—sort of. Instead of dividing

by the number of observations n, we divide by n  1:

16 1 4 1 1 1 1 1 1 1 0 1 4 1 9 1 16 52
“average” squared deviation  5 5 6.5
921 8

Because we squared all the deviations, our units are in “squared pets.” That’s no good.
We’ll take the square root to get back to the correct units—pets. The resulting value
 2. 2 Describing Distributions with Numbers 69

is the standard deviation:

standard deviation  !“average” squared deviation 5 !6.5 5 2.55
This 2.55 is “sort of like” the average distance of the values in the data set from the mean.

As you saw in Example 2.15, the “average distance” in the standard deviation is
found in a rather unexpected way. You may want to just think of the standard devi-
ation as “average distance from the mean” and leave the details to your calculator.

The standard deviation s

The standard deviation s measures the average distance of the obser-

vations from their mean. It is calculated by finding an average of the
squared distances and then taking the square root. To find the stan-
dard deviation of n observations:

1. Find the distance of each observation from the mean and square
each of these distances.
2. Average the distances by dividing their sum by n  1. This average
squared distance is called the variance.
3. The standard deviation s is the square root of this average squared
distance.

In practice, you can key the data into your calculator and get the mean and standard
deviation by asking for one-variable statistics. (See the Calculator Corner on page 72.)
Or you can enter the data into a spreadsheet or other software to find x and s. It is usual,
for good but somewhat technical reasons, to average the squared distances by dividing
their total by n  1 rather than by n. Many calculators report two standard deviations,
giving you a choice between dividing by n and dividing by n  1. The former is usually
labeled s, the symbol for the standard deviation of a population. If your data set con-
sists of the entire population, then it’s appropriate to use s. More often, the data we’re
examining come from a sample. In that case, we should use s and divide by n  1.
More important than the details of the calculation are the properties that show
how the standard deviation measures spread.

Properties of the standard deviation s

• s measures spread about the mean x. Use s to describe the spread of a

distribution only when you use x to describe the center.
• s  0 only when there is no variability. This happens only when all
observations have the same value. So standard deviation zero means
no spread at all. Otherwise, s  0. As the observations become more
spread out about their mean, s gets larger.
70 CHAPTER 2 Describing Distributions of Data

© 2008 ZITS Partnership. Distributed by King Features Syndicate.

Exercises 2.33 Metabolism, I A person’s metabolic rate is the rate at which the body
consumes energy. Metabolic rate is important in studies of weight gain, dieting,
and exercise. Here are the metabolic rates of seven men who took part in a study
of dieting. (The units are calories per 24 hours. These are the same calories used to
describe the energy content of foods.)
1792 1666 1362 1614 1460 1867 1439
Use the formula to calculate the mean. Interpret this value.
2.34 Metabolism, II Refer to the previous exercise.
(a) Use the method in Example 2.15 to calculate the deviation of each observation
from the mean. Show that the sum of the deviations is 0.
(b) Calculate the standard deviation. Show your work.
2.35 Mean income You read that the mean income of U.S. households in 2007
was $67,609. Recall from Exercise 2.27 (page 65) that the median U.S. household
income in 2007 was $50,233. Explain why the mean household income is so much
higher than the median household income.
2.36 Feeling sleepy? The first four students to arrive for a first-period statistics
class were asked how much sleep (to the nearest hour) they got last night. Their
responses were 7, 7, 9, and 9.
(a) Use the formula to calculate the mean. Interpret this value.
(b) Use the method in Example 2.15 to calculate the deviation of each observation
from the mean. Then compute the standard deviation. Explain what this value means.
2.37 A matchup Match the summary statistics with the histograms. Explain how
you made your decision.

25 10 8
6
15 6
4
5 2 2

0 10 20 3.5 6.0 8.5 15 5 5 15 25

Variable 1 Variable 2 Variable 3

(a) mean  6.6, median  6.8, standard deviation  1.3, variable  .

(b) mean  6.6, median  6.0, standard deviation  8.65, variable  .
(c) mean  6.6, median  3.75, standard deviation  7.4, variable  .
 2. 2 Describing Distributions with Numbers 71

2.38 Properties of the standard deviation

(a) Juan says that, if the standard deviation of a list is zero, then all the numbers
on the list are the same. Is Juan correct? Explain your answer.
(b) Letishia alleges that, if the means and standard deviations of two different lists
of numbers are the same, then all of the numbers in the two lists are the same. Is
Letishia correct? Explain your answer.

CALCULATOR CORNER Boxplots and summary statistics

You can use the TI-84 or TI-Nspire to make boxplots and calculate numerical
summaries. We’ll show you how using the text message data from Example
2.10 (page 61).

I. MAKING A BOXPLOT
TI-84
1. Enter the texting data: Press A and choose 1:Edit… Then type the values
into list L1.
2. Define Plot 1 as a boxplot of the data in L1.
• Press yo (STAT PLOT).
• Press b or À to go into Plot 1.
• Adjust your settings as shown.

3. Use the calculator’s zoom feature to display the boxplot.

• Press q and choose 9:ZoomStat.
• To see the five-number summary for the texting data, press the r key.

TI-Nspire
1. Enter the texting data in a Lists & Spreadsheet page.
• Press c and choose 6:New Document.
• Press b; choose 3:Add Lists & Spreadsheet.
• In the name box at the top of column A, type “texts.”
• Type the values into list “texts.”
2. Insert a new Data & Statistics page. Press /I and choose 5:Add Data
& Statistics.
CALCULATOR CORNER (continued)
72 CHAPTER 2 Describing Distributions of Data

CALCULATOR CORNER (continued)

3. Now make your boxplot.
• Use the “NavPad” to move your cursor over the “Click to add variable” box at
the bottom of the screen.
• Click the mouse x. You should see the variable “texts” highlighted. Press · or
click xto make a dotplot.
• Press b; choose 1:Plot Type and 2:Box Plot to change the graph to
a boxplot.
• Move your cursor over the boxplot to see the five-number summary.
II. CALCULATING SUMMARY STATISTICS
TI-84
• Press yz (QUIT) to go to the home screen.
• Press A, arrow-right ~ to CALC, and choose 1:1-Var Stats.
• Complete the command 1-Var Stats L1 and press b. (Press yÀ
to get L1.)
• Notice the down-arrow on the left side of the display. Press the down-arrow
key † several times to see more.

TI-NSPIRE
• Press /¡ to return to the Lists & Spreadsheet page.
• Press b; choose 4:Statistics, 1:Stat Calculations…, and
1:One-Variable Statistics.
• When the dialog box appears, press ·. The One-Variable Statistics dialog box
will appear.
• Use the drop-down menu for X1 List: to choose “texts.” Then press ·.
• Use the NavPad to scroll down the values in column C.

Choosing numerical descriptions

The five-number summary is easy to understand and is the best short description
for most distributions. The mean and standard deviation are harder to understand
but are more common. How can we decide which of these two descriptions of cen-
ter and spread we should use?
 2. 2 Describing Distributions with Numbers 73

Example 2.16 First-test scores

Figure 2.15 shows a dotplot of students’ scores on Mrs. Liao’s first test (see Exercise 2.28,
page 65). The median is M  87.75; the mean is x  84.8. Which of these values more ac-
curately summarizes how a “typical” student did on this test? The mean is clearly affected
by the two extremely low scores. Only 9 of the 30 students in the class earned a score be-
low 84.8, so the median does a better job of describing a typical student’s performance.

Figure 2.15 Minitab dotplot of students’ scores on their first statistics test.

To investigate the effect of the two low outliers on the summary statistics, we cal-
culated the mean, standard deviation, and five-number summary with and without
these two scores. The table below summarizes the results. When the outliers are
removed, the mean increases to be almost equal to the median. Also notice the large
effect that the two extreme scores have on the standard deviation. What about the
IQR? With all 30 observations, IQR  93  82  11. With the two outliers removed,
IQR  93.25  83.75  9.5. The IQR is fairly resistant to the extreme scores.
n x s Min. Q1 M Q3 Max.
Poor New York? All data 30 84.8 12.857 43 82 87.75 93 98
Is New York a rich state? No outliers 28 87.7 6.735 72 83.75 88 93.25 98
New York’s mean income
per person ranks fourth
among the states, right up
there with its rich neigh-
bors Connecticut and New
The mean and median of a symmetric distribution are close to each other. In
Jersey, which rank first and fact, x and M are exactly equal if the distribution is exactly symmetric. In skewed
second. But while Con- distributions, however, the mean runs away from the median toward the long
necticut and New Jersey tail. Many distributions of monetary values—incomes, house prices, wealth—are
rank seventh and second in strongly skewed to the right. The mean may be much larger than the median. For
median household income,
example, we saw at the beginning of the chapter that the median income of people
New York stands at twenty-
ninth, well below the na- with advanced degrees in the Current Population Survey sample is $62,000. The
tional average. What’s going distribution is skewed to the right, and the long right tail pulls the mean up to
on? Just another example of $82,729. Because monetary data often have a few extremely high observations, de-
mean versus median. New scriptions of these distributions usually employ the median.
York has many very highly The standard deviation is pulled up by outliers or the long tail of a skewed distribu-
paid people who pull up its
mean income per person.
tion even more strongly than the mean. The standard deviation of students’ test scores
But it also has a higher pro- in Mrs. Liao’s class is s  12.857 for all 30 students and only s  6.735 when the two out-
portion of poor households liers are removed. The quartiles are much less sensitive to a few extreme observations.
than do New Jersey and There is another reason to avoid the standard deviation in describing skewed
Connecticut, and this brings distributions. Because the two sides of a strongly skewed distribution have different
the median down. New
spreads, no single number such as s describes the spread well. The five-number sum-
York is not a rich state—it’s
a state with extremes of mary, with its two quartiles and two extremes, usually does a better job. In most situ-
wealth and poverty. ations, it is wise to use x and s only for distributions that are roughly symmetric.
74 CHAPTER 2 Describing Distributions of Data

Choosing a summary

The mean and standard deviation are strongly affected by outliers and
by the long tail of a skewed distribution. The median and quartiles are
less affected.

The median and IQR are usually better than the mean and standard devia-
tion for describing a skewed distribution or a distribution with outliers. Use
x and s only for reasonably symmetric distributions that are free of outliers.

Why do we bother with the mean and standard deviation at all? One answer
appears in the next chapter: the mean and standard deviation are the natural mea-
sures of center and spread for an important kind of symmetric distribution, called
a normal distribution.

Transforming data
After looking at the distribution of test scores in Example 2.16, Mrs. Liao is consid-
ering “adjusting” the scores. She has two ideas in mind:
1. Add 5 points to each student’s test score.
2. Double each student’s score and grade the test out of 200 points.
Before she implements either idea, she wants to know how each possible adjust-
ment would affect the shape, center, and spread of the test score distribution.

ACTIVITY 2.2C
First-test scores: making an adjustment
In this Activity, you will investigate the effect of Mrs. Liao’s two ideas for adjusting her class’s test scores on
the mean, standard deviation, and five-number summary. Here are the data again:

93 93 87.5 91 94.5 72 96 95 93.5 93.5 73 82 45 88 80

86 85.5 87.5 81 78 86 89 92 91 98 85 82.5 88 94.5 43

1. On your paper, make a chart like the one shown below to record your results.

n x s Min. Q1 M Q3 Max.
Original scores 30 84.8 12.857 43 82 87.75 93 98
Our guess: add 5 30
Our guess:
multiply by 2 30
Actual result:
add 5 30
Actual result:
multiply by 2 30
 2. 2 Describing Distributions with Numbers 75

2. Discuss with a partner: what do you think will 4. Start over with the original data. This time,
happen to each of the numerical summary values multiply each student’s score by 2. Proceed as
in the table for Mrs. Liao’s two ideas? Record your in Step 3 to complete the final row of the table.
educated guesses in the appropriate rows. How do the results compare with your educated
guesses?
AP 3. Start with the original data and add 5 to
PLET

each student’s test score. Use One-Variable 5. Summarize what you have learned about the
Statistics on your calculator or the One effects of (i) adding a constant and (ii) multiplying
Variable Statistical Calculator applet at the book’s by a constant on our popular measures of center
Web site (www.whfreeman.com/sta2e) to fill in the and spread.
values in the fourth row of the table. How do the 6. Investigate: What happens to the shape of the
results compare with your educated guesses? distribution with each of these two transformations?

Activity 2.2C shows what happens to the shape, center, and spread of a dis-
tribution if the original values are adjusted in a particular way. Adding a constant
shifts the center of the distribution by that amount but doesn’t affect the shape or
spread. Multiplying all of the original values by a constant multiplies the mean,
standard deviation, and every value in the five-number summary by that constant.
Again, the shape of the distribution stays the same.

Comparing distributions
Some of the most interesting statistics questions involve comparing two or more
groups. Which of two popular diets leads to greater long-term weight loss? How
does income relate to educational level? Who texts more—males or females? Does
seat belt use differ across regions of the United States? To make such comparisons,
start with a graph. For categorical variables, use a bar graph or a pie chart. For
quantitative variables, make a dotplot, stemplot, or histogram. Then add numerical
summaries and a boxplot.

Example 2.17 Habits of mind for college freshmen

The 2007 CIRP Freshman Survey asked first-year college students about their
“habits of mind”—specific behaviors that college faculty have identified as being
important for student success. One question asked students, “How often in the
past year did you revise your papers to improve your writing?” Another asked,
“How often in the past year did you seek feedback on your academic work?”
Figure 2.16 (on the next page) is a bar graph comparing male and female re-
sponses to these two questions.13
What does the bar graph tell us about the habits of mind of male and female
college freshmen? Over half of females surveyed said that they had frequently
revised their papers to improve their writing as high school seniors. Only 37% of
the males who were surveyed answered “frequently.” Nearly half of females said
that they had frequently sought feedback on their work in the past year, com-
pared to about 38% of males who were surveyed. It appears that females engaged
76 CHAPTER 2 Describing Distributions of Data

100
Men
80 Women

% “Frequently”
60 54.9
49.0
40 36.9 37.6

20

0
Revise papers to Seek feedback
improve writing on work
Figure 2.16 Comparative bar graph of male and female students’ responses to the 2007
CIRP Freshman Survey about the frequency of specific learning behaviors in the past year.

in these two specific learning behaviors more often than males did as high school
seniors.

Example 2.17 shows you how to compare distributions of a categorical variable.

The following Data Exploration involves a quantitative variable.

DATA EXPLORATION
Mercury in tuna (continued)
In Application 2.1 (page 56), you examined data
from a study of mercury concentration in canned
tuna. You should reread the details of that study
before you answer the questions that follow.
Is there a difference in the mercury concentration
(parts per million) in light tuna and albacore tuna?
The accompanying parallel boxplots and summary
chart provide information to help you answer this
question.

Type Count x s Min. Q1 M Q3 Max.

Albacore 20 0.401 0.152 0.17 0.295 0.4 0.46 0.73
Light 144 0.269 0.312 0.012 0.0595 0.16 0.345 1.5

1. Explain why the mean mercury content is much 3. Write a few sentences comparing the distributions
higher than the median for the light tuna but not of mercury concentration for these two types of
for the albacore tuna. canned tuna.
2. Confirm that the outliers shown in the boxplots
are identified by the 1.5(IQR) rule.
 2. 2 Describing Distributions with Numbers 77

As the Data Exploration suggests, you should always discuss shape, center,
spread and any unusual values whenever you compare distributions of a quantita-
tive variable.

Exercises 2.39 Music CDs How many music CDs do students own? The 24 members of
a college statistics class provided data in response to this question.14 Figure 2.17
is a dotplot of the data. Which would better summarize the center and spread of
the distribution: (i) the mean and standard deviation or (ii) the median and IQR?
Justify your answer.

Figure 2.17 Minitab dotplot showing the number of music CDs owned by students in a
college statistics class.

2.40 Domain names When it comes to Internet domain names, is shorter

www
better? According to one ranking of Web sites in 2008, the top 8 sites (by number
of “hits”) were yahoo.com, google.com, youtube.com, live.com, msn.com,
myspace.com, wikipedia.org, and facebook.com. These familiar sites certainly
have short domain names. Figure 2.18 is a histogram of the domain name lengths
for the 500 most popular Web sites.
(a) Estimate the mean and median of the distribution. Explain your method
clearly.
(b) If you wanted to argue that shorter domain names were more popular, which
measure of center would you choose—the mean or the median? Justify your answer.

120

100

80
Frequency

60

40

20

0
2 4 6 8 10 12 14 16
Domain name length

Figure 2.18 Minitab histogram showing lengths (number of letters) of domain names
for the 500 most popular Web sites.
78 CHAPTER 2 Describing Distributions of Data

2.41 Tall or short, I Mr. Walker measures the heights (in inches) of the students in
one of his classes. He uses a computer to calculate the following numerical summaries:
Mean Std. dev. Min. Q1 Med. Q3 Max.
69.188 3.20 61.5 67.75 69.5 71 74.5
Next, Mr. Walker has his entire class stand on their chairs, which are 18 inches off the
ground. Then he measures the distance from the top of each student’s head to the floor.
(a) Find the mean and median of these measurements. Show your work.
(b) Find the standard deviation and IQR of these measurements. Show your work.
2.42 Tall or short, II Refer to the previous exercise. Mr. Walker converts his
students’ original heights from inches to feet.
(a) Find the mean and median of the students’ heights in feet. Show your work.
(b) Find the standard deviation and IQR of the students’ heights in feet. Show your work.
2.43 Cool car colors Table 2.8 gives information about the most popular
colors of vehicles purchased in 2007. Make a bar graph that compares the color
distributions for full size/intermediate cars and SUVs/trucks. Describe any
similarities and differences between the two distributions.
Table 2.8 Most Popular Vehicle Colors in 2007

Color Full size/intermediate cars SUVs/trucks

Silver 21% 14%

Blue 14% 11%
White 14% 26%
Black 13% 16%
Red 13% 13%
Gray 12% 12%
Beige/brown 7% 4%

Source: 2009 World Almanac and Book of Facts.

2.44 Tuna from where? Does mercury content in canned tuna differ by country
of origin? The Fathom boxplots and summary chart that follow provide information
to help you answer this question. Write a few sentences comparing the distributions
of mercury concentration based on country of origin.
 2. 2 Describing Distributions with Numbers 79

Country Count x s Min. Q1 M Q3 Max.

Costa Rica 23 0.281 0.243 0.079 0.16 0.23 0.32 1.3
Ecuador 18 0.754 0.367 0.3 0.45 0.68 0.98 1.5
Malaysia 2 0.33 0.057 0.29 0.29 0.33 0.37 0.37
Mexico 33 0.31 0.285 0.064 0.14 0.18 0.38 1.4
Phillipines 3 0.031 0.006 0.025 0.025 0.034 0.035 0.035
Thailand 50 0.126 0.148 0.012 0.043 0.065 0.16 0.73
United States 35 0.268 0.242 0.023 0.052 0.2 0.4 0.99

ws
Ne
ws
Ne

APPLICATION 2.2 B]
RO
g
<
Se
a

Did Mr. Starnes stack his class?

Mr. Starnes teaches AP Statistics, but he also does To help resolve the dispute, the teachers
the class scheduling for the high school. There collected data on the cumulative grade point
are two AP Statistics classes—one taught by averages and SAT Math scores of their students.
Mr. Starnes and one taught by Ms. McGrail. The Mr. Starnes provided the GPA data from his
two teachers give the same first test to their computer. The students reported their SAT Math
classes and grade the test together. Mr. Starnes’s scores. The table below shows the data for
students earned an average score that was 8 points each student in the two classes. Note that the
higher than the average for Ms. McGrail’s class. two data values in each row come from a single
Ms. McGrail wonders whether Mr. Starnes might student.
have “adjusted” the class rosters from the computer Did Mr. Starnes stack his class? Give appropri-
scheduling program. In other words, she thinks ate graphical and numerical evidence to support
he might have “stacked” his class with better your conclusion.
students. He denies this, of course.
Starnes GPA Starnes SAT-M McGrail GPA McGrail SAT-M
2.9 670 2.9 620
2.86 520 3.3 590
2.6 570 3.98 650
3.6 710 2.9 600
3.2 600 3.2 620
2.7 590 3.5 680
3.1 640 2.8 500
3.085 570 2.9 502.5
3.75 710 3.95 640
3.4 630 3.1 630
3.338 630 2.85 580
3.56 670 2.9 590
3.8 650 3.245 600
3.2 660 3.0 600
3.1 510 3.0 620
2.8 580
2.9 600
3.2 600
80 CHAPTER 2 Describing Distributions of Data

Section 2.2 Summary

To describe a set of data, always start with graphs. Then add well-chosen num-
bers that summarize specific aspects of the data. If we have data on a single
quantitative variable, we start with a dotplot, stemplot, or histogram to display
the distribution. Then we add numbers to describe the center and spread of the
distribution.
There are two common descriptions of center and spread: the five-number
summary and the mean and standard deviation. The five-number summary
consists of the median to measure center and the interquartile range (IQR) to
describe spread, along with the minimum and maximum. We can use a boxplot to
display the five-number summary in graphical form. The median is the midpoint
of the observations. The mean is the average of the observations. The standard
deviation measures spread as a kind of average distance from the mean, so use it
only with the mean.
The mean and standard deviation are not resistant—they can be changed a lot
by a few outliers. The mean and median agree for symmetric distributions, but the
mean moves farther toward the long tail of a skewed distribution. In general, use
the five-number summary to describe most distributions and the mean and stan-
dard deviation only for roughly symmetric distributions.

Section 2.2 2.45 Electoral votes, I To become president of the United States, a candidate
Exercises does not have to receive a majority of the popular vote. The candidate does, how-
ever, have to win a majority of the 538 electoral votes that are cast in the Electoral
College. Figure 2.19 is a stemplot of the number of electoral votes for each of the
50 states and the District of Columbia.
(a) Find the five-number summary for this distribution.
(b) Are there any outliers? Justify your answer with a calculation.
0 3333333344444 (c) Make a boxplot for these data.
0 55555666777788 9 9 9 2.46 Electoral votes, II Refer to the previous exercise. Which measure of center
1 0000111123
and spread would you use to summarize the distribution in Figure 2.19: (i) the
mean and standard deviation or (ii) the median and IQR? Justify your answer.
1 5557
2.47 Which is easier—AP Calculus AB or AP Statistics? The table below gives
2 011 the distribution of grades earned by students taking the AP Calculus AB and AP
2 7 Statistics exams in 2008.15
3 14
3 Grade
4 No. of exams 5 4 3 2 1
4 AP Calculus AB 222,835 22.1% 21.2% 17.9% 15.2% 23.7%
5 AP Statistics 108,284 12.9% 22.7% 23.7% 18.8% 21.8%
5 5

Figure 2.19 Stemplot (a) Make an appropriate graphical display to compare the grade distributions for
of the number of electoral AP Calculus AB and AP Statistics.
votes for the 50 states and (b) Write a few sentences comparing the two distributions of exam grades. Can
the District of Columbia. we tell which exam is easier? Explain why or why not.
 2. 2 Describing Distributions with Numbers 81

2.48 Mean or median? You are planning a party and

16.0
want to know how many cans of soda to buy. A genie
offers to tell you either the mean number of cans guests
will drink or the median number of cans. Which mea-
sure of center should you ask for? Why? To make your
Poverty rate (%)

12.0 answer concrete, suppose that there will be 30 guests

∗
and the genie will tell you either x  5 cans or M  3
cans. How many cans should you have on hand?
8.0 2.49 Poverty in the eastern states The poverty rates
for states east of the Mississippi have been divided
into northern and southern states, according to the
geographic divisions used by the Census Bureau.
4.0 Figure 2.20 shows boxplots of the poverty rates for the
states in each region. Write a few sentences compar-
South North
ing the distributions.
Region
2.50 Phosphate levels The level of various substances
in the blood influences our health. Here are measure-
Figure 2.20 Boxplots comparing the poverty rates ments of the level of phosphate in the blood of a patient,
of southern and northern states east of the Mississippi in milligrams of phosphate per deciliter of blood, made
in the same year. on six consecutive visits to a clinic:
5.6 5.2 4.6 4.9 5.7 6.4
(a) Find the mean from its definition. Show your work.
(b) Find the standard deviation from its definition.
Show your work.
(c) Now enter the data into your calculator and use it
to obtain x and s. Do the results agree with your hand
calculations?
2.51 Teacher raises, I A school system employs
teachers at salaries between $18,000 and $40,000. The
teachers’ union and the school board are negotiating the
form of next year’s increase in the salary scale.
(a) If every teacher is given a $1000 raise, what will this
do to the mean salary? To the median salary? Explain
your answers.
(b) What would an across-the-board $1000 raise do to
the extremes and quartiles of the salary distribution? To
the standard deviation of teachers’ salaries? Explain your
answers.
2.52 Teacher raises, II Refer to the previous exercise.
If each teacher receives a 5% raise instead of a $1000
raise, the amount of the raise will vary from $900 to
“Yup, Old Bob drowned due to being lgnorant of $2000, depending on the present salary.
statistics. He thought it was enough to know the (a) What will this do to the mean salary? To the median
average depth of the river.” salary? Explain your answers.
(b) Will a 5% raise increase the IQR? Will it increase the
standard deviation? Explain your answers.
82 CHAPTER 2 Describing Distributions of Data

2.3 Use and Misuse of Statistics

ACTIVITY 2.3
Who buys iMacs?
After Apple Computer introduced the iMac, the 2. Two possible bar graphs of the data are shown
company wanted to know whether this new computer below. Which one could be considered deceptive?
was expanding Apple’s market share. Was the iMac Why?
mainly being bought by previous Macintosh owners,
or was it being purchased by first-time computer 80
buyers and by previous PC users who were switching 70
over? To find out, Apple hired a firm to conduct a 60
50

Percent
survey of 500 iMac customers. Each customer was
categorized as a new computer purchaser, a previous 40
PC owner, or a previous Macintosh owner. The table 30
below summarizes the survey results.16 20
10
Previous ownership Count Percent
0
None 85 17.0 None PC Macintosh
PC 60 12.0 Previous computer
Macintosh 355 71.0
Total 500 100.0 80
Discuss each of the following questions with 70
two or three classmates. Be prepared to present your 60
Percent

answers to the class. 50

1. Here’s a clever graph of the data that uses pictures 40
instead of the more traditional bars. How is this 30
graph misleading?
20
400 10
None PC Macintosh
Numbers of buyers

300 Previous computer

200 3. A clever marketing consultant wants to advertise

the results of the survey as: “Over 40% of new iMac
100 owners are former PC users.” The consultant got this
figure as follows: 60/145  0.414, or 41.4%. Explain
how this value distorts the truth.
None Windows Macintosh
Previous computer

In his Chapters from My Autobiography, Mark Twain wrote: “There are three kinds
of lies: lies, damned lies, and statistics.” Several popular books have been written
that build on this theme. There’s Darryl Huff ’s classic How to Lie with Statistics
and Joel Best’s Damned Lies and Statistics. Sometimes the “word on the street”
about statistics focuses mainly on how people can make data say anything they
want. Activity 2.3 gives you a few illustrations of this kind of deceptive statistics.
 2. 3 Use and Misuse of Statistics 83

In this section, we’ll introduce you to some of the most common ways that
graphs and numbers can be used to deceive people. We’ll also show you how to use
graphs and numbers to display data honestly and effectively.

Change over time: line graphs

Many quantitative variables are measured at intervals over time. We might, for ex-
ample, measure the height of a growing child or the price of a stock at the end of
each month. In these examples, our main interest is change over time. To display
change over time, make a line graph.

Line graph

A line graph of a variable plots each observation against the time at

which it was measured. Always put time on the horizontal scale of
your plot and the variable you are measuring on the vertical scale.
Connect the data points by lines to display the change over time.

Example 2.18 The price of gasoline

How has the price of gasoline at the pump changed over time? Figure 2.21 is a line graph
of the average price of regular unleaded gasoline each month from January 2000 to January
2009. For reference, the point for January 2000 is at 128.9 on the vertical scale.17

450
400
Price per gallon (cents)

350
300
250
200
150
100
50
0
Jan-00

Jan-01

Jan-02

Jan-03

Jan-04

Jan-05

Jan-06

Jan-07

Jan-08

Jan-09

Year

Figure 2.21 A line graph of the average cost of regular unleaded gasoline each month
between January 2000 and January 2009.

It would be difficult to see patterns in a long table of monthly prices. Figure 2.21
makes the patterns clearer. What should we look for?
First, look for an overall pattern. For example, a trend is a long-term upward
or downward movement over time. There was an overall upward trend in gasoline
prices over this time period. This upward trend began as OPEC countries drove the
price up by reducing their oil production. A major strike in Venezuela, the U.S. war
84 CHAPTER 2 Describing Distributions of Data

with Iraq, and unrest in Nigeria are just a few of the events that have contributed
to the upward trend.
Next, look for striking deviations from the overall pattern. The 2001 price drop
was due to weakening global demand. In fall 2008, a serious financial crisis in the
United States led to decreased demand and lower prices.
Change over time often has a regular pattern of seasonal variation that repeats
each year. Gasoline prices are usually highest during the summer driving season
and lowest in the winter, when there is less demand for gasoline. You can see this
up-in-summer then down-in-the-fall pattern for many years in Figure 2.21.

Watch those scales!

Because graphs speak so strongly, they can mislead the unwary. The careful reader
of a line graph looks closely at the scales marked off on the axes.

Example 2.19 Living together

The number of unmarried couples living together has increased in recent years, to the
point that some people say that cohabitation is delaying or even replacing marriage.
Figure 2.22 presents two line graphs of the number of unmarried-couple households
in the United States. The data once again come from the Current Population Survey.
The graph on the left suggests a steady but moderate increase. The right-hand graph
says that cohabitation is rapidly increasing.
7000

8000 6000
Unmarried couples
Unmarried couples

5000
(thousands)

6000
(thousands)

4000 4000

3000
2000
2000
0
1975 1980 1985 1990 1995 2000 2005 2010
1000
Year
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008

Year
Figure 2.22 The effect of changing the scales in a line graph. Both graphs plot the
same data, but the right-hand graph makes the increase appear much more rapid.

The secret is in the scales. You can transform the left-hand graph into the right-
hand graph by stretching the vertical scale, squeezing the horizontal scale, and cut-
ting off the vertical scale just above and below the values to be plotted. Now you
know how to either exaggerate or play down a trend in a line graph.
Which graph is correct? Both are accurate graphs of the data, but both have
scales chosen to create a specific effect. Because there is no one “right” scale for a
line graph, correct graphs can give different impressions by their choices of scale.
Watch those scales!
 2. 3 Use and Misuse of Statistics 85

Did you notice the sudden increase in the number of unmarried couples living to-
gether from 2006 to 2007? In previous years, the Current Population Survey only counted
couples that included the head of household. Starting in 2007, the CPS changed its sur-
vey question to “Do you have a boyfriend/girlfriend or partner in the household?” They
found about 1.1 million more unmarried couples with this direct question.

Beware the pictogram

Bar graphs compare several quantities by comparing the heights of bars that rep-
resent the quantities. Our eyes, however, react to the area of the bars as well as to
their height. When all bars have the same width, the area (width 3 height) varies in
proportion to the height and our eyes receive the right impression. When you draw
a bar graph, make the bars equally wide. Artistically speaking, bar graphs are a bit
dull. It is tempting to replace the bars with pictures for greater eye appeal.

Example 2.20 Favorite pets: a misleading graph

All 900 students in an elementary school responded to the following survey ques-
tion: “What is your favorite animal to have as a pet?” Most students said “dog,” “cat,”
or “horse,” but a few named another animal, like a guinea pig or a rabbit. Rather
than use a bar graph to display the data, one of the students suggested using the
pictogram shown in Figure 2.23. Do you see how this graph is misleading?

Key:

50 people 50 people 50 people 50 people

Dog

Cat

Horse

Other

Figure 2.23 Pictogram showing the favorite pets of students in an elementary school.
86 CHAPTER 2 Describing Distributions of Data

The pictogram makes it look like horses are the most popular pets. In fact,
only 150 students chose “horse” as their favorite pet. But the pictures of the horses
in Figure 2.23 are so large that even three of them extend past the images for all
the other pet categories. To make a pictogram that isn’t deceptive, we need to use
pictures of the animals that are all the same size. Alternatively, we could just make
a bar graph, like the one shown in Figure 2.24. Now we see that cats were actually
the most popular pet for students at this elementary school.

400
350
300
250
Count
200
150
100
50
0
Dog Cat Horse Other
Favorite pet

Figure 2.24 Bar graph that displays the data in Figure 2.23 more accurately.

Making good graphs

Graphs are the most effective way to communicate using data. A good graph fre-
quently reveals facts about the data that would be difficult or impossible to detect
from a table. In addition, the immediate visual impression of a graph is much stron-
ger than the impression made by data in numerical form. Here are some principles
for making good graphs:
• Make sure labels and legends tell what variables are plotted, what their
units are, and the source of the data.
• Make the data stand out. Be sure that the actual data, not labels, grids, or
background art, catch the viewer’s attention. You are drawing a graph, not a
piece of creative art.
• Pay attention to what the eye sees. Avoid pictograms and be careful
choosing scales. Avoid fancy “three-dimensional” effects that confuse the
eye without adding information. Ask if a simple change in a graph would
make the message clearer.

Exercises 2.53 I’d die without my phone! In a July 2008 survey of over 2000 U.S. teenagers by
Harris Interactive, 47% said that “their social life would end or be worsened without
their cell phone.”18 One survey question asked the teens how important it is for their
phone to have certain features. Figure 2.25 displays data on the percent who indicated
that a particular feature is vital. Explain how the graph gives a misleading impression.
 2. 3 Use and Misuse of Statistics 87

60
50 46%
Figure 2.25 A graph 40 39%
showing the percent of 30
teens who said that a 19% 17%
20
particular feature is es-
10
sential to have on their
cell phones in a July 2008 0 Make/receive Send/receive Camera Send/receive
survey. calls text messages pictures

2.54 The cost of fresh oranges Figure 2.26 is a line graph of the average cost of
fresh oranges each month from January 2000 to December 2008. These data, from
the Bureau of Labor Statistics monthly survey of retail prices, are “index numbers”
rather than prices in dollars and cents. That is, they give each month’s price as a
percent of the price in a base period (in this case, the years 1982 to 1984). So 250
means “250% of the base price.”

500
Orange price index

400

300

200
01/00 01/01 01/02 01/03 01/04 01/05 01/06 01/07 01/08
Date

Figure 2.26 The price (index numbers; base period 1982 to 1984) of fresh oranges,
January 2000 to December 2008.

(a) The graph shows strong seasonal variation. How is this visible in the graph?
Why would you expect the price of fresh oranges to show seasonal variation?
(b) What is the overall trend in orange prices during this period, after we take
account of the seasonal variation?
2.55 Support the court? In 2005, CNN reported the results of a survey about
a Florida court’s decision to remove the feeding tube from coma patient Terry
Schiavo, effectively ending her life. Figure 2.27 (on the next page) shows a bar
graph of the data that CNN initially posted on its Web site.
(a) What visual impression does the graph give about support for the court’s decision?
(b) Make your own graph that displays the data in a less misleading way.
(Note: When notified about the misleading nature of their graph, CNN posted a
corrected version.)
88 CHAPTER 2 Describing Distributions of Data

Figure 2.27 Bar graph that displays the results of a survey about Terry Schiavo and
was posted on the CNN Web site.

2.56 Lottery ticket sales, I Figure 2.28 is a line graph of the total lottery ticket
sales (in millions of dollars) in the United States from 1980 to 2007. What’s wrong
with this picture?

60,000
Total lottery ticket sales ($ millions)

50,000

40,000

30,000

20,000

10,000

0
1980 1990 1995 2000 2001 2002 2003 2004 2005 2006 2007
Year

Figure 2.28 Line graph of lottery ticket sales in the United States, in millions of dollars,
from 1980 to 2007.

2.57 Lottery ticket sales, II Refer to the previous exercise. Draw a new line graph
of the data displayed in Figure 2.28 that isn’t deceptive. Describe what you see.
2.58 Getting to school Students in a high school statistics class were given data
about the primary method of transportation to school for a group of 30 students.
They produced the pictograph shown on the facing page.
(a) How is this graph misleading?
(b) Make a graph of the data that is not misleading.
 2. 3 Use and Misuse of Statistics 89

Key:
Cycle

Mode of transportation
= 1 Cycler
= 2 Cars
Car = 7 Bus takers
= 2 Walkers
Bus

Walk

Do the numbers make sense?

According to the July 21 to 28, 2008, edition of the Nation, “Roughly one-third of
all eligible Americans, 64 million people, are not registered to vote. This percentage
is even higher for African-Americans (30 percent) and Hispanics (40 percent).” Did
you notice the error in that quote? If not, read it again now.
Some people lack the skills needed to use numbers carefully. Others use data to
argue a cause and care more for the cause than for the accuracy of the data. Some
people feed us fake data. We know that we should always ask:
• How were the data produced?
• What exactly was measured?
To help develop number sense, we will look at how bad data, or good data used
wrongly, can trick the unwary.
What didn’t they tell us? The most common way to mislead with data is to cite
correct numbers that don’t tell the whole story. The numbers are not made up, so
the fact that the information is a bit incomplete may be an innocent oversight. Or
maybe not.

Example 2.21 Snow! Snow! Snow!

Crested Butte attracts skiers by advertising that it has the highest average snowfall
of any ski town in Colorado. That’s true. But skiers want snow on the ski slopes, not
in the town—and many other Colorado resorts get more snow on the slopes.19

Example 2.22 We attract really good students

Colleges know that many prospective students look at popular guidebooks to decide
where to apply for admission. The guidebooks print information supplied by the col-
leges themselves. Surely no college would simply lie about, say, the average SAT score
of its entering students. But they do want their scores to look good. How about leav-
ing out the scores of international and remedial students? Northeastern University
did this, making the average SAT score of its freshman class 50 points higher than if
all students were included.
90 CHAPTER 2 Describing Distributions of Data

If the school admits economically disadvantaged students

under a special program sponsored by the state, surely no one
will complain if the school leaves their SAT scores out of the
average. New York University did this.20

The point of these examples is that numbers have a con-

text. If you don’t know the context, the lonely, isolated, naked
number doesn’t tell you much.
Are the numbers plausible? You can often detect suspicious
numbers simply because they don’t seem plausible. Sometimes
you can check an implausible number against data in reliable
‘‘Sure your patients have 50% fewer cavities.
That’s because they have 50% fewer teeth!” sources like the Statistical Abstract of the United States. Some-
©Schochet/Artizans.com times, as the next example illustrates, you can do a calculation
to show that a number can’t be right.

Example 2.23 The abundant melon field

The very respectable journal Science, in an article on insects that attack plants, men-
tioned a California field that produces 750,000 melons per acre. A reader responded,
“I learned as a farm boy that an acre covers 43,560 square feet, so this remarkable
field produces about 17 melons per square foot. If these are cantaloupes, with each
fruit covering about 1 square foot, I guess they must grow in a stack 17 deep.’’ Here
is the calculation the reader did:
melons per acre 750,000
melons per square foot 5 5 5 17.2
square feet per acre 43,560
The editor, a bit embarrassed, replied that the correct figure was about 11,000 melons
per acre.21

Oh, those percents! Rates and percents seem to cause particular trouble, as the
following example illustrates.

Example 2.24 Something doesn’t add up!

The Canberra Times reported, “Of those aged more than 60 living alone, 34% are women
and only 15% are men.” That’s 49% of those living alone. What about the other 51%?

Calculating the percent increase or decrease in some quantity seems particu-

larly prone to mistakes. The percent change in a quantity is found by
amount of change
percent change 5 3 100
starting value
 2. 3 Use and Misuse of Statistics 91

Example 2.25 Stocks go up, stocks go down

In 2008, the NASDAQ composite index of stock prices dropped from 2653.91 to
1577.03. What percent decrease was this?
amount of change
percent change 5 3 100
starting value
1577.03 2 2653.91
5 3 100
2653.91
21076.88
5 3 100 5 20.406 3 100 5 240.6%
2653.91
That’s a big drop! Remember to always use the starting value, not the smaller value,
in the denominator of your fraction.

A quantity can increase by any amount—a 100% increase just means it has
doubled. But nothing can go down more than 100%—it has then lost 100% of its
value, and 100% is all there is.

ws
Ne
ws
Ne

APPLICATION 2.3 B]
RO
g
<
Se
a

You can’t fool me!

In this section, you have seen examples of proper % of Traffic
and improper use of statistics. This Application
gives you a chance to show what you have learned. 0.36
0.34
0.32
QUESTIONS
0.30
1. The two graphs that follow display information 0.28
about the percent of a Web site’s traffic that 0.26
comes from each of three search engines: Google,
Yahoo, and MSN.22 Explain how each graph is Googl e
Yahoo
misleading. Make a graph of these data that isn’t
M SN
misleading.
% of Traffic 2. Continental Airlines once advertised that it had
“decreased lost baggage by 100% in the past six
months.” Do you believe this claim?
Yah oo MSN
3. Here’s a quotation from a book review in a
scientific journal:

Goo g l e
. . . a set of 20 studies with 57 percent reporting
significant results, of which 42 percent agree on
one conclusion while the remaining 15 percent
favor another conclusion, often the opposite one.23
APPLICATION 2.3 (continued)
92 CHAPTER 2 Describing Distributions of Data

APPLICATION 2.3 (continued)

Do the numbers given in this quotation make Check the newspaper’s arithmetic. What percent
sense? Explain. of scheduled flights from O’Hare were actually
4. The pointy-haired boss in the Dilbert cartoon canceled?
once noted that 40% of all sick days taken by the staff 6. A report on the problem of vacation cruise ships
are Fridays and Mondays. Is this evidence that the polluting the sea by dumping garbage overboard
staff are trying to take long weekends? said:
5. An article in a midwestern newspaper about flight On a seven-day cruise, a medium-size ship
delays at major airports said:
(about 1,000 passengers) might accumulate
According to a Gannett News Service study of U.S. 222,000 coffee cups, 72,000 soda cans, 40,000
airlines’ performance during the past five months, beer cans and bottles, and 11,000 wine bottles.25
Chicago’s O’Hare Field scheduled 114,370 flights. Are these numbers plausible? Do some arithmetic
Nearly 10 percent, 1,136, were canceled.24 to back up your conclusion.

Section 2.3 Summary

To show how a quantitative variable changes over time, use a line graph that plots
values of the variable (vertical scale) against time (horizontal scale). Look for an
upward or downward trend and for any seasonal variation. Graphs can mislead
the eye. Look at the scales of a line graph to see if they have been stretched or
squeezed to create a particular impression. Avoid pictograms, which replace the
bars of a bar graph with pictures whose height and width both change. Avoid clut-
ter that makes the data hard to see.
The aim of statistics is to provide insight by means of numbers. Numbers
are most likely to yield their insights to those who examine them closely. Ask
exactly what a number measures. Look for the context of the numbers and ask
if there is important missing information. Look for inconsistencies, numbers
that don’t agree as they should, and check for incorrect arithmetic. Compare
numbers that are implausible—surprisingly large or small—with numbers you
know are right.

Section 2.3 2.59 The rise in college education The line graph on the facing page shows
Exercises the rise in the percent of women 25 years old and over who have at least a
bachelor’s degree. Identify at least two features of this graph that make it
confusing or misleading.
2.60 Greeks on campus The question-and-answer column of a college campus
newspaper was asked what percent of the campus was “Greek” (that is, members
of fraternities or sororities). The answer given was that “the figures for the fall
semester are approximately 13 percent for the girls and 15–18 percent for the
guys, which produces a ‘Greek’ figure of approximately 28–31 percent of the
undergraduates.”26 Discuss the campus newspaper’s arithmetic.
 2. 3 Use and Misuse of Statistics 93

More women have

college degrees
26.7%
23.6%

18.4%

12.8%
8.1%
5.8%

1960 1970 1980 1990 2000 2007

2.61 Oatmeal and cholesterol Does eating Quaker Oatmeal reduce cholesterol?
An advertisement included the following graph as evidence that the answer
is “Yes.”

Representative cholesterol point drop

210
208
206
Cholesterol

204
202
200
198
196
Week1 Week2 Week3 Week4

(a) How is this graph misleading?

(b) Make a new graph that isn’t misleading. What do you conclude about the
effect of eating Quaker Oats on cholesterol reduction?
2.62 Deer in the suburbs Westchester County is a suburban area covering 438
square miles immediately north of New York City. A garden magazine claimed
that the county is home to 800,000 deer.27 Do a calculation that shows this claim
to be implausible.
2.63 We can read, but can we count? The Census Bureau once gave a simple
test of literacy in English to a random sample of 3400 people. The New York Times
printed some of the questions under the headline “113% of Adults in U.S. Failed
This Test.”28 Why is the percent in the headline clearly wrong?
2.64 Online card games Figure 2.29 (on the next page) shows the number
of people playing card games at the Yahoo Web site on a Sunday and on a
Wednesday in the same week.
94 CHAPTER 2 Describing Distributions of Data

8000

7000 Wednesday
Sunday
6000

5000

4000

3000

2000

1000

Gin
Bridge
Poker

Blackjack

Cribbage

Hearts

Canasta

Pinochle

Euchre

Spades
Figure 2.29 A graph showing the number of people playing various card games at the
Yahoo Web site on Sunday and Wednesday of the same week.

(a) Explain what’s wrong with this graph.

(b) Make a more appropriate graph to display these data.
2.65 Poverty The number of Americans living below the official poverty line
increased from 32,476,000 to 37,276,000 in the 10 years between 1998 and 2007.
What percent increase was this? You should not conclude from this that poverty
grew more common in these years, however. Why not?
www
2.66 CHANCE News wiki The CHANCE Web site at Dartmouth College
contains lots of interesting stuff (at least if you are interested in statistics). In
particular, the CHANCE News wiki at http://chance.dartmouth.edu/chancewiki
features articles about published statistical results in the media, including decep-
tive statistics. Go to the site and find an example of one of the following: leaving
out essential information, implausible numbers, or faulty arithmetic. Try looking
in the “Forsooth” column of a recent issue.

CHAPTER 2 REVIEW
Data analysis is the art of describing data using graphs and numerical summaries.
The purpose of data analysis is to help us see and understand the most important
features of a set of data. Section 2.1 focused on graphical displays: pie charts, bar
graphs, dotplots, stemplots, and histograms. Section 2.2 showed how to describe
shape, center, and spread for distributions of quantitative variables. Section 2.3 ex-
amined proper and improper uses of graphs (line graphs, pictograms) and numeri-
cal summaries. Figure 2.30 organizes the big ideas. We plot our data, and then we
describe their center and spread using either the mean and standard deviation or
the median and IQR.
 Chapter 2 Review 95

Plot your data

Dotplot, Stemplot, Histogram

Interpret what you see

Shape, Center, Spread, Outliers

Numerical summary
x and s, Five-Number Summary

Figure 2.30 To describe the distribution of a quantitative variable, plot, interpret, and
summarize.

Here is a review list of the most important skills you should have developed
from your study of this chapter.
A. DISPLAYING DISTRIBUTIONS
1. Recognize when a pie chart can and cannot be used.
2. Make a bar graph of the distribution of a categorical variable or, in general, to
compare related quantities.
3. Interpret pie charts and bar graphs.
4. Make a line graph of a quantitative variable over time. Recognize patterns such
as trends and seasonal variation in line graphs.
5. Be aware of graphical abuses, especially in line graphs and pictograms.
6. Make a dotplot or stemplot of the distribution of a small set of observations.
7. Make a histogram of the distribution of a quantitative variable.

B. DESCRIBING DISTRIBUTIONS (QUANTITATIVE VARIABLE)

1. Look for the overall pattern of a dotplot, stemplot, or histogram and for major
deviations from the pattern.
2. Assess from a dotplot, stemplot, or histogram whether the shape of a distri-
bution is roughly symmetric, distinctly skewed, or neither. Assess whether the
distribution has one or more major peaks.
3. Describe the overall pattern by giving numerical measures of center and spread
in addition to a verbal description of shape.
4. Decide which measures of center and spread are more appropriate: the mean
and standard deviation (especially for symmetric distributions) or the five-number
summary (especially for skewed distributions).
5. Identify outliers and give plausible explanations for them.
6. Compare distributions of categorical or quantitative variables.

C. NUMERICAL SUMMARIES OF DISTRIBUTIONS

1. Find the median M and the quartiles Q1 and Q3 for a set of observations.
2. Give the five-number summary and draw a boxplot; use boxplots to compare
distributions.
96 CHAPTER 2 Describing Distributions of Data

3. Find the mean and (using a calculator) the standard deviation s for a small set
of observations.
4. Understand that the median is less affected by extreme observations than the
mean. Recognize that skewness in a distribution moves the mean away from the
median toward the long tail.
5. Know the basic properties of the standard deviation: s  0 only when all obser-
vations are identical; s increases as the spread increases; s has the same units as the
original measurements; s is pulled strongly up by outliers or skewness.

CHAPTER 2 REVIEW EXERCISES

2.67 Longevity of presidents Table 2.9 shows the ages at death of U.S. presidents.

Table 2.9 Age at Death of U.S. Presidents

Washington 67 Fillmore 74 T. Roosevelt 60 J. Adams 90

Pierce 64 Taft 72 Jefferson 83 Buchanan 77
Wilson 67 Madison 85 Lincoln 56 Harding 57
Monroe 73 A. Johnson 66 Coolidge 60 J. Q. Adams 80
Grant 63 Hoover 90 Jackson 78 Hayes 70
F. Roosevelt 63 Van Buren 79 Garfield 49 Truman 88
Harrison 68 Arthur 56 Eisenhower 78 Tyler 71
Cleveland 71 Kennedy 46 Polk 53 Harrison 67
L. Johnson 64 Taylor 65 McKinley 58 Nixon 81
Ford 93 Reagan 93

(a) Make a stemplot of these data. Did you decide to split the stems?
(b) Now make a histogram. Describe the shape, center, and spread of the distribution.
Are there any outliers?
(c) Which plot is better at displaying this distribution: a stemplot or a histogram?
Why?
2.68 Be creative! Make up a list of numbers of which only 10% are above the aver-
age (that is, above the mean). What percent of the numbers in your list fall above the
median?
2.69 Smoking-related deaths Below is a table from Smoking and Health Now, a
report of the British Royal College of Physicians. It shows the number and percent of
deaths among men aged 35 and over from the chief diseases related to smoking. One
of the entries in the table is incorrect and an erratum slip was inserted to correct it.
Which entry is wrong, and what is the correct value?

Lung cancer Chronic bronchitis Coronary heart disease All causes

Number 26,973 24,976 85,892 312,537
Percent 8.6% 8.0% 2.75% 100%
 Chapter 2 Review Exercises 97

2.70 Who sells cars? Figure 2.31 is a pie chart of the percent of passenger car sales in
a given year by various manufacturers.
(a) The artist has tried to make the graph more interesting by using the wheel of a car
for the “pie.” Is the graph still a correct display of the data? Explain your answer.
(b) Make a bar graph of the data. What advantage does your new graph have over the
pie chart in Figure 2.31?

Ford
19.4%
General Chrysler
Motors 8.9%
32.5%
Others
8.2%
Japanese
manufacturers
31.0%

Figure 2.31 Passenger car sales by several manufacturers in the same year.

2.71 We pay high interest Figure 2.32 shows a graph taken from an advertisement
for an investment that promises to pay a higher interest rate than bank accounts and
other competing investments. Is this graph a correct comparison of the four interest
rates? Explain your answer.

6%

4%

2%

0% 3.5% 4.4% 5.1% 5.9%

Figure 2.32 Pictogram comparing interest rates.

2.72 State SAT scores Figure 2.33 (on the next page) is a histogram of the average
scores on the mathematics part of the SAT exam for students in the 50 states and the
District of Columbia.29 The distinctive overall shape of this distribution implies that a
single measure of center such as the mean or the median is of little value in describing
the distribution. Explain why.
98 CHAPTER 2 Describing Distributions of Data

14

12

Number of states
10

0
460 480 500 520 540 560 580 600 620
State average SAT mathematics score
Figure 2.33 Histogram of the average scores on the SAT Math exam for students in
the 50 states and the District of Columbia.

2.73 Are you wired? In late 2006, the Pew Internet and American Life Project con-
ducted a telephone survey of 935 adults and their children aged 12 to 17. One ques-
tion they asked was “Do you, personally, happen to have . . . a desktop computer? A
cell phone? An iPod or other MP3 player? A laptop computer? A PDA like a Palm
Pilot or Blackberry?” The table below summarizes the responses.

Desktop Cell phone iPod/MP3 Laptop PDA

Parent and teen both own one 64% 60% 22% 18% 1%
Parent owns one, teen doesn’t 19% 29% 7% 19% 12%
Teen owns one, parent doesn’t 8% 3% 29% 7% 7%
Neither parent nor teen owns one 9% 8% 42% 56% 80%

Make a bar graph comparing teens’ and parents’ ownership of these devices.
2.74 Home run king In 1927, Babe Ruth broke Major League Baseball’s single-
season home run record by hitting 60 home runs. The Babe’s record stood until 1961,
when Roger Maris hit 61 homers in a season. Almost 40 years later, Mark McGwire
(70) and Sammy Sosa (68) excited baseball fans by smashing Maris’s record in the
same season. Barry Bonds captured the record in 2001 by hitting 73 home runs in a
season. Below are data on the number of home runs hit by Bonds and McGwire each
season during the prime of their careers. Who is the better home run hitter? Make
comparative boxplots and provide numerical evidence to support your answer.

Bonds
16 25 24 19 33 25 34 46 37 33 42 40 37 34 49 73

McGwire
49 32 33 39 22 42 39 52 58 70
 Chapter 2 Review Exercises 99

2.75 Getting more sleep An experiment was carried out with 10 patients to inves-
tigate the effectiveness of a drug that was designed to increase sleep time. The data
below show the number of additional hours of sleep gained by each subject after tak-
ing the drug.30 (A negative value indicates that the subject got less sleep after taking
the drug.) Do these data provide sufficient evidence to conclude that the drug was
effective? Follow the four-step problem-solving process from Chapter 1 in answering
this question.

1.9 0.8 1.1 0.1 0.1 4.4 5.5 1.6 4.6 3.4

2.76 Sleep: hours or minutes? Refer to the previous exercise. Suppose the sleep
increase data are converted from hours to minutes. How will this affect the mean,
median, standard deviation, and IQR? Explain.
This page intentionally left blank
 CHAPTER 3

Modeling
Distributions
of Data

AP Photo/Joseph Kaczmarek
3.1 Measuring Location in a Distribution
3.2 Normal Distributions

Do you sudoku?
The sudoku craze has officially swept the globe. Here’s what Will Shortz,
crossword puzzle editor for the New York Times, said about sudoku in August 2005:
As humans we seem to have an innate desire to fill up empty spaces. This might
explain part of the appeal of sudoku, the new international craze, with its empty
squares to be filled with digits. Since April, when sudoku was introduced to the
United States in the New York Post, more than half the leading American
newspapers have begun printing one or more sudoku a day. No puzzle has had
such a fast introduction in newspapers since the crossword craze of 1924–25.1
Since then, millions of people have made sudoku part of their daily routines.
One of the authors played an online game of sudoku at www.websudoku.com.
www
The graph below provides information about how well he did. (His time is marked with
an arrow.)

101
102 CHAPTER 3 Modeling Distributions of Data

In this chapter, you’ll learn more about how to describe the location
of an individual observation—like the author’s sudoku time—within a
distribution.

3.1 Measuring Location in a Distribution

ACTIVITY 3.1
Where do I stand?
Materials: Masking tape to mark a number line 5. What percent of the students in the class have
scale; rope (if desired) to make human boxplot heights equal to or less than yours? This is your
In this Activity, you and your classmates will percentile in the distribution of heights.
explore ways to describe where you stand (literally!) 6. Work with a partner to calculate the mean and
within a distribution. standard deviation of the class’s height distribution
1. Your teacher will mark out a number line on the from the dotplot.
floor with a scale running from about 58 to 78 inches. 7. Where does your height fall relative to the
2. Make a human dotplot. Each member of the mean: above or below? How far above or below the
class should stand above the appropriate location mean is it? How many standard deviations above
along the number line scale based on height (to the or below the mean is it? This last number is the
nearest inch). z-score corresponding to your height.
3. Count the number of people in the class that have Want to know more about where you stand—
www
heights less than or equal to your height. Remember in terms of height, weight, or even body
this value—you’ll use it shortly. mass index? Go to the National Center for Health
4. Your teacher will make a copy of the dotplot on Statistics Web site, www.cdc.gov/growthcharts/,
the board for your reference. and click on the “Clinical Growth Charts” link.

Mrs. Navard’s statistics class has just completed the first three steps of Activity 3.1.
Figure 3.1 shows a dotplot of the class’s height distribution, along with summary
statistics from computer output. Lynette, who is a student in the class, is 65 inches
tall. Is she tall or short relative to her classmates?

Variable n Mean Std. dev. Min. Q1 M Q3 Max.

Height 25 67 4.29 60 63 66 69 75
Figure 3.1 Dotplot and summary statistics for the heights of Mrs. Navard’s statistics
students.
 3. 1 Measuring Location in a Distribution 103

Measuring location: percentiles

One way to describe Lynette’s location within the distribution of heights is to tell
what percent of students in the class are her height or shorter. That is, we can cal-
culate Lynette’s percentile.

Percentile

The pth percentile of a distribution is the value with p percent of the

observations less than or equal to it.

Using the dotplot in Figure 3.1, we see that Lynette’s 65-inch height is tenth
from the bottom. Since 10 of the 25 observations (40%) are at or below her height,
Lynette is at the 40th percentile in the class’s height distribution.

Example 3.1 Where do they stand: percentiles?

For Brett, who is 74 inches tall, only two students in the class are taller than he is.
Brett’s percentile is computed as follows: 23/25 5 0.92, or 92%. So Brett’s height
puts him at the 92nd percentile in the class’s height distribution.
The two students whose heights are 75 inches are by definition at the 100th per-
centile, since 25 out of 25 students are that height or less. Note that some people
define the pth percentile of a distribution as the value with p percent of observa-
tions below it. Using this alternative definition of percentile, it is never possible for
an individual to fall at the 100th percentile. That is why you never see a standard-
ized test score reported above the 99th percentile.
Consider the three students who are 66 inches tall. Thirteen of the 25 students
in the class are this height or shorter, which, by our definition of percentile, places
these three students at the 52nd percentile. If we used the alternative definition of
percentile, these three students would fall at the 40th percentile (10 of the 25 heights
are below 66 inches). But we can see from the computer output in Figure 3.1 that
66 inches is the median height for the class. So we’d prefer to think of these heights as
being at the 50th percentile. Calculating percentiles is not an exact science!

Measuring location: z-scores

Where does Lynette’s height of 65 inches fall relative to the mean of this distribu-
tion? Since the mean height of the class is 67 inches, we can see that Lynette’s height
is “below average.” But how much below average is it?
We can describe Lynette’s location within the class’s height distribution by telling
how many standard deviations above or below the mean her height is. Since the mean
is 67 inches and the standard deviation is a little over 4 inches, Lynette’s height is about
one-half standard deviation below the mean. Converting observations like this from
original values to standard deviation units is known as standardizing. To standardize a
value, subtract the mean of the distribution and then divide by the standard deviation.
104 CHAPTER 3 Modeling Distributions of Data

Standardized values and z-scores

If x is an observation from a distribution that has known mean and

standard deviation, the standardized value of x is
x 2 mean
z5
standard deviation
A standardized value is often called a z-score.

A z-score tells us how many standard deviations from the mean the original
observation falls, and in what direction. Observations larger than the mean have
positive z-scores; observations smaller than the mean have negative z-scores.

Example 3.2 Where do they stand: z-scores?

Lynette’s height is 65 inches. Her standardized height is
x 2 mean 65 2 67
z5 5 5 20.47
standard deviation 4.29
In other words, her height is about one-half of a standard deviation below the mean
height of the class.
Brett, who is 74 inches tall, has a corresponding z-score of
x 2 mean 74 2 67
z5 5 5 1.63
standard deviation 4.29
His height is 1.63 standard deviations above the mean height of the class.

We can also use z-scores to compare the locations of individuals in different

distributions, as the following example illustrates.

Example 3.3 SAT versus ACT

Sofia scores 660 on the SAT Math test. The distribution of SAT scores in the population
is roughly symmetric and single-peaked with mean 500 and standard deviation 100.
Jim takes the ACT Math test and scores 26. ACT scores follow a symmetric, single-
peaked distribution with mean 18 and standard deviation 6. Assuming that both
tests measure the same kind of ability, who did better?
Sofia’s standardized score on the SAT Math test is
x 2 mean 660 2 500
z5 5 5 1.60
standard deviation 100
So her score is 1.6 standard deviations above the mean. Jim’s 26 on the ACT Math
test converts to a z-score of
x 2 mean 26 2 18
z5 5 5 1.33
standard deviation 6
That is, Jim’s score is 1.33 standard deviations above the mean. Sofia performed
slightly better on the SAT than Jim did on the ACT.
 3. 1 Measuring Location in a Distribution 105

We often standardize observations from symmetric distributions to express them

on a common scale. We might, for example, compare the heights of two children of
different ages by calculating their z-scores. The standardized scores tell us where
each child stands (pun intended!) in the distribution for his or her age group.

Exercises 3.1 Interpreting percentiles

(a) Mrs. Munson is concerned about how her daughter’s height and weight
compare with those of other girls her age. She uses an online calculator to
determine that her daughter is at the 87th percentile for weight and the 67th
percentile for height. Explain to Mrs. Munson what this means.
(b) According to the Los Angeles Times, speed limits on California highways are
set at the 85th percentile of vehicle speeds on those stretches of road. Explain to
someone who knows little statistics what that means.
3.2 Comparing performance: percentiles Peter is a star runner on the track
team, and Molly is one of the best sprinters on the swim team. Both athletes
qualify for the league championship meet based on their performance during the
regular season.
(a) In the track playoffs, Peter records a time that would fall at the 80th
percentile of all his race times that season. But his performance places him at the
50th percentile in the league championship meet. Explain how this is possible.
(b) Molly swims a bit slowly for her in the league swim meet, recording a time
that would fall at the 50th percentile of all her meet times that season. But her
performance places Molly at the 80th percentile in this event at the league meet.
Explain how this could happen.
3.3 Measuring bone density Individuals with low bone density have a high
risk of broken bones (fractures). Physicians who are concerned about low
bone density (osteoporosis) in patients can refer them for specialized testing.
Currently, the most common method for testing bone density is dual-energy
X-ray absorptiometry (DEXA). A patient who undergoes a DEXA test usually
gets bone density results in grams per square centimeter (g/cm2) and also in
standardized units.
Francine, who is 25 years old, has her bone density measured using DEXA.
Her results indicate a bone density in the hip of 948 g/cm2, which converts to
a standardized score of z 5 21.45. In the reference population of 25-year-old
women like Francine,2 the mean bone density in the hip is 956 g/cm2.
(a) Francine has not taken a statistics class in a few years. Explain to her in simple
language what the standardized score tells her about her bone density.
(b) Use the information provided to calculate the standard deviation of bone
density in the reference population.
3.4 Comparing bone density Refer to the previous exercise. One of Francine’s
friends, Louise, has the bone density in her hip measured using DEXA. Louise is
35 years old. Her bone density is also reported as 948 g/cm2, but her standardized
score is z 5 0.50. The mean bone density in the hip for the reference population of
35-year-old women is 944 g/cm2.
106 CHAPTER 3 Modeling Distributions of Data

(a) Whose bones are healthier—Francine’s or Louise’s? Justify your answer.

(b) Calculate the standard deviation of bone density in Louise’s reference
population. How does this compare with your answer to Exercise 3.3(b)? Does
this make sense to you?
Exercises 3.5 and 3.6 refer to information found in Table 3.1 and Figure 3.2 about
the salaries of the 2008 World Champion Philadelphia Phillies baseball team.
Table 3.1 shows the salaries for each member of the Phillies baseball team
on the opening day of the 2008 season. Figure 3.2 gives a dotplot and summary
statistics for the salary data.

Table 3.1 Opening-Day Salaries for the Philadelphia Phillies, 2008

Player Salary ($) Player Salary ($) Player Salary ($)

Burrell, Pat 14,250,000 Romero, J. C. 3,250,000 Kendrick, Kyle 445,000

Howard, Ryan 10,000,000 Feliz, Pedro 3,000,000 Dobbs, Greg 440,000
Myers, Brett 8,583,333 Helms, Wes 2,400,000 Ruiz, Carlos 425,000
Rollins, Jimmy 8,000,000 Werth, Jayson 1,700,000 Condrey, Clayton 420,000
Eaton, Adam 7,958,333 Madson, Ryan 1,400,000 Coste, Chris 415,000
Utley, Chase 7,785,714 Durbin, Chad 900,000 Rosario, Francisco 395,000
Lidge, Brad 6,350,000 Taguchi, So 900,000 Zagurski, Mike 392,500
Moyer, Jamie 6,000,000 Bruntlett, Eric 600,000 Lahey, Tim 390,000
Gordon, Tom 5,500,000 Hamels, Cole 500,000 Mathieson, Scott 390,000
Jenkins, Geoff 5,000,000 Victorino, Shane 480,000

Source: USA Today online salary data base, http://content.usatoday.com/sports/baseball/salaries/default.aspx.

0 2 4 6 8 10 12 14 16
Salary (millions)

Variable n Mean Std. dev. Min. Q1 M Q3 Max.

Salary 29 3388617 3767484 390000 440000 1400000 6000000 14250000
Figure 3.2 Dotplot of salaries for the 2008 Philadelphia Phillies baseball team, along
with summary statistics for these data.

3.5 Baseball salaries, I Brad Lidge played a crucial role as the Phillies’ “closer”;
that is, he pitched the end of many games throughout the season.
(a) Find the percentile corresponding to Lidge’s salary. Explain what this value means.
Elsa/Getty Images (b) Find the z-score corresponding to Lidge’s salary. Explain what this value means.
 3. 1 Measuring Location in a Distribution 107

3.6 Baseball salaries, II Did Ryan Madson have a high salary or a low salary
compared with the rest of the team? Justify your answer using Madson’s percentile
and z-score.

Density curves
We now have a kit of graphical and numerical tools for describing distributions. What’s
more, we have a clear strategy for exploring data on a single quantitative variable:
1. Always plot your data: make a graph, usually a histogram or a stemplot.
2. Look for the overall pattern (shape, center, spread) and for striking deviations
such as outliers.
3. Choose either the five-number summary or the mean and standard deviation
to briefly describe center and spread in numbers.
Here is one more step to add to this strategy:
4. Sometimes the overall pattern of a large number of observations is so regular
that we can describe it by a smooth curve.

Example 3.4 Histograms and smooth curves

Figure 3.3 is a histogram of the scores of all 947 seventh-grade students in Gary,
Indiana, on the vocabulary part of the Iowa Test of Basic Skills.3 Scores on this
national test have a very regular distribution.

Figure 3.3 Histogram of the

vocabulary scores of all seventh- 2 4 6 8 10 12
grade students in Gary, Indiana. Vocabulary scores

The histogram is symmetric, and both tails fall off smoothly from a single cen-
ter peak. There are no large gaps or obvious outliers. The smooth curve drawn
through the tops of the histogram bars in Figure 3.3 is a good description of the
overall pattern of the data. It is called a density curve.
108 CHAPTER 3 Modeling Distributions of Data

Figure 3.3 shows a curve used in place of a histogram to picture the overall
shape of a distribution of data. You can think of drawing a curve through the tops
of the bars in a histogram and smoothing out the irregular ups and downs of the
bars. There is one important distinction between histograms and these curves. Most
histograms show the counts of observations in each class by the heights of their bars
and therefore by the areas of the bars. We set up curves to show the proportion of
observations in any region by areas under the curve. To do that, we choose the scale
so that the total area under the curve is exactly 1. We then have a density curve.

Example 3.5 Seventh-grade vocabulary scores

and density curves
Our eyes respond to the areas of the bars in a histogram. The bar areas represent
proportions of the observations. Figure 3.4(a) is a copy of Figure 3.3 with the left-
most bars shaded dark blue.
(a)
The blue bars
represent scores
 6.0

2 4 6 8 10 12
Vocabulary scores
(b)
The blue area
represents scores
 6.0

Figure 3.4 (a) The

proportion of scores
less than 6.0 from the
histogram is 0.303. (b)
The proportion of scores
less than 6.0 from the 2 4 6 8 10 12
density curve is 0.293. Vocabulary scores
 3. 1 Measuring Location in a Distribution 109

The area of the dark blue bars in Figure 3.4(a) represents the students with
vocabulary scores lower than 6.0. There are 287 such students, who make up the
proportion 287/947 5 0.303 of all Gary seventh-graders. In other words, a score of
6.0 corresponds to about the 30th percentile.
Now concentrate on the curve drawn through the bars. In Figure 3.4(b), the
area under the curve to the left of 6.0 is shaded. Adjust the scale of the graph so that
the total area under the curve is exactly 1. This area represents the proportion 1, that
is, all the observations. Areas under the curve then represent proportions of the
observations. The curve is now a density curve. The dark blue area under the density
curve in Figure 3.4(b) represents the proportion of students with scores lower than 6.0.
This area is 0.293, only 0.010 away from the histogram result. So our estimate based
on the density curve is that a score of 6.0 falls at about the 29th percentile. You can
see that areas under the density curve are quite good approximations of areas given
by the histogram.

The area under the density curve in Figure 3.4(b) is not

exactly equal to the true proportion, because the curve is an
idealized picture of the distribution. For example, the curve is
exactly symmetric but the actual data are only approximately
symmetric. Because density curves are smoothed-out, ideal-
ized pictures of the overall shapes of distributions, they are
most useful for describing large numbers of observations.
The density curve in Figures 3.3 and 3.4 is a Normal
curve. The Normal curve has a distinctive, symmetric,
single-peaked bell shape. The Normal curve is easy to
work with and does not require clever software. We will
© Dana Fradon from cartoonbank.com. All rights reserved. see in Section 3.2 that Normal curves have special prop-
erties that help us use them and think about them. Only
some kinds of data fit Normal curves, however, so keep the technology handy for
those that don’t.
The center and spread of a density curve
Density curves help us better understand our measures of center and spread. The
median and quartiles are easy. Areas under a density curve represent proportions of
the total number of observations. The median is the point with half the observations
on either side. So the median of a density curve is the equal-areas point, the point with
half the area under the curve to its left and the remaining half of the area to its right.
The quartiles divide the area under the curve into quarters. One-fourth of the area
under the curve is to the left of the first quartile, and three-fourths of the area is to
the left of the third quartile. You can roughly locate the median and quartiles of any
density curve by eye by dividing the area under the curve into four equal parts.
Because density curves are idealized patterns, a symmetric density curve is exactly
symmetric. The median of a symmetric density curve is therefore at its center. Figure
3.5(a) (on the next page) shows the median of a symmetric curve. Even on a skewed
curve like that in Figure 3.5(b), we can roughly locate the equal-areas point by eye.
What about the mean? The mean of a set of observations is their arithmetic
average. If we think of the observations as weights stacked on a seesaw, the mean is
110 CHAPTER 3 Modeling Distributions of Data

Mean
Median and Mean Median
(a) (b)

Figure 3.5 The median and mean for two density curves: a symmetric Normal curve
and a curve that is skewed to the right.

the point at which the seesaw would balance. This fact is also true of density curves.
The mean is the point at which the curve would balance if made of solid material.
Figure 3.6 illustrates this fact about the mean. A symmetric curve balances at its
center because the two sides are identical. The mean and median of a symmetric
density curve are equal, as in Figure 3.5(a). We know that the mean of a skewed
distribution is pulled toward the long tail. Figure 3.5(b) shows how the mean of a
skewed density curve is pulled toward the long tail more than is the median.

Figure 3.6 The mean of a density curve is the point at which it would balance.

Median and mean of a density curve

The median of a density curve is the equal-areas point, the point that
divides the area under the curve in half.

The mean of a density curve is the balance point, the point at which
the curve would balance if made of solid material.

The median and mean are the same for a symmetric density curve.
They both lie at the center of the curve. The mean of a skewed curve
is pulled away from the median in the direction of the long tail.

Exercises 3.7 A uniform distribution Figure 3.7 (on the facing page) shows the density
curve for a uniform distribution. This curve has height 1 over the interval
from 0 to 1 and is zero outside that range.
Use this density curve to answer the following questions.
 3. 1 Measuring Location in a Distribution 111

height = 1

0 1
Figure 3.7 The density curve of a uniform distribution.

(a) Why is the height of the curve equal to 1?

(b) Recall that the mean of a density curve is the “balance point.” What is the
value of the mean of this uniform distribution?
(c) What is the median? What property of this density curve tells you that the
mean and median are related?
(d) What percent of the observations lie between 0 and 0.4? Explain.
3.8 Another uniform distribution Refer to the previous exercise. Now let’s
consider a uniform distribution over the interval from 0 to 2.
(a) Sketch a graph of the density curve. What is the height of the curve? Why?
(b) What percent of the observations lie between 1 and 1.4? Explain.
(c) Find the median and the quartiles for this distribution. Show your work.
3.9 From histogram to density curve, I Copy the distribution in Figure 3.8 onto
your paper. Then sketch a smooth curve that describes the distribution well. Mark
your best guess for the mean and median of the distribution.
Percent of Shakespeare’s words

25 0.125
Relative frequency

20 0.100

15 0.075

10 0.050

5 0.025

0 0
1 2 3 4 5 6 7 8 9 101112 0 1 2 3 4 5 6 7 8 9
Number of letters in word Last digit of cell phone number

Figure 3.8 A histogram of the Figure 3.9 A histogram of the

distribution of lengths of words used distribution of last digits in students’
in Shakespeare’s plays. cell phone numbers.

3.10 From histogram to density curve, II Copy the distribution in Figure 3.9
onto your paper. Then sketch a smooth curve that describes the distribution well.
Mark your best guess for the mean and median of the distribution.
3.11 Mean and median Figure 3.10 (on the next page) shows two density
curves. Briefly describe the overall shape of each distribution. Two points are
112 CHAPTER 3 Modeling Distributions of Data

marked on each curve to help you locate the mean and the median. For each
curve, give the letter that corresponds to (i) the median and (ii) the mean.

(a) (b)

A B AB
Figure 3.10 Two density curves of different shapes.

3.12 Percentiles and density curves Joey told his dad that he scored at the 98th
percentile on a national standardized test. The scores on this test are approximately
Normally distributed. Sketch a Normal density curve and show the approximate
location of Joey’s score in the distribution.

ws
Ne
ws
Ne

APPLICATION 3.1 B]
RO
g
<
Se
a

A student survey
Recently, students in an introductory statistics class at a large state university responded to a survey. The first
question instructed students to roll a six-sided die and record the value on the up-face. Other questions on
the survey asked students to report their gender, birth month, the amount they spent on textbooks at the
beginning of the term, the number of sodas they consumed during the previous week, and their high school
grade point average (GPA).

QUESTIONS 2. For each of your choices in Question 1, decide

1. Three of the variables recorded in the survey whether the mean of the distribution is probably
were outcome of die roll, amount spent on larger than, smaller than, or about the same as the
textbooks ($), and number of sodas consumed median. Justify your answers.
during the previous week. For each of the three 3. Your teacher will provide you with the data,
variables, determine which of the three density either in print form or electronically. Use your
curves below will approximately describe the calculator or software to draw separate histograms
distribution. Explain your choices. for the three variables in Question 1 from the

(a) (b) (c)

 3. 1 Measuring Location in a Distribution 113

actual data. How do the histograms compare with Variable Actual Actual Actual
what you expected? median mean standard deviation
4. Look at the three histograms for die roll, Die roll
textbook cost, and soda consumption. Guess the Textbooks
value of the median, the mean, and the standard Sodas
deviation for each distribution. Record your values
in a table like the one below. 6. Compare the actual values with your guesses.
Variable My guess at My guess at My guess at the Which variable’s distribution was most difficult to
the median the mean standard deviation judge? Explain.
Die roll 7. One student reported spending $300 on books
Textbooks and consuming 12 sodas in the previous week.
Sodas
(a) Find the z-score for each of these values.
5. Use your calculator or software to compute the (b) Find the percentile corresponding to each of
actual values of the median, mean, and standard these values.
deviation for the three distributions. Record your (c) In which distribution is this student’s value
results in a table like the one that follows. more surprising? Justify your answer.

Section 3.1 Summary

Two ways of describing an individual’s location within a distribution are percentiles and
z-scores. An observation’s percentile is the percent of the distribution that is at or below
the value of that observation. To standardize any observation x, subtract the mean of
the distribution and then divide by the standard deviation. The resulting z-score

x 2 mean
z5
standard deviation

says how many standard deviations x lies from the distribution mean. We can also
use z-scores and percentiles to compare the relative standing of individuals in dif-
ferent distributions.
Stemplots, histograms, and boxplots all describe the distributions of quanti-
tative variables. Density curves are another kind of graph that serves the same
purpose. A density curve is a curve with area exactly 1 underneath it whose shape
describes the overall pattern of a distribution. An area under the curve gives the
proportion of the observations that fall in an interval of values. You can roughly
locate the median (equal-areas point) and the mean (balance point) by eye on a
density curve.
The mean and median are equal for symmetric density curves. The mean of a
skewed curve is located farther toward the long tail than is the median.

Section 3.1 3.13 Finding means and medians Figure 3.11 (on the next page) displays
Exercises three density curves, each with three points indicated. At which of these points on
each curve do the mean and the median fall?
114 CHAPTER 3 Modeling Distributions of Data

AB C A BC AB C
(a) (b) (c)
Figure 3.11 Three density curves: can you locate the mean and median?

3.14 Cholesterol: good or bad? Martin came home very excited after a visit to his
doctor. He announced proudly to his wife, “My doctor says my cholesterol level is at
the 90th percentile among men like me. That means I’m better off than about 90% of
similar men.” How should his wife, who is a statistician, respond to Martin’s statement?
3.15 Unemployment in the states, I Each month the Bureau of Labor Statistics
announces the unemployment rate for the previous month. Unemployment rates are
economically important and politically sensitive. Unemployment may vary greatly
among the states because types of work are unevenly distributed across the country.
Table 3.2 presents the unemployment rates for each of the 50 states in September 2008.

Table 3.2 Unemployment Rates by State, September 2008

State Percent State Percent State Percent

Alabama 5.3 Louisiana 5.2 Ohio 7.2

Alaska 6.8 Maine 5.6 Oklahoma 3.8
Arizona 5.9 Maryland 4.6 Oregon 6.4
Arkansas 4.9 Massachusetts 5.2 Pennsylvania 5.7
California 7.7 Michigan 8.7 Rhode Island 8.8
Colorado 5.2 Minnesota 5.9 South Carolina 7.3
Connecticut 6.1 Mississippi 7.8 South Dakota 3.2
Delaware 4.8 Missouri 6.4 Tennessee 7.2
Florida 6.6 Montana 4.6 Texas 5.1
Georgia 6.5 Nebraska 3.5 Utah 3.5
Hawaii 4.5 Nevada 7.3 Vermont 5.2
Idaho 5.0 New Hampshire 4.1 Virginia 4.3
Illinois 6.9 New Jersey 5.8 Washington 5.8
Indiana 6.2 New Mexico 4.0 West Virginia 4.5
Iowa 4.2 New York 5.8 Wisconsin 5.0
Kansas 4.8 North Carolina 7.0 Wyoming 3.3
Kentucky 7.1 North Dakota 3.6

Source: Bureau of Labor Statistics Web site, www.bls.gov.

(a) Make a histogram of these data. Be sure to label and scale your axes.
(b) Calculate numerical summaries for this data set. Describe the shape, center,
and spread of the distribution of unemployment rates.
 3. 1 Measuring Location in a Distribution 115

(c) Determine the percentile for Illinois. Explain in simple terms what this says
about the unemployment rate in Illinois relative to the other states.
(d) Which state is at the 40th percentile? Calculate the z-score for this state.
3.16 Unemployment in the states, II Refer to the previous exercise. The
December 2000 unemployment rates for the 50 states had a symmetric, single-
peaked distribution with a mean of 3.47% and a standard deviation of about
1%. The unemployment rate for Illinois that month was 4.5%. There were 42
states with lower unemployment rates than Illinois.
(a) Write a sentence comparing the actual rates of unemployment in Illinois in
December 2000 and September 2008.
(b) Compare the percentiles for the Illinois unemployment rate in these same two
months in a sentence or two.
(c) Compare the z-scores for the Illinois unemployment rate in these same two
months in a sentence or two.
Exercises 3.17 to 3.20 refer to the following information. According to the National
Center for Health Statistics, the distribution of heights for 15-year-old males
is modeled well by a Normal density curve. For this distribution, a z-score of 0
corresponds to a height of 170 centimeters (cm), and a z-score of 1 corresponds
to a height of 177.5 cm. The distribution of heights for 16-year-old females is
also modeled well by a Normal density curve. For this distribution, a z-score of 0
corresponds to a height of 162.5 cm, and a z-score of 1 corresponds to a height of
169 cm.

Stone/Getty Images

3.17 Male heights Consider the height distribution for 15-year-old males.
(a) Find its mean and standard deviation. Show your method clearly.
(b) What height would correspond to a z-score of 2.5? Show your work.
3.18 Female heights Consider the height distribution for 16-year-old females.
(a) Find its mean and standard deviation. Show your method clearly.
(b) What height would correspond to a z-score of 1.5? Show your work.
3.19 Is Paul tall? Paul is 15 years old and 175 cm tall.
(a) Find the z-score corresponding to Paul’s height. Explain what this value
means.
(b) Paul’s height puts him at the 75th percentile among 15-year-old males. Explain
what this means to someone who knows no statistics.
116 CHAPTER 3 Modeling Distributions of Data

3.20 Is Miranda taller? Miranda is 16 years old and 170 cm tall.

(a) Find the z-score corresponding to Miranda’s height. Explain what this value means.
(b) Miranda’s height puts her at the 88th percentile among 16-year-old females.
Explain what this means to someone who knows no statistics.
(c) Refer to Exercise 3.19. Who is taller relative to their peers—Miranda or Paul?
Justify your answer.

3.2 Normal Distributions

ACTIVITY 3.2A
Tennis, anyone?
Materials: Tennis ball and ruler with centimeter two measurements are similar (but not necessarily
scale for each pair of students the same). Record these values.
What’s the diameter of a tennis ball? In this 3. Your teacher will draw a number line on the
Activity, you and your classmates will make some board scaled in millimeters. Put your final two
measurements to try to answer this question. Your values on the class dotplot.
teacher will provide you and a partner with a tennis 4. Once the dotplot is complete, discuss the
ball and a ruler. shape, center, and spread of the distribution of
1. Each of you should measure the ball’s diameter measurements with your classmates. Are there any
to the nearest millimeter. Do not tell your partner outliers?
what value you got until both of you have measured. 5. As a class, identify as many possible sources of
2. Compare results with your partner. If your variation in this Activity as you can.
measurements are very different, discuss possible 6. If you had to report a single value for the
reasons for this. Then make two new measurements diameter of a tennis ball based on your class’s
of the tennis ball’s diameter. Continue until your results, what value would you choose? Explain.

Normal distributions as density curves

One particularly important class of density curves has already appeared in Fig-
ures 3.3 (page 107) and 3.5(a) (page 110). These density curves are symmetric,
single-peaked, and bell-shaped. They are called Normal curves, and they describe
Normal distributions. Normal distributions play a large role in statistics, but they
are rather special and not at all “normal” in the sense of being average or natural.
We capitalize “Normal” to remind you that these curves are special.
Why are the Normal distributions important in statistics? Here are three rea-
sons. First, Normal distributions are good descriptions for some distributions of
real data. Distributions that are often close to Normal include
• scores on tests taken by many people (such as SAT exams and IQ tests)
• repeated careful measurements of the same quantity (like the diameter of a
tennis ball)
• characteristics of biological populations (such as yields of corn and lengths
of animal pregnancies)
 3. 2 Normal Distributions 117

From Gaussian to Second, Normal distributions are good approximations to the results of many kinds
Normal of chance outcomes, such as tossing a coin many times. Third, and most important, we
Normal curves were first will see that many statistical inference procedures are based on Normal distributions.
applied to data by the great Figure 3.12 presents two more Normal density curves. Normal curves are sym-
mathematician Carl Friedrich metric, single-peaked, and bell-shaped. Their tails fall off quickly, so that we do
Gauss (1777–1855), who not expect outliers. Because Normal distributions are symmetric, the mean and
used them to describe
the small errors made by
median lie together at the peak in the center of the curve.
astronomers and survey- standard
ors in repeated careful deviation
measurements of the same
quantity.You will sometimes
see Normal distributions
mean
labeled “Gaussian” in honor
of Gauss. For much of the
standard
nineteenth century Normal
deviation
curves were called “error
curves” because they were
first used to describe the
distribution of measure-
ment errors. As it became mean
clear that the distributions Figure 3.12 Two Normal curves. The standard deviation fixes the spread of a Normal curve.
of some biological and
psychological variables were Normal curves also have the special property that we can locate the standard
at least roughly Normal, the deviation of the distribution by eye on the curve. This isn’t true for most other den-
“error curve” terminology
was dropped. The curves sity curves. Here’s how to do it. Imagine that you are skiing down a mountain that
were first called “normal” has the shape of a Normal curve. At first, you descend at an ever-steeper angle as
by Francis Galton in 1889. you go out from the peak:
Galton, a cousin of Charles
Darwin, pioneered the sta-
tistical study of inheritance.

Fortunately, before you find yourself going straight down, the slope begins to grow
flatter rather than steeper as you go out and down.

The points at which this change of curvature takes place are located one standard
deviation on either side of the mean. The standard deviations are marked on the two
curves in Figure 3.12. You can feel the change as you run a pencil along a Normal
curve, and so find the standard deviation.
Normal curves have the special property that giving the mean and the standard
deviation completely specifies the curve. The mean fixes the center of the curve,
and the standard deviation determines its shape. Changing the mean of a Normal
distribution does not change its shape, only its location on the axis. Changing the
standard deviation does change the shape of a Normal curve, as Figure 3.12 illus-
trates. The distribution with the smaller standard deviation is less spread out and
more sharply peaked. Here is a summary of basic facts about Normal curves.
118 CHAPTER 3 Modeling Distributions of Data

Normal density curves

The Normal curves are symmetric, bell-shaped curves that have these
properties:

• A specific Normal curve is completely described by giving its mean

and its standard deviation.
• The mean determines the center of the distribution. It is located at
the center of symmetry of the curve.
• The standard deviation determines the shape of the curve. It is the
distance from the mean to the change-of-curvature points on either side.

Caution! Even though many sets of data follow a Normal distribution, many
do not. Most income distributions, for example, are skewed to the right and so are
not Normal. Some distributions are symmetric but not Normal or even close to
Normal. The uniform distribution of Exercise 3.7 (page 110) is one such example.
Non-Normal data, like nonnormal people, not only are common but also are some-
times more interesting than their Normal counterparts.

The 68–95–99.7 rule

ACTIVITY 3.2B
The Normal Curve applet
PLET
AP Directions: the bright yellow shaded region you create. If you drag
Go to the texbook’s Web site, one flag past the other, the applet will show the area
www.whfreeman.com/sta2e, and select the under the curve between the two flags. When the
Normal Curve applet. Drag the green flags to have the “2-Tail” box is checked, the applet calculates symmetric
applet compute the part of the Normal curve that is in areas around the mean.
 3. 2 Normal Distributions 119

Think: APPLET TIP: The direction in which the green flags

point is the side of the flag poles that is shaded.
Answer the following without using the applet.
1. If you were to put one flag at the extreme left of Experiment:
the curve and the second flag exactly in the middle,
5. What percent of the area under the Normal
what would be the proportion reported by the
curve lies within two standard deviations of the
applet?
mean?
2. If you were to place the two flags exactly one
6. Change the mean to 100 and the standard
standard deviation on either side of the mean, what
deviation to 15. Then click “Update.” What percent
would the applet say is the area between them?
of the area under this Normal density curve lies
3. As you can see from the figure, about 99.7% of within one, two, and three standard deviations of
the area under the Normal density curve lies within the mean?
three standard deviations of the mean. Does this
mean that about 99.7%/2  49.85% will lie within 7. Change the mean to 500 and the standard
one and a half standard deviations? deviation to 100. Answer the question in Step 6.

Verify: Summarize:
4. Use the applet to test your answers to the three 8. Complete the following sentence: “For any
questions above. Did the proportion of shaded area Normal density curve, the area under the curve
under the curve behave as you thought it would in within one, two, and three standard deviations of
each case? the mean is about ___%, ___%, and ___%.”

There are many Normal curves, each described by its mean and standard deviation.
All Normal curves share many properties. In particular, the standard deviation is
the natural unit of measurement for Normal distributions. This fact is reflected in
the following rule.

The 68–95–99.7 rule

In any Normal distribution, approximately

• 68% of the observations fall within one standard deviation of the

mean
• 95% of the observations fall within two standard deviations of the
mean
• 99.7% of the observations fall within three standard deviations of
the mean

Figure 3.13 (on the next page) illustrates the 68–95–99.7 rule. (Some people
refer to this result as the “empirical rule.”) By remembering these three numbers,
you can think about Normal distributions without constantly making detailed
calculations. Remember also, though, that no set of data is exactly described by a
Normal curve. The 68–95–99.7 rule will be only approximately true for SAT scores
or the lengths of crickets, for example.
120 CHAPTER 3 Modeling Distributions of Data

68% of data

95% of data

99.7% of data

3 2 1 0 1 2 3
Figure 3.13 The 68–95–99.7 rule for Normal distributions.

Here’s an example that shows how we can use the 68–95–99.7 rule to find a
percentile.

Example 3.6 Performing well on the SAT

Jennie scored 600 on the Critical Reading section of the SAT Reasoning test. How
good a score is this? That depends on where a score of 600 lies in the distribution of all
scores. The SAT exams are scaled so that scores on the Critical Reading section should
roughly follow the Normal distribution with mean 500 and standard deviation 100.

iStockphoto
50% 34%
of scores of scores

200 300 400 500 600 700 800

Mean
Scores on the SAT Critical Reading Test
Figure 3.14 The 68–95–99.7 rule shows that 84% of any Normal distribution lies to the left
of the point one standard deviation above the mean. Here, this fact is applied to SAT scores.
 3. 2 Normal Distributions 121

Jennie’s 600 is one standard deviation above the mean. The 68–95–99.7 rule
now tells us just where she stands (see Figure 3.14 on the facing page). Half of all
scores are below 500, and another 34% are between 500 and 600. So Jennie did bet-
ter than 84% of the students who took the SAT. Her score report not only will say
that she scored 600 but will add that this is at the “84th percentile.” As you learned
in Section 3.1, that’s statistics speak for “You did better than 84% of those who took
the test.”

Because the standard deviation is the natural unit of measurement for Normal
distributions, we restated Jennie’s score of 600 as “one standard deviation above the
mean.” As you learned in Section 3.1, observations expressed in standard devia-
tions above or below the mean of a distribution are called z-scores or standardized
scores. To obtain the standardized score for a particular observation, we compute
x 2 mean
z5
standard deviation
The usual notation for the mean of a density curve is m (the Greek letter “mu”).
We write the standard deviation of a density curve as s (the Greek letter “sigma”).
These are the same symbols we use to represent the mean and standard deviation of
a population distribution. This makes sense because density curves are quite often
used to describe population distributions.
With our new symbols, the standardized score formula becomes
x2m
z5
s
For Jennie’s SAT Critical Reading score, x  600, m  500, and s  100. Her z-score is
600 2 500
z5 51
100

Exercises Use the distribution in Figure 3.15 and the 68–95–99.7 rule for Exercises 3.21 to
3.23. Figure 3.15 is a stemplot of the IQ test scores of 74 seventh-grade
students.4 This distribution is very close to Normal with mean 111 and
standard deviation 11. It includes all the seventh-graders in a rural
Midwest school. Take the Normal distribution with mean 111 and
standard deviation 11 as a description of the IQ test scores of all rural
Midwest seventh-grade students.
3.21 IQ test scores, I Between what values do the IQ scores of the
middle 95% of all rural Midwest seventh-graders lie? Explain.
3.22 IQ test scores, II What percent of IQ scores for all rural
Midwest seventh-graders are greater than 100? Explain. How does this
compare with the percent in our sample?
3.23 IQ test scores, III What percent of all students have IQ
Figure 3.15 Stemplot of the IQ scores of 144 or higher? Explain. None of the 74 students in our
scores of 74 seventh-grade students, for sample school had scores this high. Are you surprised at this?
Exercises 3.21 to 3.23. Why?
122 CHAPTER 3 Modeling Distributions of Data

Figure 3.16 What are the

mean and standard deviation of
this Normal density curve? 3 4 5 6 7 8 9 1011121314151617

3.24 A Normal curve Estimate the mean and standard deviation of the Normal
curve in Figure 3.16.
3.25 Horse pregnancies Bigger animals tend to carry their young longer before
birth. The length of horse pregnancies from conception to birth varies according to
a roughly Normal distribution with mean 336 days and standard deviation 3 days.
Use the 68–95–99.7 rule to answer the following questions.
(a) Almost all (99.7%) horse pregnancies fall in what range of lengths?
(b) What percent of horse pregnancies are longer than 339 days? Show your work.
3.26 Eggs A truck is loaded with cartons of eggs that weigh an average of 2
pounds each with a standard deviation of 0.1 pound. A histogram of these weights
looks very much like a Normal distribution. Use the 68–95–99.7 rule to answer
the following questions.
(a) What percent of the cartons weigh less than 2.1 pounds? Show your work.
(b) What percent of the cartons weigh less than 1.8 pounds? Show your work.
(c) What percent of the cartons weigh more than 1.9 pounds? Show your work.

DATA EXPLORATION
Presidents’ Ages at Inauguration
How can we decide whether a given data set follows a Normal distribution? We can start by using graphical
and numerical methods from Chapter 2 to analyze the data in Table 3.3.

Table 3.3 Ages of Presidents at Inauguration

President Age President Age President Age

Washington 57 J. Q. Adams 57 Polk 49

J. Adams 61 Jackson 61 Taylor 64
Jefferson 57 Van Buren 54 Fillmore 50
Madison 57 W. Harrison 68 Pierce 48
Monroe 58 Tyler 51 Buchanan 65
 3. 2 Normal Distributions 123

President Age President Age President Age

Lincoln 52 T. Roosevelt 42 L. Johnson 55

A. Johnson 56 Taft 51 Nixon 56
Grant 46 Wilson 56 Ford 61
Hayes 54 Harding 55 Carter 52
Garfield 49 Coolidge 51 Reagan 69
Arthur 51 Hoover 54 G. H. Bush 64
Cleveland 47 F. Roosevelt 51 Clinton 46
B. Harrison 55 Truman 60 G. W. Bush 54
Cleveland 55 Eisenhower 61 Obama 47
McKinley 54 Kennedy 43

1. Plot a histogram of the data. Your teacher will 5. Use the method in Step 4 to find the
decide whether you may use technology. percent of presidents’ ages at inauguration
within two standard deviations of the mean and
2. Describe the distribution. Include the shape,
the percent within three standard deviations of
center, spread, and any deviations from the
the mean.
pattern.
Commentary: Although the distribution
3. Which units of center and spread would be of presidents’ ages at inauguration does not
more appropriate: mean and standard deviation follow a Normal distribution, we can say that
or median and interquartile range? Explain your it is approximately Normal. Why? First, the
choice. histogram you created in Step 1 is single-
4. There are 44 data points. What percent of them peaked, roughly symmetric, and looks bell-
are within one standard deviation of the mean? shaped. Second, the percents of observations
Use the One-Variable Statistics command on your within one, two, and three standard deviations
calculator or the One-Variable Statistical Calculator of the mean fall fairly close to those specified by
applet to compute the mean and standard the 68–95–99.7 rule for Normal distributions.
deviation. Count the number of observations that When both of these conditions are met, we
fall within one standard deviation of the mean and can describe the distribution as approximately
convert to a percent. Normal.

The standard Normal distribution

As the 68–95–99.7 rule suggests, all Normal distributions are the same if we mea-
sure in units of size s about the mean m as center. Changing to these units requires
us to standardize:

x2m
z5
s

If the variable we standardize has a Normal distribution, then so does the new vari-
able z. This new distribution is called the standard Normal distribution.
124 CHAPTER 3 Modeling Distributions of Data

Standard Normal distribution

The standard Normal distribution is the Normal distribution with

mean 0 and standard deviation 1 (Figure 3.17).

The bell curve?

Does the distribution of
human intelligence follow 3 2 1 0 1 2 3
the “bell curve” of a Normal
distribution? Scores on IQ Figure 3.17 Standard Normal distribution.
tests do roughly follow a If a variable x has any Normal distribution with mean m and stan-
Normal distribution. That
is because a test score is dard deviation s, then the standardized variable
calculated from a person’s x2m
answers in a way that is de- z5
s
signed to produce a Normal
distribution. To conclude has the standard Normal distribution.
that intelligence follows a
bell curve, we must agree
that the test scores directly
measure intelligence. Many
An area under a density curve is a proportion of the observations in a distri-
psychologists don’t think bution. Any question about what proportion of observations lie in some range of
there is one human charac- values can be answered by finding an area under the curve. In a standard Normal
teristic called “intelligence” distribution, the 68–95–99.7 rule tells us that about 68% of the observations fall
that can be measured by a between and z  1 and z  1 (that is, within one standard deviation of the mean).
single test score.
What if we want to find the percent of observations that fall between z  1.25 and
z  1.25? The 68–95–99.7 rule can’t help us.
Because all Normal distributions are the same when we standardize, we can find ar-
eas under any Normal curve from a single table, a table that gives areas under the curve
for the standard Normal distribution. Table A, the standard Normal table, gives areas
under the standard Normal curve. You can find Table A at the back of the book.

The standard Normal table

Table A is a table of areas under the standard Normal curve. The table
entry for each value z is the area under the curve to the left of z.

Table entry is area

to left of z

z
 3. 2 Normal Distributions 125

The next few examples show you how to use Table A.

Example 3.7 Using the standard Normal table, I

Problem: Find the proportions of observations from the standard Normal distri-
bution that are (a) less than 21.25 and (b) greater than 0.81.

Solution:
(a) To find the area to the left of 21.25, locate 21.2 in the left-hand column of
Table A, then locate the remaining digit 5 as .05 in the top row. The entry opposite
21.2 and under .05 is .1056. This is the area we seek. Here is a reproduction of
the relevant portion of Table A, which appears on the inside back cover of this
book:

z .00 .01 .02 .03 .04 .05

21.3 .0968 .0951 .0934 .0918 .0901 .0885
21.2 .1151 .1131 .1112 .1093 .1075 .1056
21.1 .1357 .1335 .1314 .1292 .1271 .1251

Figure 3.18 illustrates the relationship between the value z  1.25 and the
area 0.1056.

Table entry for z is Table entry  0.1056

always the area under for z  1.25.
the curve to the
left of z.

3 2 1 0 1 2 3
z  1.25

Figure 3.18 The area under the standard Normal curve to the left of the point
z 5 21.25 is 0.1056.

(b) To find the area to the right of z 5 0.81, locate 0.8 in the left-hand column of
z .00 .01 .02 Table A, then locate the remaining digit 1 as .01 in the top row. The entry opposite
0.8 and under .01 is .7910. This is the area to the left of z 5 0.81. To find the area to
0.7 .7580 .7611 .7642
the right of z 5 0.81, we use the fact that the total area under the standard Normal
0.8 .7881 .7910 .7939
density curve is 1. So the desired proportion is 1  0.7910  0.2090. The relevant
0.9 .8159 .8186 .8212 portion of Table A is reproduced in the margin.
Figure 3.19 (on the next page) illustrates the relationship between the value z = 0.81
and the area 0.2090.
126 CHAPTER 3 Modeling Distributions of Data

Table entry  0.7910

for z  0.81. This is
the area to the The area to the
left of z  0.81. right of z  0.81 is:
1  0.7910  0.2090.

Figure 3.19 The area

under the standard Normal
curve to the right of the 3 2 1 0 1 2 3
point z 5 0.81 is 0.2090. z  0.81

Example 3.8 Using the standard Normal table, II

Problem: Find the proportion of observations from the standard Normal distribu-
tion that are between 21.25 and 0.81.

Solution: In Example 3.7, we found the area to the left of z 5 21.25 and the area
to the right of z 5 0.81 under the standard Normal curve. Now we want to find
the area between these two z-scores. Figure 3.20 shows one way to do this. From
Example 3.7, the area to the left of z 5 21.25 is 0.1056 and the area to the right of
z 5 0.81 is 0.2090. So the area between these two z-scores is
1 2 (0.1056 1 0.2090) 5 1 2 0.3146 5 0.6854
The area to the
left of z  1.25
is 0.1056.
The area to the
right of z  0.81
is 0.2090.
Figure 3.20 The area
under the standard Normal
curve between z 5 21.25
and z 5 0.81 is 0.6854. 3 2 1 0 1 2 3

Here’s another way to find the shaded area in Figure 3.20. This is especially
helpful if you haven’t already found the areas in the two tails of the distribution.
From Table A, the area to the left of z 5 0.81 is 0.7910 and the area to the left of
z 5 21.25 is 0.1056. So the area under the standard Normal curve between these two
z-scores is 0.7910 2 0.1056 5 0.6854. Figure 3.21 shows why this approach works.

 

3 2 1 0 1 2 3 3 2 1 0 1 2 3 3 2 1 0 1 2 3

Figure 3.21 Another way to find the area between z 5 21.25 and z 5 0.81.
 3. 2 Normal Distributions 127

Caution! A common student mistake is to look up a z-value in Table A and

report the entry corresponding to that z-value, regardless of whether the prob-
lem asks for the area to the left or to the right of that z-value. Always sketch the
standard Normal curve, mark the z-value, and shade the area of interest. And before
you finish, make sure your answer is reasonable in the context of the problem.
In Examples 3.7 and 3.8, we used Table A to find areas under the standard
Normal curve from z-scores. What if we want to find the z-score that corresponds
to a particular area?

Example 3.9 Finding z-scores from percentiles

Problem: Find the z-score that corresponds to the 90th percentile of the standard
Normal curve.

Solution: We’re looking for the z-score that has 90 percent of the area to its left, as
shown in Figure 3.22. Since Table A gives areas to the left of a specified z-score, all
we need to do is find the value closest to 0.90 in the middle of the table. From the
The area to the
left of z is 0.90.
What’s z ?

z  ???
z .07 .08 .09
1.1 .8790 .8810 .8830 Figure 3.22 The z-score with area 0.90 to its left under the standard Normal curve.
1.2 .8980 .8997 .9015 reproduced portion of Table A, you can see that the desired z-score is z 5 1.28. That
1.3 .9147 .9162 .9177 is, the area to the left of z 5 1.28 is as close to 0.90 as the table gets.

Exercises For Exercises 3.27 to 3.30, use Table A to find the proportion of observations from
PLET
AP the standard Normal distribution that satisfies each of the following statements.
In each case, sketch a standard Normal curve and shade the area under the curve
that is the answer to the question. Use the Normal Curve applet to check your
answers.
3.27 Table A practice, I
(a) z is less than 20.37 (b) z is greater than 20.37
(c) z is less than 2.15 (d) z is greater than 2.15
3.28 Table A practice, II
(a) z is less than 21.58 (b) z is greater than 1.58
(c) z is greater than 20.46 (d) z is less than 0.93
128 CHAPTER 3 Modeling Distributions of Data

3.29 Table A practice, III

(a) z is between 21.33 and 1.65 (b) z is between 0.50 and 1.79
3.30 Table A practice, IV
(a) z is between 22.05 and 0.78 (b) z is between 21.11 and 20.32
AP
PLET For Exercises 3.31 and 3.32, use Table A to find the value z from the standard
Normal distribution that satisfies each of the following conditions. (Use the value
of z from Table A that comes closest to satisfying the condition.) In each case,
sketch a standard Normal curve with your value of z marked on the axis. Use the
Normal Curve applet to check your answers.
3.31 Working backward, I
(a) The 20th percentile of the standard Normal distribution.
(b) 45% of all observations are greater than z.
3.32 Working backward, II
(a) The 63rd percentile of the standard Normal distribution.
(b) 75% of all observations are greater than z.

Normal distribution calculations

We can answer a question about proportions of observations in any Normal dis-
tribution by standardizing and then using the standard Normal table. Here is an
outline of the method for finding the proportion of the distribution in any region.

Solving problems involving Normal distributions

Step 1: State the problem in terms of the observed variable x.

Step 2: Standardize and draw a picture. Standardize x to restate the
problem in terms of a standard Normal variable z. Draw a picture to
show the area of interest under the standard Normal curve.
Step 3: Use the table. Find the required area under the standard
Normal curve using Table A and the fact that the total area under the
curve is 1.
Step 4: Conclusion. Write your conclusion in the context of the
problem.

Here’s an example of the method at work.

Example 3.10 Tiger on the range

On the driving range, Tiger Woods practices his swing with a particular club by
hitting many, many balls. When Tiger hits his driver, the distance the ball travels
follows a Normal distribution with mean 304 yards and standard deviation 8 yards.
What percent of Tiger’s drives travel at least 290 yards?
Step 1: State the problem. Let x 5 the distance that Tiger’s ball travels. The variable
David Cannon/Getty Images x has a Normal distribution with m 5 304 and s 5 8. We want the proportion of
 3. 2 Normal Distributions 129

Tiger’s drives with x $ 290. Figure 3.23(a) is a picture of the distribution with the
area of interest shaded.

(a)

272 280 288 296 304 312 320 328 336

Figure 3.23(a) Distance traveled by Tiger Woods’s drives on the range.

Step 2: Standardize and draw a picture. For x 5 290, we have

x2m 290 2 304
z5 5 5 21.75
s 8
So x $ 290 corresponds to z $ 21.75 under the standard Normal curve. Figure
3.23(b) shows the standard Normal distribution with the area of interest shaded.

(b)

3 2 1 0 1 2 3
Figure 3.23(b) Area under the standard Normal curve corresponding to drives of
length 290 yards or more in Figure 3.23(a).

Step 3: Use the table. From Table A, we see that the proportion of observations
less than 21.75 is 0.0401. About 4% of Tiger’s drives travel less than 290 yards.
The area to the right of 21.75 is therefore 1 2 0.0401 5 0.9599. This is about
0.96, or 96%.
Step 4: Conclusion. About 96% of Tiger Woods’s drives on the range travel at least
290 yards.

In a Normal distribution, the proportion of observations with x $ 290 is the

same as the proportion with x . 290. There is no area under the curve exactly above
the point 290 on the horizontal axis, so the areas under the curve with x $ 290 and
x . 290 are the same. This isn’t true of the actual data. Tiger may hit a drive exactly
290 yards. The Normal distribution is just an easy-to-use approximation, not a de-
scription of every detail in the actual data.
The key to doing a Normal calculation is to sketch the area you want, then
match that area with the area that the table gives you. Here is another example.
130 CHAPTER 3 Modeling Distributions of Data

Example 3.11 Tiger on the range (continued)

What percent of Tiger’s drives travel between 305 and 325 yards?
Step 1: State the problem. We want the proportion of Tiger’s drives with 305 # x # 325.
Step 2: Standardize and draw a picture. When x  305,

305 2 304
z5 5 0.13
8

The standardized score for a 325-yard drive (x  325) is

325 2 304
z5 5 2.63
8

Figure 3.24 shows the standard Normal curve with the area of interest shaded.

3 2 1 0 1 2 3

Figure 3.24 Standard Normal curve showing the area between z 5 0.13 and z 5 2.63.

Step 3: Use the table. The area between z  0.13 and z  2.63 is the area to the left
of 2.63 minus the area to the left of 0.13. Look at Figure 3.21 to see why this makes
sense. From Table A,

area between 0.13 and 2.63  area to the left of 2.63  area to the left of 0.13
 0.9957  0.5517  0.4440
Step 4: Conclusion. About 44% of Tiger’s drives travel between 305 and 325 yards.

What if we meet a z that falls outside the range covered by Table A? For ex-
ample, the area to the left of z  4 does not appear in the table. But since 4 is
less than 3.4, this area is smaller than the entry for z  3.40, which is 0.0003.
There is very little area under the standard Normal curve outside the range cov-
ered by Table A. You can take this area to be approximately zero with little loss of
accuracy.

Finding a value, given a proportion

Examples 3.10 and 3.11 illustrate the use of Table A to find what proportion of
the observations satisfies some condition, such as “Tiger’s drive travels between
 3. 2 Normal Distributions 131

305 and 325 yards.” Sometimes, we may want to find the observed value that
corresponds to a given percentile. To do this, use Table A backward. Find the
given proportion in the body of the table, read the corresponding z from the left
column and top row, then “unstandardize” to get the observed value. Here is an
example.

Example 3.12 Tiger on the range: using Table A

in reverse
What distance would a ball have to travel to be at the 80th percentile of Tiger’s drive
lengths?
Step 1: State the problem and draw a picture. We want to find the drive distance x
with area 0.8 to its left under the Normal curve with mean m 5 304 and standard
deviation s 5 8. Figure 3.25 poses the question in graphical form.

Area = 0.80

304 ? x

0 0.84 z

Figure 3.25 Locating the point on a Normal curve with area 0.80 to its left.

Step 2: Use the table. Look in the body of Table A for the entry closest to 0.8. It is
0.7995. This is the entry corresponding to z 5 0.84. So z 5 0.84 is the standardized
value with area 0.8 to its left.
Step 3: “Unstandardize” to transform the solution from a z-score to a value of x. We
know that the standardized value of the unknown x is z 5 0.84. So x satisfies the
equation

x 2 304
5 0.84
8

Solving this equation for x gives

x 5 304 1 (0.84)(8) 5 310.7

This equation should make sense: it finds the x that lies 0.84 standard deviations
above the mean on this particular Normal curve. That is the “unstandardized”
meaning of z 5 0.84.
Step 4: Conclusion. A ball would have to travel 310.7 yards to be at the 80th percen-
tile of Tiger’s drive length distribution.
132 CHAPTER 3 Modeling Distributions of Data

Exercises For Exercises 3.33 to 3.38, follow the four-step method of Examples 3.10 to 3.12 to
answer the questions.
3.33 Tiger prowls After hitting plenty of balls on the practice range, Tiger
Woods heads out to the first tee to begin a golf tournament. A large creek crosses
the fairway 317 yards from the tee. Assume that the distance traveled by Tiger’s
ball follows a Normal distribution with m  304 yards and s  8 yards, as in
Example 3.10. Is Tiger safe hitting his driver? Show your work.
3.34 High IQ scores Scores on the Wechsler Adult Intelligence Scale for 20- to
34-year-olds are approximately Normally distributed with mean 110 and standard
deviation 25. How high must a person score to be in the top 25% of all scores?
Show your work.
3.35 SAT scores, I The average performance of females on the SAT, especially
the Math section, is lower than that of males. The reasons for this gender gap are
controversial. In 2007, female scores on the SAT Math test followed a Normal
distribution with mean 500 and standard deviation 111. Male scores had a mean
of 533 and a standard deviation of 118. What percent of females scored higher
than the male mean? Show your work.
3.36 SAT scores, II Refer to the previous exercise.
(a) Find the 85th percentile of the SAT Math score distribution for males. Show
your work.
(b) To what percentile in the female score distribution does your answer to
(a) correspond? Show your work.
3.37 Potatoes Bags of potatoes in a shipment averaged 10 pounds with a
standard deviation of 0.5 pounds. A histogram of these weights followed a
Normal curve quite closely.
(a) What percent of the bags weighed less than 10.25 pounds? Show your work.
(b) What percent weighed between 9.5 and 10.25 pounds? Show your work.
3.38 Brush your teeth The amount of time Ricardo spends brushing his teeth
follows a Normal distribution with unknown mean m and standard deviation
s 5 20 seconds. Ricardo spends less than 60 seconds brushing his teeth about
40% of the time. Use this information to determine the mean of this distribution.
Show your method clearly.

CALCULATOR CORNER Normal curve areas

The TI-84 and TI-Nspire can be used to find the area under a Nor-
mal curve above any interval on the horizontal axis. We’ll show you
how to use the normalcdf (normCdf on the TI-Nspire) com-
mand to find the percent of Tiger Woods’s drives that travel at least
290 yards. Recall from Example 3.10 (page 128) that m  304 yards and
s  8 yards.
 3. 2 Normal Distributions 133

We want to find the area to the right of 290 under a Normal curve with mean
304 and standard deviation 8. The screen shot shows the desired area. Notice that
we chose an “upper bound” of 400 yards on Tiger’s drive lengths, which is more
than 10 standard deviations above the mean. To find this area, we’ll execute the
calculator command normalcdf(290,400,304,8), where 290 is the “lower
bound,” 400 is the “upper bound,” 304 is the mean, and 8 is the standard deviation.

TI-84
• Find the command normalcdf by pressing yÏ[DISTR] and
choosing 2:normalcdf(.
• Complete the command normalcdf(290,400,304,8) and press b.
The result should be 0.9599, after rounding to four decimal places.

TI-Nspire
• Open a Calculator page. Go into the catalog (press k) and then press N. Use
the NavPad to arrow down to normCdf and click the mouse (x) or press · to
select the command. A dialog box should appear.
• Enter these values in the dialog box: Lower Bound, 290; Upper Bound, 400;
, 304; , 8. (Use the e key to move between the fields in the dialog box.)
Then press ·.
• Compare your results with the screen shot shown.

These results agree with what we got by using Table A earlier. The first two
numbers that you entered (290 and 400) specify the interval corresponding to
the shaded area under the curve. The last two numbers (304 and 8) identify the
particular Normal curve by specifying the mean and standard deviation.
Now see if you can calculate the area in Example 3.11 (page 130).

CALCULATOR CORNER “Backward” Normal calculations

You can use the TI-84’s or the TI-Nspire’s invNorm command to
find the value with a specified area to its left under a Normal curve. To
determine the 80th percentile of Tiger’s drive distances from Example 3.12
(page 131), we need to execute the command invNorm(0.8,304,8).
CALCULATOR CORNER (continued)
134 CHAPTER 3 Modeling Distributions of Data

CALCULATOR CORNER (continued)

Notice that the inputs are the area to the left of our desired value (which is 0.8 for
the 80th percentile), the mean (304), and the standard deviation (8).

TI-84
• Find the command invNorm by pressing yÏ [DISTR] and choosing
3:invNorm(.
• Complete the command invNorm(0.8,304,8) and press b. The
result should be about 310.73.

TI-Nspire
• In a Calculator page, go into the catalog (press k) and then press I. Use the
NavPad to arrow down to invNorm and click the mouse (x) or press · to
select the command. A dialog box should appear.
• Enter these values in the dialog box: Area, 0.8; m, 304; s, 8. (Use the e key to
move between the fields in the dialog box.) Then press ·.
• Compare your results with the screen shots shown.

ws
Ne
ws
Ne

APPLICATION 3.2 B]
RO
g
<
Se
a

The vending machine problem

Have you ever purchased a hot drink from a vend- The vending machine company wants to keep
ing machine? The intended sequence of events runs customers happy. So they have decided to hire you
something like this. You insert your money into the as a statistical consultant. They provide you with
machine and select your preferred beverage. A cup the following summary of important facts about the
falls out of the machine, landing upright. Liquid vending machine:
pours out until the cup is nearly full. You reach in, • Cups will hold 8 ounces.
grab the piping hot cup, and drink happily. • The amount of liquid dispensed varies
Sometimes, things go wrong. The machine according to a Normal distribution centered at
might take your money and not give you anything the mean m that is set in the machine.
in return. Or the cup might fall over. More fre- • s  0.2 ounces
quently, everything goes smoothly until the liquid
begins to flow. It might stop flowing when the cup QUESTIONS
is only half full. Or the liquid might keep coming 1. Suppose the company sets the machine’s mean
until your cup overflows. Neither of these results at m  7.5 ounces. What percent of cups would
leaves you satisfied. overflow? Show your work.
 3. 2 Normal Distributions 135

2. If a cup contains too much liquid, a customer may get should the machine’s mean be set to ensure that
burned from a spill. This could result in an expensive 90% of cups contain at least 7 ounces?
lawsuit for the company. Explain to the company 4. Given the issues raised in the previous two
president why setting the machine’s mean at 7.0 ounces questions, what mean setting for the machine
might be smarter than setting it at 7.5 ounces. would you recommend? Write a brief report to the
3. Customers may be irritated if they get a cup with vending machine company president that explains
too little liquid from the machine. At what value your answer.

Section 3.2 Summary

Normal curves are a special kind of density curve that describe the overall pat-
tern of some sets of data. Normal curves are symmetric and bell-shaped. A specific
Normal curve is completely described by its mean and standard deviation. You can
locate the mean (center point) and the standard deviation (distance from the mean
to the change-of-curvature points) on a Normal curve. All Normal distributions
obey the 68–95–99.7 rule, which describes what percent of observations lie within
one, two, and three standard deviations of the mean.
All Normal distributions are the same when measurements are standardized.
If x follows a Normal distribution with mean m and standard deviation s, we can
standardize using
x2m
z5
s
The variable z has the standard Normal distribution with mean 0 and standard
deviation 1.
Table A gives percentiles for the standard Normal curve. By standardizing, we
can use Table A to determine the percentile for a given score or the score corre-
sponding to a given percentile in any Normal distribution. You can use the Normal
Curve applet or your calculator to perform Normal calculations quickly.

Section 3.2 3.39 Heights of young women The distribution of the heights of young women
Exercises aged 18 to 24 is approximately Normal with mean 65 inches and standard devia-
tion 2.5 inches. Sketch a picture of a Normal curve and then use the 68–95–99.7
rule to show what the rule states about these women’s heights.
3.40 Normal curve properties, I Figure 3.26 is a Normal density curve. Estimate
the mean and the standard deviation of this distribution.

Figure 3.26 A Normal

density curve. 24 25 26 27 28 29 30 31 32
136 CHAPTER 3 Modeling Distributions of Data

3.41 Normal curve properties, II Explain why the point that is one standard devia-
tion below the mean in a Normal distribution is always the 16th percentile. Explain
why the point that is two standard deviations above the mean is the 97.5th percentile.
3.42 More Table A practice, I Use Table A to find the proportion of observations
from a standard Normal distribution that satisfies each of the following statements.
In each case, sketch a standard Normal curve and shade the area under the curve that
is the answer to the question. Use an applet or your calculator to check your answers.
(a) z . 21.81 (b) z , 2.29
(c) 21.81 # z # 20.47 (d) 21.02 # z # 0.65
3.43 More Table A practice, II Use Table A to find the value z from a standard
Normal distribution that satisfies each of the following conditions. In each case,
sketch a standard Normal curve with your value of z marked on the axis. Use an
applet or your calculator to check your answers.
(a) The point z with 32% of the observations falling to its left.
(b) The point z with 40% of the observations falling to its right.
(c) The 10th percentile of the standard Normal distribution.
(d) The 83rd percentile of the standard Normal distribution.
3.44 NCAA rules for athletes The National Collegiate Athletic Association
(NCAA) requires Division I athletes to score at least 820 on the combined
Mathematics and Critical Reading parts of the SAT exam in order to compete
in their first college year. (Higher scores are required for students with poor
high school grades.) In 2007, the combined scores of the millions of students
taking the SATs were approximately Normal with mean 1017 and standard
deviation 211. What percent of all students had scores less than 820? Show
your work.
3.45 Are we getting smarter? When the Stanford-Binet “IQ test” came into
use in 1932, it was adjusted so that scores for each age group of children
followed roughly the Normal distribution with mean 100 and standard devia-
tion 15. The test is readjusted from time to time to keep the mean at 100. If
present-day American children took the 1932 Stanford-Binet test, their mean
score would be about 120. The reasons for the increase in IQ over time are not
known but probably include better childhood nutrition and more experience
in taking tests.5
(a) IQ scores above 130 are often called “very superior.” What percent of children
had very superior scores in 1932?
(b) If present-day children took the 1932 test, what percent would have very superior
scores? (Assume that the standard deviation 15 does not change.)
3.46 Get smart Refer to the previous exercise. What IQ score was at the 98th
percentile in 1932? Show your work.

CHAPTER 3 REVIEW
This chapter focused on two big issues: describing an observation’s location with-
in a distribution and modeling with Normal distributions. Both percentiles and
z-scores provide easily calculated measures of relative standing for individuals.
Density curves come in assorted shapes, but all share the property that the area beneath
the curve is 1. We can use areas under density curves to estimate the proportion of
individuals in a distribution whose values fall in a specified range.
 Chapter 3 Review Exercises 137

Sometimes the overall pattern of a large number of observations is so regular

that we can describe it by a smooth curve, called a Normal curve. Normal curves are
handy models because they have properties that make them easy to use and think
about. When data are approximately Normally distributed, the mean and standard
deviation are natural measures of center and spread. In such cases, we can use the
standard Normal curve and Table A to calculate areas.

Here is a review list of the most important skills you should have developed
from your study of this chapter.
A. MEASURING LOCATION IN A DISTRIBUTION
1. Use percentiles to locate individual values within a distribution of data.
2. Find the standardized value (z-score) of an observation. Interpret z-scores in
context.

B. DENSITY CURVES
1. Know that areas under a density curve represent proportions of all observations
and that the total area under a density curve is 1.
2. Approximately locate the median (equal-areas point) and the mean (balance
point) on a density curve.

C. NORMAL DISTRIBUTIONS
1. Recognize the shape of Normal curves and estimate by eye both the mean and
the standard deviation from such a curve.
2. Use the 68–95–99.7 rule and symmetry to state what percent of the observa-
tions from a Normal distribution fall between two points when both points lie at
the mean or one, two, or three standard deviations on either side of the mean.
3. Use Table A to find the percentile of a value from any Normal distribution and
the value that corresponds to a given percentile.

CHAPTER 3 REVIEW EXERCISES

3.47 Textbook costs Students taking a college introductory statistics class re-
ported spending an average of $305 on textbooks that quarter with a standard
deviation of $90. Here’s a rough sketch of a Normal density curve that fitted the
histogram well:

(a) Approximately what percent of the students spent between $215 and $395 on text-
books that quarter? Explain how you got this answer.
(b) One student spent $287 on textbooks. What was her standard score? What per-
cent of the students spent less than she did on textbooks that quarter?
3.48 Standardized test scores as percentiles Joey received a report that he scored in
the 97th percentile on a national standardized math test but in the 72nd percentile on
the reading portion of the test.
138 CHAPTER 3 Modeling Distributions of Data

(a) Explain to Joey’s grandmother, who knows little statistics, what these numbers
mean.
(b) Can we determine Joey’s z-scores for his math and reading performance? Why or
why not?
3.49 Finding areas Use Table A to find the proportion of observations from a
standard Normal distribution that falls in each of the following regions. In each case,
sketch a standard Normal curve and shade the area representing the region. Use an
applet or your calculator to check your answers.
(a) z $ 22.25 (b) 22.25 , z , 1.77
(c) z , 2.89 (d) 1.44 , z , 2.89
3.50 Finding z-scores Use Table A to find the value z from a standard Normal dis-
tribution that satisfies each of the following conditions. In each case, sketch a standard
Normal curve with your value of z marked on the axis. Use an applet or your calculator
to check your answers.
(a) The point z with 70% of the observations falling below it.
(b) The point z such that 90% of all observations are greater than z.
(c) The 46th percentile of the standard Normal distribution.
3.51 Low-birth-weight babies Researchers in Norway analyzed data on the birth
weights of 400,000 newborns over a 6-year period. The distribution of birth weights
is approximately Normal with a mean of 3668 grams and a standard deviation of
511 grams.6 Babies that weigh less than 2500 grams at birth are classified as “low
birth weight.”
(a) What percent of babies will be identified as low birth weight? Show your work.
(b) Find the quartiles of the birth weight distribution. Show your work.
3.52 Bacteria in milk A study of bacterial contamination in milk counted the number
of coliform bacteria per milliliter in 100 specimens of milk purchased in East Coast
grocery stores. The U.S. Public Health Service recommends no more than 10 coliform
bacteria per milliliter. Here are the data:

5 8 6 7 8 3 2 4 7 8 6 4 4 8 8 8 6 10 6 5
6 6 6 6 4 3 7 7 5 7 4 5 6 7 4 4 4 3 5 7
7 5 8 3 9 7 3 4 6 6 8 7 4 8 5 7 9 4 4 7
8 8 7 5 4 10 7 6 6 7 8 6 6 6 0 4 5 10 4 5
7 9 8 9 5 6 3 6 3 7 1 6 9 6 8 5 2 8 5 3

(a) Enter the data into your calculator and plot a histogram. Does the distribution of
coliform counts appear to be approximately Normal?
(b) Calculate the mean and standard deviation for the distribution.
(c) What percent of the observations fall within one, two, and three standard devia-
tions of the mean?
3.53 Density curves Figure 3.27 (on the facing page) shows density curves of two
different shapes. Briefly describe the overall shape of each distribution. Two or three
points are marked on each curve. The mean and the median are among these points.
For each curve, which point is the median and which is the mean?
 Chapter 3 Review Exercises 139

(a) (b)

A B C AB

Figure 3.27 Two density curves of different shapes.

3.54 Helmet sizes The army reports that the distribution of head circumference among
soldiers is approximately Normal with mean 22.8 inches and standard deviation 1.1
inches. Helmets are mass-produced for all except the smallest 5% and the largest 5% of
head sizes. Soldiers in the smallest or largest 5% get custom-made helmets. What head
sizes get custom-made helmets? Show your work.

FOXTROT © 2008 Bill Amend. Reprinted with permission of Universal Press Syndicate. All rights reserved.

3.55 The stock market The annual rate of return on stock indexes (which combine
many individual stocks) is very roughly Normal. Since 1945, the Standard & Poor’s 500
index has had a mean yearly return of 12%, with a standard deviation of 16.5%. Take
this Normal distribution to be the distribution of yearly returns over a long period.
(a) In what range do the middle 95% of all yearly returns lie? Explain.
(b) The market is down for the year if the return on the index is less than zero. In
what proportion of years is the market down? Show your work.
(c) In what proportion of years does the index gain between 15% and 25%? Show
your work.
140 CHAPTER 3 Modeling Distributions of Data

3.56 Do you sudoku? In the chapter opening story (page 101), one of the authors
played an online game of sudoku. At the end of his game, this graph was displayed.
The density curve shown was constructed from a histogram of times from 4,000,000
games played in one week at this Web site.

(a) How would you describe the shape of the density curve? Explain why this shape
makes sense in this setting.
(b) Use what you have learned in this chapter to describe the author’s performance in
a few sentences.
 CHAPTER 4

Describing
Relationships
4.1 Scatterplots and Correlation
4.2 Regression and Prediction

How faithful Is Old Faithful?

A geyser is a natural phenomenon: a type of hot spring that
shoots hot water and steam into the air. One of the best-

Mark Rassmussen/FeaturePics
known geysers is Old Faithful, located in Yellowstone National
Park in Wyoming. Accurate predictions of eruption times
allow the National Park Service to inform visitors of the
approximate time of the next eruption. Visitors can then adjust
their schedules accordingly.1 How can park geologists make
predictions with any certainty? To see how, let’s look at some data.
Figure 4.1 (on the following page) shows the distribution of intervals
between eruptions of Old Faithful in July 1995. The shortest interval was
47 minutes and the longest was 113 minutes. That’s quite a range! The distribution
has an unusual shape: it has two clear peaks—one at about 60 minutes and the
other at about 90 minutes. Let’s put this in context. On many occasions, the time
between eruptions of Old Faithful is about an hour. Even more often, the interval
between eruptions is around 90 minutes. Can park geologists make predictions
based on this double-peaked distribution?
If geologists predict a 60-minute gap between eruptions, but the actual
interval is 90 minutes, some visitors will become impatient. If they predict a
90-minute interval, but the gap is closer to 60 minutes, some visitors may go
elsewhere in the park and return after the eruption has occurred. What should
the geologists do?
Later in the chapter, we’ll answer this question. For now, keep this in mind:
to understand one variable, you often have to look at how it is related to other
variables.

141
142 CHAPTER 4 Describing Relationships

50

40

Frequency
30

20

10

0
40 50 60 70 80 90 100 110 120
Interval (minutes)

Figure 4.1 Histogram of the interval (in minutes) between eruptions of the Old Faithful
geyser in July 1995.

4.1 Scatterplots and Correlation

ACTIVITY 4.1
Tedious tasks take time
Materials: Graphing calculator, graph paper, and cursor, then pressing M ¸. Clear lists L2
stopwatch for each pair of students through L6 in the same way.
Many jobs require employees to perform TI-Nspire: Open a new Lists & Spreadsheet page (press
repetitive tasks. Fast-food restaurants rely c and choose 3:Lists & Spreadsheet).
on their employees to fill customers’ orders 2. Your first task is to enter the digits 0, 1, 2, . . . , 9
quickly and accurately. Bookkeepers and in list L1/column A. Your partner will time how
accountants enter pages of numerical data into long it takes you to complete the task. When your
Excel spreadsheets. Elementary school students partner says “Go!” enter the digits one at a time
practice their multiplication facts. Cashiers at into the list. As soon as you have finished, say
retail stores determine the total price of goods “Done!” so your partner will stop timing. If you
that are purchased. Quality control inspectors entered all of the numbers correctly, record your
examine a sample of the items produced. In each time to the nearest second. If you made a mistake,
of these situations, it is important to know how repeat the task.
long the task will take. This Activity will give you
3. Your second task is to enter the numbers
a chance to see how long a tedious task
10, 11, 12, . . . , 19 in list L2/column B. As in the
takes you.
previous task, begin when your partner says “Go!”
1. TI-84: Clear your lists. Go into your Statistics/ and say “Done!” when you are finished. If you
List Editor (press A, then choose 1:Edit entered all the numbers correctly, record your time
...). Clear list L1 by highlighting L1 with your to the nearest second. Otherwise, repeat the task.
 4. 1 Scatterplots and Correlation 143

4. You can probably guess what the third, fourth, Total no. Time to
fifth, and sixth tasks are now. Task no. of digits typed nearest second

Task 3: Enter the numbers 100, 101, 102, . . . , 109 in 1 10

list L3/column C. 2 20
...
Task 4: Enter the numbers 1000, 1001, 1002, . . . , 1009
in list L4/column D.
Task 5: Enter the numbers 10000, 10001, 10002, . . . , 7. Make a scatterplot of your data as follows:
10009 in list L5/column E.
Task 6: Enter the numbers 100000, 100001, 100002, • On graph paper, draw x and y axes. Label the
. . . , 100009 in list L6/column F. x axis “Total digits typed” and the y axis “Time.”
Scale the horizontal axis by adding tick marks at 0,
For each task, start typing when your partner says 10, 20, . . . , 70. Scale the vertical axis in half-second
“Go!” and be sure to say “Done!” when you finish. increments beginning with 0.
Once you enter all the numbers correctly for each
task, record your time to the nearest second. • Plot each of your six data points on the graph.

5. Switch roles with your partner and repeat Steps 2 8. Describe what the graph tells you about the
through 4. relationship between the two variables.
6. Summarize the data for your six tasks in a table 9. Compare graphs with your partner. What
like this one: similarities and differences do you see?

A medical study finds that short women are more likely to have heart attacks than
women of average height, while tall women have the fewest heart attacks. An in-
surance group reports that heavier cars have fewer deaths per 10,000 vehicles reg-
istered than lighter cars do. These and many other statistical studies look at the
relationship between two variables. To understand such a relationship, we must
often examine other variables as well. To conclude that shorter women have higher
risk of heart attacks, for example, the researchers had to eliminate the effect of other
variables such as weight and exercise habits.
Our topic in this chapter is relationships between variables. One of our main
themes is that the relationship between two variables can be strongly influenced by
other variables that are lurking in the background. Most statistical studies examine
data on more than one variable. Fortunately, statistical analysis of several-variable
data builds on the tools we used to examine individual variables. The principles
that guide our work also remain the same:
• First plot the data, then add numerical summaries.
• Look for overall patterns and deviations from those patterns.
• When the overall pattern is quite regular, there is sometimes a way to
describe it very briefly.

Scatterplots
The most common way to display the relationship between two quantitative vari-
ables is a scatterplot.
144 CHAPTER 4 Describing Relationships

Scatterplot

A scatterplot shows the relationship between two quantitative vari-

ables measured on the same individuals. The values of one variable ap-
pear on the horizontal axis, and the values of the other variable appear
on the vertical axis. Each individual in the data appears as a point in
the plot fixed by the values of both variables for that individual.

Here’s an example that shows how a scatterplot can illuminate the relationship
between two variables.

Example 4.1 Waiting on Old Faithful

In July 1995, the Starnes family visited Yellowstone in hopes of seeing Old Faith-
ful erupt. They had only about four hours available in the park before they had to
drive on to their hotel for the night. When they pulled into the parking lot at Old
Faithful, a large crowd of people was headed back to their cars from the geyser. Old
Faithful had just finished erupting.
How long would it be until the next eruption? From the histogram in the
opening story, Figure 4.1 (page 142), we see that the Starnes family might have to
wait anywhere from about 50 to about 100 minutes. Can we narrow it down a bit?
Figure 4.2 helps us do that.

115

105

95
Interval (minutes)

85

75

65

55

45

1 2 3 4 5
Duration (minutes)

Figure 4.2 Scatterplot of the interval between eruptions of Old Faithful against the
duration of the previous eruption.

Figure 4.2 is a scatterplot that plots the interval between successive eruptions
of Old Faithful on the vertical axis against the duration of the previous eruption
 4. 1 Scatterplots and Correlation 145

on the horizontal axis. The plot clearly shows two groups of eruptions. In one
group, the duration of the eruptions is about 2 minutes. In the other group of
eruptions, most of them last at least 4 minutes. The plot as a whole shows that
the interval between eruptions increases as the duration of the previous eruption
increases.
Since a long eruption depletes more water from a geyser’s reservoir than a short
eruption, we would expect that it would take more time to refill the reservoir after
a longer eruption. Consequently, we would expect a longer time interval until the
next eruption. Thus, the physical properties of the geyser suggest a relationship
between the duration of an eruption and the time until the next eruption. Such a
relationship may be used to predict geyser eruption times.

The scatterplot in Figure 4.2 shows how the interval between eruptions of Old
Faithful is related to the duration of the previous eruption. We think that “dura-
tion” will help explain “interval.” That is, “duration” is the explanatory variable and
“interval” is the response variable. We want to see how the interval changes when
duration changes, so we put duration (the explanatory variable) on the horizontal
axis. We can then see that as the duration of the previous eruption increases, the
interval between eruptions increases. Each point on the plot represents one erup-
tion. For one eruption that lasted 1.6 minutes, there was an interval of 74 minutes
until the next eruption. Find 1.6 on the x (horizontal) axis to locate this eruption.

Explanatory and response variables

A response variable is a variable that measures an outcome or result

of a study.

An explanatory variable is a variable that we think explains or causes

changes in the response variable.

Always plot the explanatory variable, if there is one, on the horizontal axis (the
x axis) of a scatterplot. As a reminder, we usually call the explanatory variable x and
the response variable y. If there is no explanatory-response distinction, either vari-
able can go on the horizontal axis.

©2005 Zits Partnership, distributed by King Features Syndicate

146 CHAPTER 4 Describing Relationships

Example 4.2 Heavy backpacks

Ninth-grade boys at the Webb School go on a backpacking trip each fall. Students
are divided into hiking groups of size 8 by selecting names from a hat. Before
leaving, students and their backpacks are weighed. Here are data from one hiking
group in a recent year:

Body weight (lb): 120 187 109 103 131 165 158 116
Backpack weight (lb): 26 30 26 24 29 35 31 28

Let’s make a scatterplot to investigate the relationship between body weight and
pack weight. To construct a scatterplot:
Step 1: Identify explanatory and response variables. The amount of weight a stu-
dent can comfortably carry depends on his body weight. Therefore, we’ll use
body weight as the explanatory variable and backpack weight as the response
variable.
Step 2: Label and scale axes. We put the explanatory variable, body weight, on the
horizontal axis and the response variable, pack weight, on the vertical axis. Based
on the data in the table, we decide to start labeling the horizontal axis at 100 pounds,
with tick marks every 10 pounds. For the vertical axis, we start at 20 pounds and
add tick marks every 2 pounds.
Step 3: Plot individual data values. The first student in the group weighs 120 pounds
and his pack weighs 26 pounds. We plot this point directly above 120 on the hori-
zontal axis and to the right of 26 on the vertical axis, as shown in Figure 4.3. For the
second student in the group, we add the point (187, 30) to the graph. By adding the
points for the remaining six students in the group, we get the completed scatterplot
shown in Figure 4.3.

36
34
32
Pack weight (lb)

30
28
26
24
Figure 4.3 Com- 22
pleted scatterplot of 20
body weight and
backpack weight for 100 110 120 130 140 150 160 170 180 190
the 8 students. Body weight (lb)

In general, it appears as though lighter students are carrying lighter backpacks

and heavier students are carrying heavier packs.
 4. 1 Scatterplots and Correlation 147

Exercises 4.1 Explanatory or response, I In each of the following situations, is it more

reasonable to simply explore the relationship between the two variables or to
view one of the variables as an explanatory variable and the other as a response
variable? In the latter case, which is the explanatory variable and which is the
response variable?
(a) The amount of time spent studying for a statistics exam and the grade on the exam
(b) The weight in kilograms and height in centimeters of a person
(c) Inches of rain in the growing season and the yield of corn in bushels per acre
(d) A student’s scores on the SAT Math test and the SAT Critical Reading test
4.2 Explanatory or response, II In each of the following situations, is it more
reasonable to simply explore the relationship between the two variables or to
view one of the variables as an explanatory variable and the other as a response
variable? In the latter case, which is the explanatory variable and which is the
response variable?
(a) A family’s income and the years of education their eldest child completes
(b) Price of a house and square footage of the house
(c) The arm span and height of a person
(d) Amount of snow in the Colorado mountains and the volume of water in
area rivers
4.3 IQ and GPA Figure 4.4 is a scatterplot of school grade point average versus
IQ score for all 78 seventh-grade students in a rural midwestern school.

10

8
C
School GPA

2
B

A
Figure 4.4 School 0
grade point averages and IQ 70 80 90 100 110 120 130 140
test scores for 78 seventh- IQ test score
grade students.

(a) Describe the overall pattern of the relationship in words. (Points A, B, and C
might be called outliers.)
(b) About what are the IQ and GPA for Student A?
(c) For each point A, B, and C, say how it is unusual (for example, “low GPA but a
moderately high IQ score”).
148 CHAPTER 4 Describing Relationships

4.4 SAT scores by state More than 1.5 million students took the SAT during the
2007–2008 school year. For the entire group of test takers, the mean score on the
Math test was 515. If we look at the data on a state-by-state basis, we see quite a
different picture. Figure 4.5 is a scatterplot of the percent of high school graduates
who took the SAT and the mean Math test score in each state.

625
600

Mean Math score

575
550
525
500
475
450
0 10 20 30 40 50 60 70 80 90
Percent taking SAT
Figure 4.5 Scatterplot showing the mean SAT Math score and the percent of high
school graduates who took the SAT in each state during the 2007–2008 school year.

(a) Describe the overall relationship between the two variables. Why are there two
distinct clusters of points?
(b) Two states stand out in Figure 4.5: West Virginia at (19, 501) and Maine at (87, 466).
In what way is each of these states “unusual”?
4.5 Healthy breeding Often the percent of an animal species in the wild
that survive to breed again is lower following a successful breeding season.
This is part of nature’s self-regulation, tending to keep population size stable.
A study of merlins (small falcons) in northern Sweden observed the number
of breeding pairs in an isolated area and the percent of males (banded for
identification) who returned the next breeding season. Here are the data for
nine years:2

Breeding pairs Percent of males returning

28 82
29 83, 70, 61
30 69
32 58
33 43
38 50, 47

(a) Why is the response variable the percent of males that return rather than the
number of males that return?
(b) Make a well-labeled scatterplot.
(c) Describe the pattern you see. Do the data support the theory that a smaller
percent of birds survive following a successful breeding season?
 4. 1 Scatterplots and Correlation 149

4.6 Fast cars Interested in a sporty car? Worried that it might use too much
gas? The Environmental Protection Agency lists most such vehicles in its
“minicompact” or “two-seater” categories. Table 4.1 gives city and highway gas
mileages (in miles per gallon) for all model year 2009 cars in these two groups.

Table 4.1 Gas Mileages (mpg) for Model Year 2009 Cars

Mini/subcompact cars Two-seater cars

Model City Highway Model City Highway

Audi TT coupe 23 31 Aston Martin DBS Coupe 17 24

BMW 328CI Convertible 18 27 Aston Martin V8 Vantage 13 19
BMW 335CI Convertible 17 26 Audi TT Roadster 22 30
BMW M3 Convertible 14 20 Cadillac XLR 14 23
Jaguar XK Convertible 16 25 Chevrolet Corvette 15 25
Jaguar XKR Convertible 15 23 Dodge Viper 13 22
Mercedes-Benz CLK350 17 25 Ferrari GTB Fiorano 11 15
Mercedes-Benz CLK550 15 22 Ferrari F430 11 16
Mitsubishi Eclipse Spyder 19 26 Honda S2000 18 25
Porsche 911 Carrera 18 26 Lamborghini Gallardo 14 20
Porsche 911 Turbo 15 23 Coupe
Mercedes-Benz SL500 13 21
Mercedes-Benz SL600 11 18
Mercedes-Benz SLK320 19 26
Nissan 350Z Roadster 17 23
Pontiac Solstice 19 27
Porsche 911 GT2 16 23
Saturn Sky 19 27
Smart fortwo convertible 33 41
Spyker C8 13 18
Source: Environmental Protection Agency, 2008 Fuel Economy Guide, www.fueleconomy.gov.

(a) Make a scatterplot that shows the relationship between city and highway
mileage for minicompact cars using city mileage as the explanatory variable. Be
sure to label your axes.
(b) On the same graph, make a scatterplot that shows the relationship between
city and highway mileage for two-seater cars. Use a different color or plotting
symbol.
(c) Describe what you see in the scatterplot. Is the form of the relationship
similar for the two types of car? What is the most important difference between
the two types?
150 CHAPTER 4 Describing Relationships

Interpreting scatterplots
To interpret a scatterplot, apply the usual strategies of data analysis.

Examining a scatterplot

As in any graph of data, look for the overall pattern and for striking
deviations from that pattern.

You can describe the overall pattern of a scatterplot by the direction,

form, and strength of the relationship.

An important kind of deviation is an outlier, an individual value that

falls outside the overall pattern of the relationship.

Example 4.3 Health and wealth

The Gapminder Web site, www.gapminder.org, provides loads of data on the health
www
and well-being of the world’s inhabitants. Figure 4.6 is a scatterplot of data from
Gapminder.3 The individuals are all the world’s nations for which data are available.
The explanatory variable is a measure of how rich a country is: the income per per-
son, also known as the gross domestic product (GDP) per person. GDP is the total
value of the goods and services produced in a country, converted into dollars. The
response variable is life expectancy at birth.

Japan Qatar

United States
Gabon

Equatorial Guinea

Botswana

South Africa

Figure 4.6 Scatterplot of the life expectancy of people in many nations against each
nation’s gross domestic product per person.
 4. 1 Scatterplots and Correlation 151

We expect people in richer countries to live longer. The overall pattern of

the scatterplot does show this, but the relationship has an interesting shape. Life
expectancy rises very quickly as GDP increases and then levels off. People in very
rich countries such as the United States live no longer than people in poorer but
not extremely poor nations. In some less wealthy countries, such as Japan, people
live longer than in the United States.
Four African nations are outliers. Their life expectancies are similar to those of their
neighbors but their GDP is higher. Gabon and Equatorial Guinea produce oil, and South
Africa and Botswana produce diamonds. It may be that income from mineral exports
goes mainly to a few people and so pulls up GDP per person without much effect on
either the income or the life expectancy of ordinary citizens. That is, GDP per person is a
mean, and we know that mean income can be much higher than median income.

Both Figures 4.2 (page 144) and 4.6 have a clear direction: the time between
eruptions goes up as the length of the previous eruption increases, and life expec-
tancy generally goes up as GDP increases. We say that Figures 4.2 and 4.6 show a
positive association between the variables.

Positive association, negative association

Two variables are positively associated when above-average values

of one tend to accompany above-average values of the other and
below-average values also tend to occur together. The scatterplot
slopes upward as we move from left to right.

Two variables are negatively associated when above-average values

of one tend to accompany below-average values of the other, and vice
versa. The scatterplot slopes downward from left to right.

Each of our scatterplots has a distinctive form. Figure 4.2 shows two clusters of
points, and Figure 4.6 shows a curved relationship. The strength of a relationship in
a scatterplot is determined by how closely the points follow a clear form. The rela-
tionships in Figures 4.2 and 4.6 are not strong. Eruptions of similar lengths show
quite a bit of scatter in times until the next eruption, and nations with similar GDPs
can have quite different life expectancies. Here is an example of two variables with
a strong negative relationship and a simple form.

Example 4.4 How much natural gas does a

household use?
Joan is concerned about the amount of energy she uses to heat her home in the
Midwest. She keeps a record of the natural gas she consumes each month over
one year’s heating season. Because the months are not equally long, she divides
each month’s consumption by the number of days in the month to get the average
number of cubic feet of gas used per day. Demand for heating is strongly influenced
152 CHAPTER 4 Describing Relationships

by the outside temperature. From local weather records, Joan obtains the average
temperature for each month, in degrees Fahrenheit. Here are Joan’s data:

Month: Oct. Nov. Dec. Jan. Feb. Mar. Apr. May

Temperature, x: 49.4 38.2 27.2 28.6 29.5 46.4 49.7 57.1
Gas consumed, y: 520 610 870 850 880 490 450 250

A scatterplot of these data appears in Figure 4.7. Temperature is the explanatory

variable x (plotted on the horizontal axis) because outside temperature helps explain
gas consumption. The scatterplot shows a strong negative straight-line association.
The straight-line form is important because it is common and simple. The associa-
tion is strong because the points lie close to a line. It is negative because, as tempera-
ture increases, gas consumption goes down because less gas is used for heating.

1000

800
Gas consumed (cubic feet)

600

400

200

0
Figure 4.7 Home
20 25 30 35 40 45 50 55 60
consumption of natural gas Temperature (degrees Fahrenheit)
versus outdoor temperature.

Making scatterplots with technology is much easier than constructing them by hand.

CALCULATOR CORNER How to make a scatterplot

In this Calculator Corner, you will make a scatterplot of the natural-gas
consumption data from Example 4.4.
1. Enter the data in two lists.

TI-84
• Go into your Statistics/List Editor (press A, then choose 1:Edit...).
Clear lists L1 and L2 as you did in Activity 4.1.
• Enter the temperature data in list L1 and the corresponding gas consumption
values in list L2.
 4. 1 Scatterplots and Correlation 153

TI-Nspire
• Open a new Lists & Spreadsheet page (Press c and choose 3:Lists &
Spreadsheet).
• Name column A “Temp” and column B “Gas.” Enter the temperature data in
column A and the corresponding gas consumption values in column B.

2. Make a scatterplot.
TI-84
• Go to the statistics plots menu (press yo) and select 1:Plot1. Adjust
your settings as shown below.
• Make sure that Plot2 and Plot3 are Off and that all functions are deactivated in
the Y = menu. Then use ZoomStat (press q and choose 9:ZoomStat)
to produce the graph.

TI-Nspire
• Insert a Data & Statistics page (press /I and choose 5:Add Data &
Statistics).
• Use the NavPad to move your cursor to “Click to add variable” at the bottom
of the screen, and press the “mouse” button (a). From the drop-down menu,
choose Temp as the explanatory variable. A dotplot should appear. To choose
Gas as the response variable, move your cursor to the left-hand side of the
screen until you see “Click to add variable,” press a, and choose Gas from the
drop-down menu.

3. Adjust window settings as desired.

CALCULATOR CORNER (continued)
154 CHAPTER 4 Describing Relationships

CALCULATOR CORNER (continued)

TI-84
• Press p and modify the settings to values that seem reasonable. Press
s to obtain the updated scatterplot.
TI-Nspire
• Press b, choose 5:Window/Zoom, 1:Window Settings, enter
desired values, and choose OK. The graph should update automatically.

Exercises 4.7 Rich states, poor states One measure of a state’s prosperity is the median
income of its households. Another measure is the mean personal income
per person in the state. Figure 4.8 is a scatterplot of these two variables, both
measured in dollars.

90,000
Median household income (dollars)

80,000 Conn.

70,000
Utah D.C.
60,000

50,000

40,000

30,000
Figure 4.8 Median
household income and 20,000
mean income per person 20,000 30,000 40,000 50,000 60,000 70,000 80,000 90,000
in the states. Mean income per person (dollars)

(a) Explain why you expect a positive association between these variables. Also, explain
why you expect household income to be generally higher than income per person.
(b) Describe the overall pattern of the plot, ignoring the outliers.
(c) We have labeled some interesting states on the graph. Describe what’s unusual
about each one in terms of mean income per person and median household
income.
4.8 Crawl before you walk At what age do babies learn to crawl? Does it
take longer to learn in the winter, when babies are often bundled in clothes
that restrict their movement? Perhaps there might even be an association
between babies’ crawling age and the average temperature during the month
they first try to crawl (around six months after birth). Data were collected from
parents who brought their babies into the University of Denver Infant Study
Center to participate in one of a number of experiments. Parents reported the
 4. 1 Scatterplots and Correlation 155

birth month and the age at which their child was first able to creep or crawl
a distance of four feet within one minute. Information was obtained on 414
infants, 208 boys and 206 girls. Crawling age is given in weeks, and the average
temperature (in 8F) is for the month that is six months after the birth month.4
Average temp. (8F)
Birth month Average crawling age (wk) 6 mo after birth
January 29.84 66
February 30.52 73
March 29.70 72
April 31.84 63
May 28.58 52
June 31.44 39
July 33.64 33
August 32.82 30
September 33.83 33
October 33.35 37
November 33.38 48
December 32.32 57

(a) Make a scatterplot that displays the relationship between average crawling age
and average temperature. Be sure to think about which is the explanatory variable.
(b) Describe the direction, form, and strength of the relationship. What do you
conclude about when babies learn to crawl?
4.9 Interested in stocks? When interest rates are high, investors may avoid
stocks because they can get high returns with less risk. Figure 4.9 plots the annual
returns on U.S. common stocks over a 50-year period against the returns on
Treasury bills for the same years.5 (The interest paid by Treasury bills is a measure
of how high interest rates were that year.)

60

50

40
Percent return on stocks

30

20

10

0
–10

–20

–30

Figure 4.9 Percent –40

returns on Treasury bills 0 2 4 6 8 10 12 14
and common stocks over a Percent return on Treasury bills
50-year period.
156 CHAPTER 4 Describing Relationships

(a) Approximately what were the highest and lowest annual percent returns on stocks
during this period? What were the highest and lowest returns for Treasury bills?
(b) Describe the pattern you see. Are high interest rates bad for stocks? Is the
relationship between interest rates and stock returns strong or weak?
4.10 Teen drivers Thirty students in a high school statistics class were asked to report
the age (in years) and odometer reading of their primary vehicle. Here are their data:6

Age Mileage Age Mileage Age Mileage

3 40,300 6 85,000 8 110,000
2 11,912 4 20,000 4 30,323
3 30,000 3 30,000 7 100,000
4 40,000 2 17,000 10 98,000
8 98,000 3 25,000 2 12,000
3 48,000 7 150,000 5 53,000
8 120,000 2 10,000 7 40,000
11 185,000 10 110,000 4 76,000
4 40,000 5 103,000 2 3,000
1 1,050 5 66,610 7 75,000

(a) Enter the data into your calculator lists. Refer to the Calculator Corner (page 152).
(b) Make a scatterplot of the data. Adjust the viewing window, then transfer your
graph to paper.
(c) Describe what the scatterplot tells you about the direction, form, and strength
of the relationship. Are there any outliers?
4.11 NAEP for fourth-graders The National Assessment of Educational
Progress (NAEP) assesses what students know in several subject areas based on
large representative samples. Table 4.2 (on the facing page) reports some findings
of the NAEP year 2007 Mathematics Assessment for fourth-graders in the United
States. For each state we give the mean NAEP math score (out of 500) and also the
percent of students who were at least “proficient” in the sense of being able to use
math skills to solve real-world problems. Nationally, about 25% of students are
“proficient” by NAEP standards. We expect that average performance and percent
of proficient performers will be strongly related.
(a) Make a scatterplot, using mean NAEP scores as the explanatory variable.
Notice that there are several pairs of states with identical values. Use a different
symbol for points that represent two states.
(b) Describe the direction, form, and strength of the relationship.
(c) Circle your home state’s point in the scatterplot. Although there are no clear
outliers, there are some points that you may consider interesting, perhaps because
they are on the edge of the pattern. Choose one such point: which state is this, and
in what way is it interesting?
4.12 NAEP and poverty Refer to the previous exercise. What is the
relationship between the percent of students in a state who were eligible for
free or reduced-price school lunches and the percent who scored proficient
on the NAEP mathematics test? Support your answer with a well-labeled
scatterplot.
 4. 1 Scatterplots and Correlation 157

Table 4.2 State Performance on the 2007 NAEP Mathematics Assessment of Fourth-Graders

Mean Mean
NAEP Percent Percent free/ NAEP Percent Percent free/
State score proficient reduced lunch State score proficient reduced lunch

AL 229 23 55 MT 244 39 38
AK 237 32 44 NE 238 33 39
AZ 232 27 52 NV 232 27 45
AR 238 32 57 NH 249 44 19
CA 230 25 53 NJ 249 42 29
CO 240 35 40 NM 228 22 67
CT 243 37 31 NY 243 37 49
DE 242 36 39 NC 242 35 48
FL 242 34 48 ND 245 41 32
GA 235 28 52 OH 245 39 37
HI 234 29 42 OK 237 30 55
ID 241 35 44 OR 236 31 44
IL 237 31 44 PA 244 40 35
IN 245 40 41 RI 236 31 40
IA 243 38 34 SC 237 31 53
KS 248 42 41 SD 241 37 36
KY 235 27 53 TN 233 26 49
LA 230 22 70 TX 242 35 55
ME 242 36 36 UT 239 35 37
MD 240 32 34 VT 246 42 31
MA 252 47 27 VA 244 35 30
MI 238 32 38 WA 243 37 39
MN 247 41 30 WV 236 30 50
MS 228 20 69 WI 244 40 34
MO 239 33 42 WY 244 40 36
Source: National Center for Education Statistics Web site, nces.ed.gov/nationsreportcard/.

Correlation
A scatterplot displays the direction, form, and strength of the relationship between
two variables. Straight-line relations are particularly important because a straight
line is a simple pattern that is quite common. A straight-line relation is strong if the
points lie close to a straight line, and it is weak if they are widely scattered about a
line. Our eyes are not good judges of how strong a relationship is. The two scatter-
plots in Figure 4.10 depict the same data, but the right-hand plot is drawn smaller
158 CHAPTER 4 Describing Relationships

in a large field. The right plot seems to show a stronger straight-line relationship.
Our eyes can be fooled by changing the plotting scales or the amount of blank space
around the cloud of points in a scatterplot.

80 120

100
70
80
60
60
50
40

40 20

30 0

30 40 50 60 70 80 0 20 40 60 80 100 120

Figure 4.10 Two scatterplots of the same data. The right-hand plot suggests a stronger
relationship between the variables because of the surrounding space.

We need to follow our strategy for data analysis by using a numerical measure
to supplement the graph. Correlation is the measure we use in this setting.

Correlation

The correlation describes the direction and strength of a straight-line

relationship between two quantitative variables. Correlation is usually
written as r.

To calculate correlation, use the formula

x2x y2y
ca ba bd
1
r5 a
n21 sx sy

Calculating a correlation takes a bit of work. You can usually think of r as the
result of pushing a calculator button or giving a command in software and concen-
trate on understanding its properties and use. Knowing how we obtain r from data
does help us understand how correlation works, however, so here we go.

Example 4.5 Back to the backpackers

In Example 4.2 (page 146), we examined data on the relationship between students’ body
© Monkeybusiness/Dreamstime.com weights and their backpack weights prior to a hiking trip. Here are the data once again:
 4. 1 Scatterplots and Correlation 159

Body weight (lb): 120 187 109 103 131 165 158 116
Backpack weight (lb): 26 30 26 24 29 35 31 28

To calculate the correlation:

Step 1: Find the mean and standard deviation for both x and y. Use your calculator
to check that the mean and standard deviation for each of the two variables are as
follows:

Body weight: x 5 136.125 pounds sx 5 30.296 pounds

Pack weight: y 5 28.625 pounds sy 5 3.462 pounds

Step 2: Standardize the x- and y-values. Using the means and standard deviations
from Step 1, find the standardized score (z-score) for each x-value and for each
y-value. (Note that we used the unrounded values of zx and zy to compute the
products in the final column.)

Value Standardized score Value Standardized score Product of

of x zx  (x 2 x)ysx of y zy  (y 2 y)ysy standardized scores
120 1120 2 136.125 2 y30.296 5 20.532 26 1 26 2 28.625 2 y3.462 5 20.758 1 20.532 2 1 20.758 2 5 0.404
187 1 187 2 136.125 2 y30.296 5 1.679 30 130 2 28.625 2 y3.462 5 0.397 1 1.679 2 1 0.397 2 5 0.667
109 1109 2 136.125 2 y30.296 5 20.895 26 126 2 28.625 2 y3.462 5 20.758 1 20.895 2 1 20.758 2 5 0.679
103 1103 2 136.125 2 y30.296 5 21.093 24 124 2 28.625 2 y3.462 5 21.336 121.093 2 1 21.336 2 5 1.461
131 1131 2 136.125 2 y30.296 5 20.169 29 129 2 28.625 2 y3.462 5 0.108 120.169 2 1 0.108 2 5 20.018
165 1165 2 136.125 2 y30.296 5 0.953 35 135 2 28.625 2 y3.462 5 1.841 1 0.953 2 1 1.842 2 5 1.755
158 1158 2 136.125 2 y30.296 5 0.722 31 131 2 28.625 2 y3.462 5 0.686 1 0.722 2 1 0.686 2 5 0.495
116 1116 2 136.125 2 y30.296 5 20.664 28 128 2 28.625 2 y3.462 5 20.181 120.664 2 120.181 2 5 0.120
Sum  5.563

The first hiker’s body weight is about one-half standard deviation below the
mean body weight for the group of hikers. His pack weight is about three-fourths
of a standard deviation below the mean backpack weight for the group. See if you
can describe how the second hiker is “unusual” using the standardized values of his
body weight and pack weight.
Step 3: Multiply the standardized scores for each individual, then “average” the
products. As with the standard deviation, we “average” by dividing by n  1,
one fewer than the number of individuals. In this case, since n  8, we divide
by 7:

x2x y2y
ca ba bd 5 1 5.563 2 5 0.795
1 1
r5 a
n21 s x s y 7

Notice that all the products are positive except for the hiker we identify in Figure 4.11
(on the next page), who had slightly below-average weight in the group but a pack
weight that was slightly above average.
160 CHAPTER 4 Describing Relationships

36
34
32

Pack weight (lb)

30
28
26
24
Figure 4.11 Scatterplot
of backpack weight versus 22
body weight for 8 hikers 20
with reference lines added
at the mean x-value and 100 110 120 130 140 150 160 170 180 190
mean y-value. Body weight (lb)

Understanding correlation
More important than calculating r (a task better suited to a machine) is under-
standing how correlation measures association. Here are the facts:
• Positive r indicates positive association between the variables, and negative r
indicates negative association. Figure 4.11 shows the original scatterplot of the
backpacker data from Example 4.5 with two reference lines added: a horizontal
line at the mean y-value, y 5 28.625, and a vertical line at the mean x-value,
x 5 136.125. The scatterplot reveals a moderately strong positive association
between body weight and backpack weight. Three hikers have above-average
body weights and pack weights, so their standardized scores are positive for both
x and y. Four hikers have below-average body weights and backpack weights, so
both standardized scores are negative. The products of these standardized scores
are all positive, however. Only one hiker contributes a small negative product of
standardized scores—the one with slightly below-average body weight but
slightly above-average pack weight.
With so many positive products, we
get a positive value for r.
• The correlation r always falls
between 21 and 1. Values of r near
0 indicate a very weak straight-line
relationship. The strength of the
relationship increases as r moves
away from 0 toward either 1 or 1.
Values of r close to 1 or 1 indicate
that the points lie close to a straight
line. The extreme values r  1 and
r  1 occur only when the points
in a scatterplot lie exactly along a
straight line.
The result r 5 0.795 in Example
4.5 reflects the moderately strong, posi-
"He says we've ruined his positive correlation between height and weight." tive, straight-line pattern in Figure 4.11.
 4. 1 Scatterplots and Correlation 161

Correlation r = 0 Correlation r = –0.3 Correlation r = 0.5

Figure 4.12 How

correlation indicates the
direction and strength of a
straight-line relationship.
Patterns closer to a straight Correlation r = –0.7 Correlation r = 0.9 Correlation r = –0.99
line have correlations
closer to 1 or 1.

The scatterplots in Figure 4.12 illustrate how r indicates both the direction
and the strength of a straight-line relationship. Study them carefully. Note
that the sign of r matches the direction of the slope in each plot, and that r
approaches 1 or 1 as the pattern of the plot comes closer to a straight line.

www
How good are you at estimating the correlation by eye for a scatterplot? Try the
Correlations game at www.stat.uiuc.edu/courses/stat100/cuwu/Games.html.
• Because r uses standardized scores for the observations, the correlation does
not change when we change the units of measurement of x, y, or both.
Measuring weight in kilograms rather than pounds in Example 4.5 would not
change the correlation r 5 0.795.
Our descriptive measures for one variable all share the same units as the original
observations. If we measure length in centimeters, the median, quartiles, mean, and
standard deviation are all in centimeters. The correlation between two variables, how-
ever, has no unit of measurement; it is just a number between 1 and 1.
• Correlation ignores the distinction between explanatory and response
variables. If we reverse our choice of which variable to call x and which to call
y, the correlation does not change. Can you see why from the formula?
x2x y2y
ca ba bd
1
r5 a
n21 sx sy

• Correlation measures the strength of only straight-line association between

two variables. Correlation does not describe curved relationships between
variables, no matter how strong they are.
• Like the mean and standard deviation, the correlation is strongly affected
by a few outlying observations. Use r with caution when outliers appear
in the scatterplot. Look, for example, at Figure 4.13. We removed the point
162 CHAPTER 4 Describing Relationships

corresponding to the hiker who weighed 187 pounds but whose pack
weighed only 30 pounds. The remaining 7 points have a stronger straight-
line relationship. Deleting one unusual point increases the correlation from
r 5 0.795 for the original data to r 5 0.946.

36
34
32

Pack weight (lb)

30
28
26
After you plot your
24
data, think!
22
Abraham Wald (1902–1950),
like many statisticians, 20
worked on war problems
during World War II. Wald 100 110 120 130 140 150 160 170 180 190
invented some statistical Body weight (lb)
methods that were military Figure 4.13 Hiker data from Figure 4.11 with the unusual point at (187, 30) deleted—
secrets until the war ended.
note the strong, positive, linear relationship for the remaining 7 points.
Here is one of his simpler
ideas. Asked where extra
armor should be added to There are many kinds of relationships between variables and many ways to
airplanes, Wald studied the measure them. Although correlation is very common, remember its limitations.
location of enemy bullet Correlation makes sense only for quantitative variables—we can speak of the
holes in planes returning
relationship between the sex of voters and the political party they prefer, but not
from combat. He plotted
the locations on an outline of the correlation between these variables. Even for quantitative variables such
of the plane. As data accu- as the length of bones, correlation measures only straight-line association.
mulated, most of the outline Remember also that correlation is not a complete description of two-variable
filled up. Put the armor in data, even when there is a straight-line relationship between the variables. You
the few spots with no bul- should give the means and standard deviations of both x and y along with the cor-
let holes, said Wald. That’s
where bullets hit the planes
relation. Because the formula for correlation uses the means and standard devia-
that didn’t make it back. tions, these measures are the proper choice to accompany a correlation.

Exercises 4.13 Dating and height, I A college student wonders if tall women tend to date
taller men than do short women. She measures herself, her dormitory roommate,
and the women in the adjoining rooms; then she measures the next man each
woman dates. Here are the data (heights in inches):

Women (x): 66 64 66 65 70 65
Men (y): 72 68 70 68 71 65

(a) Make a scatterplot of these data. Based on the scatterplot, do you expect the
correlation to be positive or negative? Near 61 or not?
(b) Find the correlation r between the heights of the men and women. Follow the
method of Example 4.5.
 4. 1 Scatterplots and Correlation 163

4.14 Dating and height, II Refer to the previous exercise.

(a) How would r change if all the men were 6 inches shorter than the heights
given in the table? Does the correlation tell us whether women tend to date men
taller than themselves?
(b) If heights were measured in centimeters rather than inches, how would the
correlation change? (There are 2.54 centimeters in an inch.)
(c) If every woman dated a man exactly 3 inches taller than herself, what would be
the correlation between male and female heights?
4.15 Explaining correlation You have data on the current grade point average
(GPA) of students in a basic statistics course and their scores on the first
examination in the course. Say as specifically as you can what the correlation r
between GPA and exam score measures.
4.16 What correlation doesn’t measure
(a) Make a scatterplot of the following data:

x: 5 3 0 3 5
y: 0 4 5 4 0

(b) Show that the correlation is zero. Use the method of Example 4.5.
(c) The scatterplot shows a strong association between x and y. Explain how it can
happen that r  0 in this case.
4.17 Rank the correlations Consider each of the following relationships: the
heights of fathers and the heights of their adult sons, the heights of husbands and
the heights of their wives, and the heights of females at age 4 and their heights
at age 18. Rank the correlations between these pairs of variables from highest to
lowest. Explain your reasoning.
4.18 Matching correlations Figure 4.14 displays five scatterplots. Match each to
the r below that best describes it. (Some r’s will be left over.)
r 5 20.9 r 5 20.7 r 5 20.3 r50 r 5 0.3 r 5 0.7 r 5 0.9

(a) (b) (c)

Figure 4.14 Match the

(d) (e)
correlations.
164 CHAPTER 4 Describing Relationships

ws
Ne
ws
Ne

APPLICATION 4.1 B]
RO
g
<
Se
a

Cars and correlation

The scatterplots below show some data from Environmental Protection Agency (EPA) fuel economy
tests on 140 different cars that are from the same model year and that run on unleaded gas. Variables
that were measured for each car include make of car, horsepower, weight (in pounds), displacement (in
cubic inches), mpg:highway (highway driving miles per gallon), mpg:city (city driving miles per gallon),
and carbon dioxide emitted.7

(a) (b)

(c) (d)

QUESTIONS
1. Describe the direction, form, and strength of the 5. How would the correlation in plot (c) change if
relationship in each plot. the x and y variables were switched? Explain.
2. Estimate the value of the correlation between the 6. Predict how the following pairs of variables
variables in each plot. would be related. Include direction, form, and
3. For each of the relationships, tell whether there strength in your predictions.
are clear explanatory and response variables, or • Horsepower and mpg:city
whether either variable could be viewed as the • Make of car and weight
explanatory variable.
• Weight and amount of carbon dioxide emitted
4. How would the correlation between the variables
in plot (b) change if the weights were measured in
kilograms instead of pounds? Explain.
 4. 1 Scatterplots and Correlation 165

Section 4.1 Summary

Most statistical studies examine relationships between two or more variables. A
scatterplot is a graph of the relationship between two quantitative variables. If you
have an explanatory and a response variable, put the explanatory variable on the
x (horizontal) axis of the scatterplot.
When you examine a scatterplot, look for the direction, form, and strength
of the relationship and also for possible outliers. If there is a clear direction, is it a
positive association (the scatterplot slopes upward from left to right) or a negative
association (the plot slopes downward)? Is the form straight or curved? Are there
clusters of observations? Is the relationship strong (a tight pattern in the plot) or
weak (the points scatter widely)?
The correlation r measures the direction and strength of a straight-line rela-
tionship between two quantitative variables. Correlation is a number between 1
and 1. The sign of r shows whether the association is positive or negative. The value
of r gets closer to 1 or 1 as the points cluster more tightly about a straight line.
The extreme values 1 and 1 occur only when the scatterplot shows a perfectly
straight line.

Section 4.1 4.19 Born to be old? Is there a relationship between the gestational period (time
Exercises from conception to birth) of an animal and its average life span? Figure 4.15 is a
scatterplot of the gestational period and average life span for 43 species of animals.8

40 A C
Life span (years)

30

20

10 B

0
0 100 200 300 400 500 600 700
Gestation (days)
Figure 4.15 Scatterplot of the average life span and gestational period for 43 species of
animals.

(a) Describe the direction, form, and strength of the scatterplot.

(b) Three “unusual” points are labeled in Figure 4.15: Point A is for the hippopot-
amus, Point B is for the giraffe, and Point C is for the Asian elephant. In what way
is each of these animals “unusual”?
(c) If you were to add a point for humans to the plot, would it be part of the general
pattern, or would it be an unusual point? Explain.
4.20 IQ and siblings The correlation between a child’s score on the vocabulary
portion of the Wechsler Intelligence Scale for Children (a standard IQ test) and
the number of siblings a child has is r  0.319.9
166 CHAPTER 4 Describing Relationships

(a) Explain in words what this r says.

(b) Can you suggest an explanation for this relationship?
4.21 What number can I be?
(a) What are all the values that a correlation r can possibly take?
(b) What are all the values that a standard deviation s can possibly take?
4.22 Correlation “facts” Which of these statements are true and which are false?
(a) A correlation of 0.8 means that 80% of the points in the scatterplot lie on a line.
(b) If the correlation between two variables is zero, then there can be no relation-
ship between them.
(c) For all books in the Library of Congress, the correlation between the thickness
of the books (in inches) and their number of pages would be positive.
(d) For all of the cars registered in the state of Ohio, the correlation between their
fuel efficiency (in miles per gallon) and their weight (in pounds) would be positive.
(e) If the correlation between height in inches and weight in pounds for a group
of people is 0.7, then the correlation between their heights in centimeters and
their weights in kilograms will still be 0.7.
4.23 Pollution from vehicle exhaust Auto manufacturers are required to test
their vehicles for the amount of each of several pollutants in the exhaust. The
amount of a pollutant varies even among identical vehicles, so that several
vehicles must be tested. Figure 4.16 is a scatterplot of the amounts of two pollut-
ants, carbon monoxide and nitrogen oxides, for 46 vehicles of the same model.
Both variables are measured in grams of the pollutant per mile driven.10

3.0

2.5
Nitrogen oxides

2.0

1.5

1.0

0.5

0
5 10 15 20 25
Carbon monoxide

Figure 4.16 Amounts of two pollutants (measured in grams per mile driven) in the
exhaust of 46 vehicles of the same model.

(a) Describe the nature of the relationship. Is the association positive or negative?
Is the relation close to a straight line or clearly curved? Are there any outliers?
(b) A car magazine says, “When a car’s engine is properly tuned, it emits few
pollutants. If the engine is out of tune, it emits more of all pollutants. You can
 4. 1 Scatterplots and Correlation 167

find out how badly a car is polluting the air by measuring any one pollutant. If that
value is acceptable, the other emissions will also be OK.” Do the data in Figure 4.16
support this claim?
4.24 Guess the correlation For each of the following pairs of variables, would
you expect a large negative correlation, a large positive correlation, or a small
correlation? Justify your answers.
(a) The age of secondhand cars and their prices
(b) The weight of new cars and their gas mileages in miles per gallon
(c) The heights and the weights of adult men
(d) The heights and the IQ scores of adult men
(e) The heights of daughters and the heights of their mothers
4.25 Foot problems Metatarsus adductus (call it MA) is a turning in of the
front part of the foot that is common in adolescents and usually corrects itself.
Hallux abducto valgus (call it HAV) is a deformation of the big toe that is not
common in youth and often requires surgery. Perhaps the severity of MA can
help predict the severity of HAV. Table 4.3 gives data on 38 consecutive patients
who came to a medical center for HAV surgery.11 Using X-rays, doctors mea-
sured the angle of deformity for both MA and HAV. They speculated that there
is a positive association: more serious MA is associated with more serious HAV.

Table 4.3 The Severity of Metatarsus Adductus and Hallux Abducto

Valgus

HAV angle MA angle HAV angle MA angle HAV angle MA angle

28 18 21 15 16 10
32 16 17 16 30 12
25 22 16 10 30 10
34 17 21 7 20 10
38 33 23 11 50 12
26 10 14 15 25 25
25 18 32 12 26 30
18 13 25 16 28 22
30 19 21 16 31 24
26 10 22 18 38 20
28 17 20 10 32 37
13 14 18 15 21 23
20 20 26 16

(a) Use your calculator to help you make a scatterplot of the data. Which is the
explanatory variable?
(b) Describe the direction, form, and strength of the relationship between MA
angle and HAV angle. Are there any clear outliers in your graph?
(c) The correlation between HAV angle and MA angle is 0.30. What does this tell you?
(d) Do you think the data confirm the doctors’ speculation?
168 CHAPTER 4 Describing Relationships

4.26 Strong relationship, low correlation Make a scatterplot that shows a strong
relationship between two variables but a low correlation.

4.2 Regression and Prediction

ACTIVITY 4.2A
Vitruvius and the ideal man
Materials: Ruler or measuring tape for each small 3. On graph paper, make a scatterplot of the class’s
group of 2 to 3 students, graph paper measurements using the distance from the elbow to
the tip of the hand as the explanatory variable. Be
According to the
sure to label and scale your axes.
ancient architect
Vitruvius, the 4. Describe the direction, form, and strength of the
measurements of relationship between these variables. Estimate the
various parts of correlation from your scatterplot.
the human body
5. Enter the data into lists on your calculator and
have been set by
make a scatterplot. How does the calculator graph
nature to follow
compare with the one you made on paper? Adjust
certain ratios.
the viewing window to make the graphs appear
Leonardo da Vinci
more similar.
paid tribute to
these claims in his 6. Calculate x, sx, y, sy, and r using your calculator.
painting entitled
7. According to da Vinci, the distance from the
Vitruvian Man.
elbow to the tip of the hand should be one-fifth
Scala/Art Resource Here is a brief
of the height. That is, the height should be five
excerpt from the text accompanying da Vinci’s
times the distance measured on the arm. To see
painting: “The length of a man’s outspread arms
how close your class’s data come to this prediction,
is equal to his height. The greatest width of the
graph the line y 5 5x on top of your scatterplot as
shoulders contains in itself the fourth part of the
follows:
man. From the elbow to the tip of the hand will be
the fifth part of the man.” In this Activity, you will TI-84
make some measurements to determine whether Press o. Define Y1=5x and then press s.
da Vinci’s statements seem to accurately reflect
reality. TI-Nspire
On the Data & Statistics page, press b, choose
1. Measure the distance from the elbow to the tip of
4:Analyze, 4:Plot Function. Define
the hand (end of the middle finger) and the height
f1(x):=5x and press ·.
for each group member to the nearest inch. Record
your data. Describe what you see.
2. Your teacher will prepare a chart for 8. Based on your scatterplot, make a prediction for
recording your data. When you have finished the height of a student whose distance measured
measuring, enter your group members’ values in on the arm is 10 inches. How does your prediction
the chart. compare with those of your classmates?
 4. 2 Regression and Prediction 169

Regression lines
If a scatterplot shows a straight-line relationship between two quantitative variables,
we would like to summarize this overall pattern by drawing a line on the graph. A re-
gression line summarizes the relationship between two variables, but only in a specific
setting: when one of the variables helps explain or predict the other. That is, regression
describes a relationship between an explanatory variable and a response variable.

Regression line

A regression line is a straight line that describes how a response vari-

able y changes as an explanatory variable x changes. We often use a
regression line to predict the value of y for a given value of x.

Example 4.6 Return of the backpackers

In Examples 4.2 and 4.5, we observed a moderately strong, positive, linear relation-
ship between the body weights and backpack weights of 8 hikers. Figure 4.17 is
another scatterplot of the data, but this time with a regression line added. The line
gives a quick summary of the overall pattern.

35.0

32.5
Pack weight (lb)

30.0

27.5

Figure 4.17 Using a

straight-line pattern for 25.0
prediction. The data are
the body weights and
backpack weights of 8 100 110 120 130 140 150 160 170 180 190
hikers. Body weight (lb)

At the last minute, another hiker is assigned to the group. This student weighs
150 pounds. Can we predict this hiker’s backpack weight? The straight-line pattern
connecting body weight to pack weight is strong enough that it seems reasonable to
use body weight to predict backpack weight. Figure 4.17 shows how: starting at the
new hiker’s body weight (150 lb) on the x axis, go up to the regression line, then over
to the pack weight axis. We predict a backpack weight of about 30 pounds. This is the
weight his pack would have if this student’s point lay exactly on the line. All the other
points are relatively close to the line, so we think the missing point would also be
fairly close to the line. That is, we think this prediction will be reasonably accurate.
170 CHAPTER 4 Describing Relationships

Example 4.7 Natural-gas consumption and temperature

In Example 4.4 (page 151), we saw a very strong, negative, linear relationship be-
tween the average monthly temperature and the average amount of gas consumed
per day in Joan’s home. Figure 4.18 is a reproduction of the original scatterplot with
a regression line added.

Figure 4.18 Minitab scatterplot showing a strong straight-line pattern. The data are
the average monthly temperature and average amount of natural gas consumed per day in
Joan’s midwestern home.

We can use the regression line drawn in Figure 4.18 to predict Joan’s monthly
natural-gas consumption from the average monthly temperature. The points in this
figure are more tightly packed around the line than are the points in the hiker scat-
terplot of Figure 4.17. The correlations, which measure the strengths of the straight-
line relationships, are r  0.795 for Figure 4.17 and r  0.983 for Figure 4.18. The
scatter of the points makes it clear that predictions of backpack weight will gener-
ally be less accurate than predictions of natural-gas consumption.

Regression equations
When a plot shows a straight-line relationship as strong as that in Figure 4.18, it
is easy to draw a line close to the points by eye. In Figure 4.17, however, different
people might draw quite different lines by eye. Because we want to predict y from
x, we want a line that is close to the points in the vertical (y) direction. It is hard
to concentrate on just the vertical distances when drawing a line by eye. What is
more, drawing by eye gives us a line on the graph but not an equation for the line.
We need a way to find from the data the equation of the line that comes closest to
the points in the vertical direction. There are many ways to make the collection
of vertical distances “as small as possible.” The most common is the least-squares
method.
 4. 2 Regression and Prediction 171

Least-squares regression line

The least-squares regression line of y on x is the line that makes the

sum of the squares of the vertical distances of the data points from the
line as small as possible.

Figure 4.19 illustrates the least-squares idea. For each point in Figure 4.17, we have
drawn a vertical line segment (in bold) from the point to the regression line. These line
segments show the vertical distances of the points from the regression line. To find the
least-squares line, we need to minimize the sum of the squares of these vertical dis-
tances. The squares in Figure 4.19 help you visualize these “squares of vertical distances”
that we’re trying to minimize. Imagine moving the line until the sum of the squares is
the smallest it can be for any line. The resulting line is the least-squares regression line.

Regression line

Vertical
distance to line Data point

Packweight  16.3  0.0908 Bodyweight; r 2  0.63

Sum of squares = 30.90

Figure 4.19 The least-squares regression line results in the smallest possible sum of
squared distances of the points from the line (in this case, 30.90). Fathom software allows
you to explore this fact dynamically.

The lines drawn on the scatterplots in Figures 4.17 and 4.18 are the least-squares
regression lines. We won’t give the formula for finding the least-squares line from
data—that’s a job for a calculator or computer. You should, however, be able to use
the equation that technology produces.
In writing the equation of a line, x stands as usual for the explanatory variable
and y for the response variable. From your previous math classes, you are probably
used to seeing the equation of a line written in the form y 5 mx 1 b. In statistics,
the equation of a line is usually written as
y 5 a 1 bx
172 CHAPTER 4 Describing Relationships

The number b is the slope of the line, the amount by which y changes when x in-
creases by one unit. The number a is the y intercept, the value of y when x  0. To
use the equation for prediction, just substitute your x-value into the equation and
calculate the resulting y-value.

Example 4.8 Using a regression equation

The equation of the least-squares line in Figure 4.19 is

Pack weight 5 16.3 1 0.09 1 Body weight 2

• The slope of this line is b  0.09. This means that for these hikers, pack
Did the vote counters weight is predicted to go up by 0.09 pounds when body weight goes up by
cheat? 1 pound. Said another way, for every 10-pound increase in body weight,
Republican Bruce Marks this equation predicts a 0.9-pound increase in backpack weight. The slope
was ahead of Democrat
William Stinson when the
of a regression line is usually important for understanding the data. The
voting machines were tal- slope is the rate of change, the amount of change in the predicted y when x
lied in their Pennsylvania increases by 1.
election. But Stinson was • The y intercept of the least-squares line is a  16.3. This is the value of the
ahead after absentee bal-
lots were counted by the
predicted y when x  0. Although we need the intercept to draw the line,
Democrats who controlled it is statistically meaningful only when x can actually take values close to
the election board. A court zero. Here, body weight 0 is impossible, so the intercept has no statistical
fight followed. The court meaning.
called in a statistician, who
used regression with data
• To use the equation for prediction, substitute the value of x and calculate y. The
from past elections to pre- predicted pack weight for a new hiker who weighs 150 pounds is
dict the counts of absentee
ballots from the voting- Pack weight 5 16.3 1 0.09 1 150 2
machine results. Marks’s
lead of 564 votes from the 5 29.8 pounds
machines predicted that he
would get 133 more absen- Compare this result with that of the “up-and-over” method we used in Figure 4.17 to
tee votes than Stinson. In predict this student’s backpack weight.
fact, Stinson got 1025 more
absentee votes than Marks. • To draw the line on the scatterplot, predict y for two different values of x. This
Did the vote counters gives two points. Plot them and draw the line through them.
cheat?

Exercises 4.27 Predicting gas usage In Example 4.7 (page 170), we examined the
relationship between the average monthly temperature and the amount of natural
gas consumed in Joan’s midwestern home. The equation of the least-squares line
shown in Figure 4.18 is y 5 1425 2 19.87x.
(a) Identify the slope of the line and explain what it means in this setting.
(b) Identify the y intercept of the line and explain what it means in this setting.
(c) Use the regression line to predict the amount of natural gas Joan will use in a
month with an average temperature of 30 F.
 4. 2 Regression and Prediction 173

Exercises 4.28 to 4.30 involve the following setting. Figure 4.20

displays data on the number of slices of pizza consumed by
players on a football team (the explanatory variable x) and
the number of laps around the block the players could run
immediately afterward (the response variable y). The line on
the scatterplot is the least-squares regression line computed
from these points for predicting y from x.
4.28 Pizza party regression
(a) Use the graph and some algebra to show that the
equation of the least-squares line is y 5 10 2 1 2 /32 x.
Tracy A. Woodward/Washington Post
(b) What’s the slope of the line? Explain what this value
means in this setting.
(c) What’s the y intercept? Explain what this
12 value means in this setting.
4.29 Pizza party prediction Two players,
John and Ezekiel, arrived late for the party.
10 John ate 8 slices of pizza. Ezekiel showed off
by eating 16 slices.
8 (a) Use the regression line to predict the
number of laps that John will complete.
y (b) Can you use the regression line to predict
6
how many laps Ezekiel will complete? Why or
why not?
4 4.30 Least-squares pizza Copy Figure 4.20 onto
your paper. Use the figure to help you explain the
meaning of “least-squares regression line.”
2
4.31 Least-squares idea The TI-Nspire screen
shot below shows a scatterplot of five data
0 points. Which of the following two lines “fits”
2 4 6 8 10 12 the data better: y 5 1 2 x or y 5 3 2 2x? Justify
x
your answer by computing the sum of the
squares of the vertical distances of each point
Figure 4.20 A football team’s pizza regression, where x 5 the
number of slices of pizza eaten and y  the number of laps the
from the two lines. (Note: Neither of these two
player could run afterward. lines is the least-squares regression line for
these data.)
4.32 Sleep debt A researcher reported that the average teenager needs
9.3 hours of sleep per night but gets only 6.3 hours of sleep per night. 12
By the end of a 5-day school week, a teenager would accumulate about
15 hours of “sleep debt.” Students in a high school statistics class were
skeptical, so they gathered data on the amount of sleep debt (in hours)
accumulated over time (in days) by a random sample of 25 high school
students. The resulting least-squares regression equation for their data
is Sleep debt 5 2.23 1 3.17(Days). Do the students have reason to be
skeptical of the research study’s reported results? Explain.
174 CHAPTER 4 Describing Relationships

CALCULATOR CORNER Regression and prediction

In this Calculator Corner, you will learn to use your calculator to
perform least-squares regression. We will use the temperature and
natural-gas data from Example 4.4 (page 151). In the previous Calculator
Corner (page 152), we showed you how to make a scatterplot of the
data. Follow the instructions provided there to reproduce that scatterplot now.
1. Calculate the least-squares regression line.
TI-84
Press A, arrow right to CALC, and choose 8:LinReg(abx). Then
execute the command LinReg(a1bx)L1,L2,Y1. (To find Y1, press L,
arrow over to Y-VARS, choose 1:Function and 1:Y1.) Compare your
result with the screen shot on the right.
(Note: If the values of r and r2 don’t appear, you need to turn your diagnos-
tics on. Go into the CATALOG by pressing 2_. Scroll down until your cursor
is next to DiagnosticOn. Then press b twice. Now calculate the regres-
sion line again—r and r2 should appear this time.)
TI-Nspire
On the Lists & Spreadsheet page, press b, and then choose 4:Statistics,
1:Stat Calculations..., and 4:Linear Regression (abx).
In the dialog box, set X list: temp and Y list: gas, and then choose
OK. Scroll down to see the value of r.

2. Graph the regression line on the scatterplot.

TI-84
Press s.
TI-Nspire
On the Data & Statistics page, press b, and then choose 4:Analyze,
6:Regression, and 2:Show Linear (a1bx).
 4. 2 Regression and Prediction 175

3. Use the regression line to make a prediction. Let’s predict the amount
of natural gas Joan will use in a month with an average temperature of
30 F.

TI-84
In the graph screen, press r and use the down-arrow key (#) to switch
from the scatterplot to the regression line. You should see the equation of the
least-squares line displayed at the top of the screen. Just type [_ b to
obtain the prediction.

TI-Nspire
On the Data & Statistics page, press b, and then choose 4:Analyze and
A:Graph Trace. Use the NavPad to move the cursor close to x  30.

This prediction is based on fitting a regression line to the data for eight
past months. Because the data points lie close to the line, we can be confident
that Joan’s gas consumption in such a month will be close to 829 cubic feet
per day.

Correlation and regression

Correlation measures the direction and strength of a straight-line relationship.
Regression draws a line to describe the relationship. Correlation and regression are
closely connected, even though regression requires choosing an explanatory vari-
able and correlation does not.

ACTIVITY 4.2B
The Correlation and Regression applet
AP
PLET In this Activity, you will use an applet to Applications Web site, www.whfreeman.com/sta2e,
investigate how correlation and regression click on the Statistical Applets link, and launch the
are related. Go to the Statistics Through Correlation and Regression applet.
ACTIVITY 4.2B (continued)
176 CHAPTER 4 Describing Relationships

ACTIVITY 4.2B (continued)

Directions APPLET TIP: You can clear the scatterplot by

Use your mouse to click on the blank plot to make red clicking on the trash can.
points, creating a scatterplot. Above the plot, the applet
Experiment
shows the “Correlation coefficient.” Below the plot are
5. Create a scatterplot of 10 points with no
options that can be checked if you wish to draw the
particular pattern at the lower-left corner of
regression line on the plot, see the average values of x
the plot. Move the points around to make the
and y, or even add your own line to the plot to see how
correlation as close to zero as possible. Have the
well your line competes with the regression line.
applet show the least-squares regression line.
Think What is its slope? Now add an 11th point at the
1. If you plot just 2 points, what would the upper-right corner and watch the behavior of the
correlation be? correlation and the regression line.
2. If you put 3 points in a tight, positively associated 6. Clear the scatterplot and make sure the “Show
linear pattern in the upper-left corner of the plot, least-squares line” box is unchecked. Create a cloud
what would the correlation be? If you then add a of about 20 points with a weak linear association
fourth point at the lower-right corner (near the (for example, a correlation of about 0.5 or 0.5).
trash can), what would happen to the correlation? Then click the “Draw line” button. Try to position
Where would the regression line fall? your line so it goes through the “center” of the cloud
of points.
3. If you put 10 points in the center of the plot in a
U-shaped pattern, what would the correlation be? 7. Now have the applet add the regression line to
Where would the regression line fall? the plot. How does your line compare?
Verify 8. Click on the box to show the “mean X” and
4. Use the applet to test your answers to the three “mean Y ” lines. Move some points in your
questions above. Did the correlation and regression scatterplot and observe how the “mean X” and
lines behave as you thought they would? “mean Y” lines relate to the least-squares line.
 4. 2 Regression and Prediction 177

Let’s summarize the important facts about correlation and regression.

• Correlation and regression describe only linear relationships. You can do
the calculations for any data set involving two quantitative variables, but the
results are useful only if the scatterplot shows a linear pattern.
• Prediction outside the range of the available data is risky. Suppose
that you have data on a child’s growth between 3 and 8 years of age. You
find a strong straight-line relationship between age x and height y. If
you fit a regression line to these data and use it to predict height at age
25 years, you will predict that the child will be 8 feet tall. Growth slows
down and stops at maturity, so extending the straight line to adult ages
is foolish.
• Both correlation and regression are strongly affected by outliers. Be
wary if your scatterplot shows strong outliers. Figure 4.21 plots the record-
high yearly precipitation in each state against that state’s record-high
24-hour precipitation. Hawaii is a high outlier, with a yearly record of
704.83 inches of rain recorded at Kukui in 1982. The correlation for all
50 states in Figure 4.21 is 0.408. If we leave out Hawaii, the correlation
drops to r  0.195. The solid line in the figure is the least-squares line for
predicting the annual record from the 24-hour record for all 50 states. If
we leave out Hawaii, the least-squares line drops down to the dotted line.
This line is nearly flat—there is little relation between yearly and 24-hour
record precipitation once we decide to ignore Hawaii.

700 Hawaii
Maximum annual precipitation (inches)

600

500

400

300

200

100

0 10 20 30 40 50
Maximum 24-hour precipitation (inches)

Figure 4.21 Least-squares regression lines are strongly influenced by outliers.

• The least-squares regression line always passes through the point (x, y).
If you completed Activity 4.2B, then you observed this interesting property
of the least-squares line for yourself. Of course, you already know that the
178 CHAPTER 4 Describing Relationships

least-squares line minimizes the sum of the squares of the vertical distances
of the points from the line. This line has other important mathematical
properties that we’ll encounter later.
PLET • Switching x and y has no effect on the correlation, but it does affect the
AP
least-squares line. To see for yourself, launch the Two Variable Statistical
Calculator applet at the book’s Web site, www.whfreeman.com/sta2e.
Choose the hiker data on the Data Sets tab. Click on the Scatterplot tab to
see the graph. Check the box to “Show least-squares line.” Now click on the
Correlation & Regression tab. Record the correlation and the equation of
the regression line. Then check the “Swap X & Y” box at the top of the page.
Observe what happens to the scatterplot, the correlation, and the least-
squares line.

Understanding prediction: residuals

One of the first principles of data analysis is to look for an overall pattern and also
for striking deviations from the pattern. A regression line describes the overall
pattern of a linear relationship between an explanatory variable and a response
variable. We see deviations from this pattern by looking at the scatter of the data
points about the regression line. The vertical distances from the points to the
least-squares regression line are as small as possible, in the sense that they have
the smallest possible sum of squares. Because they represent “leftover” variation in
the response after fitting the regression line, these distances are called residuals.

Residual

A residual is the difference between an observed value of the response

variable and the value predicted by the regression line. That is,

residual  observed y  predicted y

Example 4.9 Body weight and pack weight: a

residual plot
In several earlier examples, we looked at the relationship between body weight and
backpack weight for a group of 8 hikers. Figure 4.22 shows a scatterplot with the
least-squares line added, along with a residual plot.
One hiker weighed 165 pounds. His pack weighed 35 pounds. The predicted
pack weight for this hiker is

Pack weight 5 16.3 1 0.09 1 165 2 5 31.15

The residual for this hiker is therefore

residual  observed y  predicted y

 35  31.15  3.85 pounds
 4. 2 Regression and Prediction 179

36

Pack weight (lb)

34
32
30
28
26
24
100 120 140 160 180
Body weight (lb)

4
3
2

Residual
1
0
1
2
3
Figure 4.22 Scatterplot
4
with least-squares line and
a separate residual plot for 100 120 140 160 180
the hiker data. Body weight (lb)

That is, the regression line underpredicts this hiker’s pack weight by about 4 pounds.
Since this point is above the line, its residual is positive. Points below the line have
negative residuals.

Because the residuals show how far the data fall from our regression line, examin-
ing a residual plot helps us determine how well the line describes the data. The residual
plot magnifies the deviations from the line to make patterns easier to see. If the regres-
sion line captures the overall pattern of the data well, two things should be true:
1. The residual plot should show no obvious pattern. Figure 4.23(a) is a residual
plot with no pattern. It shows an unstructured scatter of points in a horizontal band
centered at zero. A curved pattern in the residual plot, like the one in Figure 4.23(b),
shows that the relationship is not linear.

4 4
3 3
2 2
Residual

Residual

1 1
0 0
21 1
22 2
23 3
24 4
x x
(a) (b)

Figure 4.23 Two residual plots. In (a), the unstructured scatter of points indicates
that the regression line fits the data well. In (b), the residuals have a curved pattern, so a
straight line may not be the best model for the data.
180 CHAPTER 4 Describing Relationships

In Figure 4.22, we see a slightly curved pattern like the one in Figure 4.23(b).
This may suggest that the underlying relationship between body weight and pack
weight isn’t linear.
2. The residuals should be relatively small in size. A regression line in a model
that fits the data well should come “close” to most of the points. That is, the re-
siduals should be fairly small. How do we decide whether the residuals are “small
enough”? We consider the size of a “typical” prediction error.
In Figure 4.22, the residuals are between 4 and 4. For these hikers, the
predicted backpack weights from the least-squares line are within 4 pounds of the
actual pack weights. That sounds okay. But the hikers’ packs weighed between 24
and 35 pounds, so a prediction error of 4 pounds is fairly large compared with the
actual pack weight for an individual.

Understanding prediction: r 2
A residual plot is a graphical tool for evaluating how well a linear model fits the
data. There is also a numerical quantity that tells us how well the least-squares
line predicts values of the response variable y. It turns out that the square of the
correlation, r 2, is the right measure to use.

r 2 in regression

The square of the correlation, r2, is the fraction of the variation in the
values of y that is explained by the least-squares regression of y on x.

The idea is that when there is a linear relationship, some of the variation in y
Regression toward is accounted for by the fact that, as x changes, it pulls y along the regression line
the mean with it.
To “regress” means to go
backward. Why are statisti-
cal methods for predicting a Example 4.10 Interpreting r2
response from an explana-
tory variable called “regres- Look again at Figure 4.22. There is a lot of variation in the body weights of the
sion”? Sir Francis Galton
(1822–1911), who was the
8 hikers, from a low of 103 pounds to a high of 187 pounds. The scatterplot shows a
first to apply regression to straight-line relationship between body weight and backpack weight, but the points
biological and psychological show a fairly large amount of scatter around the regression line. As body weight
data, looked at examples increases, it generally pulls pack weight up with it along the line. But there is quite
such as the heights of a bit of leftover variation in pack weight, as shown in the variation of points about
children versus the heights
the line. Because r 2 5 0.632 for these data, about 63% of the observed variation in
of their parents. He found
that the taller-than-average pack weight is explained by the straight-line pattern. What about the other 37%? It
parents tended to have chil- is due to other factors, such as physical fitness, motivation, and need for “creature
dren who were also taller comforts.”
than average but not as tall
as their parents. Galton
called this fact “regression
toward the mean,” and the In reporting a regression, it is usual to give r 2 as a measure of how successful
name came to be applied to the regression was in explaining the response. When you see a correlation, square
the statistical method. it to get a better feel for the strength of the association. Perfect correlation (r 5 21
 4. 2 Regression and Prediction 181

or r 5 1) means the points lie exactly on a line. Then r 2 5 1 and all of the variation
in one variable is accounted for by the straight-line relationship with the other vari-
able. If r 5 20.7 or r 5 0.7, r 2 5 0.49 and about half the variation is accounted for
by the straight-line relationship.

CALCULATOR CORNER Making a residual plot

Both the TI-84 and the TI-Nspire automatically compute and store re-
siduals when a regression is run. Let’s pick up where we left off with the
temperature and natural-gas data from the previous Calculator Corner
(page 174). To make a residual plot:

TI-84
Press 2o to go into the STAT PLOT menu. In Plot1, change the Ylist
from L2 to Resid. You can choose Resid as follows: press yA (LIST),
arrow down to Resid, and press b. To display the graph, press qo
(ZoomStat).

TI-Nspire
On the Data & Statistics page, press b and then choose 4:Analyze,
7:Residuals, and 2:Show Residual Plot.

Exercises 4.33 Hiker residuals Here, one more time, are the data on body and backpack
weights of the 8 hikers:

Body weight (lb): 120 187 109 103 131 165 158 116
Backpack weight (lb): 26 30 26 24 29 35 31 28

In Example 4.9, we showed you a computer-generated residual plot for these data
based on the least-squares regression equation Pack weight  16.3  0.09(Body
weight). We also computed the residual for the hiker who weighed 165 pounds.
Use the method shown there to calculate the residuals for the other 7 hikers. Then
find the sum of the 8 residuals.
Result: For the least-squares line, the sum of the residuals is always 0 (up
to roundoff error).
182 CHAPTER 4 Describing Relationships

4.34 Joan’s gas use residuals In the Calculator Corner, we constructed a residual
plot for Joan’s gas consumption data. Confirm the value of the residual shown in
the TI-84 screen shot for the month with an average temperature of 49.4°F. Show
your work.
4.35 r versus r 2 “When r 5 0.7, this means that y can be predicted from x for
70% of the individuals in the sample.” Is this statement true or false? Would it be
true if r 2 5 0.7? Explain your answers.
4.36 Correlation and regression If the correlation between two variables x and
y is r 5 0, there is no straight-line relationship between the variables. It turns out
that the correlation is 0 exactly when the slope of the least-squares regression line
is 0. Explain why slope 0 means that there is no linear relationship between x and
y. Start by drawing a line with slope 0 and explaining why in this situation x has
no value for predicting y.
4.37 Beer and BAC, I How much does drinking beer increase the alcohol
content of your blood? This question was addressed in an experiment at
Ohio State University. Sixteen students volunteered to participate. Before the
experiment, each student blew into a Breathalyzer machine to show that their
blood alcohol content (BAC) was at the zero mark. The student volunteers then
drank a varying number (between 1 and 9) of 12-ounce beers. How much each
student drank was assigned by drawing tickets from a bowl. About 30 minutes
later, an officer from the OSU Police Department measured their BAC using the
Breathalyzer. Here are the data:13

Student: 1 2 3 4 5 6 7 8
Beers: 5 2 9 8 3 7 3 5
BAC: 0.10 0.03 0.19 0.12 0.04 0.095 0.07 0.06
Student: 9 10 11 12 13 14 15 16
Beers: 3 5 4 6 5 7 1 4
BAC: 0.02 0.05 0.07 0.10 0.085 0.09 0.01 0.05

(a) Suppose you want to estimate how a person’s BAC is affected by the number of
beers he or she drinks. Make a scatterplot of BAC versus beers using the data in
the table above. Which variable did you choose to be the y variable and which did
you choose to be the x variable? Explain.
(b) Use your calculator to find the correlation between BAC and beers. Is the
correlation an appropriate measure of the strength of the association between
BAC and beers? Explain briefly.
(c) Use your calculator to find the least-squares regression line for BAC on beers.
Interpret the slope and y intercept of the regression line.
4.38 Beer and BAC, II Refer to the previous exercise.
(a) If a student drinks 5 beers, on average what do you predict the student’s BAC
will be? Show your work.
(b) Would the regression method be as accurate for predicting BAC for a person
who drinks 15 beers? Explain.
 4. 2 Regression and Prediction 183

(c) Use your calculator to construct a residual plot. How well does the regression
line fit the data?
(d) Interpret the value of r 2 in this setting.

DATA EXPLORATION
What’s Driving Car Sales?
U.S. car sales change month by month because of High Mpg%  9.63594  0.101690 gas price
S  1.06679 R-Sq  45.0% R-Sq(adj)  41.6% p-value  0.002
car prices, the economy, and other factors that affect
28
consumers’ buying habits. How do gas prices affect
car sales? To answer this question, we collected data 27
on U.S. car sales and gas prices over an 18-month

High mpg (%)

period when gas prices were generally increasing.14 26

1. As gas prices go up, people may think more 25

about gas mileage in deciding what car to buy.
Figure 4.24 shows parallel boxplots of gas mileage 24

for each size class of car—subcompact, compact, 23

midsize, and large. Write a paragraph comparing the
distributions of gas mileage for the four car classes. 130 140 150 160 170
Gas price (cents)
38
Figure 4.25 Scatterplot of monthly gas prices and
percent of cars purchased that month with gas mileage
greater than 26.75 mpg.
Gas mileage (mpg)

28 unleaded gasoline each month and the percent of

cars sold that month with gas mileage greater than
26.75 miles per gallon (mpg). The least-squares
regression line has been added to the plot, and
some output from a computer regression analysis
18 appears at the top of the graph.
Subcompact Compact Midsize Large 2. Interpret the slope and y intercept of the least-
Figure 4.24 Boxplot of gas mileage for different squares regression line in the context of the problem.
classes of car. 3. During one of the months in this study, the
A reasonable theory is that as gas prices average gas price was $1.50. Calculate the residual
increase, people will buy smaller cars. Figure 4.25 for that month. Show your method.
displays a scatterplot of the average price of regular 4. Explain what the value of r 2 means in this setting.

The question of causation

There is a strong relationship between cigarette smoking and death rate from lung
cancer. Does smoking cigarettes cause lung cancer? There is a strong association
between the availability of handguns in a nation and that nation’s homicide rate from
guns. Does easy access to handguns cause more murders? It says right on the pack
184 CHAPTER 4 Describing Relationships

that cigarettes cause cancer. Whether more guns cause more murders is hotly de-
bated. Why is the evidence for cigarettes and cancer better than the evidence for guns
and homicide? Let’s consider a few examples before we try to answer this question.

Example 4.11 Does television extend life?

In the 1990s, researchers measured the number of television sets per person x and
the life expectancy y for the world’s nations. There was a high positive correlation:
nations with many TV sets had higher life expectancies.
The basic meaning of causation is that by changing x we can bring about a
change in y. Could we lengthen the lives of people in Botswana by shipping them
TV sets? No. Rich nations have more TV sets than poor nations. Rich nations also
have longer life expectancies because they offer better nutrition, clean water, and bet-
ter health care. There is no cause-and-effect tie between TV sets and length of life.

"In a new attack on third-world poverty, aid organizations today began delivery of
100,000 television sets."

Correlations such as the one in Example 4.11 are sometimes called “nonsense
correlations.” The correlation is real. What is nonsense is the conclusion that chang-
ing one of the variables causes changes in the other. A lurking variable—such as
national wealth in Example 4.11—that influences both x and y can create a high cor-
relation even though there is no direct connection between x and y.

Lurking variable

A lurking variable is a variable that has an important effect on the

relationship among the variables in a study but is not one of the
explanatory variables studied.
 4. 2 Regression and Prediction 185

We might call a situation like the one in Example 4.11 common response: both
the explanatory and the response variables are responding to some lurking variable.
Now consider a different example.

Example 4.12 Obesity in mothers and daughters

What causes obesity in children? Inheritance from parents, overeating, lack of
physical activity, and too much television have all been named as explanatory
variables.
The results of a study of Mexican American girls aged 9 to 12 years are typical.
Measure body mass index (BMI), a measure of weight relative to height, for both
the girls and their mothers. People with high BMIs are overweight or obese. Also
measure hours of television, minutes of physical activity, and intake of several kinds
of food. Result: The girls’ BMIs were weakly correlated (r 5 20.18) with physical
activity and also with diet and television. The strongest correlation (r 5 0.506) was
between the BMI of daughters and the BMI of their mothers.15

Body type is in part determined by heredity. Daughters inherit half their genes
from their mothers. There is therefore a direct causal link between the BMI of
mothers and daughters. Of course, the causal link is far from perfect. The mothers’
BMIs explain only 25.6% (that’s r 2 again) of the variation among the daughters’
BMIs. Other factors, some measured in the study and others not measured, also
influence BMI. Even when direct causation is present, it is rarely a complete explana-
tion of an association between two variables.
Can we use r or r 2 from Example 4.12 to say how much inheritance contributes
to the daughters’ BMIs? No. The reason is confounding. It may well be that mothers
who are overweight also set an example of little exercise, poor eating habits, and lots
of television. Their daughters pick up these habits to some extent, so the influence of
heredity is mixed up with influences from the girls’ environment. We can’t say how
much of the correlation between mother and daughter BMIs is due to inheritance.

Confounding

Two variables are confounded when their effects on a response vari-

able cannot be distinguished from each other. The confounded vari-
ables may be either explanatory variables or lurking variables.

Figure 4.26 (on the next page) shows in outline form how a variety of underly-
ing links between variables can explain association. The dashed line represents an
observed association between the variables x and y.
Some associations are explained by a direct cause-and-effect link between the
variables. The first diagram in Figure 4.26 shows “x causes y” by an arrow running
from x to y. The second diagram illustrates common response. The observed associa-
tion between the variables x and y is explained by a lurking variable z. Both x and y
change in response to changes in z. This common response creates an association
186 CHAPTER 4 Describing Relationships

x y x y x y
?

z z

(a) Causation (b) Common response (c) Confounding

Figure 4.26 Some explanations for an observed association. A dashed line shows
an association. An arrow shows a cause-and-effect link. Variable x is explanatory, y is a
response variable, and z is a lurking variable.

even though there may be no direct causal link between x and y. The third diagram
in Figure 4.26 illustrates confounding. Both the explanatory variable x and the lurk-
ing variable z may influence the response variable y. Variables x and z are them-
selves associated, so we cannot distinguish the influence of x from the influence of z.
We cannot say how strong the direct effect of x on y is. In fact, it can be hard to say
if x influences y at all.
Both common response and confounding involve the influence of a lurking
variable or variables z on the response variable y. We won’t overemphasize the dis-
tinction between the two kinds of relationships. Just remember that “beware the
lurking variable” is good advice in thinking about relationships between variables.
Here is another example of common response, in a setting where we want to do
prediction.

Example 4.13 SAT scores and college grades

High scores on the SAT examinations in high school certainly do not cause high
grades in college. The moderate association (r 2 is about 27%) is no doubt explained
by common response to such lurking variables as academic ability, study habits,
and staying sober.
The ability of SAT scores to partly predict college performance doesn’t depend
on causation. We need only believe that the relationship between SAT scores and
college grades that we see in past years will continue to hold for this year’s high
school graduates. Prediction doesn’t require causation.

Evidence for causation

There’s an important fact that you’ll discover in Chapter 6: the best evidence for
causation comes from well-designed experiments. Many of the sharpest disputes
in which statistics plays a role involve questions of causation that cannot be
settled by experiment, however. Does taking the drug Bendectin cause birth
defects? Does living near power lines cause cancer? Both of these questions con-
cern associations between variables. And they both try to pinpoint cause and
effect in a setting involving complex relationships among many interacting vari-
ables. Common response and confounding, along with the number of potential
 4. 2 Regression and Prediction 187

lurking variables, make observed associations misleading. However, for moral

or practical reasons, it is not possible to conduct experiments to answer these
questions.

Example 4.14 Do power lines increase the risk of

leukemia?
Electric currents generate magnetic fields. So living with electricity exposes people
to magnetic fields. Living near power lines increases exposure to these fields. Really
strong fields can disturb living cells in laboratory studies. What about the weaker
fields we experience if we live near power lines?
It isn’t ethical to do experiments that expose children to magnetic fields. It’s
hard to compare cancer rates among children who happen to live in more and less
exposed locations, because leukemia is rare and locations vary in many ways other than
magnetic fields. We must rely on studies that compare children who have leukemia
with children who don’t.
A careful study of the effect of magnetic fields on children took 5 years and cost
$5 million. The researchers compared 638 children who had leukemia with 620
who did not. They went into the homes and actually measured the magnetic fields
in the children’s bedrooms, in other rooms, and at the front door. They recorded
facts about nearby power lines for the family home and also for the mother’s resi-
dence when she was pregnant. Result: no evidence of more than a chance connec-
tion between magnetic fields and childhood leukemia.16

Despite the difficulties, it is sometimes possible to build a strong case for causa-
tion in the absence of experiments. The evidence that smoking causes lung cancer
is about as strong as nonexperimental evidence can be.
Doctors had long observed that most lung cancer patients were smokers.
Observational studies comparing smokers and “similar” nonsmokers showed a
strong association between smoking and death from lung cancer. Could the asso-
ciation be explained by lurking variables that the studies could not measure? Might
there be, for example, a genetic factor that predisposes people both to nicotine
addiction and to lung cancer? Smoking and lung cancer would then be positively
associated even if smoking had no direct effect on the lungs. How were these objec-
tions overcome?
Let’s answer this question in general terms: what are the criteria for establishing
causation when we cannot do an experiment?
• The association is strong. The association between smoking and lung cancer is
very strong.
• The association is consistent. Many studies of different kinds of people in
many countries link smoking to lung cancer. That reduces the chance that a
lurking variable specific to one group or one study explains the association.
• Higher doses are associated with stronger responses. People who smoke more
cigarettes per day or who smoke over a longer period get lung cancer more
often. People who stop smoking reduce their risk.
188 CHAPTER 4 Describing Relationships

• The alleged cause precedes the effect in time. Lung cancer develops
after years of smoking. The number of men dying of lung cancer rose as
smoking became more common, with a lag of about 30 years. Lung cancer
kills more men than any other form of cancer. Lung cancer was rare among
women until women began to smoke. Lung cancer in women rose along
with smoking, again with a lag of about 30 years, and has now passed breast
cancer as the leading cause of cancer death among women.
• The alleged cause is plausible. Experiments with animals show that tars from
cigarette smoke do cause cancer.
Medical authorities do not hesitate to say that smoking causes lung cancer. The
U.S. Surgeon General has long stated that cigarette smoking is “the largest avoid-
able cause of death and disability in the United States.” The evidence for causation
is overwhelming—but it is not as strong as the evidence provided by well-designed
experiments.

Exercises 4.39 Ice cream and drowning There is a positive association between ice cream
sales and the number of people who drown. Does that mean eating ice cream
causes people to drown? Of course not. Give a more plausible explanation that
involves common response. Draw a picture like Figure 4.26 (page 186) to illustrate
the relationship among the variables.
4.40 Is a little alcohol good for you? A survey of 7000 California men found little
correlation between alcohol consumption and chance of dying during the 51⁄2 years
of the study. In fact, men who did not drink at all during these years had a slightly
higher death rate than did light drinkers. This lack of correlation was somewhat
surprising. Explain how common response might account for the higher death rate
among men who did not drink at all over a short period.
4.41 Education and income There is a strong positive correlation between years
of schooling completed and lifetime earnings for American men. One possible
reason for this association is causation: more education leads to higher-paying
jobs. Another explanation is confounding: men who complete many years of
schooling have other characteristics that would lead to better jobs even without
the education. Suggest several possible confounding variables.
4.42 Snow and earthquakes A study measures the average annual snowfall (in
inches) for 10 cities over the last decade along with the greatest earth movement
(on the Richter scale) over this same time period. The study included 5 cities in
California’s San Francisco Bay Area and 5 cities from Canada’s province of Ontario.
The study found a very strong negative correlation between the two variables. Does
this mean that a strong snowfall will prevent earthquakes? Explain your answer
briefly. You may want to draw a picture like Figure 4.26 (page 186) to illustrate it.
4.43 Is math the key to success in college? Here is the opening of a newspaper
account of a College Board study of 15,941 high school graduates:
Minority students who take high school algebra and geometry succeed in col-
lege at almost the same rate as whites, a new study says. The link between high
school math and college graduation is “almost magical,” says College Board
 4. 2 Regression and Prediction 189

President Donald Stewart, suggesting “math is the gatekeeper for success in

college.” “These findings,” he says, “justify serious consideration of a national
policy that all students take algebra and geometry.”17
What lurking variables might explain the association between taking several math
courses in high school and success in college? Explain why requiring algebra and
geometry may have little effect on who succeeds in college.
4.44 Predicting Old Faithful eruptions Figure 4.27 shows the results from a
least-squares regression on the Old Faithful eruption data from the beginning of the
chapter. Length of previous eruption is the explanatory variable, and time until the
next eruption is the response variable. Both variables were measured in minutes.

Figure 4.27 Fathom

scatterplot with least-
squares regression line
and residual plot for the
Old Faithful eruption data
from the beginning of the
chapter.

(a) Is the relationship between these two variables most likely a result of
causation, confounding, or common response? Justify your answer.
(b) Write a brief analysis for your teacher that demonstrates your understanding
of the material in this chapter. Follow the four-step statistical problem-solving
process (page 19).

ws
Ne
ws
Ne

APPLICATION 4.2 B]
RO
g
<
Se
a

Cricket chirps and air temperature

Can the frequency of cricket chirps be used to pre- ing, George Washington Pierce, the answer is yes.
dict the outdoor temperature? According to one of During his career, Pierce invented several pieces
the founding fathers of communications engineer- of technology that earned him patents and a lot of
APPLICATION 4.2 (continued)
190 CHAPTER 4 Describing Relationships

APPLICATION 4.2 (continued)

money from companies like RCA and AT&T. When 2. Make a well-labeled scatterplot of the data.
he retired, Pierce built a device that allowed him to Describe the direction, form, and strength of the
record the sounds made by various insects near his relationship. Are there any outliers?
New Hampshire home. In 1948, he published his re- 3. Use your calculator to find the least-squares
search findings in a book titled The Songs of Insects. regression line for these data. Record the equation.
In this Application, you’ll examine data that Pierce
collected on the number of chirps per second of the 4. Interpret the slope and the y intercept of the
striped ground cricket and the outdoor temperature least-squares line in this setting.
in degrees Fahrenheit.18 5. Use the equation to predict the temperature when
there are 15 cricket chirps per second.
Cricket chirps per second Outdoor temperature (°F)
6. Construct a residual plot. Describe what you
20.0 88.6 see.
16.0 71.6
19.8 93.3 7. Explain what the value of r 2 tells you in this
18.4 84.3 setting.
17.1 80.6 8. How well does the regression line fit the data?
15.5 75.2
Explain.
14.7 69.7
17.1 82.0 9. Is it reasonable to use the equation to predict the
15.4 69.4 temperature when there are 25 cricket chirps per
16.2 83.3 second? Explain.
15.0 79.6
17.2 82.6 10. Crickets make their chirping sounds by rapidly
16.0 80.6 rubbing their wings together. From Pierce’s data,
17.0 83.5 we see that outdoor temperature increases as
14.4 76.3 the number of cricket chirps increases. Can we
conclude that the increased number of chirps
QUESTIONS causes the temperature to increase (maybe due to
1. Enter the data into your calculator. Which is the the heat generated from wings rubbing together)?
explanatory variable? Explain.

Section 4.2 Summary

Regression is the name for statistical methods that fit some model to data in order
to predict a response variable from one or more explanatory variables. The simplest
kind of regression fits a straight line, called a regression line, on a scatterplot for
use in predicting y from x. The most common way to fit a line is the least-squares
method, which finds the line that makes the sum of the squared vertical distances
of the data points from the line as small as possible.
Least-squares regression is closely related to correlation. In particular, the
squared correlation r 2 tells us what fraction of the variation in the responses is
explained by the straight-line relationship between y and x. Residual plots also
help us assess how well a regression line fits the data. It is generally true that the
success of any statistical prediction depends on the presence of strong patterns in
the data. Prediction outside the range of the data is risky because the pattern may
be different there.
 4. 2 Regression and Prediction 191

A strong relationship between two variables is not always evidence that changes
in one variable cause changes in the other. Lurking variables can create relation-
ships through common response or confounding. If we cannot do experiments, it
is often difficult to get convincing evidence for causation.

Section 4.2 4.45 Heart attacks If you need medical care, should you go to a hospital that handles
Exercises many cases like yours? Figure 4.28 presents some data for heart attacks. The figure plots
mortality (the proportion of patients who died) against the number of heart attack
patients treated for a large number of hospitals in a recent year. The line on the plot is
the least-squares regression line for predicting mortality from number of patients.

0.4

0.3
Mortality (%)

0.2

0.1

0 100 200 300 400 500 600 700

Patients treated

Figure 4.28 Mortality of heart attack patients and number of heart attack cases treated
for a large group of hospitals.

(a) Do the plot and regression generally support the hypothesis that mortality is
lower at hospitals that treat more heart attacks? Is the relationship very strong?
(b) In what way is the pattern of the plot nonlinear? Does the nonlinearity strengthen or
weaken the conclusion that heart attack patients should avoid hospitals that treat
few heart attacks? Why?
4.46 Beer drinking and cancer A study using data from 41 states found a posi-
tive correlation between beer consumption per person and death rates from some
forms of cancer. The states with the highest death rates from these types of cancer
were Rhode Island and New York. The beer consumption in those states was 80
quarts per person per year. People in South Carolina, Alabama, and Arkansas
drank only 26 quarts of beer per person and had cancer rates less than one-third
of those in Rhode Island and New York.
Suggest some lurking variables that may be confounded with a state’s beer con-
sumption. For a clue, look at the high- and low-consumption states given above.
192 CHAPTER 4 Describing Relationships

4.47 Obesity in mothers and daughters A study found that the correla-
tion between the body mass index (BMI) of young girls and their minutes of
physical activity in a day was r 5 20.18.19 Why might we expect this correla-
tion to be negative? What percent of the variation in BMI among the girls in
the study can be explained by the straight-line relationship with minutes of
activity?
4.48 Application 4.2 follow-up Refer to Application 4.2 (page 189).
(a) Suppose we want to predict the number of cricket chirps per second when the
outdoor temperature is 80 F. Should we use the regression equation from Applica-
tion 4.2? Why or why not?
(b) If we changed the temperature data in Application 4.2 from Fahrenheit to
Celsius, how would this affect the correlation? The slope of the least-squares line?
The y intercept? Justify your answers.
From Rex Boggs in Australia comes an unusual data set. Before showering in
the morning, he weighed the bar of soap in his shower stall. The weight goes
down as the soap is used. The data appear in Table 4.4 (weights in grams).
iSt Notice that Mr. Boggs forgot to weigh the soap on some days. Exercises 4.49
oc
kph
oto to 4.51 are based on the soap data set.

Table 4.4 Weight (Grams) of a Bar of Soap Used to Shower

Day Weight Day Weight Day Weight

1 124 8 84 16 27
2 121 9 78 18 16
5 103 10 71 19 12
6 96 12 58 20 8
7 90 13 50 21 6

Source: Rex Boggs.

4.49 Scatterplot Plot the weight of the bar of soap against day. Is the overall pat-
tern roughly a straight line? Based on your scatterplot, is the correlation between
day and weight close to 1, positive but not close to 1, close to 0, negative but not
close to 1, or close to 1? Explain your answer.
4.50 Regression The equation for the least-squares regression line for the data in
Table 4.4 is

weight  133.2  6.31(day)

(a) Explain carefully what the slope b  6.31 tells us about how fast the soap
lost weight.
(b) Interpret the y intercept, 133.2.
(c) Mr. Boggs did not measure the weight of the soap on Day 4. Use the regression
equation to predict that weight.
(d) Draw the regression line on your scatterplot from the previous exercise.
 4. 2 Regression and Prediction 193

4.51 Prediction?
(a) Use the regression equation in the previous exercise to predict the weight of
the soap after 30 days. What’s wrong with using the regression line to predict
weight after 30 days?
(b) For these data, r 2  0.996. Explain what this value means.
(c) Make a residual plot for these data. Is a linear model appropriate in this case?
Explain.
4.52 More chirping crickets In late summer 2007, Dr. Peggy LeMone decided
to collect some cricket chirp and temperature data at her home in Colorado.
She made several measurements of the number of cricket chirps in a 15-second
period and the outdoor temperature in degrees Fahrenheit.20 Figure 4.29 shows a
scatterplot of Dr. LeMone’s data with a least-squares regression line added and a
residual plot.

Figure 4.29 Fathom scatterplot with least-squares line and residual plot for the num-
ber of cricket chirps in 15 seconds and the outdoor temperature at Dr. LeMone’s home in
Colorado in 2007.

(a) How well does the regression line fit the data? Give graphical and numerical
evidence to support your answer.
(b) One formula that Dr. LeMone found on the Web for predicting temperature
from cricket chirps was

temp. in degrees Fahrenheit  no. of cricket chirps in 15 seconds  37

Compare the temperature predictions for the least-squares line and this formula
when there are 10, 20, 30, and 40 cricket chirps per 15 seconds.
4.53 Miscarriages among workers A study showed that women who work in
the production of computer chips have abnormally high numbers of miscarriages.
The union claimed that exposure to chemicals used in production caused the mis-
carriages. Another possible explanation is that these workers spend most of their
194 CHAPTER 4 Describing Relationships

work time standing up. Illustrate these relationships in a diagram like one of those
in Figure 4.26 (page 186).
4.54 Calculating the least-squares line Do you like to know the mathematical de-
tails when you study something? Here is the formula for the least-squares regres-
sion line for predicting y from x. Start with the means and the standard deviations
of the two variables and the correlation r between them. The least-squares line has
equation y  a  bx with
sy
b5r and a 5 y 2 bx
sx
For the backpacker data of this chapter, we found these summary statistics:

Body weight: x 5 136.125 pounds sx 5 30.296 pounds

Pack weight: y 5 28.625 pounds sy 5 3.462 pounds

The correlation between these variables is r  0.795. Use these values in the
formulas just given to verify the equation of the least-squares line given earlier:
pack weight 5 16.3 1 0.09 1 Body weight 2 .

CHAPTER 4 REVIEW
This chapter focused on describing relationships between two quantitative vari-
ables. Figure 4.30 retraces the big ideas. We always begin by making graphs of our
data. In the case of a scatterplot, we have learned a numerical summary only for
data that show a roughly straight-line pattern on the scatterplot. This summary is
the means and standard deviations of the two variables and their correlation. A
regression line drawn on the plot gives us a compact model of the overall pattern
that we can use for prediction. Residual plots and r 2 help us assess how well the
linear model fits the data. The question marks at the last two stages remind us that
correlation and regression describe only straight-line relationships.

Plot your data

Scatterplot

Interpret what you see:

Direction, form, strength, outliers

Numerical summary
x, y, sx, sy, and r?

Compact model?
Regression line?

How well does it fit?

Residuals and r2

Figure 4.30 Flowchart for describing relationships.

 Chapter 4 Review Exercises 195

Relationships often raise the question of causation. You will see in Chapter 6
that evidence from well-designed experiments is the “gold standard” for deciding
that one variable causes changes in another variable. Section 4.2 reminded us in
more detail how strong associations can appear in data even when there is no direct
causation. We must always think about the possible effects of variables lurking in
the background.
Here is a review list of the most important skills you should have developed from
your study of this chapter.
A. SCATTERPLOTS AND CORRELATION
1. Make a scatterplot to display the relationship between two quantitative vari-
ables measured on the same subjects. Place the explanatory variable (if any) on
the horizontal scale of the plot.
2. Describe the direction, form, and strength of the overall pattern of a scatterplot.
In particular, recognize positive or negative associations and straight-line pat-
terns. Recognize outliers in a scatterplot.
3. Judge whether it is appropriate to use correlation to describe the relationship
between two quantitative variables. Use a calculator to find the correlation r.
4. Know the basic properties of correlation: r measures the strength and direc-
tion of only straight-line relationships; r is always a number between 1 and 1;
r 5 61 only for perfect straight-line relations; r moves away from 0 toward 61 as
the straight-line relation gets stronger.
B. REGRESSION LINES
1. Explain what the slope b and the intercept a mean in the equation y 5 a 1 bx
of a straight line.
2. Draw a graph of the straight line when you are given its equation.
3. Use a regression line, given on a graph or as an equation, to predict y for a given
x. Recognize the danger of prediction outside the range of the available data.
4. Use a residual plot to examine how well a regression line fits the data.
5. Use r 2, the square of the correlation, to describe how much of the variation
in one variable can be accounted for by a straight-line relationship with another
variable.
C. STATISTICS AND CAUSATION
1. Give plausible explanations for an observed association between two variables:
direct cause and effect, the influence of lurking variables, or both.
2. Assess the strength of statistical evidence for a claim of causation, especially
when experiments are not possible.

CHAPTER 4 REVIEW EXERCISES

4.55 Measuring crickets For a biology project, you measure the length (centimeters)
and weight (grams) of 12 crickets.
(a) Explain why you expect the correlation between length and weight to be positive.
(b) If you measured length in inches, how would the correlation change? (There are
2.54 centimeters in an inch.)
196 CHAPTER 4 Describing Relationships

4.56 When it rains, it pours Figure 4.31 plots the highest yearly precipitation ever
recorded in each state against the highest daily precipitation ever recorded in that
state. The points for Alaska, Hawaii, and Texas are marked on the scatterplot.

700 Hawaii

Maximum annual precipitation (inches)

600

500

400

Alaska
300

200

100 Texas

0 10 20 30 40 50
Maximum 24-hour precipitation (inches)

Figure 4.31 Record-high yearly precipitation at any weather station in each state plotted
against record-high daily precipitation for the state.

(a) About what are the highest daily and yearly precipitation values for Alaska?
(b) Alaska and Hawaii have very high yearly maximums relative to their daily maxi-
mums. Omit these two states as outliers. Describe the nature of the relationship for
the other states. Would knowing a state’s highest daily precipitation be a great help in
predicting that state’s highest yearly precipitation?
4.57 Why so small?
(a) Make a scatterplot of the following data:

x: 1 2 3 4 10 10
y: 1 3 3 5 1 11

(b) Use the method of Example 4.5 (page 158) to show that the correlation is
about 0.5.
(c) What feature of the data is responsible for reducing the correlation to this value
despite a strong straight-line association between x and y in most of the observations?
4.58 SAT preparation Can intensive preparation significantly improve students’ SAT
scores? Here’s one way to find out. Take a random sample of 100 high school students
who took the most recent SAT test and ask them how much time (in hours) they
spent preparing and what score they earned on the Math section. If we find a correla-
tion of r 5 0.79 between preparation time and SAT Math score, can we conclude that
preparation causes higher scores? Why or why not?
 Chapter 4 Review Exercises 197

Figure 4.32 plots the average brain weight in grams versus average body weight in kilo-
grams for 96 species of mammals.21 There are many small mammals whose points at the
lower left overlap. Exercises 4.59 to 4.62 are based on this scatterplot.

4500
elephant
4000

3500

Brain weight (grams)

3000

2500

2000

1500 dolphin
human
1000

500 hippo

0
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
Body weight (kilograms)

Figure 4.32 Scatterplot of the average brain weight (grams) against the average body
weight (kilograms) for 96 species of mammals.

4.59 Dolphins and hippos

(a) The points for the dolphin and hippopotamus are labeled in Figure 4.32. Read
from the graph the approximate body weight and brain weight for these two species.
(b) One reaction to this scatterplot is “Dolphins are smart, hippos are dumb.” What
feature of the plot lies behind this reaction?
4.60 Outliers The African elephant is much larger than any other mammal in the
data set but lies roughly in the overall straight-line pattern. Dolphins, humans, and
hippos lie outside the overall pattern. The correlation between body weight for the
entire data set is r 5 0.86.
(a) If we removed elephants, would this correlation increase, decrease, or not change
much? Explain your answer.
(b) If we removed dolphins, hippos, and humans, would this correlation increase,
decrease, or not change much? Explain your answer.
4.61 Brain and body The correlation between body weight and brain weight is
r = 0.86. How well does body weight explain brain weight for mammals? Give a num-
ber to answer this question, and briefly explain what the number tells us.
4.62 Prediction The line on the scatterplot in Figure 4.32 is the least-squares regres-
sion line for predicting brain weight from body weight.
(a) Suppose that a new mammal species is discovered hidden in the rain forest with
body weight 600 kilograms. Predict the brain weight for this species.
(b) The slope of this line is one of these numbers: 0.5, 1.3, or 3.2. Which number is the
slope? Why?
198 CHAPTER 4 Describing Relationships

4.63 Late bloomers? From experience, Japanese cherry trees tend to blossom early
when spring weather is warm and later when spring weather is cool. Here are some
data on the average March temperature (in C) and the day in April when the first
cherry blossom appeared over a 25-year period.22

Temperature ( C): 4.0 5.4 3.2 2.6 4.2 4.7 4.9 4.0 4.9 3.8 4.0 5.1
Days in April to 1st bloom: 14 8 11 19 14 14 14 21 9 14 13 11

Temperature ( C): 4.3 1.5 3.7 3.8 4.5 4.1 6.1 6.2 5.1 5.0 4.6 4.0
Days in April to 1st bloom: 13 28 17 19 10 17 3 3 11 6 9 11

(a) Make a well-labeled scatterplot that is suitable for predicting when the cherry trees will
bloom from the temperature. Describe the direction, form, and strength of the relationship.
(b) Use technology to find the equation of the least-squares regression line. Interpret
the slope and y intercept of the line in this setting.
(c) The average March temperature this year was 3.5 C. When do you predict that the
first cherry blossom will appear? Show your method clearly.
(d) Find the residual for the year when the average March temperature was 4.5 C.
Show your work.
(e) Use technology to construct a residual plot. Describe what you see.
(f) Find and interpret the value of r 2 in this setting.
4.64 Teacher pay and liquor sales A study found a strong positive correlation be-
tween average teacher salaries and liquor sales over a 12-month period. Does this sug-
gest that we should not pay teachers more because they would only spend the money
on more liquor? Explain your answer clearly.
 PA RT B

PRODUCING DATA

What kind of music

do you like? Rap?
Rock? Alternative?
Country? Some of
your friends probably
share your taste in
music. But you and
your friends are not
typical. Your parents
and grandparents
likely have different
tastes. They are not
PEANUTS reprinted by permission of United Feature Syndicate Inc.
typical either. To get
a true picture of the country as a whole (or even of high school students),
you must recognize that the picture may not resemble you or what you see
around you. You need data.
Data from retail sales show that the top-selling music types are rock (31.8% of
all recorded music sold in 2008) and country (11.9%).1 You might like rap (10.7%)
and I might like rhythm and blues (10.2%), but that doesn’t mean we have a clue
about the tastes of the music-buying public as a whole. If we are in the music
business, or even if we are interested in pop culture, we have to put our own
tastes aside and look at the data.
You can find data in the library or on the Internet (that’s where we found the
music sales data). How can we know whether data can be trusted? Good data are
as much a human product as shoes and MP3 players. Sloppily produced data will
frustrate you as much as a sloppily made pair of shoes.You examine shoes before you
buy, and you don’t buy if they are not well made. Neither should you use data that are
not well made. This part of this book shows how to tell if data are well made.

199
This page intentionally left blank
 CHAPTER 5

Sampling and
Surveys
5.1 Samples, Good and Bad

AP Photo/Evan Vucci
5.2 What Do Samples Tell Us?
5.3 Sample Surveys in the Real World

Why can’t pollsters get it right?

Follow any election closely, and you’ll see numerous polls that try to predict
the winner. Newscasters and political commentators frequently reference polls
taken by Gallup, Zogby, and many others, and political strategists use polling data
to help make decisions. Even experienced pollsters sometimes get it wrong,
as we saw in the 2008 New Hampshire Democratic primary involving Hillary
Clinton and Barack Obama. Heading into the primary, every poll predicted that
Obama would win—some by a very large margin. In fact, Clinton won the New
Hampshire primary by 3%. (Senator Obama did go on to win the presidential
election, however.) Why is it so difficult to get accurate poll results?
Consider this. Suppose the pollsters choose a “representative” sample of
400 people from the population and ask them which candidate they prefer. If
60% of the sample say they will vote for candidate A, what can we say about the
population percent who favor A? Is it exactly 60%? Probably not. Is it close to
60%? That depends on what we mean by “close.”
If the pollsters had chosen a different sample of 400 people, they probably
wouldn’t get exactly 60% who favor candidate A. That’s the idea of sampling
variability. The good news is that sample results vary around the population truth.
How much they vary depends mainly on the size of the sample, as you’ll see
later in the chapter. Sample surveys come with practical difficulties in addition to
sampling variability, as the New Hampshire primary results remind us.
So what happened in New Hampshire? We can’t be sure, but we can
speculate. Maybe the pollsters selected samples that weren’t representative
of the population of actual voters. Some people in the sample couldn’t be
reached or refused to answer. What effect did this nonresponse have on the
results? Those who did respond may not have told the truth about their voting

201
202 CHAPTER 5 Sampling and Surveys

intentions or might have changed their minds later. Sample surveys assume that
we can measure people’s views accurately at a moment in time. That’s easy to
say, but much harder to do.

5.1 Samples, Good and Bad

ACTIVITY 5.1A
How many fish are in the lake?
Materials: Large bag of Goldfish crackers and small different-colored fish as you had in your sample.
bag of different Goldfish crackers; brown paper Then put these tagged fish into the paper bag.
lunch bag; small paper cup. (Colored marbles can be 4. Thoroughly mix the Goldfish in the bag. (This
used instead of Goldfish.) imitates the fish swimming around in the lake
In this Activity, you will learn how scientists use after the tagged fish have been replaced.) Now use
sampling to estimate the size of an animal population, the paper cup to “recapture” a new sample of fish.
like the number of deer in a large forest or the Count the number of tagged and untagged fish in
number of fish in a lake. You will also investigate how your sample.
information obtained from a sample relates to the 5. With your classmates, discuss a method for
truth about the corresponding population. calculating the total number of fish in the lake.
1. Prior to the Activity, your teacher will count the (Here’s a hint: what should be true about the
number of “fish” in a large bag of Goldfish crackers overall proportion of tagged fish in the lake and the
and then pour all of the crackers into a brown proportion of tagged fish in your sample?)
paper lunch bag (the “lake”). Your class’s goal is to 6. Repeat Step 4 four more times to obtain a total of
determine the total number of fish in the lake. five estimates of the number of fish in the lake. Find
2. Without looking in the bag, use a small paper the average (mean) of your five estimates.
cup to “capture” a sample of fish. Count the number 7. Your teacher will now tell you the actual number
of fish in your sample. of fish in the lake. How close were your estimates to
3. We would like to find a way to “tag” the fish in the actual size of the fish population?
your sample so that you can identify them if you 8. If scientists used the capture-recapture method
capture them again later. Your teacher will provide from this Activity to estimate the size of an animal
you with some different-colored Goldfish to represent population in the wild, what practical problems
these tagged fish. Count out the same number of might they encounter?

How to sample badly

Example 5.1 The Town Talk takes an opinion poll

In Rapides Parish, Louisiana, only one company is allowed to provide ambu-
lance service. The local paper, the Town Talk, asked readers to call in to offer
their opinions on whether the company should keep its monopoly. Call-in polls
 5. 1 Samples, Good and Bad 203

are generally automated: call one telephone number to vote “Yes’’ and call an-
other number to vote “No.’’ Telephone companies often charge callers to dial
these numbers.
The Town Talk got 3763 calls, which suggests unusual interest in ambulance
service. Investigation showed that 638 calls came from the ambulance company of-
fice or from the homes of its executives. Many more no doubt came from lower-level
employees. “We’ve got employees who are concerned about this situation, their job
stability and their families and maybe called more than they should have,’’ said a
company vice president. Other sources said employees were told to “vote early and
often.’’1
We will see that a sample size of 3763 is quite respectable—if the sample is prop-
erly designed. Inviting people to call (and call, and call . . . ) isn’t proper sample de-
sign. We are about to learn how the pros take a sample, and why it beats the Town
Talk method.

As the Town Talk learned, it is easier to sample badly than to sample well. The
paper relied on voluntary response, allowing people to call in rather than actively se-
lecting its own sample. The result was biased—the sample was overweighted with
people favoring the ambulance monopoly. Voluntary response samples attract
people who feel strongly about the issue in question. These people, like the em-
ployees of the ambulance company, may not fairly represent the opinions of the
entire population.
There are other ways to sample badly. Suppose that I sell your company several
crates of oranges each week. You examine a sample of oranges from each crate to
determine the quality of my oranges. It is easy to inspect a few oranges from the
top of each crate, but these oranges may not be representative of the entire crate.
Those on the bottom are more often damaged in shipment. If I were less than
honest, I might make sure that any rotten oranges are packed on the bottom, with
some good ones on top for you to inspect. If you sample from the top, your sample
results are again biased—the sample oranges are systematically better than the
population they are supposed to represent.

Biased sampling methods

The design of a statistical study is biased if it systematically favors

certain outcomes.

A voluntary response sample chooses itself by responding to a general

appeal. Write-in or call-in opinion polls are examples of voluntary
response samples.

Selection of whichever individuals are easiest to reach is called

convenience sampling.

Convenience samples and voluntary response samples are often biased.

204 CHAPTER 5 Sampling and Surveys

Inspecting the oranges at the top of the crate is one example of convenience
sampling. Mall interviews are another.

Example 5.2 Interviewing at the mall

Manufacturers and advertising agencies often use interviews at shopping malls to
gather information about consumer habits and the effectiveness of ads. A sample
of mall shoppers is fast and cheap. But people contacted at shopping malls are
not representative of the entire U.S. population. They tend to be richer, for ex-
ample, than the population as a whole. In addition, interviewers tend to select
neat, safe-looking people to talk to. Mall samples are biased: they systematically
overrepresent some parts of the population and underrepresent others. The opin-
ions of such a convenience sample may be very different from those of the entire
population.

Example 5.3 No kids for me!

Ann Landers once asked the readers of her advice column, “If you had it to do
over again, would you have children?’’ She received nearly 10,000 responses, almost
70% saying “NO!’’ Can it be true that 70% of parents regret having children? Not
at all. This is a voluntary response sample. People who feel strongly about an issue,
particularly people with strong negative feelings, are more likely to take the trouble
to respond. Ann Landers’s results are strongly biased—the percent of parents who
would not have children again is much higher in her sample than in the population
of all parents.

“Hey, Pops, what was that letter you sent off to Ann Landers yesterday?”

Write-in and call-in opinion polls are almost sure to lead to strong bias. In fact,
only about 15% of the public has ever responded to a call-in poll, and these tend to
be the same people who call radio talk shows. That’s not a representative sample of
the population as a whole.
 5. 1 Samples, Good and Bad 205

www
The Internet brings voluntary response polls to the computer nearest you. Visit
www.misterpoll.com to become part of the sample in any of dozens of online polls.
As the site says, “None of these polls are ‘scientific,’ but do represent the collective
opinion of everyone who participates.’’

Exercises 5.1 Instant opinion A recent online poll posed the question: “Should female
athletes be paid the same as men for the work they do?’’ In all, 13,147 (44%)
said “Yes,’’ another 15,182 (50%) said “No,’’ and the remaining 1448 said
“Don’t know.’’
(a) What is the sample size for this poll?
(b) That’s a much larger sample than standard sample surveys. In spite of this,
we can’t trust the result to give good information about any clearly defined
population. Why?
5.2 An online poll In June 2008, Parade magazine posed the following question:
www
“Should drivers be banned from using all cell phones?” Readers were encouraged to
vote online at parade.com. The July 13, 2008, issue of Parade reported the results:
2407 (85%) said “Yes” and 410 (15%) said “No.”
(a) What type of sample did the Parade survey obtain?
(b) Explain why this sampling method is biased. Is 85% probably higher or lower
than the true percent of all adults who believe that cell phone use while driving
should be banned? Why?
5.3 Sleepless nights How much sleep do high school students get on a typical
school night? An interested student designed a survey to find out. To make data
collection easier, the student surveyed the first 100 students to arrive at school on
a particular morning. These students reported an average of 7.2 hours of sleep on
the previous night.
(a) What type of sample did the student obtain?
(b) Explain why this sampling method is biased. Is 7.2 hours probably higher
or lower than the true average amount of sleep last night for all students at the
school? Why?
5.4 Hand-washing habits Do adults typically wash their hands after using the
bathroom? In a telephone survey of 1001 U.S. adults, 92% said they always wash
their hands after using a public restroom.2 An observational study of 6076 adults in
public restrooms told a slightly different story: only 77% of those observed actually
washed their hands after using the restroom.3
(a) Why do you think the results of the two studies are so different?
(b) According to a description of the observational study, “Observers discreetly
watched and recorded whether or not adults using public restrooms washed their
hands. Observers were instructed to groom themselves (comb their hair, put on
make-up, etc.) while observing and to rotate bathrooms every hour or so to avoid
counting repeat users more than once. Observers were also instructed to wash
their hands no more than 10 percent of the time.” Explain why these precautions
were taken by the observers.
206 CHAPTER 5 Sampling and Surveys

5.5 Design your own bad sample A large high school wants to gather student
opinion about parking for students on campus. It isn’t practical to contact all
students.
(a) Give an example of a way to choose a sample of students that is bad because it
depends on voluntary response.
(b) Give another example of a bad way to choose a sample—but this time one that
doesn’t use voluntary response.
5.6 More capture-recapture Refer to Activity 5.1A (page 202). Mr. Washington’s
class takes an initial sample of 50 goldfish. After replacing the 50 tagged fish, the
class obtains a second sample of fish. Of the 75 fish in this sample, 25 of the tagged
fish are recaptured (along with 50 untagged fish).
(a) Compute an estimate of the total number of fish in the lake. Show your
method clearly.
(b) Is the number of fish you calculated in (a) definitely, probably, probably not,
or definitely not equal to the actual number of fish in the lake? Why?
(c) Explain why the capture-recapture sampling method used in this Activity is
not biased.

Simple random samples

In a voluntary response sample, people choose whether to respond. In a conve-
nience sample, the interviewer makes the choice. In both cases, personal choice
produces bias. The statistician’s remedy is to allow impersonal chance to choose the
sample. A sample chosen by chance allows neither favoritism by the sampler nor
self-selection by respondents. Choosing a sample by chance avoids bias by giving
all individuals an equal chance to be chosen. Rich and poor, young and old, black
and white, all have the same chance to be in the sample.
The simplest way to use chance to select a sample is to place names in a hat
(the population) and draw out a handful (the sample). This is the idea of a simple
random sample.

Simple random sample

A simple random sample (SRS) of size n consists of n individuals from

the population chosen in such a way that every set of n individuals has
an equal chance to be the sample actually selected.

An SRS not only gives each individual an equal chance to be chosen (thus
avoiding bias in the choice) but also gives every possible sample an equal chance
to be chosen. Drawing names from a hat does this. Write 100 names on identical
slips of paper and mix them in a hat. This is a population. Now draw 10 slips, one
after the other. This is an SRS, because any 10 slips have the same chance as any
other 10.
 5. 1 Samples, Good and Bad 207

Drawing names from a hat makes clear what it means to give each individual
and each possible set of n individuals the same chance to be chosen. That’s the idea
of an SRS. Of course, we would need a very big hat to get a sample of the country’s
117 million households. In practice, we use computer-generated random digits to
choose samples. If you don’t use software directly, you can use a table of random
digits to choose small samples by hand.

Random digits

A table of random digits is a long string of the digits 0, 1, 2, 3, 4, 5, 6,

7, 8, 9 with these two properties:

1. Each entry in the table is equally likely to be any of the 10 digits 0

through 9.
2. The entries are independent of each other. That is, knowledge
of one part of the table gives no information about any other
part.

Table B at the back of the book is a table of random digits. You can think of Table
B as the result of asking an assistant (or a computer) to mix the digits 0 to 9 in a hat,
draw one, then replace the digit drawn, mix again, draw a second digit, and so on.
The assistant’s mixing and drawing save us the work of mixing and drawing when
we need to randomize. Table B begins with the digits 19223950340575628713. To
make the table easier to read, the digits appear in groups of five and in numbered
Are these random rows. The groups and rows have no meaning—the table is just a long list of ran-
digits really random? domly chosen digits. Here’s how to use the table to choose an SRS.
Not a chance. The random
digits in Table B were pro-
duced by a computer pro-
gram. Computer programs Example 5.4 Choosing an SRS: teens and the Internet
do exactly what you tell
them to do. Give the pro- There has been much talk (and growing concern) about teenagers’ use of the Internet.
gram the same input and In addition to countless hours spent in chat rooms and surfing the Net, teens are
it will produce exactly the
viewing increasing amounts of pornography. Should stricter controls be imposed
same “random” digits. Of
course, clever people have on teen Internet use?
devised computer programs Let’s select a panel of teens to help answer this question. Ideally, we would like
that produce output that to have teens representing each of the 50 states on the panel. Unfortunately, budget
looks like random digits. restrictions will permit us to choose only 6 teens for the panel. So here’s what we’ll
These are called “pseudo- do. We’ll ask each state to select 1 teen nominee by a process of their choosing.
random numbers,” and
that’s what Table B contains.
Then, to avoid bias, we will select an SRS of size 6 from the 50 state nominees to
Pseudo-random numbers form the panel.
work fine for statistical
Step 1: Label. Give each state a numerical label, using as few digits as possible. Two
randomizing, but they have
hidden nonrandom patterns digits are needed to label 50 states, so we use labels
that can mess up more
advanced uses. 01, 02, 03, . . . , 48, 49, 50
208 CHAPTER 5 Sampling and Surveys

It is also correct to use labels 00 to 49 or even another choice of 50 two-digit labels.

Here is a partial list of states, with labels attached:

01 Alabama 02 Alaska 03 Arizona . . .

48 Washington 49 Wisconsin 50 Wyoming

Step 2: Table. Enter Table B anywhere and read two-digit groups. Suppose we enter
at line 103, which is

45467 71709 77558 00095 32863 29485 82226 90056

Each two-digit group in Table B is equally likely to be any of the 100 possible groups,
00, 01, 02, . . . , 99. So two-digit groups choose two-digit labels at random. That’s
just what we want.
We used only labels 01 to 50, so we ignore all other two-digit groups. The first
6 labels between 01 and 50 that we encounter in the table choose our sample. The
first 10 two-digit groups in line 103 are

45 46 77 17 09 77 55 80 00 95
U U skip U U skip skip skip skip skip

Notice that we skip six of these 10 labels because they are too high (over 50).
The others are 45, 46, 17, and 09. The nominees from the states labeled 09, 17, 45,
and 46 go into the sample. We need 2 more teenagers to complete the panel. The
remaining 10 two-digit groups in line 103 are

32 86 32 94 85 82 22 69 00 56
U skip skip skip skip skip U

The teenager from state 32 is chosen for the panel. Ignore the second 32 because
that state is already in the sample. State 22’s teen completes the panel.
The sample is composed of the states labeled 09, 17, 22, 32, 45, and 46. Our
panel consists of the teens nominated by those states.

Golfing at random
Random drawings give Using the table of random digits is much quicker than drawing names from a
everyone the same chance hat. As Example 5.4 shows, choosing an SRS has two steps.
to be chosen, so they offer
a fair way to decide who
gets a rare prize—like a
Choose an SRS in two steps
round of golf at the famous
Old Course at St. Andrews, Step 1: Label. Assign a numerical label to every individual in the
Scotland. A few can reserve
in advance. Most must hope population. Be sure that all labels have the same number of digits.
that chance favors them in Step 2: Table. Use random digits to select labels at random.
the daily random drawing
for tee times. At the height
of the summer season, only
1 in 6 wins the right to pay You can assign labels in any convenient manner, such as alphabetical order
over $100 for a round. for names of people. As long as all labels have the same number of digits, all
 5. 1 Samples, Good and Bad 209

individuals will have the same chance to be chosen. Use the shortest possible
labels: one digit for a population of up to 10 members, two digits for 11 to 100
members, three digits for 101 to 1000 members, and so on. As standard practice,
we recommend that you begin with label 1 (or 01 or 001, as needed). You can
read digits from Table B in any order—across a row, down a column, and so
on—because the table has no order. As standard practice, we recommend reading
across rows.

Example 5.5 Choosing an SRS: Congress joins the

teenagers
To add some clout to the teenagers’ recommendations about Internet use in
Example 5.4, we decide to add four members of the House of Representatives to the
panel. Again, we decide to take an SRS.
Step 1: Label. Obtain an alphabetical listing of the 435 members of the House of
Representatives. Assign numbers 001 to 435 as labels to the alphabetized list.
Step 2: Table. Enter Table B at line 134 (say) and read groups of three digits. The
first 10 three-digit groups are

278 167 841 618 329 213 373 521 337 741
U U skip skip U U

Our sample consists of the representatives labeled 278, 167, 329, and 213. These
individuals will join the teens on our panel.

Exercises 5.7 Apartment living You are planning a report on apartment living in a
college town. You decide to select three apartment complexes at random
for in-depth interviews with residents. Use Table B, starting at line 117,
to select a simple random sample of three of the following apartment
complexes. Explain your method clearly enough for a classmate to obtain
your results.
Ashley Oaks Country View Mayfair Village
Bay Pointe Country Villa Nobb Hill
Beau Jardin Crestview Pemberly Courts
Bluffs Del-Lynn Peppermill
Brandon Place Fairington Pheasant Run
Briarwood Fairway Knolls Richfield
Brownstone Fowler Sagamore Ridge
Burberry Franklin Park Salem Courthouse
Cambridge Georgetown Village Manor
Chauncey Village Greenacres Waterford Court
Country Squire Lahr House Williamsburg
210 CHAPTER 5 Sampling and Surveys

5.8 How do random digits behave? Which of the following statements are true
of a table of random digits, and which are false? Explain your answers.
(a) There are exactly four 0s in each row of 40 digits.
(b) Each pair of digits has chance 1/100 of being 00.
(c) The digits 0000 can never appear as a group, because this pattern is not
random.
5.9 An election day sample You want to choose an SRS of 25 of a city’s 440
voting precincts for special voting-fraud surveillance on election day.
(a) Explain clearly how you would label the 440 precincts. How many digits make
up each of your labels? What is the greatest number of precincts you could label
using this number of digits?
(b) Use Table B at line 107 to choose the first 10 precincts in the SRS.
5.10 Is this an SRS? A university has 1000 male and 500 female faculty
members. A survey of faculty opinion selects 100 of the 1000 men at random and
then separately selects 50 of the 500 women at random. The 150 faculty members
chosen make up the sample.
(a) Explain why this sampling method gives each member of the faculty an equal
chance to be chosen.
(b) Nonetheless, this is not an SRS. Why not?
5.11 Drug testing Use the table of random digits (Table B) to select an SRS of
3 of the following 25 members of an athletic team for a drug test. Be sure to say
where you entered the table and how you used it.
Agarwal Fuest Milhalko Shen
Andrews Fuhrmann Moser Smith
Baer Garcia Musselman Sundheim
Berger Healy Pavnica Wilson
Brockman Hixson Petrucelli
Chen Lee Reda
Frank Lynch Roberts
5.12 Not an SRS It’s sometimes not practical to take a simple random
sample. Here’s an example that shows why. Suppose 1000 iPhones are
produced at a factory today. Management would like to ensure that the
phones’ display screens meet their quality control standards before
shipping them to retail stores. Since it takes about 10 minutes to
inspect an individual phone’s display screen, managers decide to
inspect a sample of 20 phones from the day’s production.
(a) Explain why it would be difficult for managers to inspect an
SRS of 20 iPhones that are produced today.
(b) An eager employee suggests that it would be easy to
inspect the last 20 iPhones that were produced today. Why
y
lam

isn’t this a good idea?

A
rst/
Hu

(c) Another employee recommends inspecting every fiftieth iPhone

D.

that is produced. Explain carefully why this sampling method is not an SRS.
 5. 1 Samples, Good and Bad 211

Sampling with technology

Sample surveys use computer software to choose an SRS, but the software just au-
tomates the steps in Examples 5.4 and 5.5. The computer doesn’t look in a table of
random digits, because it can generate them on the spot, as Activity 5.1B illustrates.

ACTIVITY 5.1B
Simple random samples on the computer
In this Activity, you will learn how to select a simple 3. Think: If you were to sample three balls from
random sample from a population of interest using a population of size five, which is more likely to
a computer applet. You will also explore how results happen: getting the sample 1, 2, and 3 (in that
vary in repeated simple random samples from the order) or getting the sample 3, 1, and 5 (in that
same population. order)?
Materials: Computer with Internet access and Java 4. Think: If you were to repeatedly draw samples of
capability size three from a population of size five, how often
PLET would the largest of the three numbers be picked
AP 1. Go to the textbook Web site,
first?
www.whfreeman.com/sta2e. Click on the
Statistical Applets link, and then choose 5. Verify: Test your answers to Questions 3 and 4
the Simple Random Sample applet. You should see a by making repeated use of the applet. Record the
screen like Figure 5.1. results of each sample in a table. Did the samples
2. Enter the number 5 next to “Population 5 behave as you thought they would?
1 to” and click the Reset button. The “Population 6. Draw a sample of seven balls from a population
hopper” now contains five balls numbered 1 to 5. of size twenty. After the sample is drawn, what
Enter sample size 3 and click the Sample button. is left in the “Population hopper”? Is the sample
Record the numbers of the balls in the “Sample bin.” “random”? Explain your answer.

2 4 3

Population hopper Sample bin

5 2
1 4
3

Population = 1 to 5 Reset

Select a sample of size 3 Sample Figure 5.1 The Simple Random

Sample applet at the book’s Web site.
212 CHAPTER 5 Sampling and Surveys

To see how software speeds up choosing an SRS, go to the Research Randomizer

www
at www.randomizer.org. Click on “Randomize” and fill in the boxes. You can even
ask the Randomizer to arrange your sample in order.
You can also use a calculator to choose an SRS.

CALCULATOR CORNER How to choose an SRS

The TI-84 and TI-Nspire have built-in commands for generating “ran-
dom” numbers. Just how “random” are they?
TI-84
1. Press Ì, then select PRB and 5:randInt(.Complete the
command randInt(1,5) and press b.

TI-Nspire
1. Open a New Document and create a calculator page. Press c, choose 6: New
Document, and 1:Add Calculator.

Generate a “random” integer between 1 and 5. Press b and choose

4:Probability, 4:Random, and 2:Integer. Complete the
command randInt(1,5) and press ·.

Determine by a show of hands how many students in the class got each
number: 1, 2, 3, 4, and 5. Do the results seem random?
 5. 1 Samples, Good and Bad 213

2. Press b four more times. Record the sequence of five numbers you
obtain. Compare your results with those of your classmates. Do the results seem
random?
3. Calculators and computers use algorithms written by humans to generate
“random” numbers. The TI-84 and TI-Nspire randInt command uses a
starting value, called a “seed,” to determine the next “random” integer between 1
and 5. If several students in your class obtained the same values in Steps 1 and 2,
it is because their calculators were using the same seed. We can improve on this
situation by having each student input a different seed to the random number
generator. Here’s one method: use the last four digits (####) of your home phone
number as the seed (I used 0594).

TI-84
Type ####, press ¿ Ì, choose PRB and 1:rand, then press b to
generate the command #### ➔ rand.

TI-Nspire
Press b, then choose 4:Probability, 4:Random, and 6:Seed. Complete
the command RandSeed #### and press ·.
4. Now do randInt(1,100). See if anyone else in the class got the same
number.

5. Now you’re ready to choose an SRS of students from your class. Your teacher
will provide a class roster. Number the students on the roster 1, 2, 3, . . . in some
sensible way. On your calculator, execute the command randInt(1,n)
where n is the number of students in your class. Note who is selected. Press
b four more times. Did you get five different numbers? If not, press b
a few more times until you have selected five people for your sample. Note who
is selected.

Did you notice an important difference between the samples produced by the
Simple Random Sample applet of Activity 5.1B and the calculator’s RandInt com-
mand in the Calculator Corner? The applet sampled without replacement from the
“Population hopper,” but the calculator sampled with replacement from the spec-
ified population. As a result, the calculator sometimes selects the same number
more than once in a given sample. To deal with this problem, you can generate
additional random numbers as needed to replace any repeats. Alternatively, you can
214 CHAPTER 5 Sampling and Surveys

use a method other than RandInt (like RandSamp on the TI-Nspire) to sample
without replacement. Refer to your device’s reference manual or your teacher for
more details.

DILBERT reprinted by permission of United Feature Syndicate, Inc.

Can you trust a sample?

The Town Talk, Ann Landers, and mall interviews produce samples. We can’t trust
results from these samples, because they are chosen in ways that invite bias. We
have more confidence in results from an SRS, because it uses impersonal chance
to avoid bias. The first question to ask of any sample is whether it was chosen at
random. Opinion polls and other sample surveys carried out by people who know
what they are doing use random sampling.

Example 5.6 A Gallup Poll

A Gallup Poll on smoking began with the question “Have you, yourself, smoked any
cigarettes in the past week?’’ The press release reported that “just 25% of Americans
say they smoked cigarettes in the past week.’’ Can we trust this fact? Ask first how
Gallup selected its sample. Later in the press release we read this: “These results are
based on telephone interviews with a randomly selected national sample of 1,007
adults, aged 18 years and older, conducted July 6–9, 2006.’’4
This is a good start toward gaining our confidence. Gallup tells us what population
it has in mind (people at least 18 years old living anywhere in the United States). We
know that the sample from this population was of size 1007 and, most important, that it
was chosen at random. There is more to say, and we will soon say it, but we have at least
heard the comforting words “randomly selected.’’ (In the 2007 and 2008 Gallup polls on
smoking, only 21% of American adults said that they had smoked cigarettes in the past
week.)

What should a nation do if there are not enough volunteers for military ser-
vice but only a small fraction of eligible youth are needed by the military? Th at
question last arose during the Vietnam era. Beginning in 1970, a draft lottery
was used to choose draftees by random selection. Although the draft was ended
in 1976, young men must still register at age 18. The draft lottery may return if
the military cannot attract enough volunteers.
 5. 1 Samples, Good and Bad 215

Actually making a random selection in the public arena isn’t always easy. In
principle, it is just like choosing an SRS. In practice, random digits can’t be used.
Because few people understand random digits, a lottery looks fairer if a respected
person chooses capsules from a glass bowl in front of the TV cameras. This also
prevents cheating: no one can check the table of random digits in advance to see
how Cousin Joe will make out in the selection.
Physical mixing and drawing look random, but it can be hard to achieve a mixing
that really is random. There is no better illustration of this than the first Vietnam era
draft lottery, held in 1970.

ws
Ne
ws
Ne

APPLICATION 5.1 B]
RO
g
<
Se
a

The draft lottery

Because taking an SRS of all eligible men would The scatterplot in Figure 5.2 displays the re-
be hopelessly awkward, the draft lottery selected sults of the 1970 draft lottery. We have plotted the
birth dates in a random order. Men born on the variable “day of year” on the horizontal axis, with
first date chosen would be drafted first, then those Day 1  January 1 and Day 366  December 31.
born on the second date chosen, and so on. All The variable “draft number” is plotted on the vertical
men aged 19 to 25 were included in the first lot- axis.
tery, and there were 366 birth dates. The 366 dates 2. The highlighted point has coordinates (15, 17).
were placed into identical small capsules, which Explain what this point means in terms of the draft
were put in a bowl and publicly drawn one by one. lottery.
3. Describe the relationship that you see in the
QUESTIONS scatterplot between the day of the year on which
1. Explain why this method does not choose an people were born and the draft lottery number
SRS from all men eligible for the draft. from the drawing.

350

300
Draft number

250

200

150
100

50

0 50 100 150 200 250 300 350

Day of year

Figure 5.2 Scatterplot showing the results of the 1970 draft lottery.
APPLICATION 5.1 (continued)
216 CHAPTER 5 Sampling and Surveys

APPLICATION 5.1 (continued)

Figure 5.3 shows side-by-side boxplots of the were born and the draft lottery number from the
draft lottery results for each birth month. drawing.
4. Describe the relationship that you see in the
boxplots between the day of the year on which people

Jan.
Feb.
Mar.
Apr.
May
Month

June
July
Aug.
Sept.
Oct.
Nov.
Dec.

0 50 100 150 200 250 300 350 400

Draft number
Figure 5.3 Side-by-side boxplots of the draft numbers
for birth dates within each month.

As the 1970 draft lottery was taking place, news reporters quickly noticed that
men born later in the year seemed to receive lower draft numbers. Statisticians
soon showed that this trend was so strong that it would occur less than once in a
thousand truly random annual lotteries. What’s done is done, and off to Vietnam
went too many men born in December. But for 1971, statisticians were brought in
to conduct the lottery.

Exercises 5.13 SRS applet Explain how you would use the Simple Random Sample
PLET
AP
applet from Activity 5.1B (page 211) to perform each of the following
tasks. Try each of your ideas to be sure that it works. Can the applet be
used in more than one way to carry out any of these tasks? Explain.
(a) Flip a coin.
(b) Shuffle a deck of 52 cards.
(c) Pick a jury of 12 people from a small town with a population of 500 adult
citizens.
5.14 SRS on the calculator Explain how you would use your calculator’s
randInt command to perform each of the tasks in Exercise 5.13, if possible.
Try each of your ideas to be sure that it works.
 5. 1 Samples, Good and Bad 217

5.15 Rating the police The Miami Police Department wants to know how
black residents of Miami feel about police service. A researcher prepares
several questions about the police. The police department chooses an SRS of
300 mailing addresses in predominantly black neighborhoods and sends a
uniformed black police officer to each address to ask the questions of an adult
living there.
(a) What are the population and the sample?
(b) Why are the results likely to be biased even though the sample is an SRS?
5.16 A call-in opinion poll Should the United Nations continue to have its
headquarters in the United States? A television program asked its viewers to call
in with their opinions on that question. There were 186,000 callers, 67% of whom
said “No.’’ A nationwide random sample of 500 adults found that 72% answered
“Yes’’ to the same question.5 Explain to someone who knows no statistics why the
opinions of only 500 randomly chosen respondents are a better guide to what all
Americans think than the opinions of 186,000 callers.
5.17 Local draft boards Prior to 1970, young men were selected for military service
by local draft boards. There was a complex system of exemptions and quotas that
allowed, for example, farmers’ sons and married young men with children to avoid
the draft. Do you think that random selection among all men of the same age is
preferable to making distinctions based on marital status? Give your reasons.
5.18 Random selection? Choosing at random is a “fair’’ way to decide who
gets some rare prize, in the sense that everyone has the same chance to win.
Random choice isn’t always a good idea—sometimes we don’t want to treat
everyone the same, because some people have a better claim. In each of the
following situations, would you support choosing at random? Give your
reasons in each case.
(a) The basketball arena has 4000 student seats, and 7000 students want tickets.
Should we choose 4000 of the 7000 at random?
(b) The list of people waiting for liver transplants is much larger than the number
of available livers. Should we let chance decide who gets a transplant?

Section 5.1 Summary

We select a sample in order to get information about some population. How can
we choose a sample that represents the population fairly? Convenience samples
and voluntary response samples are common but do not produce trustworthy
data because these sampling methods are usually biased. That is, they systemati-
cally favor some parts of the population over others in choosing the sample.
The deliberate use of chance in producing data is one of the big ideas of sta-
tistics. Random samples use chance to choose a sample, thus avoiding bias due to
personal choice. The basic type of random sample is the simple random sample,
which gives all samples of the same size the same chance to be the sample we actu-
ally choose. To choose an SRS by hand, use a table of random digits such as Table B
in the back of the book. Alternatively, you can use a computer or calculator random
number generator to help choose an SRS.
218 CHAPTER 5 Sampling and Surveys

Section 5.1 5.19 How much do students earn? A university’s financial aid office wants
Exercises to know how much it can expect students to earn from summer employment.
This information will be used in setting the level of financial aid. The population
contains 3478 students who have completed at least one year of study but have not
yet graduated. The university will send a questionnaire to an SRS of 100 of these
students, drawn from an alphabetized list.
(a) Describe how you will label the students in order to select the sample.
(b) Use Table B, beginning at line 105, to select the first 5 students in the sample.
5.20 More randomization Most sample surveys call residential telephone num-
bers at random. They do not, however, always ask their questions of the person
who picks up the phone. Instead, they ask about the adults who live in the resi-
dence and choose one at random to be in the sample. Why is this a good idea?
5.21 A biased sample You see a female student standing in front of the cafeteria,
now and then stopping other students to ask them questions. She says that she is
conducting a survey of student opinions about the quality of food in the cafeteria.
Explain why this sampling method is almost certainly biased.
5.22 Sampling TVs An electronics company has 50 large flat-screen televisions
ready for shipment, each labeled with one of the following serial numbers:

A1109 A2056 A2219 A2381 B0001

A1123 A2083 A2336 A2382 B0012
A1186 A2084 A2337 A2383 B0046
A1197 A2100 A2338 A2384 B1195
A1198 A2108 A2339 A2385 B1196
A2016 A2113 A2340 A2390 B1197
A2017 A2119 A2351 A2396 B1198
A2020 A2124 A2352 A2410 B1199
A2029 A2125 A2367 A2410 B1200
A2032 A2130 A2372 A2500 B1201

An SRS of 5 TVs must be chosen for inspection. Use Table B to do this, beginning
at line 139. Explain your method clearly.
5.23 Gun control Since the beginning of opinion polls over 50 years ago, at least
two-thirds of those surveyed said that they favored stronger controls on firearms.
Specific gun control proposals have often been favored by 80% to 85% of respon-
dents. Yet little national gun control legislation has passed, and no major national
restrictions on firearms exist. Why do you think this has occurred?
5.24 A lottery for drugs? Only limited supplies of some experimental drugs are
available because the drugs are very difficult to make. Sometimes it is necessary to
use a lottery to decide at random which patients can receive a drug that is in short
supply. Several years ago, for example, Hoffman–La Roche had enough doses of
the drug Invirase for 2880 AIDS patients. Shortly after announcing this, the com-
pany received 10,000 calls from patients who wanted to enter the lottery.
Discuss this practice. Do you favor random selection? If not, how should
recipients be chosen?
5.25 Read anything lately? The Denver Public Library wants to estimate the
percent of Denver households with an adult who has read at least one book in the last
 5. 2 What Do Samples Tell Us? 219

month. The homes of 400 people who have library cards are sampled, and it turns out
that 90% of these households have an adult who has read a book in the past month.
Is 90% likely to be a biased estimate for the true percent of Denver households with
an adult who has read at least one book in the last month? Explain why or why not.
5.26 Women count! Broadcast and print media across the country have reported
on the government’s release of a report on the status of working women called
“Working Women Count.” This report gives findings from a survey of 250,000
women. According to one news report:
More than 1600 businesses, unions, newspapers, magazines, and community ser-
vice organizations helped distribute the survey to their members, subscribers, and
patrons, which the White House announced with much fanfare. It sought women’s
opinions on job satisfaction, pay, benefits, and opportunities for advancement.6
A second survey, asking the same questions, was conducted at the same time.
However, this survey interviewed only 1200 working women chosen at random.
Which survey is likely to give a more accurate view of working women’s
opinions on job satisfaction, pay, benefits, and opportunities for advancement?
Explain briefly.

5 .2 What Do Samples Tell Us?

ACTIVITY 5.2A
Should gambling be legal?
Figure 5.4 (on the next page) shows a small 1/10, 2/10, . . . , 9/10, and 10/10. That is, the sample
population. Each circle represents an adult. The proportion is always one of the following values: 0, 0.1,
colored circles are people who disapprove of legal 0.2, . . . , 0.9, or 1.
gambling, and the white circles are people who
4. Taking samples of size 10 from a population of size
approve. You can check that 60 of the 100 circles
100 is not a setting that occurs often in the real world,
are white, so in this population the proportion
but let’s look at your results anyway. How many of your
who approve of gambling is 60/100  0.6.
10 samples estimated the population proportion of 0.6
1. The circles are labeled 00, 01, . . . , 99. Your exactly correctly? Is the true value 0.6 roughly in the
teacher will assign each student in the class a center of your sample values? Explain why 0.6 would
different line of Table B to use for this Activity. Use be in the center of the sample values if you took a large
your assigned line to take an SRS of size 10. Explain number of samples.
your method clearly enough for a classmate to
repeat the process. What is the proportion of the 5. Calculate the average of your 10 sample proportions.
people in your sample who approve of gambling? How does this value compare with the proportion of
adults in the population who favor legal gambling?
2. Take 9 more SRSs of size 10 (10 samples in all).
Be sure to use a different starting point in Table B for 6. Pool results with your classmates. Make a large
each sample. You now have 10 values of the sample dotplot on the board. Describe what you see.
proportion. 7. Calculate the average of the class’s sample
3. Because your samples have only 10 people, the proportions. Compare with the true population
only values the sample proportion can take are 0/10, proportion.
ACTIVITY 5.2A (continued)
220 CHAPTER 5 Sampling and Surveys

ACTIVITY 5.2A (continued)

04 06 07
00 01 02 05
03
08 14
09 10 12 13
11 15
16
22
17
19 20 21 24
18 23
25
31
26 33
28 30 32
29

27 41 43
40 42
34 35 38
37 39
44 51
36 50 53
48 49
45 52
46 47
54 59 60
58 62
55
57 61
56
67 68
71
63 64 65 66 69
70

72 76
73 74 77
75 79 80
78
82 84 86
81 83
85 88
87
90 91 89
92
96 93 94
95
97 98
99

Figure 5.4 A population of 100 individuals for Activity 5.2A. Some

individuals (white circles) approve of legal gambling, and others
disapprove.

From sample to population

You know that lotteries are popular. How popular? Here’s what the Gallup Poll says:
“They can offer massive jackpots—and a ticket just costs a dollar at your neighborhood
store. For many Americans, picking up a lottery ticket has become routine, despite the
massive odds against striking it rich. A Gallup Poll Social Audit on gambling shows
 5. 2 What Do Samples Tell Us? 221

that 57% of Americans have bought a lottery ticket in the last 12 months, making
lotteries by far the favorite choice of gamblers.” Reading further, we find that Gallup
talked with “1523 randomly selected adults” to reach these conclusions.7
Gallup’s finding that “57% of Americans have bought a lottery ticket in the last
12 months” makes a claim about the population of 235 million adults. But Gallup
doesn’t know the truth about this population. The poll contacted 1523 people and
found that 57% of them said they had bought a lottery ticket in the past year. Because
the sample of 1523 people was chosen at random, it’s reasonable to think that they rep-
resent the entire population pretty well. So Gallup turns the fact that 57% of the sample
bought lottery tickets into an estimate that about 57% of all adults bought tickets. That’s
a basic strategy in statistics: use a fact about a sample to estimate the truth about the
whole population. To think about such a strategy, we must keep straight whether a
number describes a sample or a population. Here is the vocabulary we use.

Parameters and statistics

A parameter is a number that describes the population. A parameter

is a fixed number, but in practice we don’t know its value.

A statistic is a number that describes a sample. The value of a sta-

tistic is known when we have taken a sample, but it can change from
sample to sample. We often use a statistic to estimate an unknown
parameter.

So “parameter” is to “population” as “statistic” is to “sample.” Want to estimate

an unknown parameter? Choose a sample from the population and use a sample
statistic as your estimate. That’s what Gallup did.

Example 5.7 Who plays Lotto?

The proportion of all adults who bought a lottery ticket in the past 12 months is a
parameter describing the population of 235 million adults. Call it p, for “propor-
tion.” Alas, we don’t know the numerical value of p. To estimate p, Gallup took a
sample of 1523 adults. The proportion of the sample who bought lottery tickets is
a statistic. Call it p^ (read as “p-hat”). It happens that 868 of this sample of size 1523
bought tickets, so for this sample,
868
p^ 5 5 0.57 (that is, 57%)
1523
Because all adults had the same chance to be among the chosen 1523, it seems reason-
able to use the statistic p^  0.57 as an estimate of the unknown parameter p. It’s a fact that
57% of the sample bought lottery tickets—we know because we asked them. We don’t
know what percent of all adults bought tickets, but we estimate that about 57% did.
222 CHAPTER 5 Sampling and Surveys

Exercises 5.27 Stop smoking! A random sample of 1000 people who signed a card saying
they intended to quit smoking were contacted 9 months later. It turned out that
210 (21%) of the sampled individuals had not smoked over the past 6 months.
Specify the population of interest, the parameter of interest, the sample, and the
sample statistic in this problem.
Each boldface number in Exercises 5.28 to 5.30 is the value of either a parameter
or a statistic. In each case, state which it is.
5.28 Drink Arizona On Tuesday, the bottles of Arizona iced tea filled in a plant
were supposed to contain an average of 20 ounces of iced tea. Quality control
inspectors sampled 50 bottles at random from the day’s production. These bottles
contained an average of 19.6 ounces of iced tea.
5.29 Flight safety On a New York–to–Denver flight, 8% of the 125 passengers
were selected for random security screening prior to boarding. According to the
Transportation Security Administration, 10% of airline passengers are chosen for
random screening.
5.30 Sleeping ducks, I A recent report in the journal Nature examined
whether ducks keep an eye out for predators while they sleep. The researchers,
from Indiana State University, put four ducks in each of four plastic boxes,
which were arranged in a row. Ducks in the two end boxes slept with one eye
open 31.8% of the time, compared with only 12.4% of the time for the ducks in
the two center boxes.
5.31 Sleeping ducks, II Is the study described in the previous exercise an
example of an observational study or a comparative experiment? Explain
briefly.
5.32 Dead trees On the west side of Rocky Mountain National Park, many
mature pine trees are dying due to infestation by pine beetles. Scientists would
like to use sampling to estimate the proportion of all pine trees in the area that
have been infected.
(a) Explain why it wouldn’t be practical for scientists to obtain an SRS in this
setting.
(b) A possible alternative would be to use every pine tree along the park’s
main road as a sample. Why is this sampling method biased?
(c) Suppose that a more complicated random sampling plan is carried out, and
that 35% of the pine trees in the sample are infested by the pine beetle. Can
scientists conclude that 35% of all the pine trees on the west side of the park are
AP Photo/Ed Andrieski infested? Why or why not?

Sampling variability
If Gallup took a second random sample of 1523 adults, the new sample would
have different people in it. It is almost certain that there would not be exactly
868 positive responses. That is, the value of the statistic p^ will vary from sample
to sample. Could it happen that one random sample finds that 57% of adults
recently bought a lottery ticket and a second random sample finds that only 37%
had done so? Random samples eliminate bias from the act of choosing a sample,
 5. 2 What Do Samples Tell Us? 223

but they can estimate a population proportion badly because of the variability
that results when we choose at random. If the variation when we take repeated
samples from the same population is too great, we can’t trust the results of any
one sample.
We are saved by the second great advantage of random samples. The first
advantage is that choosing at random eliminates favoritism. That is, random
sampling reduces bias. The second advantage is that, if we took lots of ran-
dom samples of the same size from the same population, the variation from
sample to sample would follow a predictable pattern. This predictable pattern
shows that results of bigger samples are less variable than the results of smaller
samples.

Example 5.8 Lots and lots of samples

Here’s another big idea of statistics: to see how trustworthy one sample is likely
to be, ask what would happen if we took many samples from the same popula-
tion. Let’s try it and see. Suppose that in fact (unknown to Gallup) exactly 60%
of all adults have bought a lottery ticket in the past 12 months. That is, the truth
about the population is that p  0.6. What if Gallup used the sample propor-
tion p^ from an SRS of size 100 to estimate the unknown value of the population
proportion p?
Figure 5.5 illustrates the process of choosing many samples and finding p^ for
each one. In the first sample, 56 of the 100 people had bought lottery tickets, so
p^  56/100  0.56. Only 46 in the next sample had bought tickets, so for that
sample p^  0.46. We let technology choose 1000 samples and make a histogram of
the 1000 values of p^ . That’s the graph at the right of Figure 5.5. The different values
of p^ run along the horizontal axis. The heights of the bars show how many of our
1000 samples gave each group of values.

SRS n = 100
Number of samples

p = 0.56
400
SRS n = 100
p = 0.46 300
SRS n = 100 200
p = 0.61
100
0
0.45 0.50 0.55 0.60 0.65 0.70 0.75
p = 0.60 Values of p

Figure 5.5 The results of many SRSs have a regular pattern. Here we draw 1000 SRSs
of size 100 from the same population. The population proportion is p  0.6. The sample
proportions vary from sample to sample, but their values center at the truth about the
population.
Of course, Gallup interviewed 1523 people, not just 100. Figure 5.6 shows
the results of 1000 SRSs, each of size 1523, drawn from a population in which
the true proportion is p  0.6. Figures 5.5 and 5.6 are drawn on the same scale.
Comparing them shows what happens when we increase the size of our samples
from 100 to 1523.
224 CHAPTER 5 Sampling and Surveys

SRS n = 1523

Number of samples
p = 0.609
400
SRS n = 1523
p = 0.625 300
SRS n = 1523 200
p = 0.579
100
0
0.45 0.50 0.55 0.60 0.65 0.70 0.75
p = 0.60 Values of p

Figure 5.6 Draw 1000 SRSs of size 1523 from the same population as in Figure 5.5.
The 1000 values of the sample proportion are much less spread out than was the case for
the smaller samples.

Look carefully at Figures 5.5 and 5.6. We flow from the population, to many samples
from the population, to the many values of p^ from these many samples. Gather these
values together and study the histograms that display them.
• In both cases, the values of the sample proportion p^ vary from sample to
sample, but the values are centered at 0.6. Recall that p  0.6 is the true
population parameter. Some samples have a p^ less than 0.6 and some greater,
but there is no tendency to be always low or always high. That is, p^ has no
bias as an estimator of p. This is true for both large and small samples.
• The values of p^ from samples of size 100 are much more spread out than the
values from samples of size 1523. In fact, 95% of our 1000 samples of size
1523 have a p^ lying between 0.576 and 0.624. That’s within 0.024 on either
side of the population truth 0.6. Our samples of size 100, on the other hand,
spread the middle 95% of their values between 0.50 and 0.69. That goes out
0.1 from the truth, about four times as far as the larger samples. So larger
random samples have less variability than smaller samples.
We can rely on a sample of size 1523 to almost always give an estimate p^ that is
close to the truth about the population. Figure 5.6 illustrates this fact for just one value
of the population proportion, but it is true for any population. Samples of size 100, on
the other hand, might give an estimate of 50% or 70% when the truth is 60%.
Thinking about Figures 5.5 and 5.6 helps us restate the idea of bias when we
use a statistic like p^ to estimate a parameter like p. It also reminds us that variability
matters as much as bias.

Two types of error in estimation

Bias is consistent, repeated deviation of the sample statistic from the

population parameter in the same direction when we take many samples.

Variability describes how spread out the values of the sample statis-
tic are when we take many samples. Large variability means that the
result of sampling is not repeatable.

A good sampling method has both small bias and small variability.
 5. 2 What Do Samples Tell Us? 225

We can think of the true value of the population parameter as the bull’s-eye
on a target, and of the sample statistic as an arrow shot at the bull’s-eye. Bias
and variability describe what happens when an archer shoots many arrows at the
target. Bias means that the aim is off, and the arrows land consistently off the
bull’s-eye in the same direction. The sample values do not center about the pop-
ulation value. High variability means that repeated shots are widely scattered on
the target. Repeated samples do not give similar results but differ widely among
themselves. Figure 5.7 shows this target illustration of the two types of error.

(a) High bias, low variability ( b) Low bias, high variability

(c) High bias, high variability (d ) The ideal: low bias,

low variability

Figure 5.7 Bias and variability in shooting arrows at a target. Bias means the archer
systematically misses in the same direction. Variability means the arrows are scattered.

Notice that low variability (repeated shots are close together) can accompany
high bias (the arrows are consistently away from the bull’s-eye in one direction).
And low bias (the arrows center on the bull’s-eye) can accompany high variability
(repeated shots are widely scattered). A good sampling scheme, like a good archer,
must have both low bias and low variability. Here’s how we do this.

Managing bias and variability

To reduce bias, use random sampling. When we start with a list

of the entire population, simple random sampling produces unbiased
estimates—the values of a statistic computed from an SRS neither
consistently overestimate nor consistently underestimate the value of
the population parameter.

To reduce the variability of an SRS, use a larger sample. You can make
the variability as small as you want by taking a large enough sample.
226 CHAPTER 5 Sampling and Surveys

In practice, Gallup takes only one sample. We don’t know how close to the
truth an estimate from this one sample is, because we don’t know what the truth
about the population is. But large random samples almost always give an estimate
that is close to the truth. Looking at the pattern of many samples shows that we can
trust the result of one sample.

ACTIVITY 5.2B
Sampling heights
Materials: Identical, small pieces of paper or 4. The process in Step 3 should be repeated until
cardstock—one for each student in the class; hat every student in the class has selected an SRS of size 4.
Then, students should calculate the average height
Think of all the students in your class as the
of the 4 students in their samples.
population of interest. What is the average height
of the population? Later in the Activity, you’ll 5. As a class, make a large dotplot of the heights
compute this population parameter. But for now, let’s of all students in the class. We call this graph the
investigate how close the average height for a random population distribution. Compute the population
sample of 4 students in the class (the sample statistic) parameter—the average height of all students in the
is likely to be to the truth about the population. class. Draw a vertical line through this value.
1. Write your name and height, in inches, on one 6. Immediately below the graph from Step 5, make
side of the paper provided by your teacher. a dotplot of the average heights from the samples of
2. Put your piece of paper in the hat with those of size 4. Use the same horizontal scale as you did for
your classmates. the graph in Step 5.

3. Your teacher will mix the pieces of paper in the 7. Is the dotplot in Step 6 centered at the population
hat and then ask a student to draw 4 slips of paper parameter? Explain why this makes sense.
from the hat. Notice that the 4 students chosen are a 8. Which distribution shows more variability
simple random sample of size 4 from the population around the population parameter—the population
of interest. The student should quickly record the distribution or the distribution of sample averages?
names and heights of the 4 students in the SRS and In general, how close are the values of the sample
then return all 4 slips to the hat. statistic to the population parameter?

Exercises 5.33 Bias and variability Figure 5.8 (on the facing page) shows the behavior of a
sample statistic in many samples in four situations. These graphs are like those in
Figures 5.5 and 5.6. That is, the heights of the bars show how often the sample statistic
took various values in many samples from the same population. The true value of the
population parameter is marked by an arrow on each graph. Label each of the graphs
in Figure 5.8 as showing high or low bias and as showing high or low variability.
5.34 Sampling variability In thinking about Gallup’s sample of size 1523,
we asked, “Could it happen that one random sample finds that 57% of adults
recently bought a lottery ticket and a second random sample finds that only
37% had done so?” Look at Figure 5.6 (page 224), which shows the results of
1000 samples of this size when the population truth is 60%. Would you be
 5. 2 What Do Samples Tell Us? 227

Population parameter Population parameter

(a) (b)

Population parameter Population parameter

(c) (d)

Figure 5.8 Take many samples from the same population and make a histogram of
the values taken by a sample statistic. Here are the results for four different sampling
methods.

surprised if a sample from this population gave 57%? Would you be surprised
if a sample gave 37%?
5.35 No hands! Students from Hunter College in New York City carried out an
observational study of driver behaviors at 50 different intersections. Of the 3120
drivers they observed, 23% were talking on cell phones.8
(a) In national surveys of driver behavior, well over half of those surveyed admit
that they regularly talk on cell phones while driving. Give at least two reasons why
the Hunter College study gave such different results.
(b) If the Hunter College students had observed twice as many drivers, would the
percent of drivers talking on cell phones have been about the same as the results
from national surveys? Why or why not?
5.36 Predict the election Just before a presidential election, a national opinion poll
increases the size of its weekly sample from the usual 1500 people to 4000 people.
(a) Does the larger random sample reduce the bias of the poll result? Explain.
(b) Does it reduce the variability of the result? Explain.
5.37 A sampling simulation Let us illustrate sampling variability in a small
sample from a small population. Ten of the 25 club members listed below are
female. Their names are marked with asterisks in the list. The club chooses
5 members at random to receive free trips to the national convention.
Alonso Darwin Herrnstein Myrdal Vogt*
Binet* Epstein Jimenez* Perez* Went
Blumenbach Ferri Luo Spencer* Wilson
Chase* Gonzales Moll* Thomson Yerkes
Chen* Gupta* Morales* Toulmin Zimmer
(a) Draw 20 SRSs of size 5, using a different part of Table B each time. Record the
number of females in each of your samples.
(b) Make a dotplot to display your results. What is the average number of females
in your 20 samples?
(c) Do you think the club members should suspect discrimination if none of the
5 tickets goes to a woman? Justify your answer.
228 CHAPTER 5 Sampling and Surveys

5.38 Canada’s national health care The Ministry of Health in the Canadian
province of Ontario wants to know whether the national health care system is
achieving its goals in the province. Much information about health care comes
from patient records, but that source doesn’t allow us to compare people who use
health services with those who don’t. So the Ministry of Health conducted the
Ontario Health Survey, which interviewed a random sample of 61,239 people who
live in the province of Ontario.9
(a) What is the population for this sample survey? What is the sample?
(b) The survey found that 76% of males and 86% of females in the sample had
visited a general practitioner at least once in the past year. Do you think these
estimates are close to the truth about the entire population? Why?

Margin of error and all that

The margin of error that sample surveys announce translates sampling variability
of the kind shown in Figure 5.5 (page 223) and Figure 5.6 (page 224) into a state-
ment of how much confidence we can have in the results of a survey. Let’s start with
the kind of language we hear so often in the news.

What a margin of error means

“Margin of error plus or minus 3 percentage points” is shorthand for

this statement:

If we took many samples using the same method we used to get this
one sample, 95% of the samples would give a result within plus or
minus 3 percentage points of the truth about the population.

Take this step-by-step. A sample chosen at random will usually not estimate the
truth about the population exactly. We need a margin of error to tell us how close
our estimate comes to the truth. But we can’t be certain that the truth differs from
the estimate by no more than the margin of error. Ninety-five percent of all samples
come this close to the truth, but 5% miss by more than the margin of error. We
don’t know the truth about the population, so we don’t know if our sample is one of
the 95% that hit or one of the 5% that miss. We say we are 95% confident that the
truth lies within the margin of error.

Example 5.9 Understanding the news

Here’s what the TV news announcer says: “A new Gallup Poll finds that 57% of American
adults bought a lottery ticket in the last 12 months. The margin of error for the poll was
3 percentage points.” Of course, 57% plus or minus 3 percent is 54% to 60%. Most people
think that Gallup claims that the truth about the entire population lies in that range.
This is what Gallup actually said: “For results based on a sample of this size, one
can say with 95% confidence that the error attributable to sampling and other random
 5. 2 What Do Samples Tell Us? 229

effects could be plus or minus 3 percentage points for adults.” That is, Gallup tells us
that the margin of error works for only 95% of all its samples. And “95% confidence”
is shorthand for that. The news report left out the “95% confidence.”

Finding the margin of error exactly is a job for statisticians. You can, however, use
a simple formula to get a rough idea of the size of a sample survey’s margin of error.

A quick method for the margin of error

If you use the sample proportion p^ from a simple random sample

of size n to estimate an unknown population proportion p, then the
margin of error for 95% confidence is roughly equal to 1y!n.

Example 5.10 What is the margin of error?

The Gallup Poll in Example 5.9 interviewed 1523 people. The margin of error for
95% confidence will be about
1 1
5 5 0.026 (that is, 2.6%)
"1523 39.03
Gallup actually announced a margin of error of 3%. Our result differs a bit from
Gallup’s for two reasons. First, polls usually round their announced margin of
error to the nearest whole percent to keep their press releases simple. Second, our
rough formula works for an SRS. We will see in the next section that most national
samples are more complicated than an SRS in ways that tend to slightly increase the
margin of error. Still, our quick method comes pretty close.

“A new poll shows 67% of Americans think polls are inaccurate 52% of the time, and 84%
think the margin of error of 3% is inaccurate 71% of the time . . .”
230 CHAPTER 5 Sampling and Surveys

Example 5.11 Estimating the margin of error

In Example 5.8 we compared the results of taking many SRSs of size n  100 and
many SRSs of size n  1523 from the same population. We found that the spread of
the middle 95% of the sample results was about four times as large for the smaller
samples.
Our quick formula estimates the margin of error for SRSs of size 1523 to be
about 2.6%. The margin of error for SRSs of size 100 is about
1 1
5 5 0.1 (that is, 10%)
"100 10
Because 1523 is roughly 16 times 100 and the square root of 16 is 4, the margin of error
is about four times larger for samples of 100 people than for samples of 1523 people.

Our quick method also reveals an important fact about how margins of error
behave. Because the sample size n appears in the denominator of the fraction, larger
samples have smaller margins of error. We knew that. Because the formula uses the
square root of the sample size, however, to cut the margin of error in half, we must
use a sample four times as large.

Confidence statements
Here is Gallup’s conclusion about buying lottery tickets in short form: “The poll
found that 57% of adults bought a lottery ticket in the past 12 months. We are 95%
confident that the truth about all adults is within plus or minus 3 percentage points
of this sample result.” Here is an even shorter form: “We are 95% confident that
between 54% and 60% of all adults bought a lottery ticket in the last 12 months.”
These are confidence statements.

Confidence statements

A confidence statement has two parts: a margin of error and a level

of confidence. The margin of error says how close the sample statistic
lies to the population parameter. The level of confidence says what
percent of all possible samples satisfy the margin of error.

A confidence statement is a fact about what happens in all possible samples; it is

used to state how much we can trust the result of one sample. The phrase “95% confi-
dence” means “We used a sampling method that gives a result this close to the truth 95%
of the time.” Here are some hints for interpreting confidence statements:
• The conclusion of a confidence statement always applies to the population, not to
the sample. We know exactly how the 1523 people in the sample acted, because
Gallup interviewed them. The confidence statement uses the sample result to
say something about the population of all adults.
 5. 2 What Do Samples Tell Us? 231

• Our conclusion about the population is never completely certain. Gallup’s sample
might be one of the 5% that miss by more than 3 percentage points.
• A sample survey can choose to use a confidence level other than 95%. We pay for
higher confidence with a larger margin of error. For the same sample, a 99%
confidence level requires a larger margin of error than 95% confidence. If you
are content with 90% confidence, you get a smaller margin of error in return.
Remember that our quick method gives the margin of error only for 95%
confidence.
• It is usual to report the margin of error for 95% confidence. If a news report gives
a margin of error but leaves out the confidence level, it’s pretty safe to assume
95% confidence.
• Want a smaller margin of error with the same confidence? Take a larger sample.
Remember that larger samples yield less variable sample statistics than smaller
samples. You can get as small a margin of error as you want and still have high
confidence by paying for a large enough sample.

Example 5.12 Do you approve of gambling?

Gallup began its poll on gambling with this question: “First, generally speaking, do
you approve or disapprove of legal gambling or betting?” In addition to the 1523
adults (aged 18 and older) whose lottery habits we have explored, Gallup took a
random sample of 501 teenagers (aged 13 to 17). The sample results were

Adults: 959 out of 1523 approve p^  959/1523  0.63

Teens: 261 out of 501 approve p^  261/501  0.52

After reporting these and other results, Gallup says: “For results based on a sam-
ple of this size, one can say with 95 percent confidence that the error attributable to
sampling and other random effects could be plus or minus 3 percentage points for
adults (18), and plus or minus 5 percentage points for teens (13–17 year olds).”
There you have it: the sample of teens is smaller, so the margin of error for conclu-
sions about teens is wider. We are 95% confident that between 47% (that’s 52% minus 5%)
and 57% (that’s 52% plus 5%) of all teenagers approve of legal gambling.

Exercises 5.39 Take a bigger sample A student is planning a project on student attitudes
toward part-time work while attending school. She develops a questionnaire and plans
to ask 25 randomly selected students to fill it out. Her statistics teacher approves the
questionnaire but suggests that the sample size be increased to at least 100 students.
Why is the larger sample helpful? Back up your answer by using the quick method to
estimate the margin of error for samples of size 25 and for samples of size 100.
5.40 Find the margin of error Example 5.12 tells us that Gallup asked 501
teenagers whether they approved of legal gambling; 52% said they did. Use the quick
method to estimate the margin of error for conclusions about all teenagers. How
does your result compare with Gallup’s margin of error quoted in Example 5.12?
232 CHAPTER 5 Sampling and Surveys

5.41 The Current Population Survey Though opinion polls usually make
95% confidence statements, some sample surveys use other confidence levels.
The monthly unemployment rate, for example, is based on the Current
Population Survey of about 50,000 households. The margin of error in the
unemployment rate is announced as about two-tenths of 1 percentage point
with 90% confidence. Would the margin of error for 95% confidence be
smaller or larger? Why?
5.42 Polling women Many years ago, a New York Times Poll on women’s issues
interviewed 1025 women randomly selected from the United States, excluding
Alaska and Hawaii. One question was “Many women have better jobs and more
opportunities than they did 20 years ago. Do you think women had to give up
too much in the process, or not?” Forty-eight percent of women in the sample
said “Yes.”10
(a) The poll announced a margin of error of 63 percentage points for 95%
confidence in its conclusions. Make a 95% confidence statement about the percent
of all adult women who felt that women had to give up too much.
(b) Explain to someone who knows no statistics why we can’t just say that 48% of
all adult women felt that women had to give up too much.
(c) Explain clearly what “95% confidence” means.
5.43 Polling men and women The sample survey described in Exercise 5.42
interviewed 472 randomly selected men as well as 1025 women. The poll
announced a margin of error of 63 percentage points for 95% confidence in
conclusions about women. The margin of error for results concerning men was
65 percentage points. Why is this larger than the margin of error for women?
5.44 Is there a heaven? A news article reports that in a recent opinion poll, 81%
of a sample of 1003 adults said they believe there is a heaven.
(a) Use the quick method to estimate the margin of error for a sample of this size.
(b) Make a confidence statement about the percent of all adults who believe there
is a heaven.

Sampling from large populations

Gallup’s sample of 1523 adults is only 1 out of every 154,000 adults in the United
States. Does it matter whether 1523 is 1-in-100 individuals in the population or
1-in-154,000?

Population size doesn’t matter

The variability of a statistic from a random sample does not depend

on the size of the population, as long as the population is at least
10 times larger than the sample.

Why does the size of the population have little influence on the behavior of
statistics from random samples? Imagine sampling harvested corn by thrusting
a scoop into a large sack of corn kernels. The scoop doesn’t know whether it is
 5. 2 What Do Samples Tell Us? 233

surrounded by a bag of corn or by an entire truckload. As long as the corn is well

mixed (so that the scoop selects a random sample), the variability of the result
depends only on the size of the scoop.
You can read the Gallup Organization’s own explanation of why surpris-
www
ingly small samples can give trustworthy results about large populations at
media.gallup.com/PDF/FAQ/HowArePolls.pdf.
This is good news for national sample surveys like the Gallup Poll. A random
sample of size 1000 or 1500 has small variability because the sample size is large.
But remember that even a very large voluntary response sample or convenience
sample is worthless because of bias. Taking a larger sample doesn’t fix bias.
However, the fact that the variability of a sample statistic depends on the size of
the sample and not on the size of the population is bad news for anyone planning
a sample survey in a university or a small city. For example, it takes just as large an
SRS to estimate the proportion of Ohio State University students who blog regu-
larly as to estimate with the same margin of error the proportion of all U.S. adults
who blog regularly. We can’t use a smaller SRS at Ohio State just because there are
50,000 Ohio State students and 235 million adults in the United States.

ws
Ne
ws
Ne

APPLICATION 5.2 B]
RO
g
<
Se
a

Doping questions cloud Americans’ view of the Olympics

when a swimmer sets a world record, and 1 in 3 be-
lieve that at least some Olympians in swimming use
performance-enhancing drugs, the poll indicates.
The results reflect how scandals involving perfor-
mance-enhancing drugs—notably in Major League
Baseball, track and field, and professional cycling—
have created a credibility gap in the public’s mind.
QUESTIONS
1. The poll described in the article sampled 626
sports fans. Why do you think the pollsters asked
sports fans rather than people in general?
© Paul J. Sutton/PCN/PCN/Corbis
2. Here is the actual question that the poll asked:
Positive drug tests or admissions of doping in recent “When you see or hear about an athlete breaking
years have cost Olympic athletes their records, their a world record in swimming or track and field, are
medals, their careers, even their freedom. Just over you suspicious or not suspicious that the athlete
half of American sports fans believe that at least used performance-enhancing drugs?” Does the
some Olympic athletes in track and field use question seem clear? Is the question slanted in a
performance-enhancing drugs. A USA Today/ particular direction, or does it seem balanced?
Gallup Poll conducted just prior to the Beijing 3. The pie charts on the next page summarize the
Olympics (July 31, 2008) also shows that when a track- poll results. A bar graph might be more appropriate
and-field athlete sets a world record, more than 1 in 3 for comparing people’s opinions for these two
sports fans are suspicious that doping helped. More sports. Construct a well-labeled bar graph that
than 1 in 5 fans say they are suspicious of doping could be used for easy comparison.
APPLICATION 5.2 (continued)
234 CHAPTER 5 Sampling and Surveys

APPLICATION 5.2 (continued)

Swimming Track and field

No opinion No opinion
2% 2%

Suspicious
22%
Suspicious
35%
Not suspicious
Not suspicious 63%
76%

Along with the poll results, USA Today an- 5. Explain to someone who knows no statistics
nounced a margin of error of 64 percentage points. what the margin of error means in this case.
4. Use the quick estimate for the margin of error to
confirm this value.

Section 5.2 Summary

The purpose of sampling is to use a sample to gain information about a population.
We often use a sample statistic to estimate the value of a population parameter.
This section has one big idea: to describe how trustworthy a sample is, ask “What
would happen if we took a large number of samples from the same population?”
If almost all samples would give a result close to the truth, we can trust our one
sample even though we can’t be certain that it is close to the truth.
In planning a sample survey, first aim for small bias by using random sampling
and avoiding bad sampling methods such as voluntary response. Next, choose a large
enough random sample to reduce the variability of the result. Using a large random
sample guarantees that almost all samples will give accurate results. To say how accu-
rate our conclusions about the population are, make a confidence statement. News
reports often mention only the margin of error. Most often this margin of error is for
95% confidence. That is, if we chose many samples, the truth about the population
would be within the margin of error 95% of the time. We can estimate the margin of
error for 95% confidence based on a simple random sample of size n by the formula
1y!n. As this formula suggests, only the size of the sample, not the size of the popu-
lation, matters. This is true as long as the population is much larger than the sample.

Section 5.2 5.45 Strike three A survey is conducted in Chicago (population 2,800,000) using
Exercises random digit dialing equipment that places calls at random to residential phones,
both listed and unlisted. The purpose of the survey is to determine the percent of
Chicagoans who support the “three strikes and you’re out” provision of the crime bill
passed by Congress. One-tenth of 1% of the adult population is interviewed, and 53%
 5. 2 What Do Samples Tell Us? 235

of them favor the proposal. A second survey is taken in Dayton, Ohio (population
156,000), using the same techniques and asking the same question of one-tenth of
1% of the adults living in Dayton. Which survey is more likely to produce a sample
result that is closer to the truth about the population in that city? Explain briefly.
5.46 Presidential polls California has about four times as many voters as Ohio,
but both are very important states for the outcome of any presidential election.
One national polling organization plans to take surveys of 1000 randomly selected
voters in each of these two states on the day before the election. In which state is
the survey more likely to accurately predict the state’s election results? Explain.
Each boldface number in Exercises 5.47 and 5.48 is the value of either a parameter
or a statistic. In each case, state which it is.
5.47 Lefties According to “Real Facts” at Snapple.com, 13% of adults are left-
www
handed. At a math teacher’s conference, 16% of those attending were left-handed.
5.48 Single-sex classes In an experiment to test the effectiveness of single-sex
classrooms, girls assigned at random to a coeducational chemistry class gained an
average of 12.2 points from a pretest to a posttest. Girls assigned randomly to a
single-sex chemistry class taught by the same teacher gained 15.1 points.
5.49 Find the margin of error Exercise 5.38 (page 228) describes a sample
survey of 61,239 adults living in Ontario. About what is the margin of error for
conclusions having 95% confidence about the entire adult population of Ontario?
Show your work.
5.50 Explaining confidence A student reads that we are 95% confident that the
average score of young men on the quantitative part of the National Assessment
of Educational Progress is 267.8 to 276.2. Asked to explain the meaning of this
statement, the student says, “95% of all young men have scores between 267.8 and
276.2.” Is the student right? Explain your answer.
5.51 Teens on the Net A poll of 1070 teens aged 13 to 17 finds that 742 have
received personal messages online from people they don’t know. The announced
margin of error for this result is plus or minus 3 percentage points. The news
report does not give the confidence level, but you can be quite sure that it’s 95%.
(a) What is the value of the sample proportion p^ who have received personal mes-
sages online from people they don’t know? Explain in words what the population
parameter p is in this setting.
(b) Make a confidence statement about the parameter p.
5.52 Protecting teens The previous exercise describes a sample survey of 1070
teens, with margin of error 63% for 95% confidence.
(a) An agency that is responsible for protecting children thinks that 95% confi-
dence is not enough. They want to be 99% confident. How would the margin of
error for 99% confidence based on the same sample compare with the margin of
error for 95% confidence?
(b) A parents group is satisfied with 95% confidence but wants a smaller margin of
error than 63%. How can we get a smaller margin of error, still with 95% confidence?
5.53 Video gaming In a 2006 poll, 40% of American adults surveyed said that
they play computer or video games. Of those who played such games, 10% reported
playing for 10 or more hours per week. According to the Associated Press, “The
results are taken from a poll of 3,024 adults, including 1,046 who play electronic
games, that was conducted April 18–27. The poll, conducted by the international
236 CHAPTER 5 Sampling and Surveys

polling firm Ipsos, has a margin of sampling error of plus or minus 2 percentage
points, 3 points for the sample of those who play electronic games.”11
(a) Use the quick method to show that the margins of error quoted above are ap-
proximately correct if a simple random sample was taken.
(b) The confidence level wasn’t stated, but it’s almost certainly 95%. Make an appropri-
ate confidence statement about video gaming in the population of American adults.
(c) Identify two possible sources of error in this survey that the margin of error
does not include.
(d) If this survey were repeated today, do you think the results would be about the
same? Do some research to find whether your suspicion is correct.
5.54 Simulating summer employment Random digits can be used to simulate the
results of random sampling. Suppose that you are drawing simple random samples
of size 25 from a large number of college students and that 20% of the students are
unemployed during the summer. To simulate this SRS, let 25 consecutive digits in
Table B stand for the 25 students in your sample. The digits 0 and 1 stand for unem-
ployed students, and other digits stand for employed students. This is an accurate
imitation of the SRS because 0 and 1 make up 20% of the 10 equally likely digits.
Simulate the results of 50 samples by counting the number of 0s and 1s in the
first 25 entries in each of the 50 rows of Table B. Make a histogram like the one in
Figure 5.5 to display the results of your 50 samples. Is the truth about the popu-
lation (20% unemployed, or 5 in a sample of 25) near the center of your graph?
What are the smallest and largest counts of unemployed students you obtained in
your 50 samples? What percent of your samples had either 4, 5, or 6 unemployed?

5.3 Sample Surveys in the Real World

ACTIVITY 5.3
Design your own survey
In this Activity, you will work with two or three of 4. Exchange survey questions with another team,
your classmates to design a survey. By the end of this as instructed by your teacher. Discuss the content
section, you will be ready to revise your survey and to and clarity of the questions. Note any changes you
administer it at your school. would recommend on their question list.
1. On your own, brainstorm a list of current topics of 5. Review any suggestions made regarding your
interest to your school community. own list. Finalize your questions.
2. Pool ideas with your team members. Once you have 6. Ideally, you would select a simple random
agreed on a topic of interest, write three to five questions sample of students from your school to complete
that address the topic. Try to include one question that the survey. Give two reasons why this might not be
measures a quantitative (numerical) variable. practical.
3. Spend a few minutes editing your questions to 7. Make a list of other issues your team needs to
make them as clear as possible. Have one member of address before you can administer the survey.
the group write the revised list of questions neatly on Keep all of your work handy for later use.
a clean sheet of paper.
 5. 3 Sample Surveys in the Real World 237

The whole truth about opinion polls

An opinion poll is administered to 1000 people chosen at random, announces its
results, and announces a margin of error. Should we be happy? Maybe not. Many
polls don’t tell the whole truth about their samples. The Pew Research Center for
the People and the Press imitated the methods of the better opinion polls and did
tell the whole truth. Here it is.
Most polls are taken by telephone, dialing numbers at random to get a random
sample of households. After eliminating fax and business numbers, Pew had to call
2879 residential numbers to get their sample of 1000 people. Here’s the breakdown:
Never answered phone 938
Answered but refused 678
Not eligible: no person aged at
least 18, or language barrier 221
Incomplete interview 42
Complete interview 1000
Total called 2879

Out of 2879 good residential phone numbers, 33% never answered. Of those who
answered, 35% refused to talk. The overall rate of nonresponse (people who never
answered, refused, or would not complete the interview) was 1658 out of 2879, or
58%. Pew called every number five times over a five-day period, at different times of
day and on different days of the week. Many polls call only once, and it is usual to find
that over half of those who answer refuse to talk. In the real world, a simple random
sample isn’t simple and may not be random. That’s our issue for this section.12

How sample surveys go wrong

Random sampling eliminates bias in choosing a sample and allows control of vari-
ability. So once we see the magic words “randomly selected” and “margin of error,”
do we know we have trustworthy information before us? It certainly beats volun-
tary response, but not always by as much as we might hope. Sampling in the real
world is more complex and less reliable than choosing an SRS from a list of names
in a textbook exercise. Confidence statements do not reflect all of the sources of
error that are present in practical sampling.

Errors in sampling

Sampling errors are errors caused by the act of taking a sample. They
cause sample results to be different from the results of a census.

Random sampling error is the deviation between the sample statistic

and the population parameter caused by chance in selecting a random
sample. The margin of error in a confidence statement includes only
random sampling error.

Nonsampling errors are errors not related to the act of selecting a

sample from the population. They can be present even in a census.
238 CHAPTER 5 Sampling and Surveys

Most sample surveys are affected by errors other than random sampling errors.
These errors can introduce bias that makes a confidence statement meaningless.
Good sampling technique includes the art of reducing all sources of error. Part of
this art is the science of statistics, with its random samples and confidence state-
ments. In practice, however, good statistics isn’t all there is to good sampling. Let’s
look at sources of errors in sample surveys and at how samplers combat them.

Sampling errors
Random sampling error is one kind of sampling error. The margin of error tells us
how serious random sampling error is, and we can control it by choosing the size
of our random sample. Another source of sampling error is the use of bad sampling
methods, such as voluntary response. We can avoid bad methods. Other sampling
errors are not so easy to handle. Sampling begins with a list of individuals from
which we will draw our sample. This list is called the sampling frame. Ideally, the
sampling frame should list every individual in the population. Because a list of
the entire population is rarely available, most samples suffer from some degree of
undercoverage.

Undercoverage

Undercoverage occurs when some groups in the population are left

out of the process of choosing the sample.

If the sampling frame leaves out certain classes of people, even random samples
from that frame will be biased. Using telephone directories as the frame for a tele-
phone survey, for example, would miss everyone with an unlisted telephone num-
ber. More than half the households in many large cities have unlisted numbers, so
massive undercoverage and bias against urban areas would result. In fact, telephone
surveys usually use random digit dialing equipment that dials landline telephone
numbers in selected regions at random. In effect, the sampling frame contains all
residential telephone numbers. Of course, this sampling frame will exclude all in-
dividuals who have only a cell phone.

Example 5.13 We undercover

Most opinion polls can’t afford to even attempt full coverage of the population of all
adult residents of the United States. The interviews are done by telephone, thus miss-
ing the 6% of households without phones. Only households are contacted, so that
students in dormitories, prison inmates, and most members of the armed forces are
left out. So are the homeless and people staying in shelters. Because calls to Alaska
and Hawaii are expensive, most polls restrict their samples to the continental states.
Many polls interview only in English, which leaves some immigrant households out
of their samples. And most polls use sampling frames that consist only of residential
phone numbers, which omits people who use cell phones exclusively.
 5. 3 Sample Surveys in the Real World 239

The kinds of undercoverage found in most sample surveys are most likely to
leave out people who are young or poor or who move often. Nonetheless, random
digit dialing comes close to producing a random sample of households with phones
outside Alaska and Hawaii. Sampling errors in careful sample surveys are usually
quite small. The real problems start when someone picks up (or doesn’t pick up) the
phone. Now nonsampling errors take over.

Nonsampling errors
Nonsampling errors are those that can plague even a census. They include processing
errors, which are mistakes in mechanical tasks such as doing arithmetic or entering
responses into a computer. The spread of computer-assisted interviewing has made
processing errors less common than in the past.13

Example 5.14 Computer-assisted interviewing

The days of the interviewer with a clipboard are past. Contemporary interviewers
carry a laptop computer for face-to-face interviews or watch a computer screen
as they conduct a telephone interview. Computer software manages the interview.
The interviewer reads questions from the computer screen and uses the keyboard
to enter the responses. The computer skips irrelevant items—once a respondent
says that she has no children, further questions about her children never appear.
The computer can check that answers to related questions are consistent with each
other. It can even present questions in random order to avoid any bias due to always
asking questions in the same order.
Computer software also manages the sampling process. It keeps records of
who has responded and prepares a file of data from the responses. The tedious
process of transferring responses from paper to computer, once a source of pro-
cessing errors, has been eliminated. The computer even schedules the calls in
telephone surveys, taking into account the respondent’s time zone and honoring
appointments made by people who were willing to respond but did not have time
when first called.

Another type of nonsampling error is response error, which occurs when

someone gives an incorrect response. A person may lie about her age or income
or about whether she has used illegal drugs. She may remember incorrectly
when asked how many packs of cigarettes she smoked last week. Someone
who does not understand a question may guess at an answer rather than ap-
pear ignorant. Questions that ask people about their behavior during a fixed
time period are prone to response errors due to faulty memory. 14 For example,
the National Health Survey asks people how many times they have visited a
doctor in the past year. Checking their responses against health records found
that they failed to remember 60% of their visits to a doctor. A survey that asks
about sensitive issues can also expect response errors, as the next example
illustrates.
240 CHAPTER 5 Sampling and Surveys

Example 5.15 The effect of race

In 1989, New York City elected its first black mayor and the state of Virginia elected
its first black governor. In both cases, samples of voters interviewed as they left their
polling places predicted larger margins of victory than the official vote counts. The
polling organizations were certain that some voters lied when interviewed because
they felt uncomfortable admitting that they had voted against the black candidate.
History may have repeated itself in 2008. Preelection polls before the New
Hampshire primary projected Barack Obama as the winner over Hillary Clinton. In
fact, these polls suggested that Obama would win by about 8 percentage points—an
amount much larger than the margin of error for any of the polls. Clinton won the
New Hampshire primary, receiving 39% of the vote to Obama’s 36%.

Technology and attention to detail can minimize processing errors. Skilled inter-
viewers greatly reduce response errors, especially in face-to-face interviews. There is no
simple cure, however, for the most serious kind of nonsampling error, nonresponse.

Nonresponse

Nonresponse is the failure to obtain data from an individual selected for a

sample. Most nonresponse happens because some people can’t be con-
tacted or because some individuals who are contacted refuse to cooperate.

Nonresponse is the most serious problem facing sample

surveys. People are increasingly reluctant to answer ques-
tions, particularly over the phone. The rise of telemarket-
ing, answering machines, cell phones, and caller ID drives
down response to telephone surveys. Gated communities
and buildings guarded by doormen prevent face-to-face
interviews. Nonresponse can bias sample survey results be-
cause different groups have different rates of nonresponse.
Refusals are higher in large cities and among the elderly, for
example. Bias due to nonresponse can easily overwhelm
the random sampling error described by a survey’s margin
of error.
“You can call, you can send email, you can stand at
the door all day. The answer is still NO!”

Example 5.16 How bad is nonresponse?

The Current Population Survey has the best response rate of any poll we know:
only about 6% or 7% of the households in the CPS sample don’t respond. People
are more likely to respond to a government survey such as the CPS, and the CPS
contacts its sample in person before doing later interviews by phone.
 5. 3 Sample Surveys in the Real World 241

The General Social Survey (Example 1.8, page 11) also contacts its sample in
person, and it is run by a university. Despite these advantages, its recent surveys
have a 24% rate of nonresponse.
What about polls done by the media and by market research and opinion-polling
firms? We don’t know their rates of nonresponse, because they won’t say. That’s a
bad sign! The Pew study we looked at suggests how bad things are. Pew got 1221
responses (of whom 1000 were in the population they targeted) and 1658 who were
never at home, refused, or would not finish the interview. That’s a nonresponse rate
of 1658 out of 2879, or 58%. The Pew researchers were more thorough than many
pollsters. Insiders say that nonresponse often reaches 75% or 80% of an opinion
poll’s original sample.15

New York, New York

New York City, they say,
is bigger, richer, faster, and Sample surveyors know some tricks to reduce nonresponse. Carefully trained
sometimes ruder. Maybe interviewers can keep people on the line if they answer at all. Calling back after
there’s something to that. longer time periods helps. So do letters sent in advance. Letters and many callbacks
The sample survey firm
slow down the survey, so opinion polls that want fast answers to satisfy the media
Zogby International says
that as a national average it don’t use them. Even the most careful surveys find that nonresponse is a problem
takes 5 telephone calls to that no amount of expertise can fully overcome. That makes this reminder even
reach a live person. When more important.
calling to New York, it takes
12 calls. Survey firms assign
their best interviewers to
make calls to New York and What the margin of error doesn’t say
often pay them bonuses to
cope with the stress. The announced margin of error for a sample survey covers only
random sampling error. Undercoverage, nonresponse, and other
practical difficulties can cause large bias that is not covered by the
margin of error.

Careful sample surveys tell us this. Gallup, for example, says, “In addition
to sampling error, question wording and practical difficulties in conducting sur-
veys can introduce error or bias into the findings of public opinion polls.” How
true that is.

Exercises 5.55 Not in the margin of error A recent Gallup Poll found that 68% of adult
Americans favor teaching creationism along with evolution in public schools. The
Gallup press release says:
For results based on samples of this size, one can say with 95 percent confidence
that the maximum error attributable to sampling and other random effects is
plus or minus 3 percentage points.16
Give one example of a source of error in the poll result that is not included in this
margin of error.
242 CHAPTER 5 Sampling and Surveys

5.56 What kind of error? Which of the following are sources of sampling error
and which are sources of nonsampling error? Explain your answers.
(a) The subject lies about past drug use.
(b) A typing error is made in recording the data.
(c) Data are gathered by asking people to mail in a coupon printed in a
newspaper.
5.57 Internet users then A survey of Internet users in 1995 found that males
outnumbered females by nearly 2 to 1. This was a surprise, because earlier surveys
had put the ratio of men to women closer to 9 to 1. Later in the article we find this
information:
Detailed surveys were sent to more than 13,000 organizations on the Internet;
1,468 usable responses were received. According to Mr. Quarterman, the margin
of error is 2.8 percent, with a confidence level of 95 percent.17
(a) What was the response rate for this survey? (The response rate is the percent
of the planned sample that responded.)
(b) Use the quick method (page 229) to estimate the margin of error of this
survey. Is your result close to the 2.8% claimed?
(c) Is the small margin of error a good measure of the accuracy of the survey’s
results? Explain your answer.
5.58 Polling students A high school chooses an SRS of 100 students from the
school’s attendance list to interview about student life. If it selected two SRSs of
100 students at the same time, the two samples would give somewhat different
results. Is this variation a source of sampling error or of nonsampling error?
Will the survey’s announced margin of error take this source of error into
account?
5.59 Internet users now Who uses the Internet more today—males or females?
Find a report of a recent survey that you believe provides accurate information on
this question. Print the article if possible. Be sure to record all source information.
Write a brief report following the four-step statistical problem-solving process
(page 19).
5.60 Activity 5.3 follow-up Return to the survey your team constructed in
Activity 5.3 (page 236). Discuss how your design protects against
(a) sampling errors.
(b) nonsampling errors.

Wording questions
A final influence on the results of a sample survey is the exact wording of
questions. It is surprisingly difficult to word questions that are completely
clear. A survey that asked about “ownership of stock” found that most Texas
ranchers owned stock, though probably not the kind traded on the New York
Stock Exchange.
 5. 3 Sample Surveys in the Real World 243

The telemarketer’s
pause
People who do sample sur-
veys hate telemarketing. We
“Do I own any stock, Ma’am? Why, I've got 10,000 head out there.”
get so many unwanted sales
pitches by phone that many
people hang up before
learning that the caller is Example 5.17 A few words make a big difference
conducting a survey rather
than selling vinyl siding.
Here’s a tip. Both sample Question: How do Americans feel about government help for the poor? Only 13%
surveys and telemarketers think we are spending too much on “assistance to the poor,” but 44% think we are
dial telephone numbers spending too much on “welfare.”
at random. Telemarketers Question: How do the Scots feel about the movement to become independent
automatically dial many from England? Well, 51% would vote for “independence for Scotland,” but only
numbers, and their sellers
34% support “an independent Scotland separate from the United Kingdom.”18
come on the line only after
you pick up the phone. It seems that “assistance to the poor” and “independence” are nice, hopeful
Once you know this, the words. “Welfare” and “separate” are negative words. Small changes in how a ques-
telltale “telemarketer’s tion is worded can make a big difference in the response.
pause” gives you a chance
to hang up before the seller
arrives. Sample surveys have
a live interviewer on the The wording of questions always influences the answers. If the questions are
line when you answer.
slanted to favor one response over others, we have another source of nonsampling
error. A favorite trick is to ask if the subject favors some policy as a means to a
desirable end: “Do you favor banning private ownership of handguns in order to
reduce the rate of violent crime?” and “Do you favor imposing the death penalty in
order to reduce the rate of violent crime?” are loaded questions that draw positive
responses from people who are worried about crime.

Sample design in the real world

The basic idea of sampling is straightforward: take an SRS from the population
and use a statistic from your sample to estimate a population parameter. Unfor-
tunately, it’s usually very difficult to actually get an SRS from the population
of interest. Imagine trying to get a simple random sample of all the batteries
produced in one day at a factory. Or an SRS of all U.S. high school students.
244 CHAPTER 5 Sampling and Surveys

In either case, it’s just not practical to choose an SRS. In the real world, most
sample surveys use more complex designs.

Example 5.18 The Current Population Survey (CPS)

The CPS population consists of all households in the United States (including
Alaska and Hawaii). The sample is chosen in stages. The Census Bureau divides the
nation into 2007 geographic areas called Primary Sampling Units (PSUs). These
are generally groups of neighboring counties. At the first stage, 792 PSUs are cho-
sen. This isn’t an SRS. If all PSUs had the same chance to be chosen, the sample
might miss Chicago and Los Angeles. So 432 highly populated PSUs are auto-
matically in the sample. The other 1575 are grouped into 360 strata by combining
PSUs that are similar in various ways. One PSU is chosen at random to represent
each stratum. From there, groups of nearby households in each PSU are randomly
selected for the CPS.19

The design of the CPS illustrates several ideas that are common in real-world
samples that use face-to-face interviews. Taking the sample in several stages
with nearby households at the final stage saves travel time for interviewers. The
most important refinement mentioned in Example 5.18 is stratified random
sampling.

Stratified random sample

To choose a stratified random sample:

Step 1: Divide the sampling frame into distinct groups of individuals,

called strata. Choose the strata because you have a special interest in
these groups within the population or because the individuals in each
stratum resemble each other.
Step 2: Take a separate SRS in each stratum and combine these to
make up the complete sample.

We must of course choose the strata using facts about the population that are
known before we take the sample. You might group a high school’s students by
grade level (9, 10, 11, 12) or by gender. Stratified samples have some advantages
over an SRS. First, by taking a separate SRS in each stratum, we can set sample sizes
to allow separate conclusions about each stratum. Second, a stratified sample usu-
ally has a smaller margin of error than an SRS of the same size. The reason is that
the individuals in each stratum are more alike than the population as a whole, so
working stratum by stratum eliminates some variability in the sample.
It may surprise you that stratified samples can violate one of the most appeal-
ing properties of the SRS—stratified samples need not give all individuals in the
population the same chance to be chosen. Some strata may be deliberately over-
represented in the sample.
 5.3 Sample Surveys in the Real World 245

Example 5.19 Stratifying a sample of students

A large university has 30,000 students, of whom 3000 are graduate students. An SRS of
500 students gives every student the same chance to be in the sample. That chance is

500 1

30,000 60

We expect an SRS to contain only about 50 grad students—because grad students

make up 10% of the population, we expect them to make up about 10% of an SRS.
A sample of size 50, however, isn’t large enough to estimate grad student opinion
with reasonable accuracy. Therefore, we prefer to take a stratified random sample
of 200 grad students and 300 undergraduates.
You know how to select such a stratified sample. Label the graduate students
0001 to 3000 and use Table B to select an SRS of 200. Then label the undergraduates
00001 to 27000 and use Table B a second time to select an SRS of 300 of them. These
two SRSs together form the stratified sample.
In the stratified sample, each grad student has chance

200 1

3000 15

to be chosen. Each of the undergraduates has a smaller chance,

300 1

27,000 90

It’s clear from the last few examples that designing samples is a business for ex-
perts. We won’t worry about such details. The big idea is that good sample designs
use chance to select individuals from the population. That is, all good samples are
probability samples.

Probability sample

A probability sample is a sample chosen by chance. We must know

what samples are possible and what chance, or probability, each pos-
sible sample has. Some probability samples, such as stratified samples,
don’t allow all possible samples from the population and may not give
an equal chance to all the samples they do allow.

A stratified sample of 300 undergraduate students and 200 grad students, for ex-
ample, only allows samples with exactly that makeup. An SRS would allow any 500
students. Both are probability samples. We need know only that estimates from any
probability sample share the nice properties of estimates from an SRS. Confidence
statements can be made without bias and have smaller margins of error as the size of
the sample increases. Nonprobability samples such as voluntary response samples do
246 CHAPTER 5 Sampling and Surveys

not share these advantages and cannot give trustworthy information about a popula-
tion. Now that we know that most nationwide samples are more complicated than an
SRS, we will usually go back to acting as if good samples were SRSs. That keeps the big
idea and hides the messy details.

Exercises 5.61 A sampling paradox? Example 5.19 compares two SRSs—of a university’s
undergraduate and graduate students. The sample of undergraduates contains
a smaller fraction of the population, 1 out of 90, versus 1 out of 15 for graduate
students. Yet sampling 1 out of 90 undergraduates gives a smaller margin of error
than sampling 1 out of 15 graduate students. Explain to someone who knows no
statistics why this happens.
5.62 Scholar-athletes To ask about their future plans, you want to interview 10
students at your high school who have received athletic scholarships to attend
college. Because you believe there may be large differences among athletes in
different sports, you decide to interview a stratified random sample of 7 basketball
players and 3 golfers. Use Table B, beginning at line 101, to select your sample
from the team rosters below. Explain your method carefully enough that a
classmate could obtain your results.

BASKETBALL
Arenas Duncan Leslie Robinson
Billups Farmar McGrady Stoudamire
Brand Fowles Miller Taurasi
Bryant Gasol Nowitzki Wade
Carter Iverson Parker

GOLF
Creamer Kim Singh
Els Mickelson Wie
Gulbis Prammanasudh Woods

5.63 Genetically modified foods An article in the journal Science looks at differences
in attitudes toward genetically modified foods in Europe and the United States. This
calls for sample surveys. The European survey chose a sample of 1000 adults in each of
17 European countries. Here’s part of the description: “The Eurobarometer survey is a
multistage, random-probability face-to-face sample survey.”20
(a) What does “multistage” mean?
(b) You can see that the first stage was stratified. What were the strata?
(c) What does “random-probability sample” mean?
5.64 Wording of questions A New York Times/CBS News Poll asked a
random sample of Americans about abortion: “Do you think there should be an
amendment to the Constitution prohibiting abortions, or shouldn’t there be such an
amendment?” The same people were later asked, “Do you believe there should be an
amendment to the Constitution protecting the life of the unborn child, or shouldn’t
there be such an amendment?” For one of the questions, 50% were in favor and 39%
 5.3 Sample Surveys in the Real World 247

were opposed. For the other, 29% were in favor and 62% were opposed. (The rest
were uncertain.)21 Which question do you think yielded each result? Explain why.
5.65 Closed versus open questions Two basic types of questions are closed
questions and open questions. A closed question asks the subject for one or more
of a fixed set of responses. An open question allows the subject to answer in his or
her own words. The interviewer writes down the responses and sorts them later.
An example of an open question is

How do you feel about broccoli?

An example of a closed question is

What is your opinion about broccoli? Do you

a. like it very much? b. like it somewhat? c. neither like nor dislike it?
d. dislike it somewhat? e. dislike it very much?

What are the advantages and disadvantages of open and closed questions?
5.66 Bad survey questions Write your own examples of bad sample survey
questions.
(a) Write a biased question designed to get one answer rather than another.
(b) Write a question that is confusing, so that it is hard to answer.
5.67 Systematic random sample The final stage in a multistage sample must
choose 5 of the 500 addresses in a neighborhood. You have a list of the 500 addresses
in geographical order. To choose a systematic random sample, proceed as follows:
Step 1. Choose 1 of the first 100 addresses on the list at random. (Label them 00,
01, . . . , 99 and use a pair of digits from Table B to make the choice.)
Step 2. The sample consists of the address from Step 1 and the addresses 100,
200, 300, and 400 positions down the list from it.
If 71 is chosen at random in Step 1, for example, the systematic random sample
consists of the addresses numbered 71, 171, 271, 371, and 471.
(a) Use Table B to choose a systematic random sample of 5 from a list of 500
addresses. Enter the table at line 130.
(b) What is the chance that any specific address will be chosen? Explain your answer.
(c) Explain why this sample is not an SRS.
5.68 A stratified sample A university has 2000 male and 500 female faculty
members. The university president wants to poll the opinions of a random sample of
faculty members. In order to give adequate attention to female faculty opinion, the
president decides to choose a stratified random sample of 200 males and 200 females.
The president obtains alphabetized lists of female and male faculty members.
(a) Explain how you would assign labels and use random digits to choose the
desired sample. Enter Table B at line 122 and give the first 5 females and the first
5 males in your sample.
(b) What is the chance that any one of the 2000 males will be in your sample?
What is the chance that any one of the 500 females will be in your sample?
248 CHAPTER 5 Sampling and Surveys

(c) Each member of the sample is asked, “In your opinion, are female faculty
members in general paid less than males with similar positions and qualifications?”
180 of the 200 females (90%) say “Yes.”
60 of the 200 males (30%) say “Yes.”
In all, 240 of the sample of 400 (60%) answered “Yes.” The president therefore reports:
“Based on a sample, we can conclude that 60% of the total faculty feel that female
members are underpaid relative to males.” Explain why this conclusion is wrong.
(d) If we took a stratified random sample of 200 male and 50 female faculty
members at this university, each member of the faculty would have the same chance
of being chosen. What is that chance? Explain why this sample is not an SRS.

DATA EXPLORATION
Do Teachers Have Tattoos?
About 1100 high school teachers attended a the cafeteria for the morning snack break. Did this
weeklong summer institute for teaching AP classes. sampling method give every teacher at the institute
After studying Example 1.10 (page 20), the teachers an equal chance to be selected? Why or why not?
in the AP Statistics class wondered whether the Explain why this sampling method does not yield
results of the tattoo survey would be similar for an SRS of teachers at the institute.
teachers. They designed a survey to find out. The
4. Members of the class handed the survey to the
class opted for a sample size of 100 teachers. One of
randomly selected teachers. The respondents were
the questions on the survey was:
asked to complete the survey at a table nearby, and
Do you have any tattoos on YES NO then to fold the paper in half and place it in a box
your body? (Circle one) with a hole in the lid. Why do you think the class
decided to administer the survey in this way rather
1. One of the first decisions the class had to make than doing face-to-face interviews?
was what kind of sampling method to use. They knew
5. Two of the selected teachers refused to respond
that a simple random sample was the “preferred”
to the survey. How might this affect the results?
method. With 1100 teachers in 40 different sessions,
the class decided not to use an SRS. Give at least two 6. Some teachers did not come to the morning
reasons why you think they made this decision. snack break that day, including an entire class of
teachers who were on a field trip. How might this
2. The AP Statistics class believed that there might
affect the survey results?
be systematic differences in the proportions of
teachers who had tattoos based on the subject areas 7. Of the 98 teachers who responded, 23.5% said
that they taught. What sampling method would you that they had one or more tattoos. Although the
recommend to account for this possibility? sampling method used was not an SRS, the quick
method still gives a reasonable estimate for the
3. In the end, the class chose to use systematic
margin of error. Use the quick method to help you
random sampling (see Exercise 5.67) to choose
make a 95% confidence statement.
their sample. They used a calculator’s random
number generator to pick a number from 1 to 10. 8. Write a brief report summarizing this survey.
The result was a “7.” So the class decided to give the Follow the four-step statistical problem-solving
survey to the 7th, 17th, 27th, . . . people to arrive at process (page 19).
 5.3 Sample Surveys in the Real World 249

Questions to ask before you believe a poll

Opinion polls and other sample surveys can produce accurate and useful informa-
tion if the pollster uses good sampling techniques and also works hard at preparing
a sampling frame, wording questions, and reducing nonresponse. Many surveys,
however—especially those designed to influence public opinion rather than just
record it—do not produce accurate or useful information. Here are some questions
to ask before you pay much attention to poll results.
• Who carried out the survey? Even a political party should hire a professional sample
survey firm whose reputation demands that they follow good survey practices.
• What was the population? That is, whose opinions were being sought?
• How was the sample selected? Look for mention of random sampling.
• How large was the sample? Even better, find out both the sample size and the
margin of error within which the results of 95% of all samples drawn as this one
was would fall.
• What was the response rate? That is, what percent of the people who were
selected actually provided information?
• How were the subjects contacted? By telephone? Mail? Face-to-face interview?
• When was the survey conducted? Was it just after some event that might have
influenced opinion?
• What were the exact questions asked?
Academic survey centers and government statistical offi ces answer these ques-
tions when they announce the results of a sample survey. National opinion polls
usually don’t announce their response rate (which is often low) but do give us the
other information. Editors and newscasters have the bad habit of cutting out these
dull facts and reporting only the sample results. Many sample surveys by interest
groups and local newspapers and TV stations don’t answer these questions because
their polling methods are in fact unreliable. If a politician, an advertiser, or your
local TV station announces the results of a poll without complete information, be
skeptical.

ws
Ne
ws
Ne

APPLICATION 5.3 B]
RO
g
<
Se
a

What high school math courses help in college?

Many high school students are faced with the actually prepares more students for the majors they’ll
choice of taking a statistics course or a calculus pursue in college and for their careers. What’s a
course. Which is the better choice? The answer may student to do?
be “it depends.” College admissions staff often view A math department chair at a large private high
calculus as a more “rigorous” course when review- school wanted to get some advice from the school’s
ing applicants’ transcripts. Many of those same recent graduates about this issue. He contacted the
people concede that a high school statistics course school’s data analysis class and asked if they would
APPLICATION 5.3 (continued)
250 CHAPTER 5 Sampling and Surveys

APPLICATION 5.3 (continued)

design and carry out a survey for him. The students Survey of math course experiences
eagerly accepted the challenge. They began by get- 100%
ting a list of recent graduates (current juniors and Actually took
80%
Would take
seniors in college) with their email addresses from

Percent
60%
the school’s alumni office. After eliminating
40%
duplicate names, the students were left with a list
20%
of 428 recent graduates.
Once the survey questionnaire was finalized, 0%

us

s
s
us

tic
si
the class sent an email message to all 428 people on

B
l
l

ly

cu
cu

is
us

us
na

at
al
al

l
the list inviting them to take the survey online at

cu
A

St
cu
ec

al
Pr

al

P
at

C
www.surveymonkey.com. Some of the email mes-

A
D

P
A

A
sages bounced back, so students tried to track down
correct contact information for those individuals. QUESTIONS
With a reminder 2 weeks later, the class eventually 1. Why do you think the data analysis class decided
got 128 completed surveys. not to ask the survey questions in an email and to
The two questions of particular interest on the have people respond directly to the email?
survey were: 2. The class was unable to locate current email
• After Algebra 2, which mathematics courses did you addresses for 10 graduates. What sampling issue
take in high school? does this raise?
• If you were to design your high school course 3. What was the response rate for this survey? How
schedule again, which mathematics courses would might this nonresponse affect the results of the survey?
you take after Algebra 2? 4. Write a few sentences describing what the bar
The bar graph summarizes the 128 responses. graph tells you.

Section 5.3 Summary

Even professional sample surveys don’t give exactly correct information about the
population. There are many potential sources of error in sampling. The margin
of error announced by a sample survey covers only random sampling error, the
variation due to chance in choosing a random sample. Other types of error are in
addition to the margin of error and can’t be directly measured. Sampling errors
come from the act of choosing a sample. Random sampling error and undercoverage
are common types of sampling error. Undercoverage occurs when some members
of the population are left out of the sampling frame, the list from which the sample
is actually chosen.
The most serious errors in most careful surveys, however, are nonsampling
errors. These have nothing to do with choosing a sample—they are present
even in a census. The single biggest problem for sample surveys is nonresponse:
subjects can’t be contacted or refuse to answer. Mistakes in handling the data
(processing errors) and incorrect answers by respondents (response errors)
are other examples of nonsampling errors. Finally, the exact wording of ques-
tions has a big influence on the answers. People who design sample surveys
use statistical techniques that help correct nonsampling errors, and they also
 5.3 Sample Surveys in the Real World 251

use probability samples more complex than simple random samples, such as
stratified samples.
You can assess the quality of a sample survey quite well by just looking at the
basics: use of random samples, sample size and margin of error, the rate of nonre-
sponse, and the wording of the questions.

Section 5.3 5.69 Women engineers About 20% of the engineering students at a large uni-
Exercises versity are women. The school plans to poll a sample of 200 engineering students
about the quality of student life.
(a) If an SRS of size 200 is selected, about how many women do you expect to find
in the sample?
(b) If the poll wants to be able to report separately the opinions of male and fe-
male students, what type of sampling design would you suggest? Why?
5.70 Critiquing a poll The Wall Street Journal published an article on attitudes
toward the Social Security system based on a sample survey. It found, for example,
that 36% of people aged 18 to 34 expected Social Security to pay nothing at all
when they retire. News articles tend to be brief in describing sample surveys. Here
is part of the Wall Street Journal’s description of this poll:

The Wall Street Journal/NBC News poll was based on nationwide telephone
interviews of 2,012 adults, conducted Thursday through Sunday by the polling
organizations of Peter Hart and Robert Teeter.
The sample was drawn from 520 randomly selected geographic points in the
continental U.S. Each region was represented in proportion to its population.
Households were selected by a method that gave all telephone numbers, listed
and unlisted, an equal chance of being included.22

Page 249 lists several “questions to ask” about an opinion poll. What answers does
the Wall Street Journal give to each of these questions?
5.71 Multistage sampling The previous exercise gives part of the description of
a sample survey from the Wall Street Journal.
(a) It appears that the sample was taken in several stages. Why can we say this?
(b) The first stage no doubt used a stratified sample, though the Journal does not
say this. Explain why it would be bad practice to use an SRS from a large number of
“geographic points” across the country rather than a stratified sample of such points.
5.72 TV ratings The method of collecting the data can influence the accuracy of
sample results. The following methods have been used to collect data on television
viewing in a sample household:
(a) Diary method. The household keeps a diary of all programs watched and who
watched them for a week, then mails in the diary at the end of the week.
(b) Roster-recall method. An interviewer shows the person a list of programs for
the preceding week and asks which programs were watched.
(c) Telephone-coincidental method. The survey firm telephones the household at a
specific time and asks if the television is on, which program is being watched, and
who is watching it.
252 CHAPTER 5 Sampling and Surveys

(d) Automatic recorder method. A device attached to the set records what hours
the set is on and to which channel it is tuned. At the end of the week, this record is
removed from the recorder.
(e) People meter. Each member of the household is assigned a numbered button
on a hand-held remote control. Everyone is asked to push their button whenever
they start or stop watching TV. The remote control signals a device attached to the
set that keeps track of what channel the set is tuned to and who is watching at all
times.
Discuss the advantages and disadvantages of each of these methods, especially
the possible sources of error associated with each method. The Nielsen national
ratings use Method (e). Local ratings (there are more than 200 local television
markets) use Method (a). Do you agree with these choices? (Do not discuss
choosing the sample, just collecting the data once the sample is chosen.)
5.73 Did you vote? When the Current Population Survey asked the adults in its
sample of 50,000 households if they voted in the most recent presidential elec-
tion, 54% said they had. In fact, only 49% of the adult population voted in that
election.
(a) Use the quick method to estimate the margin of error for this survey. (Although
the CPS sample is not an SRS, the quick method still gives a reasonably accurate
estimate of the margin of error.)
(b) Why do you think the CPS result missed by much more than the margin of
error?
5.74 Telling the truth? Many people don’t give honest answers to questions
about illegal or sensitive issues. One study divided a large group of adults into
thirds at random. All were asked if they had ever used cocaine. The first group
was interviewed by telephone: 21% said “Yes.” In the group visited at home by an
interviewer, 25% said “Yes.” The final group was interviewed at home but an-
swered the question on an anonymous form that they sealed in an envelope. Of
this group, 28% said they had used cocaine.23
(a) Which result do you think is closest to the truth? Why?
(b) Give two other examples of behavior you think would be underreported in a
telephone survey.
5.75 What kind of error? Each of the following is a source of error in a sample
survey. Label each as sampling error or nonsampling error, and explain your
answers.
(a) The telephone directory is used as a sampling frame.
(b) The person cannot be contacted in five calls.
(c) Interviewers choose people walking by on the sidewalk to interview.
5.76 Survey time! Return to Activity 5.3 (page 236). Once your teacher has approved
your survey design plan, go ahead and carry it out. Write a brief report summarizing
the results. Follow the four-step statistical problem-solving process (page 19).

CHAPTER 5 REVIEW
We have discussed sampling, the art of choosing a part of a population to represent
the whole. We saw the potential for bias from voluntary response and convenience
samples. This led us to the big idea of a simple random sample, which is summa-
rized in the following figure.
 Chapter 5 Review Exercises 253

All samples of size n

Sample data
Population
are equally likely x 1, x 2, …, x n

Introducing randomness into the sampling scheme means that the things we
measure will vary from sample to sample. But this variation will have a predictable
pattern if we repeat the procedure over and over. Increasing the size of the sample
will decrease the variability of a sample statistic. But this won’t help with bias.
Two types of error can affect sample surveys: sampling errors and nonsam-
pling errors. When the sampling frame differs from the population of interest,
undercoverage results. The margin of error does not account for such sampling
errors. It also does not reflect nonresponse or question wording, two major
causes of nonsampling error.
In many settings, an SRS would be too difficult or too costly to obtain. Many
national surveys use some form of multistage sampling. If subgroups within a pop-
ulation differ in their opinions, a stratified random sample is appropriate.
Here is a review list of the most important skills you should have developed
from your study of this chapter.
1. Identify the population and parameter of interest.
2. Recognize bias due to voluntary response samples and other inferior sampling
methods.
3. Use Table B of random digits to select a simple random sample (SRS) from a
population.
4. Explain how sample surveys deal with bias and variability in their conclusions.
5. Explain in simple language what the margin of error for a sample survey result
tells us and what “95% confidence” means.
6. Use the quick method to get an approximate margin of error for 95% confidence.
Make an appropriate 95% confidence statement about a population parameter.
7. Understand the distinction between sampling errors and nonsampling errors.
Recognize the presence of undercoverage and nonresponse as sources of error in a
sample survey. Recognize the effect of the wording of questions on the responses.
8. Use random digits to select a stratified random sample from a population when
the strata are identified.

CHAPTER 5 REVIEW EXERCISES

5.77 Baseball tickets Suppose you want to know the average amount of money
spent by the fans attending opening day for the Cleveland Indians baseball season.
You get permission from the team’s management to conduct a survey at the stadium,
but they will not allow you to bother the fans in the club seating or box seat areas (the
most expensive seating). Using a computer, you randomly select 500 seats from the
rest of the stadium. During the game, you ask the fans in those seats how much they
spent that day.
(a) Provide a reason why this survey might yield a biased result.
(b) Explain whether the reason you provided in (a) is a sampling error or a nonsam-
pling error.
254 CHAPTER 5 Sampling and Surveys

5.78 Dress code? The principal has asked your statistics class to carry out a survey
of student opinion about a proposed dress code. The class decides to choose four of its
members at random to meet with the principal. The class list appears below. Choose
an SRS of 4 using Table B, beginning at line 145. Explain your method clearly enough
for a classmate to duplicate your results.

Anderson Fernandez Kempthorn Robertson

Aspin Gupta Liang Rodriguez
Benitez Gutierrez Montoya Siegel
Bock Gwynn Olds Tompkins
Breiman Harter Patnaik Vandegraff
Castillo Henderson Pirelli Wang
Dixon Hughes Rao
Edwards Johnson Rider

5.79 We don’t like one-way streets Highway planners decided to make a main street
in West Lafayette, Indiana, a one-way street. The Lafayette Journal and Courier took a
one-day poll by inviting readers to call a telephone number to record their comments.
The next day, the paper reported:

Journal and Courier readers overwhelmingly prefer two-way traffic flow in West
Lafayette’s Village area to one-way streets. By nearly a 7-1 margin, callers to the
newspaper’s Express Yourself opinion line on Wednesday complained about the one-
way streets that have been in place since May. Of the 98 comments received, all but
14 said no to one-way.

(a) What population do you think the newspaper wants information about?
(b) Is the proportion of this population who favor one-way streets almost certainly
larger or smaller than the proportion 14/98 in the sample? Why?
5.80 TV commercials The noted scientist Dr. Iconu wanted to investigate attitudes
toward television advertising among American college students. He decided to use a
sample of 100 students. Students in freshman psychology (PSY 001) are required to
serve as subjects for experimental work. Dr. Iconu obtained a class list for PSY 001
and chose a simple random sample of 100 of the 340 students on the list. He asked
each of the 100 students in the sample the following question:

Do you agree or disagree that having commercials on TV is a fair price to pay for
being able to watch it?

Of the 100 students in the sample, 82 marked “Agree.” Dr. Iconu announced the result
of his investigation by saying, “82% of American college students are in favor of TV
commercials.”
(a) What is the population in this example?
(b) What is the sampling frame in this example?
(c) Explain briefly why the sampling frame is or is not suitable for the question being
investigated.
(d) Discuss briefly the question Dr. Iconu asked. Is it a slanted question?
 Chapter 5 Review Exercises 255

(e) Discuss briefly why Dr. Iconu’s announced result is misleading.

(f) Dr. Iconu defended himself against criticism by pointing out that he had care-
fully selected a simple random sample from his sampling frame. Is this defense
relevant? Why?
5.81 Planning a survey of students The student government plans to ask a ran-
dom sample of students at a large high school about their priorities for improving
the school newspaper. A school counselor provides a list of the 3500 students at the
school to serve as a sampling frame.
(a) How would you choose an SRS of 250 students?
(b) How would you choose a systematic sample of 250 students? (See Exercise 5.67
page 247, to learn about systematic samples.)
(c) The list shows whether students are bussed to school (2400 students) or live nearby
(1100 students). How would you choose a stratified sample of 200 bussed students
and 50 local students?
(d) Which of the three sampling methods would you choose? Why?
5.82 Bigger samples, please Explain in your own words the advantages of bigger
random samples in a sample survey.
5.83 Should he go or stay? In December 1998, the House of Representatives im-
peached President Clinton, the first step in removing him from office. The Senate then
conducted a trial and found the president not guilty. Here are two opinion poll ques-
tions asked after the House had acted:

What do you think President Clinton should do: fight the charges in the Senate, or
resign from office?
What do you think President Clinton should do: continue to serve and stand trial in
the Senate, or resign from office?

In response to the first question, 58% thought the president should resign. But only
43% of those asked the second question thought he should resign.24 Why do you think
the first wording encouraged more people to favor resignation?
5.84 The Harris Poll Here is the language used by the Harris Poll to explain
the accuracy of its results: “In theory, with a sample of this size, one can say with
95 percent certainty that the results have a statistical precision of plus or minus
3 percentage points of what they would be if the entire adult population had
been polled with complete accuracy.” 25 What does Harris mean by “95 percent
certainty”?
5.85 Life and Internet A poll of 586 adults who used the Internet in the past week
were asked whether “the Internet has made your life much better, somewhat better,
somewhat worse, much worse, or has it not affected your life either way.” In all, 152 of
the 586 said “much better.”26
(a) What is the population for this sample survey?
(b) Use the quick method to find a margin of error. Then give a complete confidence
statement for a conclusion about the population.
5.86 Steroids in baseball A 2008 New York Times article on public opinion about
steroid use in baseball discussed the results of a sample survey. The survey found that
256 CHAPTER 5 Sampling and Surveys

34% of adults think that at least half of Major League Baseball (MLB) players “use
steroids to enhance their athletic performance.” Another 36% thought that about a
quarter of MLB players use steroids; 8% had no opinion. Here is part of the Times’s
statement on “How the Poll Was Conducted”:
The latest New York Times/CBS News Poll is based on telephone interviews conduct-
ed March 15 through March 18 with 1,067 adults throughout the United States. . . .
The sample of telephone numbers called was randomly selected by a computer
from a list of more than 42,000 active residential exchanges across the country. The
exchanges were chosen to ensure that each region of the country was represented in
proportion to its population.
In each exchange, random digits were added to form a complete telephone
number, thus permitting access to listed and unlisted numbers. In each house-
hold, one adult was designated by a random procedure to be the respondent for
the survey.27
Page 249 lists several “questions to ask” about an opinion poll. What answers does the
Times give to each of these questions?
 CHAPTER 6

Designing
Experiments
6.1 Experiments, Good and Bad

©BananaStock/Imagestate
6.2 Experiments in the Real World
6.3 Data Ethics

A tale of three studies

An optimistic account of online learning describes a study at Nova Southeastern
University in Fort Lauderdale, Florida. The authors of the study claim that
students taking courses online were “equal in learning” to students taking the
same courses in class. Replacing college classes with Web sites saves colleges
money, so this study suggests we should all move online.1
Ulcers seem to accompany the stress of modern life. “Gastric freezing” is
a clever treatment for stomach ulcers. The patient swallows a deflated balloon
with tubes attached; then a refrigerated solution is pumped through the balloon
for an hour. The idea is that cooling the stomach will reduce its production of
acid and so relieve ulcers. An experiment reported in the Journal of the American
Medical Association claimed that gastric freezing did relieve ulcer pain.2
Should the government provide day care for children in low-income families?
If day care helps these children stay in school and hold good jobs later in life, the
government would save money by paying less welfare and collecting more taxes.
The Carolina Abecedarian Project (the name suggests learning the ABCs) has
followed a group of children since 1972. The results show that good day care
makes a big difference in later school and work.3
Three big issues, three studies that claim to shed light on the issues. In fact, the
first of these is an observational study that can’t be trusted to give good evidence
about learning in online college courses. The gastric-freezing experiment now
appears to be plainly misleading. Most ulcers are caused by bacteria, and cooling
the stomach is not an effective treatment. The results of the Abecedarian Project,
on the other hand, are about as convincing as evidence about the long-term effects
of day care can be. What makes some studies—especially some experiments—
convincing? Why should we ignore others? This chapter shows you what to look for.

257
258 CHAPTER 6 Designing Experiments

6.1 Experiments, Good and Bad

ACTIVITY 6.1
Testing therapeutic touch
Materials: Blindfold and coin for each pair 3. The experimenter should flip a coin. If it
of students lands “heads,” place one hand about six inches
above the subject’s left hand. If it lands “tails,”
For her fourth-grade science fair project, Emily
place one hand about six inches above the
Rosa designed an experiment to test whether an
subject’s right hand. Ask the subject to identify
alternative healing method known as therapeutic
the presence of the energy field by saying either
touch actually works. Therapeutic touch requires
“left” or “right.”
the “healer” to detect and manipulate the patient’s
“human energy field.” Emily doubted that this 4. Record the outcome of the coin toss (H or T)
was possible. So she invited 21 people who had and whether the subject picked correctly (C) or
practiced therapeutic touch to participate in her incorrectly (I). Don’t tell the subject whether the
experiment. identification is correct!
Emily’s experimental design was fairly simple.
5. Repeat the previous two steps 19 more times, for
She sat at one side of a table and the subject sat
a total of 20 trials. Organize your results in a table
at the other side. A tall screen with two holes was
like this one.
placed in the middle of the table, so that the subject
could not see Emily. The subject was asked to Coin outcome Subject’s
extend one arm, palm side up, through each of the Trial no. (H or T) pick (C or I)
holes. Unknown to the subject, Emily then flipped 1 T C
a coin to decide which of the subject’s two hands
she would place her hand above. She then asked the 6. Have the subject remove the blindfold. Share the
subject to identify the location of the “energy field.” results of the experiment.
This process was repeated either 10 or 20 times with
7. Switch roles, and repeat Steps 2 through 6.
each of the 21 subjects.
Want to know what happened in Emily’s experi- 8. Pool results with your classmates. Make a dotplot
ment? We’ll tell you after you and your classmates that shows the percent of correct (C) identifications
try a version of her experiment for yourself. Work made by each subject.
with a partner as instructed by your teacher.
9. Discuss what you see in the dotplot. Does any
1. With your partner, decide which of you will be member of the class appear to be able to detect a
the “experimenter” and which will be the “subject.” human energy field? Why or why not?
You will switch roles later, so don’t spend too much
10. Calculate the average percent of correct
time deciding!
identifications made by all of the subjects. How does
2. Blindfold the subject. Have the subject sit with this compare with what you would expect if the
both arms extended, palms up. subjects were just guessing?

The results of Emily Rosa’s experiment in Activity 6.1 were striking: the subjects cor-
rectly identified the location of Emily’s hand in only 123 of 280 trials (44%). This is
pretty much what we would expect if the subjects were merely guessing. With a little
help from her mother (a nurse) and a statistician, Emily published an article summariz-
ing her research in the Journal of the American Medical Association on April 1, 1998.4
 6. 1 Experiments, Good and Bad 259

Talking about experiments

Observational studies involve passive data collection. We observe, record, or measure,
but we don’t interfere. Experiments involve active data production. Experimenters
actively intervene by imposing some treatment in order to see what happens. All
experiments and many observational studies are interested in the effect one vari-
able has on another variable. In Chapter 4, we called these the explanatory variable
and the response variable, respectively. Now we introduce two more terms that
have common use when describing experiments—subjects and treatments.

Subjects and treatments

The individuals studied in an experiment are often called subjects.

A treatment is any specific experimental condition applied to the sub-

jects. If an experiment has several explanatory variables, a treatment is
a combination of specific values of these variables.

The following example illustrates how we use the vocabulary of experiments.

Example 6.1 The effects of day care

The Carolina Abecedarian Project described in the chapter-opening story (page 257)
is an experiment. The subjects were 111 healthy, low-income black infants in Chapel
Hill, North Carolina. All the infants received nutritional supplements and help from
social workers. Half, chosen at random, were also placed in an intensive preschool
program. The experiment compares these two treatments. The explanatory variable
is just “preschool, yes or no.” There are many response variables, recorded over more
than 20 years, including academic test scores, college attendance, and employment.

You will often see explanatory variables called independent variables and re-
sponse variables called dependent variables. The idea is that the response variables
depend on the explanatory variables. We won’t use these terms, partly because
“independent” has other and very different meanings in statistics.

Designing studies: what can go wrong

Do students who take a course on the Web learn as well as those who take the
same course in a traditional classroom? The best way to find out is to assign some
students to the classroom and others to the Web. That’s an experiment. The Nova
Southeastern University study described in the chapter-opening story (page 257)
was not an experiment, because it imposed no treatment on the student subjects.
Students chose for themselves whether to enroll in a classroom or in an online ver-
sion of a course. The study simply measured their learning.
260 CHAPTER 6 Designing Experiments

The students who chose the online course were very different from the class-
room students. For example, their average score on tests on the course material
given before the courses started was 40.70, against only 27.64 for the classroom stu-
dents. It’s hard to compare in-class versus online learning when the online students
have a big head start. The effect of online versus in-class instruction is hopelessly
mixed up with influences lurking in the background. In Chapter 4, we called this
situation confounding. Figure 6.1 shows the mixed-up influences in picture form.

Online Test score

vs. after course
classroom Causes ?
(explanatory (response
variable) variable)

Student
preparation
(lurking
variable)

Figure 6.1 Confounding in the Nova Southeastern University study. The influence
of course setting (the explanatory variable) can’t be distinguished from the influence of
student preparation (a lurking variable).

In the Nova Southeastern University study, student preparation (a lurking

variable) is confounded with the explanatory variable. The study report claims that
the two groups did equally well on the final test. We can’t say how much of the on-
line group’s performance is due to their head start. Since the online group started
with a big advantage but did no better than the more poorly prepared classroom
students on the final, the study gives little evidence to support the effectiveness of
Web-based instruction.
Here is another example, one in which a second experiment untangled the
confounding.

Example 6.2 Gastric freezing flunks

Experiments that study the effectiveness of medical treatments on actual patients
are called clinical trials. The clinical trial that made gastric freezing a popular
treatment for stomach ulcers had this “one-track” design:

Impose treatment S Measure response

Gastric freezing S Reduced pain?

The patients did report reduced pain, but we can’t say that gastric freezing caused
the reduced pain. It might be just the placebo effect. A placebo is a dummy treatment
with no active ingredients. Many patients respond favorably to any treatment,
even a placebo. This response to a dummy treatment is the placebo effect. Perhaps
the placebo effect is in our minds, based on trust in the doctor and expectations
 6. 1 Experiments, Good and Bad 261

of a cure. Maybe it is just a name for the fact that many patients improve for no
visible reason. The one-track design of the experiment meant that the placebo
effect was confounded with any effect gastric freezing might have.
A second clinical trial, done several years later, divided ulcer patients into
two groups. One group was treated by gastric freezing as before. The other group
received a placebo treatment in which the solution in the balloon was at body tem-
perature rather than freezing. The results: 34% of the 82 patients in the treatment
group improved, but so did 38% of the 78 patients in the placebo group. This and
other properly designed experiments showed that gastric freezing was no better
than a placebo, and doctors stopped using it.5

Both observation and one-track experiments often yield useless data because
of confounding with lurking variables. It is hard to avoid confounding when only
observation is possible. Experiments offer better possibilities, as the second gastric-
freezing experiment shows. This experiment included a group of subjects who re-
ceived only a placebo. This allows us to see whether the treatment being tested does
better than the placebo and so has more than the placebo effect going for it. Effec-
tive medical treatments pass the placebo test. Gastric freezing flunks the test.

Exercises 6.1 On sale A researcher wants to study the effect of price promotions on
consumers’ expectations. The researcher makes up two different histories of the
store price of a video game for the past year. Students in an economics course
view one or the other price history on a computer. Some students see a steady
price, while others see regular promotions that temporarily cut the price. Then the
students are asked what price they would expect to pay for the video game.
(a) Is this study an experiment? Why?
(b) Identify the explanatory and response variables, the subjects, and the treatments.
6.2 Public housing A study of the effect of living in public housing on family
stability in low-income households was carried out as follows: A list of applicants
accepted for public housing was obtained, together with a list of families who
applied but were rejected by the housing authorities. A random sample was drawn
from each list, and the two groups were observed for several years.
(a) Is this an experiment? Why?
(b) What are the explanatory and response variables?
(c) Does this study contain confounding that may prevent valid conclusions on
the effects of living in public housing? Explain.
6.3 Nursing your baby An article in a women’s magazine says that women
who nurse their babies feel warmer and more receptive toward the infants than
mothers who bottle-feed. The author concludes that nursing has desirable effects
on the mother’s attitude toward the child. But women choose whether to nurse
or bottle-feed. Explain why this fact makes any conclusion about cause and effect
untrustworthy. Use the language of lurking variables and confounding in your
explanation, and draw a picture like Figure 6.1 to illustrate it.
262 CHAPTER 6 Designing Experiments

6.4 Vitamin C and colds Last year only 10% of a group of adult men did not
have a cold at some time during the winter. This year all the men in the group
took 1 gram of vitamin C each day, and 20% had no colds. A writer claims that
this shows that vitamin C helps prevent colds. Explain why this conclusion is
shaky at best. Use the language of lurking variables and confounding in your
explanation, and draw a picture like Figure 6.1 to illustrate it.
6.5 Activity 6.1 follow-up
(a) Identify the subjects, treatments, and explanatory and response variables in
the therapeutic touch experiment.
(b) After Emily’s results were published, therapeutic touch practitioners argued that
her results were not valid. Discuss one or two arguments they might have made.
6.6 Do placebos really work? Researchers in Japan conducted an experiment on
13 individuals who were extremely allergic to poison ivy. On one arm, each subject
was rubbed with a poison ivy leaf and told the leaf was harmless. On the other arm,
each subject was rubbed with a harmless leaf and told it was poison ivy. All of the
subjects developed a rash on the arm where the harmless leaf was rubbed. Of the
13 subjects, 11 did not have any reaction to the real poison ivy leaf.6
(a) What was the placebo in this experiment?
(b) Explain how the results of this study support the idea of a placebo effect.

By permission of John L. Hart FLP and Creators Syndicate, Inc.

Randomized comparative experiments

The first goal in designing an experiment is to ensure that it will show us the effect
of the explanatory variables on the response variables. Confounding often prevents
one-track experiments from doing this. The remedy is to compare two or more
treatments. Here is an example of a new medical treatment that passes the placebo
test in a direct comparison.

Example 6.3 Sickle-cell anemia

Sickle-cell anemia is an inherited disorder of the red blood cells that in the United
States affects mostly blacks. It can cause severe pain and many complications. The
National Institutes of Health carried out a clinical trial of the drug hydroxyurea for
treatment of sickle-cell anemia. The subjects were 299 adult patients who had had
at least three episodes of pain from sickle-cell anemia in the previous year.
 6. 1 Experiments, Good and Bad 263

Simply giving hydroxyurea to all 299 subjects would confound the effect of the
medication with the placebo effect and other lurking variables such as the effect
of knowing that you are a subject in an experiment. Instead, half of the subjects
received hydroxyurea, and the other half received a placebo that looked and tasted
the same. All subjects were treated exactly the same (same schedule of medical
checkups, for example) except for the content of the medicine they took. Lurking
variables therefore affected both groups equally and should not have caused any
differences between their average responses.
The two groups of subjects must be similar in all respects before they start tak-
ing the medication. Just as in sampling, the best way to avoid bias in determining
which subjects get hydroxyurea is to allow impersonal chance to make the choice.
An SRS of 152 of the subjects formed the hydroxyurea group; the remaining 147
subjects made up the placebo group. Figure 6.2 outlines the experimental design.

Group 1 Treatment 1
152 patients Hydroxyurea Compare
Random pain
assignment episodes
Group 2 Treatment 2
147 patients Placebo

Figure 6.2 The design of a randomized comparative experiment to compare hydroxyurea

with a placebo for treating sickle-cell anemia.

The experiment was stopped ahead of schedule because the hydroxyurea group
had many fewer pain episodes than the placebo group. This was compelling evi-
dence that hydroxyurea is an effective treatment for sickle-cell anemia, good news
for those who suffer from this serious illness.7

Figure 6.2 illustrates the simplest randomized comparative experiment, one

that compares just two treatments. The diagram outlines the essential information
about the design: random assignment; one group for each treatment; the number of
subjects in each group (it is generally best to keep the groups similar in size); what
treatment each group gets; and the response variable we compare. You know how
to carry out the random assignment of subjects to groups. Label the 299 subjects
001 to 299, then read three-digit groups from the table of random digits (Table B)
until you have chosen the 152 subjects for Group 1. The remaining 147 subjects
form Group 2.
The placebo group in Example 6.3 is called a control group because com-
paring the treatment and control groups allows us to control the effects of lurking
variables. A control group need not receive a dummy treatment such as a placebo.
Clinical trials often compare a new treatment for a medical condition, not with
a placebo, but with a treatment that is already on the market. Patients who are
randomly assigned to the existing treatment form the control group. To compare
more than two treatments, we can randomly assign the available experimental
subjects to as many groups as there are treatments. Here is an example with three
groups.
264 CHAPTER 6 Designing Experiments

Example 6.4 Stopping drunk drivers

Most people know that drunk driving is dangerous. Even so, some drivers are
cited multiple times for driving while intoxicated. Is there anything that can be
done to change their behavior? An experiment was designed to help answer this
question.
The subjects were 300 people convicted of drunk driving three times in one
year. The treatments, imposed after the third conviction, were a fine plus a sus-
pended jail sentence plus one of (1) no treatment, (2) attend an alcoholism clinic,
or (3) participate in Alcoholics Anonymous. The subjects assigned to Treatment 1
served as a control group. One response variable was whether or not the subject
was arrested again in the following year. Figure 6.3 outlines the design.

Group 1 Treatment 1
80 drunk Control
drivers
Compare
Random Group 2 Treatment 2 arrests the
assignment 98 drunk Alcoholism following
drivers clinic year
Group 3 Treatment 3
122 drunk Alcoholics
drivers Anonymous

Figure 6.3 The design of a randomized comparative experiment to compare three

programs to deter drunk drivers.

Ideally, 100 subjects would be assigned at random to each of the three treat-
ments. This is probably not practical, since a treatment must be assigned immedi-
ately after the subject’s third drunk-driving conviction. Instead, enter Table B at any
row and begin reading off single digits. The digits 1, 2, and 3 represent treatment
option 1; the digits 4, 5, and 6 indicate treatment option 2; and the digits 7, 8, and
9 signal treatment option 3. Ignore the digit 0. The first subject selected for the ex-
periment is assigned the treatment corresponding to the first nonzero digit in the
row. The second subject’s treatment is determined by the second nonzero digit in
the row, and so on. For example, if we use row 105 in Table B, our assignment of
subjects to treatments proceeds as follows:

Subject: 1 2 3 4 5 6 7 8 9 10
Nonzero digit: 9 5 5 9 2 9 4 7 6 9
Treatment: 3 2 2 3 1 3 2 3 2 3

By the end of row 105, we have assigned 10 subjects to Treatment 1, 9 to Treatment

2, and 18 to Treatment 3. Continuing in this fashion through rows 106, 107, . . . ,
113, we end up with 80 subjects receiving Treatment 1, 98 receiving Treatment 2,
and 122 receiving Treatment 3.
 6. 1 Experiments, Good and Bad 265

The logic of experimental design

The randomized comparative experiment is one of the
most important ideas in statistics. It is designed to al-
low us to draw cause-and-effect conclusions. Be sure you
understand the logic:
• Random assignment produces groups of
subjects that should be similar in all respects
before we apply the treatments.
• A proper comparative design ensures that
influences other than the experimental
treatments operate equally on all groups.
• Therefore, differences in the response variable
must be due to the effects of the treatments.
We use chance to choose the groups in order to elimi-
nate any systematic bias in assigning the subjects to groups.
In the sickle-cell study, for example, a doctor might sub-
consciously assign the most seriously ill patients to the hy-
droxyurea group, hoping that the untested drug will help
them. That would bias the experiment against hydroxyurea.
"OK, Mr. Simms. Now it's time to test your pacemaker
Choosing an SRS of the subjects to be Group 1 gives every-
against a control subject." one the same chance to be in either group. We expect the
two groups to be similar in all respects—age, seriousness of
illness, smoker or not, and so on. Chance tends to assign equal numbers of smokers to
both groups, for example, even if we don’t know which subjects are smokers.
There is one important caution about randomized experiments. Like random
samples, they are subject to the laws of chance. Just as an SRS of voters might by bad
luck choose nearly all Republicans, a random assignment of subjects might by bad
luck put nearly all the smokers in one group. We know that, if we choose large ran-
dom samples, it is very likely that the sample will match the population well. In the
same way, if we use many experimental subjects, it is very likely that random assign-
ment will produce groups that match closely. More subjects means that there is less
chance variation among the treatment groups and less chance variation in the out-
comes of the experiment. “Use enough subjects” joins “control for lurking variables”
and “randomize” as a basic principle of statistical design of experiments.

Principles of experimental design

The basic principles of statistical design of experiments are

1. Control the effects of lurking variables on the response. Use a

comparative design and ensure that the only systematic difference
between the groups is the treatment administered.
2. Randomize—use impersonal chance to assign subjects to treatments.
3. Use enough subjects in each group to reduce chance variation in
the results.
266 CHAPTER 6 Designing Experiments

Exercises 6.7 Aspirin and heart attacks, I Can aspirin help prevent heart attacks? The
Physicians’ Health Study, a large medical experiment involving 22,071 male
physicians, attempted to answer this question. One randomly selected group
of 11,037 physicians took an aspirin every second day, while the rest took a
placebo. After several years the study found that subjects in the aspirin group had
significantly fewer heart attacks than subjects in the placebo group.
(a) Identify the subjects, the explanatory variable and the values it can take, and
the response variable.
(b) Use a diagram like the one in Figure 6.2 or Figure 6.3 to outline the design of
the Physicians’ Health Study. (When you outline the design of an experiment, be
sure to indicate the size of the treatment groups and the response variable.)
6.8 Aspirin and heart attacks, II Refer to the previous exercise. Discuss
how each of the three principles of experimental design was addressed in the
Physicians’ Health Study.
Note: You can get all the details of the Physicians’ Health Study described in
www
Exercises 6.7 and 6.8 at the Web site http://phs.bwh.harvard.edu/.
6.9 Internet phone calls It is possible to use a computer to make telephone calls
over the Internet. How will the cost affect the behavior of users of this service?
You will offer the service to all 200 rooms in a college dormitory. Some rooms will
pay a flat rate. Others will pay higher rates at peak periods and very low rates off-
peak. You are interested in the amount and time of use.
(a) Outline the design of an experiment to study the effect of rate structure. Use
the diagrams in Figures 6.2 and 6.3 as models.
(b) Use Table B, starting at line 125, to assign the first 5 rooms to the flat-rate group.
6.10 Healthy turkeys Turkeys raised commercially for food are often fed the
antibiotic salinomycin to prevent infections from spreading among the birds.
Salinomycin can damage the birds’ internal organs, especially the pancreas. A
researcher believes that adding vitamin E to the diet may prevent injury. He wants to
explore the effects of three levels of vitamin E added to the diet of turkeys along with
the usual dose of salinomycin. There are 30 turkeys available for the study. At the end
of the study, the birds will be killed and each pancreas examined under a microscope.
(a) Give a careful outline of the design of a randomized comparative experiment
in this setting. Use the diagrams in Figures 6.2 and 6.3 as models.
(b) Use Table B, beginning at line 118, to carry out the random assignment
required by your design.
6.11 Learning on the Web The discussion following Example 6.1 (page 259)
notes that the Nova Southeastern University
study doesn’t tell us much about Web versus
classroom learning because the students
ellner/iStockphoto

who chose the Web version were much better

prepared. Describe the design of an experiment to
get better information.
Z
Joe

6.12 Randomization at work To demonstrate how

randomization reduces confounding,
consider the following situation. A
 6. 1 Experiments, Good and Bad 267

nutrition experimenter intends to compare the weight gain of newly weaned male
rats fed Diet A with that of rats fed Diet B. To do this, she will feed each diet to 10
rats. She has available 10 rats of genetic Strain 1 and 10 of Strain 2. Strain 1 is more
vigorous, so if the 10 rats of Strain 1 were fed Diet A, the effects of strain and diet
would be confounded, and the experiment would be biased in favor of Diet A.
(a) Label the rats 00, 01, . . . , 19. Use Table B to assign 10 rats to Diet A. Do this
four times, using different parts of the table, and write down the four groups
assigned to Diet A.
(b) Unknown to the experimenter, the rats labeled 00, 02, 04, 06, 08, 10, 12, 14, 16, and
18 are the 10 Strain 1 rats. How many of these rats were in each of the four Diet A groups
that you generated? What was the average number of Strain 1 rats assigned to Diet A?

Statistical significance
The presence of chance variation requires us to look more closely at the logic of
randomized comparative experiments. We cannot say that any difference in the
average number of pain episodes between the hydroxyurea and control groups in
the sickle-cell study must be due to the effect of the drug. There will be some differ-
ences even if both treatments are the same because there will always be some dif-
ferences among the individuals who are our subjects. Even though randomization
eliminates systematic differences between the groups, there will still be chance dif-
ferences. We should insist on a difference in the responses so large that it is unlikely
to happen just because of chance variation.

Statistical significance

An observed effect so large that it would rarely occur by chance is

called statistically significant.

The difference between the average number of pain episodes for subjects in the
hydroxyurea group and the average for the control group was “highly statistically
significant.” That means that a difference this large would almost never happen just
by chance. We do indeed have strong evidence that hydroxyurea beats a placebo
in helping sickle-cell disease sufferers. You will often see the phrase “statistically
significant” in published reports. It tells you that the investigators found good evi-
dence for the effect they were seeking.
Of course, the actual results of an experiment are more important than the seal of ap-
proval given by statistical significance. The treatment group in the sickle-cell experiment
had an average of 2.5 pain episodes per year, against 4.5 per year in the control group.
That’s a big enough difference to be important to people with the disease. A difference of
2.5 versus 2.8 would be much less interesting even if it were statistically significant.

How to live with observational studies

Does regular church attendance lengthen people’s lives? Do doctors discriminate
against women in treating heart disease? Does talking on a cell phone while driving
increase the risk of having an accident? These are cause-and-effect questions, so
we reach for our favorite tool, the randomized comparative experiment. Sorry. We
268 CHAPTER 6 Designing Experiments

can’t randomly assign people to attend church or not, because going to religious
services is an expression of beliefs or their absence. We can’t use random digits to
assign heart disease patients to be men or women. We are reluctant to require driv-
ers to use cell phones in traffic, because talking while driving may be risky.
The best data we have about these and many other cause-and-effect questions
come from observational studies. We know that observation is a weak second best
to experiments, but good observational studies are far from worthless. What makes
a good observational study?
First, good studies are comparative even when they are not experiments. We
compare random samples of people who do and who don’t attend religious services
regularly. We compare how doctors treat men and women patients. We might compare
drivers talking on cell phones with the same drivers when they are not on the phone.
We can often combine comparison with matching in creating a control group.

Example 6.5 Painkillers and pregnancy

To see the effects of taking a painkiller during pregnancy, we compare women who did
so with women who did not. From a large pool of women who did not take the drug,
we select individuals who match the drug group in age, education, number of chil-
dren, and other lurking variables. We now have two groups that are similar in all these
ways, so that these lurking variables should not affect our comparison of the groups.

Comparison does not eliminate confounding. People who attend church or

synagogue or mosque take better care of themselves than nonattenders. They are
less likely to smoke, more likely to exercise, and less likely to be overweight. Match-
ing can reduce some but not all of these differences. A direct comparison of the ages
at death of attenders and nonattenders would confound any effect of religion with
the effects of healthy living. So a good comparative study measures and adjusts
for confounding variables. If we measure weight, smoking, and exercise, there are
statistical techniques that reduce the effects of these variables on length of life so
that (we hope) only the effect of religion itself remains.

Example 6.6 Living longer through religion

One of the better studies of the effect of regular attendance at religious services gath-
ered data from a random sample of 3617 adults. Random sampling is a good start.
The researchers then measured lots of variables, not just the explanatory variable
(religious activities) and the response variable (length of life). A news article said:
Churchgoers were more likely to be nonsmokers, physically active, and at their right
weight. But even after health behaviors were taken into account, those not attend-
ing religious services regularly still were about 25% more likely to have died.8
That “taken into account” means that the final results were adjusted for differences
between the two groups. Adjustment reduced the advantage of religion but still left
a large benefit.
 6. 1 Experiments, Good and Bad 269

Example 6.7 Sex bias in treating heart disease?

Doctors are less likely to give aggressive treatment to women with symptoms of
heart disease than to men with similar symptoms. Is this because doctors are sex-
ist? Not necessarily. Women tend to develop heart problems much later than men,
so that female heart patients are older and often have other health problems. That
might explain why doctors proceed more cautiously in treating them.
This is a case for a comparative study with statistical adjustments for the effects
of confounding variables. There have been several such studies, and they produce
conflicting results. Some show, in the words of one doctor, “When men and women
are otherwise the same and the only difference is gender, you find that treatments are
very similar.”9 Other studies find that women are undertreated even after adjusting
for differences between the female and male subjects.

As Example 6.7 suggests, statistical adjustment is tricky.

Randomization creates groups that are similar in all variables
known and unknown. Matching and adjustment, on the other
hand, can’t work with variables the study didn’t think to mea-
sure. Even if you believe that the researchers thought of every-
thing, you should be a bit skeptical about statistical adjustment.
There’s lots of room for cheating in deciding which variables
to adjust for. And the “adjusted” conclusion is really something
like this:

If female heart disease patients were younger and healthier

than they really are, and if male patients were older and less
healthy than they really are, then the two groups would get
the same medical care.

“Statistics say that religious people live longer, This may be the best we can get, and we should thank statistics
so I practice a different religion every day of for making such wisdom possible. But we end up longing for the
the week to be sure I'm covered.” clarity of a good experiment.

Exercises 6.13 Statistical significance, I A randomized comparative experiment examines

whether a calcium supplement in the diet reduces the blood pressure of healthy men.
The subjects receive either a calcium supplement or a placebo for 12 weeks. The
researchers conclude that “the blood pressure of the calcium group was significantly
lower than that of the placebo group.” “Significant” in this conclusion means
statistically significant. Explain what statistically significant means in the context of
this experiment, as if you were speaking to someone who knows no statistics.
6.14 Statistical significance, II The financial aid office of a university asks a
sample of students about their employment and earnings. The report says that “for
academic year earnings, a significant difference was found between the sexes, with
men earning more on the average. No significant difference was found between
270 CHAPTER 6 Designing Experiments

the earnings of black and white students.” Explain the meaning of “a significant
difference” and “no significant difference” in plain language.
6.15 Prayer and meditation You read in a magazine that “nonphysical
treatments such as meditation and prayer have been shown to be effective in
controlled scientific studies for such ailments as high blood pressure, insomnia,
ulcers, and asthma.” Explain in simple language what the article means by
“controlled scientific studies” and why such studies might show that meditation
and prayer are effective treatments for some medical problems.
6.16 TV harms children Observational studies suggest that children who watch
many hours of television get lower grades in school and are more likely to commit
crimes than those who watch less TV. Explain clearly why these studies do not
show that watching TV causes these harmful effects. In particular, suggest some
lurking variables that may be confounded with heavy TV viewing.
6.17 Family dinners and better grades? Does eating dinner with their families
improve students’ academic performance? According to an ABC News article,
“Teenagers who eat with their families at least five times a week are more likely
to get better grades in school.”10 This finding was based on an observational study
conducted by researchers at Columbia University. Explain clearly why such a
study cannot establish a cause-and-effect relationship. Suggest a lurking variable
that may be confounded with whether families eat dinner together.
6.18 Significant doesn’t mean important Suppose we’re testing a new antibacterial
cream, “Formulation NS.” We know from previous research that, with no medication,
the mean healing time (defined as the time for the scab to fall off ) is 7.6 days.
We make a small cut on the inner forearm of 25 volunteer college students and
apply Formulation NS to the wounds. The mean healing time for these subjects is
x 5 7.1 days. A statistical test shows that this decrease in healing time is statistically
significant. Explain why this result has little practical importance. Moral: Statistical
significance is not the same as practical importance.

ws
Ne
ws
Ne

APPLICATION 6.1 B]
RO
g
<
Se
a

Good for your heart

Heart Treatment’s
Value Doubted
By Michael Specter
Washington Post Staff Writer

In a study with far-reaching implications for the routine treatment of heart attack
patients, researchers have found that the immediate use of special balloons to force
open clogged arteries is unnecessary in the vast majority of cases if the patient is
treated with a clot-dissolving drug.
 6. 1 Experiments, Good and Bad 271

Most heart specialists have assumed that the balloon treatment, an expensive and
increasingly popular procedure called balloon angioplasty, should routinely follow
the use of drugs, such as TPA, that dissolve the blood clots that cause most heart
attacks.
But in a study of 3,262 heart attack patients that is expected to transform the
standards for treatment of the nation’s leading killer, researchers at medical centers
across the country found the extra measures were rarely needed. Half of the ran-
domly selected patients were treated solely with the clot-dissolving drug TPA, while
the other half were given TPA followed by angioplasty.
The results, reported in today’s issue of The New England Journal of Medicine,
showed that for most people adding angioplasty was no better than relying on the
less complicated and less costly drug treatment.
Only 10 percent of the group assigned solely to drug treatment died or had
another heart attack in the six weeks following treatment. By comparison, 11 per-
cent of the group assigned to receive the combination drug and angioplasty treat-
ment suffered the same fate. The difference could have occurred by chance. But the
fact that angioplasty did not prove to be the better option was a shock even to some
of the nation’s most renowned heart experts.
“There is no question that it bucks the trend,” said Eugene Braunwald, profes-
sor of medicine at Harvard Medical School and the study chairman for the research
project. “It will spare many patients unnecessary surgery and reduce the cost of
medical care greatly.”11

QUESTIONS
1. Identify the subjects, treatments, and 4. What other information would you like to have
explanatory and response variables in this study. to evaluate the design of the experiment?
2. Describe the design of the experiment in as Comment: As Application 6.1 shows, the objective of some
much detail as you can. clinical trials is to show that there is no significant differ-
ence in the effects of the treatments, especially when one of
3. Explain how the principles of control, the treatments is less risky or less expensive.
randomization, and using enough subjects were
applied in this experiment.

Section 6.1 Summary

Statistical studies often try to show that changing one variable (the explanatory
variable) causes changes in another variable (the response variable). In an experi-
ment, we actually set the explanatory variables ourselves rather than just observing
them. Observational studies and one-track experiments that simply apply a single
treatment often fail to produce useful data because confounding with lurking vari-
ables makes it impossible to say what the effect of the treatment was. The remedy is
to use a randomized comparative experiment that compares two or more treat-
ments. Use chance to decide which subjects get each treatment, control for lurking
variables, and use enough subjects so that the effects of chance variations between
the groups are small. Comparing two or more treatments helps avoid the placebo
effect because it will act on all the treatment groups.
Differences among the effects of the treatments so large that they would rarely
happen just by chance are called statistically significant. Statistically significant
272 CHAPTER 6 Designing Experiments

results from randomized comparative experiments are the best available evidence
that changing the explanatory variable really causes changes in the response.
Observational studies of cause-and-effect questions are more impressive if they
compare matched groups and measure as many lurking variables as possible to
allow statistical adjustment. For answering questions about causation, observa-
tional studies remain a weak second best to experiments.

Section 6.1 6.19 Does job training work? A state institutes a job-training program for
Exercises manufacturing workers who lose their jobs. After five years, the state reviews how
well the program works. Critics claim that, because the state’s unemployment rate
for manufacturing workers was 6% when the program began and 10% five years
later, the program is ineffective. Explain why higher unemployment does not nec-
essarily mean that the training program failed. In particular, identify some lurking
variables whose effect on unemployment may be confounded with the effect of
the training program. Draw a picture like Figure 6.1 (page 260) to illustrate your
explanation.
6.20 What a headache! Doctors identify “chronic tension–type headaches”
as headaches that occur almost daily for at least six months. Can antidepres-
sant medications or stress management training reduce the number and
severity of these headaches? Are both together more effective than either
alone? Investigators compared four treatments: antidepressant alone, placebo
alone, antidepressant plus stress management, and placebo plus stress
management.
(a) Outline the design of the experiment. Use the diagrams in Figures 6.2 (page 263)
and 6.3 (page 264) as models.
(b) The headache sufferers named below have agreed to participate in the study.
Use Table B at line 130 to randomly assign the subjects to the treatments.

Acosta Duncan Han Liang Padilla Valasco

Asihiro Durr Howard Maldonado Plochman Vaughn
Bennett Edwards Hruska Marsden Rosen Wei
Bikalis Farouk Imrani Montoya Solomon Wilder
Chem Fleming James O’Brien Trujillo Willis
Clement George Kaplan Ogle Tullock Zhang

6.21 Learning about markets, I An economics teacher wonders if playing market

games online will help students understand how markets set prices. You suggest an
experiment: have some students use the online games, while others talk about mar-
kets in small group discussions. The course has two sections of 30 students each,
at 8:30 a.m. and 2:30 p.m. The teacher says, “Let’s just have the 8:30 class do online
work and the 2:30 class do discussion.” Why is this a bad idea?
6.22 Learning about markets, II Refer to the previous exercise.
(a) Outline a better design for an experiment to compare the two methods of
learning about economic markets. What do you suggest as a response variable?
Use the diagrams in Figures 6.2 (page 263) and 6.3 (page 264) as models.
(b) Use Table B, starting at line 116, to do the randomization your design requires.
 6. 2 Experiments in the Real World 273

6.23 Healthy nurses? The Nurses’ Health Study has surveyed a sample of over
100,000 female registered nurses every 2 years since 1976. Beginning in 1980,
the study asked questions about diet, including alcohol consumption. After looking at
all deaths among these nurses, the researchers concluded: “As compared with non-
drinkers and heavy drinkers, light-to-moderate drinkers had a significantly lower risk
of death.”12 The word “significantly” in this conclusion has a technical meaning. Explain
to someone who knows no statistics what “significant” means in a statistical study.
6.24 Clumsy men? A doctor at a veterans hospital examines all of the patient
records over a 9-year period and finds that twice as many men as women fell out
of their hospital beds during their stay. This is put forward as evidence that men
are clumsier than women.
(a) Is this an experiment or an observational study? Why?
(b) Give an example of a possible confounding variable and a clear explanation of
why it is a possible confounding variable.
6.25 More statistics majors A university’s statistics department wants to attract
more majors. It prepares two advertising brochures. Brochure A stresses the intellec-
tual excitement of statistics. Brochure B stresses how much more money statisticians
make. Which will be more attractive to first-year students? You have a questionnaire
to measure interest in majoring in statistics, and you have 50 first-year students to
work with. Outline the design of an experiment to decide which brochure works
better. Use the diagrams in Figures 6.2 (page 263) and 6.3 (page 264) as models.
6.26 Acupuncture and pregnancy A study sought to determine if the ancient
Chinese art of acupuncture could help infertile women become pregnant.13 One
hundred sixty healthy women undergoing treatment with artificial insemina-
tion were recruited for the study. Half of the subjects were randomly assigned to
receive acupuncture treatment 25 minutes before embryo transfer and again
25 minutes after the transfer. The remaining 80 subjects were instructed to lie
still for 25 minutes after the embryo transfer. Results: In the acupuncture group,
34 women became pregnant. In the control group, 21 women became pregnant.
(a) Explain how the three principles of experimental design were addressed in this study.
(b) The difference in the percent of women who became pregnant in the two
groups is statistically significant. Explain to someone who knows little statistics
what this means.
(c) Why can’t we conclude that acupuncture caused the difference in pregnancy rates?

6.2 Experiments in the Real World

ACTIVITY 6.2A
Mozart and mazes
Materials: Two mazes per student (provided by In this Activity, you will take part in an
your teacher), classical music CD or cassette, experiment to determine whether listening to
contemporary music CD or cassette, CD/cassette either classical music or contemporary music
player, stopwatch, and coin for each pair of students helps students complete a maze task more quickly.
ACTIVITY 6.2A (continued)
274 CHAPTER 6 Designing Experiments

ACTIVITY 6.2A (continued)

You will work with a partner during the data remaining member times, using the procedure
collection. described in Step 2.
1. Your teacher will distribute two copies of Maze 6. Your teacher will play a short selection of
1 face-down to each pair of students. With your contemporary music. When the music stops, the
partner, decide who will work the maze first and remaining member of each pair will work Maze 2
who will time. while the partner times.
2. When the timer says “go,” the subject should turn 7. With your partner, compare your times on Maze
the paper over and connect the circles in the maze, 1 and Maze 2. Did the music appear to help? If so,
beginning at 1 (the start) and finishing at 25 (the did the classical music or the contemporary music
end). The subject should say “done” when finished. produce more improvement?
Record the time it takes for the subject to complete
the maze to the nearest tenth of a second. (Note: The 8. Pool results with your classmates. In one column,
timer should not look at the maze!) list the difference in times (Maze 2  Maze 1)
for each student who received the classical music
3. Switch roles, and repeat Step 2. treatment. In a second column, list the difference
4. Next, the members of each pair must be assigned for each student who listened to contemporary
at random to the two treatments: classical music or music. Construct a graph to compare the results.
contemporary music. Determine this with a coin flip. 9. Compute the average difference for each of
5. Your teacher will play a short selection of the two lists. Do you have reason to believe that
classical music. When the music stops, the chosen classical music helped more than contemporary
member of each pair will work Maze 2 while the music? Explain.

In 1993, three researchers conducted an experiment

to determine whether listening to Mozart’s music
would improve performance on spatial-reasoning
tasks. Using 36 college students as subjects, Frances
Rauscher and her colleagues randomly assigned 12
students to each of three treatment groups. Group 1
listened to a 10-minute selection from Mozart’s Sonata
for Two Pianos in D Major. Group 2 listened to a
relaxation tape with mixed sounds for 10 minutes.
Group 3 sat in silence for 10 minutes and served as a
control group. Each subject completed a pretest two
days before the treatment was given and a posttest
©Paha/Dreamstime.com immediately following the treatment.
The results of the experiment were surprising:
students who listened to Mozart showed significant gains in their scores on spatial-
reasoning tasks.14 Rauscher, Shaw, and Ky published their findings in Nature, and
the so-called Mozart Effect was born. Parents looking for ways to improve their
child’s IQ scores started buying Mozart tapes. One state provided Mozart cassettes
and CDs to parents of newborn babies. Another passed legislation requiring that
preschools play 30 minutes of classical music a day. Meanwhile, other researchers
attempted to verify the Mozart Effect in experiments of their own. A few obtained
 6. 2 Experiments in the Real World 275

similar results. Many, however, found no evidence of a Mozart Effect. The contro-
versy continues, and Mozart plays on.

Equal treatment for all

Probability samples are a big idea, but sampling in practice has difficulties that
just using random samples doesn’t solve. Randomized comparative experiments
are also a big idea, but they don’t solve all the difficulties of experimenting. A sam-
pler must know exactly what information is wanted and must compose questions
that extract that information from the sample. An experimenter must know exactly
what treatments and responses information is wanted about and must construct
the apparatus needed to apply the treatments and measure the responses. This is
what psychologists, medical researchers, and engineers mean when they talk about
“designing an experiment.” We are concerned with the statistical side of designing
experiments, ideas that apply to experiments in psychology, medicine, engineering,
and other areas as well. Even at this general level, you should understand the practi-
cal problems that can prevent an experiment from producing useful data.
The logic of a randomized comparative experiment assumes that all the sub-
jects are treated alike except for the treatments that the experiment is designed to
compare. Any other unequal treatment can cause bias. But treating subjects exactly
alike is hard to do.

Example 6.8 Mice, rats, and rabbits

Mice, rats, and rabbits that are specially bred to be uniform in their inherited char-
acteristics are the subjects in many experiments. Animals, like people, can be quite
sensitive to how they are treated. Here are two amusing examples of how unequal
treatment can create bias.
Does a new breakfast cereal provide good nutrition? To find out, compare the
weight gains of young rats fed the new product and rats fed a standard diet. The
rats are randomly assigned to diets and are housed in large racks of cages. It turns
out that rats in upper cages grow a bit faster than rats in bottom cages. If the ex-
perimenters put rats fed the new product at the top and those fed the standard diet
below, the experiment would be biased in favor of the new product.15 Solution: Assign
the rats to cages at random.
Another study looked at the effects of human affection on the cholesterol level
of rabbits. All of the rabbit subjects ate the same diet. Some (chosen at random)
were regularly removed from their cages to have their furry heads scratched by
friendly people. The rabbits who received affection had lower cholesterol. So af-
fection for some but not other rabbits could bias other experiments in which the
rabbits’ cholesterol level is a response variable.

Double-blind experiments
Placebos work. That bare fact means that medical studies must take special care
to show that a new treatment is not just a placebo. Part of equal treatment for all
subjects is to be sure that the placebo effect operates on all subjects.
276 CHAPTER 6 Designing Experiments

Example 6.9 The powerful placebo

Want to help balding men keep their hair? Give them a placebo—one study found
that 42% of balding men maintained or increased the amount of hair on their heads
when they took a placebo. In another study, researchers zapped the wrists of 24 test
subjects with a painful jolt of electricity. Then they rubbed a cream with no active
medicine on subjects’ wrists and told them the cream should help soothe the pain.
When researchers shocked them again, 8 subjects said they experienced signifi-
cantly less pain.16
When the ailment is vague and psychological, like depression, some experts
think that about three-quarters of the effect of the most widely used drugs is just
the placebo effect.17 Others disagree. The strength of the placebo effect in medical
treatments is hard to pin down because it depends on the exact environment. How
enthusiastic the doctor is seems to matter a lot. But “placebos work” is a good place
to start when you think about planning medical experiments.

The strength of the placebo effect is a strong argument for randomized com-
parative experiments. In the baldness study, 42% of the placebo group kept or in-
creased their hair, but 86% of the men getting a new drug to fight baldness did so.
The drug beats the placebo, so it has something besides the placebo effect going for
it. Of course, the placebo effect is still part of the reason this and other treatments
work.
Because the placebo effect is so strong, it would be foolish to tell subjects in a
medical experiment whether they are receiving a new drug or a placebo. Know-
ing that they are getting “just a placebo” might weaken the placebo effect and
bias the experiment in favor of the other treatments. It is also foolish to tell doc-
tors and other medical personnel what treatment each subject received. If they
know that a subject is getting “just a placebo,” they may expect less than if they
know the subject is receiving a promising experimental drug. Doctors’ expecta-
tions change how they interact with patients and even the way they diagnose a
patient’s condition. Whenever possible, experiments with human subjects should
be double-blind.

Double-blind experiments

In a double-blind experiment, neither the subjects nor the people

who work with them know which treatment each subject is receiving.

Until the study ends and the results are in, only the study’s statistician knows
for sure which treatment a subject is receiving. Reports in medical journals regu-
larly begin with words like these, from a study of a flu vaccine given as a nose spray:
“This study was a randomized, double-blind, placebo-controlled trial. Participants
were enrolled from 13 sites across the continental United States between mid-
September and mid-November.”18 Doctors are supposed to know what this means.
Now you also know.
 6. 2 Experiments in the Real World 277

“Dr. Burns, are you sure this is what the statisticians

call a double-blind experiment?”

Refusals, nonadherers, and dropouts

Sample surveys suffer from nonresponse due to failure to contact some people se-
lected for the sample and the refusal of others to participate. Experiments with
human subjects suffer from similar problems.

Example 6.10 Minorities in clinical trials

Refusal to participate is a serious problem for medical experiments on treatments
for serious diseases such as cancer. As with samples, bias can result if those who
refuse are systematically different from those who cooperate.
Minorities, women, the poor, and the elderly have long been underrepresented
in clinical trials. In many cases, they weren’t asked. The law now requires represen-
tation of women and minorities, and data show that most clinical trials now have
fair representation. But refusals remain a problem. Minorities, especially blacks,
are more likely to refuse to participate. The government’s Office of Minority Health
says, “Though recent studies have shown that African Americans have increasingly
positive attitudes toward cancer medical research, several studies corroborate that
they are still cynical about clinical trials. A major impediment for lack of partici-
pation is a lack of trust in the medical establishment.”19 Some remedies for lack of
trust are complete and clear information about the experiment, insurance coverage
for experimental treatments, participation of black researchers, and cooperation
with doctors and health organizations in black communities.

Subjects who participate but don’t follow the experimental treatment, called
nonadherers, can also cause bias. AIDS patients who participate in trials of a new
drug sometimes take other treatments on their own, for example. What’s more,
some AIDS subjects have their medication tested privately and drop out or add
other medications if they were not assigned to the new drug. This may bias the trial
against the new drug.
278 CHAPTER 6 Designing Experiments

Experiments that continue over an extended period of time also suffer dropouts,
subjects who begin the experiment but do not complete it. If the reasons for drop-
ping out are unrelated to the experimental treatments, no harm is done other than
reducing the number of subjects. If subjects drop out because of their reaction to
one of the treatments, bias can result.

Example 6.11 Dropouts in a medical study

Orlistat is a drug that may help reduce obesity by preventing absorption of fat from
the foods we eat. As usual, the drug was compared with a placebo in a double-blind
randomized trial. Here’s what happened.
Researchers started with 1187 obese subjects. They gave a placebo to all the sub-
jects for 4 weeks and dropped the subjects who didn’t take a pill regularly. This at-
tacked the problem of nonadherers. There were 892 subjects left. Researchers ran-
domly assigned these subjects to orlistat or a placebo, along with a weight-loss diet.
After a year devoted to losing weight, 576 subjects were still participating. On the
average, the orlistat group lost 3.15 kilograms (about 7 pounds) more than the pla-
cebo group. The study kept going for another year, emphasizing maintaining the
weight loss from the first year. At the end of the second year, 403 subjects were left.
That’s only 45% of the 892 who were randomized. Orlistat again beat the placebo,
reducing the weight regained by an average of 2.25 kilograms (about 5 pounds).
Can we trust the results when so many subjects dropped out? The overall
dropout rates were similar in the two groups: 54% of the subjects taking orlistat
and 57% of those in the placebo group dropped out. Were dropouts related to
the treatments? Placebo subjects in weight-loss experiments often drop out be-
cause they aren’t losing weight. This would bias the study against orlistat because
the subjects in the placebo group at the end may be those who could lose weight
just by following a diet. The researchers looked carefully at the data available
for subjects who dropped out. Dropouts from both groups had lost less weight
than those who stayed, but careful statistical study suggested that there was little
bias. Perhaps so, but the results aren’t as clean as our first look at experiments
promised.20

Exercises 6.27 Medical news When it was found that hydroxyurea reduced the symptoms
of sickle-cell anemia, the National Institutes of Health released a medical
bulletin. The bulletin said, “These findings are the results of data analyzed from
the Multicenter Study of Hydroxyurea in Sickle Cell Anemia (MSH), which was
a double-blind, placebo-controlled trial in which half of the patients received
hydroxyurea and half received a placebo capsule.” Explain to someone who
knows no statistics what the terms “placebo-controlled” and “double-blind”
mean here.
6.28 Emergency room care An article in a medical journal reports on
an experiment to see if injecting an oxygen-carrying fluid in addition to
performing standard emergency room procedures would help patients in shock
 6. 2 Experiments in the Real World 279

from loss of blood. The article describes the experiment as a “randomized,

controlled, single-blinded efficacy trial conducted at 18 U.S. trauma centers.”21
What do you think “single-blinded” means here? Why isn’t a double-blind
experiment possible?
6.29 Testing a natural remedy Although the law doesn’t require it, we decide to
subject Dr. Moore’s Old Indiana Extract to a clinical trial. We hope to show that
the extract reduces pain from arthritis. Sixty patients suffering from arthritis and
needing pain relief are available. We will give a pill to each patient and ask them
an hour later, “About what percent of pain relief did you experience?”
(a) Why should we not simply give the extract to all 60 patients and record the
responses?
(b) Outline the design of an experiment to compare the extract’s effectiveness
with that of aspirin and of a placebo.
(c) Should patients be told which remedy they are receiving? How might this
knowledge affect their reactions?
(d) If patients are not told which treatment they are receiving, the experiment is
single-blind. Should this experiment be double-blind? Explain.
6.30 Real-world problems Explain the difference between refusals,
nonadherers, and dropouts in an experimental study. Why must researchers be
concerned about these three groups?
6.31 Ultrasound and birth weight In this study, researchers examined the
effect of ultrasound on birth weight. Pregnant women participating in the study
were randomly assigned one of two groups. The first group of women received
an ultrasound; the second group did not. When the subjects’ babies were born,
their birth weights were recorded. The women who received the ultrasounds had
heavier babies.22
(a) Was the experiment double-blind? Why is this important?
(b) Did the experimental design take the placebo effect into account? Why is this
important?
(c) Based on your answers to (a) and (b), describe an improved design for this
experiment.
6.32 Activity 6.2A follow-up Refer to the Mozart Effect experiment from
Activity 6.2A (page 273).
(a) Was the experiment double-blind? Why is this important?
(b) Did the experimental design take the placebo effect into account? Why is this
important?
(c) Why was the coin flip important in this experiment?

Can we generalize?
A well-designed experiment tells us that changes in the explanatory variable cause
changes in the response variable. More exactly, it tells us that this happened for spe-
cific subjects in the specific environment of this specific experiment. No doubt we
had bigger things in mind. We want to proclaim that our new method of teaching
math does better for high school students in general or that our new drug beats a
280 CHAPTER 6 Designing Experiments

placebo for some broad class of patients. Can we generalize our conclusions from
our little group of subjects to a wider population?
The first step is to be sure that our findings are statistically significant, that they
are too strong to often occur just by chance. That’s important, but it’s a technical
detail that the study’s statistician can reassure us about. The serious threat is that
the treatments, the subjects, or the environment of our experiment may not be re-
alistic. Let’s look at some examples.

Example 6.12 Studying frustration

A psychologist wants to study the effects of failure and frustration on the relation-
ships among members of a work team. She forms a team of students, brings them to
the psychology laboratory, and has them play a game that requires teamwork. The
game is rigged so that they lose regularly. The psychologist observes the students
through a one-way window and notes the changes in their behavior during an eve-
ning of game playing.
Playing a game in a laboratory for small stakes, knowing that the session will
soon be over, is a long way from working for months developing a new product that
never works right and is finally abandoned by your company. Does the behavior of
the students in the lab tell us much about the behavior of the team whose product
failed?

In Example 6.12, the subjects (students who know they are subjects in an ex-
periment), the treatment (a rigged game), and the environment (the psychology
lab) are all unrealistic if the psychologist’s goal is to reach conclusions about the
effects of frustration on teamwork in the workplace. Psychologists do their best
to devise realistic experiments for studying human behavior, but lack of realism
limits the usefulness of experiments in this area.

Example 6.13 Center brake lights

Cars sold in the United States since 1986 have been required to have a high cen-
ter brake light in addition to the usual two brake lights at the rear of the vehicle.
This safety requirement was justified by randomized comparative experiments with
fleets of rental and business cars. The experiments showed that the third brake light
reduced rear-end collisions by as much as 50%.
After almost a decade of actual use of center brake lights, the Insurance Insti-
tute found only a 5% reduction in rear-end collisions, helpful but much less than
the experiments predicted. What happened? Most cars did not have the extra brake
light when the experiments were carried out, so it caught the eye of following driv-
ers. Now that almost all cars have the third light, it no longer captures attention.
The experimental conclusions did not generalize as well as safety experts hoped
because the environment changed.
 6. 2 Experiments in the Real World 281

Example 6.14 Are subjects treated too well?

Surely medical experiments are realistic? After all, the subjects are real patients in
real hospitals really being treated for real illnesses.
Even here, there are some questions. Patients participating in medical trials
get better medical care than most other patients, even if they are in the placebo
group. Their doctors are specialists doing research on their specific ailment. They are
watched more carefully than other patients. They are more likely to take their pills
regularly because they are constantly reminded to do so. Providing “equal treatment
for all” except for the experimental and control therapies translates into “provide the
best possible medical care for all.” The result: Ordinary patients may not do as well as
the clinical trial subjects when the new therapy comes into general use. It’s likely that
a therapy that beats a placebo in a clinical trial will beat it in ordinary medical care,
but “cure rates” or other measures of success from the trial may be optimistic.

The Carolina Abecedarian Project (Section 6.1) faces the same “too good to be re-
alistic” question. That long and expensive experiment does show that intensive day care
has substantial benefits in later life. The day care in the study was intensive indeed—lots
of highly qualified staff, lots of parent participation, and detailed activities starting at a
very young age, all costing about $11,000 per year for each child. It’s unlikely that society
will decide to offer such care to all low-income children. The unanswered question is a
big one: how good must day care be to really help children succeed in life?
When experiments are not fully realistic, statistical analysis of the experimental data
cannot tell us how far the results will generalize. Experimenters generalizing from stu-
dents in a lab to workers in the real world must argue based on their understanding of
how people function, not based just on the data. It is even harder to generalize from rats in
a lab to people in the real world. This is one reason why a single experiment is rarely com-
pletely convincing, despite the compelling logic of experimental design. The true scope of
Meta-analysis
a new finding must usually be explored by a number of experiments in various settings.
A single study of an impor-
A convincing case that an experiment is sufficiently realistic to produce useful
tant issue is rarely decisive.
We often find several information is based not on statistics but on the experimenter’s knowledge of the
studies in different settings, subject matter of the experiment. The attention to detail required to avoid hidden
with different designs, and bias also rests on subject-matter knowledge. Good experiments combine statistical
of different quality. Can we principles with understanding of a specific field of study.
combine their results to get
an overall conclusion? That
is the idea of “meta-analysis.” Experimental design in the real world
Of course, differences The experimental designs we have met all have the same pattern: divide the subjects
among the studies prevent at random into as many groups as there are treatments, then apply each treatment
us from just lumping them
to one of the groups. These are completely randomized designs.
together. Statisticians have
more sophisticated ways
of combining the results.
Completely randomized design
Meta-analysis has been
applied to issues from In a completely randomized experimental design, all the experimental
the effect of secondhand
smoke to whether coaching
subjects are allocated at random among all the treatments.
improves SAT scores.
282 CHAPTER 6 Designing Experiments

Our examples to this point have had only a single explanatory variable
(drug versus placebo, classroom versus Web instruction). A completely random-
ized design can have any number of explanatory variables. Here is an example
with two.

Example 6.15 Durable fabric

A fabrics researcher is studying the durability of a fabric under repeated wash-
ings. Because the durability may depend on the water temperature and the type of
cleansing agent used, the researcher decides to investigate the effect of these two
explanatory variables on durability. Variable A is water temperature and has three
levels: hot (1458F), warm (1008F), and cold (508F). Variable B is the cleansing agent
and also has three levels: regular Tide, low-phosphate Tide, and Ivory Liquid. A
treatment consists of washing a piece of fabric (a unit) 50 times in a home automatic
washer with a specific combination of water temperature and cleansing agent. The
response variable is strength after 50 washes, measured by a fabric-testing machine
that forces a steel ball through the fabric and records the fabric’s resistance to break-
ing. Figure 6.4 shows the layout of the treatments.

Variable B
Cleansing Agent
Regular Tide Low-phosphate Tide Ivory Liquid

Hot (145°F) Treatment 1 Treatment 2 Treatment 3

Variable A
Warm (100°F) Treatment 4 Treatment 5 Treatment 6
Temperature

Cold (50°F) Treatment 7 Treatment 8 Treatment 9

Figure 6.4 The treatments in the fabric durability experiment. Combinations of two
explanatory variables form nine treatments.

In Example 6.15, there are nine possible treatments (combinations of a tem-

perature and a cleansing agent). By using them all, the researcher obtains a wealth
of information on how temperature alone, cleansing agent alone, and the two in
combination affect the durability of the fabric. For example, water temperature may
have no effect on the strength of the fabric when regular Tide is used, but after
50 washings in low-phosphate Tide, the fabric may be weaker when cold water is
used instead of hot water. This kind of combination effect is called an interaction
between cleansing agent and water temperature. Interactions can be important, as
when a drug that ordinarily has no unpleasant side effects interacts with alcohol
to knock out the patient who drinks a beer. Because an experiment can combine
levels of several explanatory variables, interactions between the variables can be
observed.
 6. 2 Experiments in the Real World 283

Exercises 6.33 Testing a natural remedy The National Institutes of Health is at last
sponsoring proper clinical trials of some natural remedies. In one study at Duke
University, 330 patients with mild depression are enrolled in a trial to compare
Saint-John’s-wort with a placebo and with Zoloft, a common prescription drug for
depression. The Beck Depression Inventory is a common instrument that rates the
severity of depression on a 0-to-3 scale.
(a) What would you use as the response variable to measure change in depression
after treatment?
(b) Outline the design of a completely randomized clinical trial for this study.
(c) What other precautions would you take in this trial?
6.34 Daytime running lights Canada requires that cars be equipped with
“daytime running lights,” headlights that automatically come on at a low level
when the car is started. Some manufacturers are now equipping cars sold in the
United States with running lights. Will running lights reduce accidents by making
cars more visible?
(a) Briefly discuss the design of an experiment to help answer this question. In
particular, what response variables will you examine?
(b) Example 6.13 discusses center brake lights. What cautions do you draw
from that example that apply to an experiment on the effects of running
lights?
6.35 Diet and cancer Substances that cause cancer should not appear in our
food. We don’t want to experiment on people to learn what substances cause
cancer, so we experiment on rats instead. The rats are specially bred to have more
tumors than humans do. They are fed large doses of the test chemical for most of
their natural lives, about 2 years. Briefly discuss the questions that arise in using
these experiments to decide what is safe in human diets.
6.36 Dealing with cholesterol Clinical trials have shown that reducing blood
cholesterol using either drugs or diet reduces heart attacks. The first researchers
followed their subjects for 5 to 7 years. In order to see results as quickly as
possible, the subjects were chosen from the group at greatest risk, middle-aged
men with high cholesterol levels or existing heart disease. The experiments
generally showed that reducing blood cholesterol does decrease the risk of a heart
attack. Some doctors questioned whether these experimental results applied to
many of their patients. Why?
6.37 Clean clothes Which of two brands of laundry detergent—Brand A or
Brand B—cleans clothes better? Does one brand work better in both hot and cold
water? Let’s design an experiment to find out. A basket containing 120 pieces of
dirty laundry is available for the experiment.
(a) Identify the subjects, explanatory variable(s), and response variable(s) for this
experiment.
(b) Make a diagram like Figure 6.4 to describe the treatments. How many
treatment combinations are there? How many pieces of laundry should be
assigned to each treatment combination?
(c) Describe how you would use randomization in your experimental design.
284 CHAPTER 6 Designing Experiments

6.38 Baking cakes A food company is preparing to market a new

cake mix. It is important that the taste of the cake not be changed
by small variations in baking time or temperature. In an
experiment, cakes made from the mix are baked at 3008F, 3208F,
or 3408F, and for 1 hour or for 1 hour and 15 minutes. Ten
cakes are baked at each combination of temperature and time.
A panel of tasters scores each cake for texture and taste.
(a) What are the explanatory variables and the response
variables for this experiment?
(b) Make a diagram like Figure 6.4 to describe the treatments.
How many treatments are there? How many cakes are needed?
M

ha
ic

el
Va
ld
ez
/iS
(c) Explain why it is a bad idea to bake all 10 cakes for
toc
kp
ho
to
one treatment at once, then bake the 10 cakes for the second
treatment, and so on. Instead, the experimenters will bake the cakes
in a random order determined by the randomization in your design.

Matched pairs and block designs

Completely randomized designs are the simplest statistical designs for experiments.
They illustrate clearly the principles of control and randomization. However, com-
pletely randomized designs are often inferior to more elaborate statistical designs.
In particular, matching the subjects in various ways can produce more precise re-
sults than simple randomization.
One common design that combines matching with randomization is the matched
pairs design. A matched pairs design compares just two treatments. Choose pairs of
subjects that are as closely matched as possible. Randomly assign the two treatments to
the subjects in each pair by tossing a coin or reading odd and even digits from Table B.
Sometimes each “pair” in a matched pairs design consists of just one subject,
who gets both treatments one after the other. Each subject serves as his or her own
control. The order of the treatments can influence the subject’s response, so we ran-
domize the order for each subject, again by a coin toss.

ACTIVITY 6.2B
Stepping up your heart rate
Materials: Stopwatch, two steps of height about 1. One option for designing the experiment would
6 and 12 inches (you may be able to use some be to randomly assign half of the students in your
steps near your classroom), metronome (if class to use 6-inch steps and the other half to use
available) 12-inch steps. Explain why this is not a good idea.
How is heart rate affected by physical activity? The 2. Instead, we’ll use a matched pairs design in
more work your body is doing, the higher your which each subject receives both treatments.
heart rate will be. In this Activity, you will perform Randomly assign the students in your class to
an experiment to examine the effect on people’s one of two treatment groups, with about equal
heart rates of repeatedly stepping up and down on numbers in each group. You can draw names
steps of two different heights. from a hat or assign numbers and use the random
 6. 2 Experiments in the Real World 285

digit table or your calculator. Students assigned to 7. For the second trial, you will switch to the other
Group 1 will use the 6-inch step on the first trial step height for 3 minutes. When you have finished,
and the 12-inch step on the second trial. Students take your pulse one more time and record the result.
assigned to Group 2 will use the 12-inch step first
8. Calculate the difference in your stepping pulse rate
and then the 6-inch step.
and your resting pulse rate for each of the two trials. Did
3. When your teacher instructs you to, take your your pulse increase more when you used the higher step?
pulse for 60 seconds. Record this “resting” pulse
9. Make a chart on the board that shows the
rate.
differences in pulse rate for each subject, like this:
4. Your teacher will tell you when to begin the first
6 step 12 step Diff.
trial. Be sure to step up and down according to the
Subject Group diff. diff. (12  6)
pace set by your teacher’s voice (or the metronome).
You will step up and down for 3 minutes.
5. At the end of the first trial, take your pulse for 10. Construct a graph to display the differences in
60 seconds. Record the data. pulse rate for the 6- and 12-inch steps (last column
in the chart). Describe what you see.
6. Your teacher will give you a few minutes
to “recover.” Afterward, take your pulse again 11. Calculate the average difference in pulse rate for
prior to beginning the second trial. Record this your class. Is there evidence that heart rate increases
measurement. more with greater step height? Explain.

Activity 6.2B shows you how a matched pairs experimental design can be ef-
fectively used to answer a question of interest. Here’s another example of a matched
pairs design that produced less clear results.

Example 6.16 Coke versus Pepsi

Pepsi wanted to demonstrate that Coke drinkers prefer Pepsi when they taste both
colas blind. The subjects, all people who said they were Coke drinkers, tasted both
colas from glasses without brand markings and said which they liked better. This
is a matched pairs design in which each subject compares the two colas. Because
responses may depend on which cola is tasted first, the order of tasting should be
chosen at random for each subject.
When more than half the Coke drinkers chose Pepsi, Coke claimed that the ex-
periment was biased. The Pepsi glasses were marked M and the Coke glasses were
marked Q. Aha, said Coke, the results could just mean that people like the letter M
better than the letter Q. The matched pairs design is adequate, but a more careful
experiment would avoid any distinction other than Coke versus Pepsi.23

Matched pairs designs use the principles of comparison of treatments and ran-
domization. However, the randomization is not complete—we do not randomly
assign all the subjects at once to the two treatments. Instead, we randomize only
within each matched pair. Matching reduces the effect of variation among the sub-
jects. Matched pairs are an example of block designs.
286 CHAPTER 6 Designing Experiments

Block design

A block is a group of experimental subjects that are known before the

experiment to be similar in some way that is expected to affect the
response to the treatments. In a block design, the random assignment
of subjects to treatments is carried out separately within each block.

A block design combines the idea of creating equivalent treatment groups by match-
ing with the principle of forming treatment groups at random. Blocks are another form
of control. They control the effects of some outside variables by bringing those variables
into the experiment to form the blocks. Here is a typical example of a block design.

Example 6.17 Men, women, and advertising

Women and men respond differently to advertising. An experiment to compare the ef-
fectiveness of 3 television commercials for the same product will want to look separately at
the reactions of men and women, as well as assess the overall response to the ads.
A completely randomized design considers all subjects, both men and women, as
a single pool. The randomization assigns subjects to three treatment groups with-
out regard to their sex. This ignores the differences between men and women. A
better design considers women and men separately. Randomly assign the women
to three groups, one to view each commercial. Then separately assign the men at
random to three groups. Figure 6.5 outlines this improved design.

Group 1 Ad 1

Women Random Group 2 Ad 2 Compare

assignment reaction
Group 3 Ad 3
Subjects

Group 1 Ad 1

Men Random Group 2 Ad 2 Compare

assignment reaction
Group 3 Ad 3

Figure 6.5 A block design to compare the effectiveness of 3 TV advertisements.

Female and male subjects form 2 blocks.

A block is a group of subjects formed before an experiment starts. We reserve

the word “treatment” for a condition that we impose on the subjects. We don’t speak
of 6 treatments in Example 6.17 even though we can compare the responses of
6 groups of subjects formed by the 2 blocks (men, women) and the 3 commercials.
Block designs are similar to stratified samples. Blocks and strata both group
similar individuals together. We use two different names only because the idea
 6. 2 Experiments in the Real World 287

developed separately for sampling and experiments. The advantages of block de-
signs are the same as the advantages of stratified samples. Blocks allow us to draw
separate conclusions about each block—for example, about men and women in
the advertising study in Example 6.17. Blocking also allows more precise overall
conclusions, because the systematic differences between men and women can be
removed when we study the overall effects of the three commercials. The idea of
blocking is an important additional principle of statistical design of experiments.
A wise experimenter will form blocks based on the most important unavoid-
able sources of variability among the experimental subjects. Randomization will
then average out the effects of the remaining variation and allow an unbiased com-
parison of the treatments.

DATA EXPLORATION
Get Your Heart Beating
Are standing pulse rates generally higher than subjects. This time, random assignment was used
sitting pulse rates? A high school statistics class to decide which 7 students would stand first and
performed a set of two experiments involving 14 which 7 would sit first. Once again, the students
students to help answer this question. In this Data completed their assigned treatments in the same
Exploration, you’ll analyze their results. room at the same time. Then they measured
1. In the first experiment, random assignment their pulses for 60 seconds. Following a brief
was used to divide the subjects into two groups adjustment period, the subjects who sat first stood
of 7 students—a sitting group and a standing up, and those who stood first sat down. Then the
group. All of the students completed their assigned subjects measured their pulse rates again for
treatments in the same room at the same time. 60 seconds.
Then they measured their pulses for 60 seconds. A
comparative dotplot and numerical summaries for
these data are shown below. Write a few sentences
comparing the pulse rates in the two treatment
groups. What conclusion do you draw?

Group n x s Min. Q1 M Q3 Max.

2. The scatterplot displays data from the second
Sit 7 74.86 6.09 66 72 76 76 86
experiment.
Stand 7 78 19.49 60 66 72 82 118
(a) What does each point represent?
A second experiment was conducted using a (b) Describe the pattern you see in the scatterplot.
matched pairs design with the same 14 students as Explain what it means.
DATA EXPLORATION (continued)
288 CHAPTER 6 Designing Experiments

DATA EXPLORATION (continued)

3. For each subject, the difference between standing

and sitting pulse rate was calculated (difference 5
standing 2 sitting). A dotplot and numerical
summaries of the difference values are shown here.
Write a few sentences describing the distribution of
difference in pulse rates. What conclusion do you
draw now?
n x s Min. Q1 M Q3 Max.
Difference 14 11.57 7.37 2 6 11 18 26

Exercises 6.39 In the cornfield An agriculture researcher wants to compare the yield of
5 corn varieties. The field in which the experiment will be carried out increases
in fertility from north to south. The researcher therefore divides the field into
30 plots of equal size, arranged in 6 east-west rows of 5 plots each, and employs a
block design with the rows of plots as the blocks.
(a) Draw a sketch of the field, divided into 30 plots. Label the rows Block 1 to Block 6.
(b) Do the randomization required by the block design. That is, randomly assign
the 5 corn varieties A, B, C, D, and E to the 5 plots in each block. Mark on your
sketch which variety is planted in each plot.
6.40 Comparing cancer treatments The progress of a type of cancer differs in
women and men. A clinical experiment to compare three therapies for this cancer
therefore treats sex as a blocking variable.
(a) You have 500 male and 300 female patients who are willing to serve as subjects.
Use a diagram to outline a block design for this experiment. Use Figure 6.5 as a model.
(b) What are the advantages of a block design over a completely randomized
design using these 800 subjects? What are the advantages of a block design over a
completely randomized design using 800 male subjects?
6.41 Doctors and nurses Nurse-practitioners are nurses with advanced
qualifications who often act much like primary-care physicians. Are they as
effective as doctors at treating patients with chronic conditions? An experiment
was conducted with 1316 patients who had been diagnosed with asthma, diabetes,
or high blood pressure. Within each condition, patients were randomly assigned to
either a doctor or a nurse-practitioner. The response variables included measures of
the patients’ health and of their satisfaction with their medical care after 6 months.24
(a) Is the diagnosis (asthma, etc.) a treatment variable or a block? Why?
(b) Is the type of care (nurse or doctor) a treatment variable or a block? Why?
6.42 More rats In Exercise 6.12 (page 266), a nutritionist had 10 rats of each
of two genetic strains. The effect of genetic strain can be controlled by treating
the strains as blocks and randomly assigning 5 rats of each strain to Diet A. The
remaining 5 rats of each strain receive Diet B.
 6. 2 Experiments in the Real World 289

(a) Describe the design of this experiment.

(b) Use Table B, beginning at line 111, to do the required randomization.
6.43 Activity 6.2B follow-up Refer to the heart rate experiment from Activity 6.2B.
(a) Why did we randomize the order in which each student used the two different
step heights?
(b) Examine the class data from a different angle. Do you see any evidence of a
“learning effect” from the first step height to the second? Follow the four-step
statistical problem-solving process (page 19) as you answer this question.
6.44 Data Exploration follow-up Refer to the Data Exploration (page 287) on
standing and sitting pulse rates.
(a) What statistical advantages did the class’s matched pairs design have over its
completely randomized design?
(b) In the matched pairs experiment, why did the class randomize who stood first
and who sat first?

ws
Ne
ws
Ne

APPLICATION 6.2 B]
RO
g
<
Se
a

Design your own study

In this Application, you will design a taste-testing • how you will determine which subjects are to
experiment in which subjects will be asked which be blinded and which will be told what they are
of two similar items tastes better to them. However, tasting, and
the issue is not which item the subjects prefer, but • what other special precautions you will take to
rather whether or not preferences are different when avoid confounding.
you know what is being tasted. “Blinding” subjects
is an important precaution in testing a new medi- Incorporate these items into a written report
cal treatment because subjects might be influenced describing your experiment.
by the idea of being treated. We want to design a Here are some examples of items you might
study to see if blinding is also an important issue taste-test:
for taste-testing experiments. For example, perhaps • a nonfat yogurt versus a regular yogurt of the
people will tend to state a preference for a more ex- same brand,
pensive item because they think they are supposed
to like it better. • Kellogg’s corn flakes versus a store brand of corn
Your group will develop a complete experimen- flakes,
tal plan to determine whether blinding affects sub- • orange juice from a carton versus freshly squeezed
jects’ preferences. Your group’s plan should describe orange juice, or
the following: • french fries cooked in a microwave versus the
• precisely what the two items being tested are, same brand cooked in a conventional oven.
• how the items will be served (for example, will Remember the issue at hand: whether or not prefer-
corn flakes be served with nonfat, low-fat, or ences are different when you know what is being tasted.
regular milk?), Once your teacher approves your design, carry out
• how you will decide which item to present first your experiment and analyze the data. Follow the four-
and which to present second, step statistical problem-solving process (page 19).
290 CHAPTER 6 Designing Experiments

Section 6.2 Summary

As with samples, experiments require a combination of good statistical design and
steps to deal with practical problems. Because the placebo effect is strong, clinical
trials and other experiments with human subjects should be double-blind whenever
possible. The double-blind method helps achieve a basic requirement of compara-
tive experiments: equal treatment for all subjects except for the actual treatments
the experiment is comparing.
Just as samples suffer from nonresponse, experiments suffer from uncoopera-
tive subjects. Some subjects refuse to participate; others drop out before the experi-
ment is complete; others, called nonadherers, don’t follow instructions, as when
some subjects in a drug trial don’t take their pills. The most common weakness in
experiments is that we can’t generalize the conclusions widely. Some experiments
apply unrealistic treatments, some use subjects from some special group such as
college students, and all take place at some specific place and time. We need similar
experiments at other places and times to confirm important findings.
Many experiments use designs that are more complex than the basic completely
randomized design that divides all the subjects among all the treatments in one ran-
domization. Matched pairs designs compare two treatments by giving one to one mem-
ber of a pair of similar subjects and the other to the other member or by giving both to
the same subject in random order. Block designs form blocks of similar subjects and
assign treatments at random separately in each block. The big ideas of randomization,
control, and using enough subjects remain the keys to convincing experiments.

Section 6.2 6.45 Better air bags Car makers are interested in developing safer air bags that
Exercises reduce injuries in collisions. An automobile manufacturer is trying to decide between two
possible air bags for use in the company’s vehicles. Six reusable crash test dummies—large
male, average male, small male, large female, average female, and small female—are
available for use on the company’s indoor “test track.” Carefully describe the design
of an experiment to help the company decide which air bag is safer.
6.46 Meditation lowers anxiety An experiment that was publicized as showing
that a meditation technique lowered the anxiety level of subjects was conducted
as follows: The experimenter interviewed the subjects and assessed their levels
of anxiety. The subjects then learned how to meditate and did so regularly for a
month. The experimenter reinterviewed them at the end of the month and
assessed whether their anxiety levels had decreased or not.
(a) There was no control group in this experiment. Why is this a blunder? What
lurking variables may be confounded with the effect of meditation?
(b) The experimenter who diagnosed the effect of the treatment knew that the
subjects had been meditating. Explain how this knowledge could bias the experi-
mental conclusion.
(c) Briefly discuss a proper experimental design, with controls and blind diagnosis,
to assess the effect of meditation on anxiety level.
6.47 Ugly french fries Few people want to eat discolored french fries. Potatoes are
kept refrigerated before being cut for french fries to prevent spoiling and preserve
flavor. But immediate processing of cold potatoes causes discoloring due to complex
chemical reactions. The potatoes must therefore be brought to room temperature
 6. 2 Experiments in the Real World 291

before processing. Design an experiment in which tasters will rate the color and flavor of
french fries prepared from several groups of potatoes. The potatoes will be fresh picked
or stored for a month at room temperature or stored for a month refrigerated. They will
then be sliced and cooked either immediately or after an hour at room temperature.
(a) What are the explanatory variables, the treatments, and the response variables?
(b) Describe and outline the design of this experiment.
(c) It is efficient to have each taster rate fries from all treatments. How will you use
randomization in presenting fries to the tasters?
6.48 Drug improves coordination A drug is suspected of affecting the coordination of
subjects. The drug can be administered in three ways: orally, by injection under the skin,
or by injection into a vein. The potency of the drug probably depends on the method
of administration as well as on the dose administered. A researcher therefore wishes to
study the effects of the two explanatory variables: dose at two levels and method of ad-
ministration by the three methods mentioned. The response variable is the score of the
subjects on a standard test of coordination. Ninety subjects are available.
(a) List the treatments that can be formed from the two explanatory variables.
(b) Describe an appropriate completely randomized design. (Just outline the
design; don’t do any randomization.)
(c) The researcher could study the effect of the dose in an experiment comparing
two dose levels for one method of administration. The researcher then could sepa-
rately study the effect of administration by comparing the three methods for one
dose level. What advantage does the experiment you designed in (a) have over
these two separate experiments taken together?
6.49 Harmful auto emissions Most motor vehicles are equipped with catalytic
converters to reduce harmful emissions. The ceramic used to make the convert-
ers must be baked to a certain hardness. The manufacturer must decide which of
three temperatures (5008F, 7508F, and 10008F) is best. The position of the converter
in the oven (front, middle, or back) also affects the hardness. So there are two
explanatory variables: temperature and placement.
(a) List the treatments in this experiment if all possible treatment combinations are used.
(b) Design a completely randomized experiment with 5 units in each group.
(c) Using Table B, beginning at line 101, do the randomization for your experiment.
6.50 Morphine and cancer Health care providers are giving more attention to
relieving the pain of cancer patients. An article in the journal Cancer reviewed
a number of studies and concluded that controlled-release morphine tablets,
which release the painkiller gradually over time, are more effective than giving
standard morphine when the patient needs it.25 The “methods” section of the
article begins: “Only those published studies that were controlled (i.e., ran-
domized, double-blind, and comparative), repeated dose studies with CR mor-
phine tablets in cancer pain patients were considered for this review.” Explain
the terms in parentheses to someone who knows nothing about medical trials.
6.51 Is caffeine dependence real? Many people start their day with a jolt of
caffeine from coffee or a soft drink. Most experts agree that people who take in
large amounts of caffeine each day may suffer from physical withdrawal symp-
toms if they stop ingesting their usual amounts of caffeine. Researchers recruited
11 volunteers who were caffeine dependent and who were willing to take part in
a caffeine withdrawal experiment. The experiment was conducted on two 2-day
periods that occurred one week apart. During one of the 2-day periods, each
292 CHAPTER 6 Designing Experiments

subject was given a capsule containing the amount of caffeine normally ingested
by that subject in one day. During the other study period, the subjects were given
placebos. The order in which each subject received the two types of capsules was
randomized. The subjects’ diets were restricted during each of the study periods. At
the end of each 2-day study period, subjects were evaluated using a tapping task in
which they were instructed to press a button 200 times as fast as they could.26
(a) Carefully describe the design of this experiment. How was blocking incorpo-
rated, and why?
(b) Why did researchers randomize the order in which subjects received the two
treatments?
(c) Could this experiment have been carried out in a double-blind manner? Explain.
6.52 Bee healed! “Bee pollen is effective for combating fatigue, depression, can-
cer, and colon disorders.” So says a Web site that offers the pollen for sale. We won-
der if bee pollen really does prevent colon disorders. Here are two ways to study
this question. Explain why the second design will produce more trustworthy data.
• Find 200 women who take bee pollen regularly. Match each with a woman
of the same age, race, and occupation who does not take bee pollen. Follow
both groups for 5 years.
• Find 400 women who do not have colon disorders. Assign 200 to take bee
pollen capsules and the other 200 to take placebo capsules that are identical
in appearance. Follow both groups for 5 years.

6.3 Data Ethics

ACTIVITY 6.3
Is it ethical?
Medical professionals are taught to follow the
basic principle “First, do no harm.” Shouldn’t those
who carry out statistical studies follow the same
principle? Most reasonable people think so. But
this may not always be as simple as it sounds. Not
convinced? Discuss each of the following statistical
practices with your classmates. Decide whether each
is ethical or unethical.
1. A promising new drug has been developed for
treating cancer in humans. Researchers want to
administer the drug to animal subjects to see if
there are any potentially serious side effects before
giving the drug to human subjects.
2. Are companies discriminating against some
individuals in the hiring process? To find out,
researchers prepare several equivalent résumés for
fictitious job applicants, with the only difference
being the sex of the applicant. They send the fake By permission of Dave Coverly and Creators Syndicate, Inc.
 6. 3 Data Ethics 293

résumés to the companies advertising positions and whether they are taking a placebo or the active
keep track of the number of males and females who drug. Neither do any of the physicians who are
are contacted for interviews. interacting with the subjects know.
3. In a medical study of a new drug for migraine 4. Will people try to stop someone from driving
sufferers, volunteer subjects are randomly assigned drunk? A television news program hires an actor
to two groups. Members of the first group are to play a drunk driver and records the behavior of
given a placebo pill. Subjects in the second group individuals who encounter the driver on hidden
are given the new drug. None of the subjects know camera.

Are sham surgeries ethical?

“Randomized, double-blind, placebo-controlled trials are the gold standard for
evaluating new interventions and are routinely used to assess new medical thera-
pies.”27 So says an article in the New England Journal of Medicine that discusses the
treatment of Parkinson’s disease. The article isn’t about the new treatment, which
offers hope of reducing the tremors and lack of control brought on by the disease,
but about the ethics of studying the treatment.
The law requires well-designed experiments to show that new drugs work
and are safe. Not so with surgery—only about 7% of studies of surgery use
randomized comparisons. Surgeons think their operations succeed—but in-
novators always think their innovations work. Even if the patients are helped,
the placebo effect may deserve most of the credit. So we don’t really know
whether many common surgeries are worth the risk they carry. To find out,
do a proper experiment. That includes a “sham surgery” to serve as a placebo.
In the case of Parkinson’s disease, the promising treatment involves surgery to
implant new cells. The placebo subjects get the same surgery, but the cells are
not implanted.
Is this ethical? One side says: Only a randomized comparative experiment can
tell us if the new treatment works. If it doesn’t work, thousands of patients will be
spared operations that will do no good. If it does work, we can do those thousands
of operations knowing that we offer more than a placebo. The other side says: Any
surgery carries risks. Sham operations place patients at risk without the hope that
the operation will help them. Doctors should not risk harm to some patients today
just because other patients may benefit tomorrow.
Reasonable people disagree about sham surgery as part of a test of real surgery
for Parkinson’s disease. But there is wide agreement on some basic ethical prin-
ciples for statistical studies. To think about the hard cases, we should first think
about these principles.

First principles
The most complex issues of data ethics arise when we collect data from people. The
ethical difficulties are more severe for experiments that impose some treatment on
people than for sample surveys that simply gather information. Trials of new medi-
cal treatments, for example, can do harm as well as good to their subjects. Here are
294 CHAPTER 6 Designing Experiments

some basic standards of data ethics that must be obeyed by any study that gathers
data from human subjects, whether sample survey or experiment.

Basic data ethics

The organization that carries out the study must have an institutional
review board that reviews all planned studies in advance in order to
protect the subjects from possible harm.

All individuals who are subjects in a study must give their informed
consent before data are collected.

All individual data must be kept confidential. Only statistical summa-

ries for groups of subjects may be made public.

The law requires that studies funded by the federal government obey these prin-
ciples. But neither the law nor the consensus of experts is completely clear about the
details of their application.

Institutional review boards

The purpose of an institutional review board is not to decide whether a proposed
study will produce valuable information or whether it is statistically sound. The
board’s purpose is, in the words of one university’s board, “to protect the rights and
welfare of human subjects (including patients) recruited to participate in research
activities.” The board reviews the plan of the study and can require changes. It re-
views the consent form to be sure that subjects are informed about the nature of
the study and about any potential risks. Once research begins, the board monitors
its progress at least once a year.
The most pressing issue concerning institutional review boards is whether
their workload has become so large that their effectiveness in protecting subjects
drops. When the government temporarily stopped human-subject research at
Duke University Medical Center in 1999 due to inadequate protection of subjects,
more than 2000 studies were going on. That’s a lot of review work. There are shorter
review procedures for projects that involve only minimal risks to subjects, such
as most sample surveys. When a board is overloaded, there is a temptation to put
more proposals in the minimal-risk category to speed the work.

Informed consent
Both words in the phrase “informed consent” are important, and both can be contro-
versial. Subjects must be informed in advance about the nature of a study and any risk
of harm it may bring. In the case of a sample survey, physical harm is not possible, but
there could be emotional harm (like asking a rape victim questions about rape). The
subjects should be told what kinds of questions the survey will ask and about how
much of their time it will take. Experimenters must tell subjects the nature and pur-
pose of the study and outline possible risks. Subjects must then consent in writing.
 6. 3 Data Ethics 295

Example 6.18 Who can consent?

Are there some subjects who can’t give informed consent? It was once common, for
example, to test new vaccines on prison inmates who gave their consent in return
for good-behavior credit. Now we worry that prisoners are not really free to refuse,
and the law forbids medical experiments in prisons.
Children can’t give fully informed consent, so the usual procedure is to ask
their parents. A study of new ways to teach reading is about to start at a local
elementary school, so the study team sends consent forms home to parents. Many
parents don’t return the forms. Can their children take part in the study because the
parents did not say “No,” or should we allow only children whose parents returned
the form and said “Yes”?
What about research into new medical treatments for people with mental dis-
orders? What about studies of new ways to help emergency room patients who may
be unconscious or have suffered a stroke? In most cases, there is not time even to
get the consent of the family. Does the principle of informed consent bar realis-
tic trials of new treatments for unconscious patients? These are questions without
clear answers. Reasonable people differ strongly on all of them. There is nothing
simple about informed consent.28

The difficulties of informed consent do not vanish even for capable subjects.
Some researchers, especially in medical trials, regard consent as a barrier to getting
patients to participate in research. They may not explain all possible risks; they may
not point out that there are other therapies that might be better than those being
studied; they may be too optimistic when talking with patients even when the con-
sent form has all the right details. On the other hand, mentioning every possible
risk leads to very long consent forms that really are barriers. “They are like rental
car contracts,” one lawyer said. Some subjects don’t read forms that run five or six
printed pages. Others are frightened by the large number of possible (but unlikely)
disasters that might happen and so refuse to participate. Of course, unlikely di-
sasters sometimes happen. When they do, lawsuits follow and the consent forms
become even longer and more detailed.

Confidentiality
Ethical problems do not disappear once a study has been cleared by the review
board, has obtained consent from its subjects, and has actually collected data about
the subjects. It is important to protect the subjects’ privacy by keeping all data about
individuals confidential. The report of an opinion poll may say what percent of the
1500 respondents felt that legal immigration should be reduced. It may not report
what you said about this or any other issue.
Confidentiality is not the same as anonymity. Anonymity means that subjects
are anonymous—their names are not known even to the director of the study. Ano-
nymity is rare in statistical studies. Even where anonymity is possible (mainly in
surveys conducted by mail), it prevents any follow-up to improve nonresponse or
inform subjects of results.
296 CHAPTER 6 Designing Experiments

Any breach of confidentiality is a serious viola-

tion of data ethics. The best practice is to separate the
identity of the subjects from the rest of the data at
once. Sample surveys, for example, use the identifica-
tion only to check on who did or did not respond. In
an era of advanced technology, however, it is no lon-
ger enough to be sure that each individual set of data
protects people’s privacy.
The government, for example, maintains a vast
amount of information about citizens in many
separate data bases—census responses, tax returns,
Social Security information, data from surveys such
as the Current Population Survey, and so on. Many
“I realize the participants in this study are to be of these data bases can be searched by computers
anonymous, but you're going to have to expose your eyes.” for statistical studies. A clever computer search of
several data bases might be able, by combining in-
formation, to identify you and learn a great deal about you even if your name
and other identification have been removed from the data available for search.
Privacy and confidentiality of data are hot issues among statisticians in the
computer age.

Exercises Most of the exercises in this section pose issues for discussion. There are few right or
wrong answers, but there are more and less thoughtful answers.
6.53 Opinion polls The presidential election campaign is in full swing, and
the candidates have hired polling organizations to take regular polls to find out
what the voters think about the issues. What information should the pollsters
be required to give out?
(a) What does the standard of informed consent require the pollsters to
tell potential respondents?
(b) The standards accepted by polling organizations also require giving
respondents the name and address of the organization that carries out
the poll. Why do you think this is required?
(c) The polling organization usually has a professional name such as
“Samples Incorporated,” so respondents don’t know that the poll is being
paid for by a political party or candidate. Would revealing the sponsor
to respondents bias the poll? Should the sponsor always be announced
s

whenever poll results are made public?

aturePic

6.54 Minimal risk? You have been invited to serve on a college’s institutional
ilder/Fe

review board. You must decide whether several research proposals qualify
David G

for lighter review because they involve only minimal risk to subjects. Federal
regulations say that “minimal risk” means the risks are no greater than “those
ordinarily encountered in daily life or during the performance of routine
physical or psychological examinations or tests.” That’s vague. Which of these do
you think qualifies as “minimal risk”?
 6. 3 Data Ethics 297

(a) Draw a drop of blood by pricking a finger in order to measure blood sugar.
(b) Draw blood from the arm for a full set of blood tests.
(c) Insert a tube that remains in the arm, so that blood can be drawn regularly.
6.55 Is consent needed? In which of the circumstances below would you allow
collecting personal information without the subjects’ consent?
(a) A government agency takes a random sample of income tax returns to obtain
information on the average income of people in different occupations. Only the
incomes and occupations are recorded from the returns, not the names.
(b) A social psychologist attends public meetings of a religious group to study the
behavior patterns of members.
(c) The social psychologist pretends to be converted to membership in a religious
group and attends private meetings to study the behavior patterns of members.
6.56 Informed consent A researcher suspects that traditional religious
beliefs tend to be associated with an authoritarian personality. She prepares
a questionnaire that measures authoritarian tendencies and also asks many
religious questions. Write a description of the purpose of this research to be
read by subjects in order to obtain their informed consent. You must balance the
conflicting goals of not deceiving the subjects as to what the questionnaire will tell
about them and of not biasing the sample by scaring off religious people.
6.57 Anonymous or confidential? Texas A&M, like many universities, offers
free screening for HIV, the virus that causes AIDS. The announcement says,
“Persons who sign up for the HIV Screening will be assigned a number so that
they do not have to give their name.” They can learn the results of the test by
telephone, still without giving their name. Does this practice offer anonymity or
just confidentiality?
6.58 Not really anonymous Some common practices may appear to offer
anonymity while actually delivering only confidentiality. Market researchers
often use mail surveys that do not ask the respondent’s identity but contain
hidden codes on the questionnaire that identify the respondent. A false claim
of anonymity is clearly unethical. If only confidentiality is promised, is it also
unethical to say nothing about the identifying code, perhaps causing respondents
to believe their replies are anonymous?

Clinical trials
Clinical trials are experiments that study the effectiveness of medical treatments
on actual patients. Medical treatments can harm as well as heal, so clinical trials
spotlight the ethical problems of experiments with human subjects. Here are the
starting points for a discussion:
• Randomized comparative experiments are the only way to see the true
effects of new treatments. Without them, risky treatments that are no better
than placebos will become common.
• Clinical trials produce great benefits, but most of these benefits go to future
patients. The trials also pose risks, and these risks are borne by the subjects
of the trial. So we must balance future benefits against present risks.
298 CHAPTER 6 Designing Experiments

• Both medical ethics and international human rights standards say that “the
interests of the subject must always prevail over the interests of science and
society.”
The quoted words are from the 1964 Helsinki Declaration of the World
Medical Association, the most respected international standard. The most out-
rageous examples of unethical experiments are those that ignore the interests of
the subjects.

Example 6.19 The Tuskegee syphilis study

In the 1930s, syphilis was common among black men in the rural South, a group
that had almost no access to medical care. The Public Health Service recruited
Statisticians, honest 399 poor black sharecroppers with syphilis and 201 others without the disease in
and dishonest order to observe how syphilis progressed when no treatment was given. Begin-
Developed nations rely on ning in 1943, penicillin became available to treat syphilis. The study subjects were
government statisticians to not treated. In fact, the Public Health Service prevented any treatment until word
produce honest data. We leaked out and forced an end to the study in the 1970s.
trust the monthly unem- The Tuskegee study is an extreme example of investigators following their own
ployment rate, for example,
to guide both public and
interests and ignoring the well-being of their subjects. A 1996 review said, “It has
private decisions. Honesty come to symbolize racism in medicine, ethical misconduct in human research,
can’t be taken for granted paternalism by physicians, and government abuse of vulnerable people.”29 In 1997,
everywhere, however. In President Clinton formally apologized to the surviving participants in a White
1998, the Russian govern- House ceremony.
ment arrested the top
For additional details of the Tuskegee syphilis study, visit www.cdc.gov/tuskegee/.
statisticians in the State www

Committee for Statistics.

They were accused of tak-
ing bribes to fudge data to
help companies avoid taxes. Because “the interests of the subject must always prevail,” medical treatments
“It means that we know can be tested in clinical trials only when there is reason to hope that they will help
nothing about the perfor-
mance of Russian compa- the patients who are subjects in the trials. Future benefits aren’t enough to justify
nies,” said one newspaper experiments with human subjects. Of course, if there is already strong evidence
editor. that a treatment works and is safe, it is unethical not to give it. Here are the words of
Dr. Charles Hennekens of Harvard Medical School, who directed the large clinical
trial that showed that aspirin reduces the risk of heart attacks:

There’s a delicate balance between when to do or not do a randomized trial. On

the one hand, there must be sufficient belief in the agent’s potential to justify
exposing half the subjects to it. On the other hand, there must be sufficient doubt
about its efficacy to justify withholding it from the other half of subjects who
might be assigned to placebos.30

Why is it ethical to give a control group of patients a placebo? Well, we know that
placebos often work. What is more, placebos have no harmful side effects. So in
the state of balanced doubt described by Dr. Hennekens, the placebo group may
be getting a better treatment than the drug group. If we knew which treatment was
better, we would give it to everyone. When we don’t know, it is ethical to try both
and compare them.
 6. 3 Data Ethics 299

Behavioral and social science experiments

When we move from medicine to the behavioral and social sciences, the direct
risks to experimental subjects are less acute, but so are the possible benefits to the
subjects. Consider, for example, some experiments conducted by psychologists in
their study of human behavior.

Example 6.20 Group pressure

Stanley Milgram of Yale conducted a famous experiment “to see if a person will
perform acts under group pressure that he would not have performed in the ab-
sence of social inducement.”31
The subject arrives with three others who (unknown to the subject) are confed-
erates of the experimenter. The experimenter explains that he is studying the effects
of punishment on learning. The Learner (one of the confederates) is strapped into
an electric chair, mentioning in passing that he has a mild heart condition. The
subject and two other confederates are Teachers. They sit in front of a panel with
switches with labels ranging from “Slight Shock” to “Danger: Severe Shock” and are
told that they must shock the Learner whenever he fails in a memory learning task.
How badly the Learner is shocked is up to the Teachers and will be the lowest level
suggested by any Teacher on that trial.
All is rigged. The Learner answers incorrectly on 30 of the 40 trials. The two
Teacher confederates call for a higher shock level at each failure. As the shock in-
creases, the Learner protests, asks to be let out, shouts, complains of heart trouble,
and finally screams in pain. (The “shocks” are phony and the Learner is an actor,
but the subject doesn’t know this.) What will the subject do? He can keep the shock
at the lowest level, but the two other Teachers are pressuring him to punish the
Learner more for each failure.
The subjects usually give in to the pressure. “While the experiment yields wide
variation in performance, a substantial number of subjects submitted readily to
pressure applied to them by the confederates.” Milgram noted that, in questioning
after the experiment, the subjects often admitted that they had acted against their
own principles and were upset by what they had done. “The subject was then de-
hoaxed carefully and had a friendly reconciliation with the victim.”32
www
You can find more information about Stanley Milgram’s research, including the
infamous shock therapy experiment, at www.stanleymilgram.com.

This experiment was acceptable in the early 1960s, when Milgram did his work.
Psychologists felt it gave insight into the power of group pressure to coerce indi-
viduals into acts they would not otherwise commit. Now, however, such a study
would be considered clearly unethical.
Milgram’s experiment illustrates the difficulties facing those who plan and re-
view behavioral studies.
• There is no risk of physical harm to the subjects, but they would certainly
risk embarrassment and lingering emotional effects. What should we
protect subjects from when physical harm is unlikely? Possible emotional
harm? Undignified situations? Invasion of privacy?
300 CHAPTER 6 Designing Experiments

• What about informed consent? Many behavioral

experiments rely on hiding the true purpose of the
study. Some, like Milgram’s, actively deceive subjects.
Since the subjects might change their behavior if
told in advance what the investigators are looking
for, they are asked to consent on the basis of vague
information. They receive full information only after
the experiment.
The “Ethical Principles” of the American Psycho-
logical Association require consent unless a study merely
observes behavior in a public place. They allow decep-
tion only when it is necessary to the study, does not hide
information that might influence a subject’s willingness
“I’m doing a little study on the effects of emotional to participate, and is explained to subjects as soon as pos-
stress. Now, just take the axe from my assistant.”
sible. Milgram’s study (from the 1960s) does not meet
current ethical standards because subjects did not give informed consent to take
part in an experiment.
We see that the basic requirement for informed consent is understood differ-
ently in medicine and psychology. Here is an example of another setting with yet
another interpretation of what is ethical. The subjects get no information and give
no consent. They don’t even know that an experiment may be sending them to jail
for the night.

Example 6.21 Domestic violence

How should police respond to domestic-violence calls? In the past, the usual
practice was to remove the offender and order him to stay out of the house-
hold overnight. Police were reluctant to make arrests because the victims rarely
pressed charges. Women’s groups argued that arresting offenders would help
prevent future violence even if no charges were filed. Is there evidence that ar-
rest will reduce future offenses? That’s a question that experiments have tried to
answer.
A typical domestic-violence experiment compares two treatments: arrest the
suspect and hold him overnight, or warn the suspect and release him. When police
officers reach the scene of a domestic-violence call, they calm the participants and
investigate. Weapons or death threats require an arrest. If the facts permit an arrest
but do not require it, an officer radios headquarters for instructions. The person
on duty opens the next envelope in a file prepared in advance by a statistician.
The envelopes contain the treatments in random order. The police either arrest the
suspect or warn and release him, depending on the contents of the envelope. The
researchers then monitor police records and visit the victim to see if the domestic
violence reoccurs.
The first such experiment appeared to show that arresting domestic-violence
suspects does reduce their future violent behavior. As a result of this evidence,
arrest has become the common police response to domestic violence.
 6. 3 Data Ethics 301

The domestic-violence experiments shed light on an important issue of public

policy. Because there is no informed consent, the ethical rules that govern clinical
trials and most social science studies would forbid these experiments. They were
cleared by review boards because, in the words of one domestic-violence researcher,
“These people became subjects by committing acts that allow the police to arrest
them. You don’t need consent to arrest someone.”

Exercises 6.59 Sham surgery? Clinical trials like the Parkinson’s disease study cited at the
beginning of this section are becoming more common. One medical researcher
says, “This is just the beginning. Tomorrow, if you have a new procedure, you
will have to do a double-blind placebo trial.”33 Arguments for and against testing
surgery just as drugs are tested are given on page 293. When would you allow
sham surgery in a clinical trial of a new surgery?
6.60 AIDS clinical trials Now that effective treatments for AIDS are
at last available, is it ethical to test treatments that may be less effective?
Combinations of several powerful drugs reduce the level of the virus in the
blood and at least delay illness and death from AIDS. But effectiveness depends
on how damaged the patient’s immune system is and what drugs he or she
has previously taken. There are strong side effects, and patients must be able
to take more than a dozen pills on time every day. Because AIDS is often fatal
and the combination therapy works, we might argue that it isn’t ethical to test
any treatment for AIDS that doesn’t give the full combination. But not doing
further tests might prevent discovery of better treatments. This is a strong
example of the conflict between doing the best we know for patients now and
finding better treatments for other patients in the future. How can we ethically
test new drugs for AIDS?34
6.61 Equal treatment Researchers on aging proposed to investigate the effect
of supplemental health services on the quality of life of older people. Eligible
patients on the rolls of a large medical clinic were to be randomly assigned to
treatment and control groups. The treatment group would be offered hearing
aids, dentures, transportation, and other services not available without charge
to the control group. The review board felt that providing these services to some
but not other persons in the same institution raised ethical questions. Do you
agree?
6.62 Placebos At present there is no vaccine for a serious viral disease. A
vaccine is developed and appears effective in animal trials. Only a comparative
experiment with human subjects in which a control group receives a placebo can
determine the true worth of the vaccine. Is it ethical to give some subjects the
placebo, which cannot protect them against the disease?35
6.63 Tempting subjects A psychologist conducts the following experiment:
she measures the attitude of subjects toward cheating, then has them play a
game rigged so that winning without cheating is impossible. The computer that
organizes the game also records—unknown to the subjects—whether or not they
cheat. Then attitude toward cheating is retested.
302 CHAPTER 6 Designing Experiments

Subjects who cheat tend to change their attitudes to find cheating more ac-
ceptable. Those who resist the temptation to cheat tend to condemn cheating more
strongly on the second test of attitude. These results confirm the psychologist’s theory.
This experiment tempts subjects to cheat. The subjects are led to believe that
they can cheat secretly when in fact they are observed. Is this experiment ethically
objectionable? Explain your position.
6.64 Students as subjects Students taking Psychology 001 are required to
serve as experimental subjects. Students in Psychology 002 are not required to
serve, but they are given extra credit if they do so. Students in Psychology 003
are required either to sign up as subjects or to write a term paper. Serving as an
experimental subject may be educational, but current ethical standards frown on
using “dependent subjects” such as prisoners or charity medical patients. Students
are certainly somewhat dependent on their teachers. Do you object to any of these
course policies? If so, which ones, and why?

ws
Ne
ws
Ne

APPLICATION 6.3 B]
RO
g
<
Se
a

Hope for sale?

We have pointed to the ethical problems of ex- standard treatments rather than to BMT. Results
periments with human subjects, clinical trials in from such trials were delayed for years by the dif-
particular. Not doing proper experiments can also ficulty in recruiting subjects. Of the first five trials
pose problems. Here is an example. Women with reported, four found no significant difference
advanced breast cancer will eventually die. A prom- between BMT and standard treatments. The fifth
ising but untried treatment appears. Should we wait favored BMT—but the researcher soon admitted
for controlled clinical trials to show that it works, “a serious breach of scientific honesty and integ-
or should we make it available right now? rity.” The New York Times put it more bluntly: “he
The promising treatment is “bone marrow falsified data.”36
transplant” (BMT for short). The idea of BMT is
to harvest a patient’s bone marrow cells, blast the QUESTIONS
cancer with very high doses of drugs, then return 1. Outline the design of an experiment that
the harvested cells to keep the drugs from killing compares BMT to an existing treatment.
the patient. BMT has become popular. It is painful, 2. Describe how you would incorporate each of the
expensive, and dangerous. three basic principles of data ethics in your design
New anticancer drugs are first available from Question 1.
through clinical trials, but there is no constraint
on therapies such as BMT. When small, un- 3. Explain what “no significant difference between
controlled trials seemed to show success, BMT BMT and standard treatments” means.
became widely available. But does BMT keep 4. Compassion seems to support making untested
patients alive longer than standard treatments? We treatments available to dying patients. Reason
don’t know, but the answer appears to be “prob- responds that this approach gives patients false
ably not.” The patients naturally want to try any- hope and delays development of treatments that
thing that might keep them alive. Patients would really work. Which position do you support?
not join controlled trials that might assign them to Why?
 6. 3 Data Ethics 303

Section 6.3 Summary

Ordinary honesty says you shouldn’t make up data or claim to be taking a survey
when you are really selling security alarms. Data ethics begin with some principles
that go beyond just being honest. Studies with human subjects must be screened in
advance by an institutional review board. All subjects must give their informed
consent before taking part. All information about individual subjects must be kept
confidential. These principles are a good start, but many ethical debates remain,
especially in the area of experiments with human subjects. Many of the debates
concern the right balance between the welfare of the subjects and the future ben-
efits of the experiment. Remember that randomized comparative experiments can
answer questions that can’t be answered without them. Also remember that “the in-
terests of the subject must always prevail over the interests of science and society.”

Section 6.3 6.65 Surveys of youth A survey asked teenagers whether they had ever con-
Exercises sumed an alcoholic beverage. Those who said “Yes” were then asked,
How old were you when you first consumed an alcoholic beverage?
Should consent of parents be required to ask minors about alcohol, drugs, and
other such issues, or is consent of the minors themselves enough? Give reasons
for your opinion.
6.66 Who serves on the review board? Government regulations require that
institutional review boards consist of at least five people, including at least one sci-
entist, one nonscientist, and one person from outside the institution. Most boards
are larger, but many contain just one outsider.
(a) Why should review boards contain people who are not scientists?
(b) Do you think that one outside member is enough? How would you choose
that member? (For example, would you prefer a medical doctor? A member of the
clergy? An activist for patients’ rights?)
6.67 Animal welfare Many people are concerned about the ethics of experimen-
tation with living animals. Some go as far as to regard any animal experiments as
unethical, regardless of the benefits to human beings. Briefly explain your position
on each of the following uses of animal subjects.
(a) Military doctors use goats that have been deliberately shot (while completely
anesthetized) to study and teach the treatment of combat wounds. Assume that
there is no equally effective way to prepare doctors to treat human wounds.
(b) Several states are considering legislation that would end the practice of using
cats and dogs from pounds in medical research. Instead, the animals will be killed
at the pounds.
(c) The cancer-causing potential of chemicals is assessed by exposing lab rats
to high concentrations. The rats are bred for this specific purpose. (Would your
opinion differ if dogs or monkeys were used?)
6.68 Informed consent The information given to potential subjects in a clini-
cal trial before asking them to decide whether or not to participate might include
any of the following. Do you feel that all of this information is ethically required?
Discuss.
304 CHAPTER 6 Designing Experiments

(a) The basic statement that an experiment is being conducted; that is, something
beyond simply treating your medical problem occurs in your therapy.
(b) A statement of any potential risks from any of the experimental treatments.
(c) An explanation that random assignment will be used to decide which treat-
ment you get.
(d) An explanation that one “treatment” is a placebo and a statement of the prob-
ability that you will receive the placebo.
6.69 AIDS trials in Africa Effective drugs for treating AIDS are very expensive,
so most African nations cannot afford to give them to large numbers of people.
Yet AIDS is more common in parts of Africa than anywhere else. Several clinical
trials are looking at ways to prevent pregnant mothers infected with HIV from
passing the infection to their unborn children, a major source of HIV infections
in Africa. Some people say these trials are unethical because they do not give ef-
fective AIDS drugs to their subjects, as would be required in rich nations. Others
reply that the trials are looking for treatments that can work in the real world in
Africa and that they promise benefits at least to the children of their subjects.
What do you think?
6.70 Anonymous or confidential? One of the most important nongovernment
surveys in the United States is the General Social Survey (Example 1.7, page 9).
The GSS regularly monitors public opinion on a wide variety of political and
social issues. Interviews are conducted in person in the subject’s home. Are a
subject’s responses to GSS questions anonymous, confidential, or both? Explain
your answer.
6.71 Human biological materials Long ago, doctors drew a blood specimen
from you as part of treating minor anemia. Unknown to you, the sample was
stored. Now researchers plan to use stored samples from you and many other
people to look for genetic factors that may influence anemia. It is no longer possi-
ble to ask your consent. Modern technology can read your entire genetic makeup
from the blood sample.
(a) Do you think it violates the principle of informed consent to use your blood
sample if your name is on it but you were not told that it might be saved and
studied later?
(b) Suppose that your identity is not attached. The blood sample is known only to
come from (say) “a 20-year-old white female being treated for anemia.” Is it now
OK to use the sample for research?
(c) Perhaps we should use biological materials such as blood samples only
from patients who have agreed to allow the material to be stored for later use in
research. It isn’t possible to say in advance what kind of research, so this falls short
of the usual standard for informed consent. Is it nonetheless acceptable, given
complete confidentiality and the fact that using the sample can’t physically harm
the patient?
6.72 Deceiving subjects Students sign up to be subjects in a psychology
experiment. When they arrive, they are told that interviews are running late
and are taken to a waiting room. The experimenters then stage a theft of a valu-
able object left in the waiting room. Some subjects are alone with the thief, and
others are in pairs—these are the treatments being compared. Will the subject
report the theft?
 Chapter 6 Review 305

The students had agreed to take part in an unspecified study, and the true nature of
the experiment is explained to them afterward. Do you think this study is ethically OK?

CHAPTER 6 REVIEW
Sections 6.1 and 6.2 dealt with the statistical aspects of designing experiments,
studies that impose some treatment in order to learn about the response. The
big idea is the randomized comparative experiment. Figure 6.6 outlines the
simplest design.

Group 1
Treatment 1
n1 subjects

Random Compare
allocation response

Group 2
n2 subjects Treatment 2

Figure 6.6 The idea of a randomized comparative experiment.

If the subjects in an experiment differ in some systematic way that may affect
their responses to the treatments, blocking can be used to reduce unwanted vari-
ability. A matched pairs design, in which each subject serves as his or her own
control, is a special form of blocking.
Well-designed experiments include random allocation of subjects to treat-
ments, comparison, and use of enough subjects. To reduce the potential for bias,
many designs are double-blind—neither the subjects nor those working with them
know who is receiving which treatment. Confounding can blur the relationship
between the explanatory and response variables and must be actively prevented.
When we collect data about people, ethical issues can be important. Section
6.3 discussed these issues and introduced three principles that apply to any study
with human subjects.

Here is a review list of the most important skills you should have developed
from your study of this chapter.
1. Identify the explanatory variables, treatments, response variables, and subjects
in an experiment.
2. Recognize bias due to confounding of explanatory variables with lurking vari-
ables in either an observational study or an experiment.
3. Outline the design of a completely randomized experiment using a diagram
like that in Figure 6.6. Such a diagram should show the sizes of the groups, the
specific treatments, and the response variable.
4. Use Table B of random digits to carry out random assignment of subjects to
groups in a completely randomized experiment.
5. Make use of matched pairs or other block designs when appropriate.
306 CHAPTER 6 Designing Experiments

6. Recognize the placebo effect. Recognize when the double-blind technique

should be used. Be aware of weaknesses in an experiment, especially in the ability to
generalize its conclusions.
7. Explain why a randomized comparative experiment can give good evidence for
cause-and-effect relationships.
8. Understand key concepts of data ethics: informed consent, confidentiality of
personal data, and review of ethical considerations by an institutional review
board. Discuss how these apply in specific settings.

© 2008 Creators Syndicate, Inc. www.creators.com

CHAPTER 6 REVIEW EXERCISES

6.73 Testing a new drug A drug manufacturer is studying how a new drug behaves
in patients. Investigators compare 2 doses: 5 milligrams (mg) and 10 mg. The drug
can be administered by injection, by a skin patch, or by intravenous drip. Concentra-
tion in the blood after 30 minutes (the response variable) may depend both on the
dose and on the method of administration.
(a) Make a sketch that describes the treatments formed by combining dose and
method. Then use a diagram to outline a completely randomized design for this
experiment.
(b) “How many subjects?” is a tough issue. What is the advantage of using larger
groups of subjects?
(c) The drug may behave differently in men and women. How would you modify your
experimental design from (a) to take this into account?
6.74 Should I take the class? A college allows students to choose either classroom or
self-paced instruction in a basic economics course. The college wants to compare the
effectiveness of self-paced and regular instruction. A professor proposes giving the
same final exam to all students in both versions of the course and comparing the aver-
age score of those who took the self-paced option with the average score of students
in the regular sections.
(a) Explain why confounding would make the results of this study worthless.
(b) Given 30 students who are willing to use either regular or self-paced instruction,
outline an experimental design to compare the two methods of instruction. Then use
Table B, starting at line 108, to carry out the randomization.
6.75 Surgery or not? To compare surgical and nonsurgical treatments of intestinal
cancer, researchers examined the records of a large number of patients. Patients who
received surgery survived much longer (on the average) than patients who were treated
 Chapter 6 Review Exercises 307

without surgery. The study concludes that surgery is more effective than nonsurgical
treatment.
(a) What are the explanatory and response variables in this study?
(b) Is this study an experiment? Why or why not?
(c) Discussion with medical experts reveals that surgery is reserved for relatively
healthy patients; patients who are too ill to tolerate surgery receive nonsurgical
treatment. Explain why the conclusion of the study is not supported by the
data.
(d) Outline the design of a randomized comparative experiment.
6.76 Reducing health care spending Will people spend less on health care if
their health insurance requires them to pay some part of the cost themselves? An
experiment on this issue asked if the percent of medical costs paid by health insur-
ance has an effect either on the amount of medical care that people use or on their
health. The treatments were four insurance plans. Each plan paid all medical costs
above a ceiling. Below the ceiling, the plans paid 100%, 75%, 50%, or 0% of costs
incurred.37
(a) Outline the design of a randomized comparative experiment suitable for this
study.
(b) Briefly describe the practical and ethical difficulties that might arise in such an
experiment.
6.77 Do the twist The design of controls and instruments has a large effect on
how easily people can use them. A student investigates this effect by asking right-
handed students to turn a knob (with their right hands) that moves an indicator by
screw action. There are two identical instruments, one with a right-hand thread
(the knob turns clockwise) and the other with a left-hand thread (the knob must
be turned counterclockwise). The response variable is the time required (in
seconds) to move the indicator a fixed distance. Thirty right-handed students are
available to serve as subjects. You have been asked to lay out the statistical design
of this experiment. Describe your design, and carry out any randomization that is
required.
Exercises 6.78 to 6.81 are based on an article in the Journal of the American Medical
Association that asks if flu vaccine works.38 The article reports on a study of the
effectiveness of a nasal spray vaccine called “trivalent LAIV.” Here is part of the article’s
summary:
Design: Randomized, double-blind, placebo-controlled trial.
Participants: A total of 4561 healthy, working adults aged 18 to 64 years recruited
through health insurance plans, at work sites, and from the general population.
Intervention: Participants were randomized 2:1 to receive intranasally administered
trivalent LAIV vaccine (n = 3041) or placebo (n = 1520).
Results: Vaccination also led to fewer days of work lost (17.9% reduction for severe
febrile illnesses; 28.4% reduction for febrile upper-respiratory-tract illnesses) and
fewer days with health-care-provider visits (24.8% reduction for severe febrile
illnesses; 40.9% reduction for febrile upper-respiratory-tract illnesses).
6.78 Know these terms Explain in one sentence each what “randomized,”
“double-blind,” and “placebo-controlled” mean in the description of the design
of this study.
308 CHAPTER 6 Designing Experiments

6.79 Experiment basics Identify the subjects, the explanatory variable, and several
response variables in this study.
6.80 Design an experiment Use a diagram to outline the design of the experiment in
this medical study.
6.81 Ethics What are the three “first principles” of data ethics? Explain briefly what
the flu vaccine study must do to apply each of these principles.
6.82 Prison experiments The decision to ban medical experiments on federal pris-
oners followed the uncovering of experiments in the 1960s that exposed prisoners
to serious harm. But experiments such as the vitamin C test of Example 1.9 (page 13)
are also banned from federal prisons. Because of the difficulty of obtaining truly
voluntary consent in a prison, is it necessary to ban even experiments in which
all treatments appear harmless? What is your overall opinion of this ban on
experimentation?
 PA RT C

CHANCE

Reproduced by permission of Bob Schochet.

Chance is all around
us.You and your friend
play rock-paper-scissors
to determine who gets
the remaining slice
of pizza. A coin toss
decides which team
gets to receive the “What kind of childish nonsense are you working on now?”
ball first in a football
game. Many adults regularly play the lottery, hoping to win a big jackpot with
a few lucky numbers. Others head to casinos or racetracks, hoping that some
combination of luck and skill will pay off. People young and old play games of
chance involving cards or dice or spinners. The traits that children inherit—
gender, hair color, eyesight, blood type, handedness, dimples, whether they can
roll their tongue—are determined by the chance involved in which genes get
passed along by their parents.
A roll of dice, a simple random sample, and even the inheritance of eye color
or blood type represent chance behavior that we can understand and work with.
We can roll the dice again, and again, and again. The outcomes are governed
by chance, but in many repetitions a pattern emerges. We use mathematics to
understand the regular patterns of chance behavior when we can repeat the
same chance phenomenon again and again.
The mathematics of chance is called probability. Probability is the topic of
this part of the book.

309
This page intentionally left blank
 CHAPTER 7

Probability: What
Are the Chances?
7.1 Randomness, Probability, and Simulation
7 .2 Probability Rules
7.3 Conditional Probability and Independence

AP Photo/Mark Humphrey
Playing the lottery
Many people play the lottery in hopes of striking it rich.
They pick a few “lucky” numbers, buy a lottery ticket, and wait.
The winning numbers are chosen using a chance process.
Picture a bunch of numbered balls bubbling about in a
container and then being randomly popped out one at a time by air pressure.
Even though most people don’t win, they view the lottery as “fair” because
chance decides the winner. Once in a while, the lottery isn’t as fair as it seems.
Several years ago, the Pennsylvania lottery was rigged by the smiling host and
several stagehands. They injected paint into all balls bearing 8 of the 10 digits. This
weighed them down and guaranteed that all three balls for the winning number
would have the remaining 2 digits. The perpetrators then bet on all combinations of
these digits. When 6-6-6 popped out, they won $1.2 million. Yes, they were caught.
To avoid similar incidents, some states switched from live drawings using
numbered balls to computer-generated random numbers. The state of Tennessee
did just that. They hired one company to program the random number generator
and a second company to certify that the program was working properly. In 2007,
a problem emerged with their Cash 3 drawing. For several consecutive weeks,
the computer generated no repeated digits among the three winning numbers.
Any player who had picked numbers with repeated digits was guaranteed to lose.
Investigation revealed that the computer programmer entered one incorrect
character that told the computer not to repeat digits. The mistake was fixed, but
public trust was lost.
In 2008, the Cash 3 computer program spat out a surprising set of winning
numbers over a three-day period. On Thursday, the winning number was 0-7-7.
On Friday, the winning number was 7-0-7. On Saturday, the winning number was
again 0-7-7. Lottery officials had the computer equipment checked again. Their
conclusion: “The repeating numbers are a statistical coincidence.” With a little 311
312 CHAPTER 7 Probability: What Are the Chances?

deeper understanding of random numbers, simulation, and probability, you can

decide for yourself whether the lottery was conducted fairly.

7.1 Randomness, Probability, and Simulation

ACTIVITY 7.1A
The Probability applet
AP
PLET Go to the textbook Web site APPLET TIP: You can toss the coin a maximum of
www.whfreeman.com/sta2e, click on “Statistical 40 times at once. To make longer sequences, toss
Applets,” and select the Probability applet. again without resetting.
Directions When you click the Toss button, the
Verify
applet will toss a coin the number of times you
Use the applet to test your answers to
specify. You can set the probability that the coin
questions 1–3. Did the proportion of heads
lands heads by entering a new value and clicking the
behave as you predicted?
Reset button. A sample screen shot from the applet
is shown. Before you begin using the applet, try to
Experiment
answer the following three questions.
4. Set the probability of heads at 0.3 and have
Think the applet repeatedly toss the coin until you
1. If you toss the coin just once, what are the accumulate several hundred tosses. How does
possible values for the proportion of heads? the behavior of the proportion of heads at the
2. If you toss the coin 400 times with probability beginning of the plot differ from that at the end of
0.5 for heads, which is more likely to happen: (a) the plot?
getting exactly 200 heads or (b) getting within a few 5. Check the “Show true probability” box to have
percent of 200 heads? the applet draw a horizontal line on the plot at 0.3.
3. What will the plot of the proportion of heads How does the behavior of the plot of the proportion
look like as you accumulate more and more tosses of heads relate to this line?
with a probability of 0.7 for heads?

# Heads = 7/10 = 0.70

1.0

Proportion
heads

0.5

0
1 10
# Tails = 3/10 = 0.30
Tosses

Probability of heads = 0.5 Reset

Show true probability
Toss 10 times Toss
 7. 1 Randomness, Probability, and Simulation 313

6. Change the probability of heads to 0.7 and have heads compare with the plot when the probability
the applet toss the coin several hundred times was 0.3? In what ways is it similar? In what ways
again. How does the plot of the proportion of does it differ?

Probability: what happens in the long run

In football, a coin toss helps determine which team gets the ball first. Why do the
rules of football require a coin toss? Because tossing a coin seems a “fair” way to
decide. That is, tossing a coin avoids favoritism. Favoritism is as undesirable in
choosing people for a sample survey or in assigning patients to treatment and
placebo groups in a medical experiment as it is in awarding first possession in
a football game. That’s why statisticians recommend random samples and ran-
domized experiments. They avoid favoritism by letting chance decide who gets
selected or who receives which treatment. A big fact emerges when we watch coin
tosses or the results of random samples closely: chance behavior is unpredictable
in the short run but has a regular and predictable pattern in the long run.
Toss a coin, or choose a simple random sample. The result can’t be predicted in
advance, because the result will vary when you toss the coin or choose the sample repeat-
edly. But there is still a regular pattern in the results, a pattern that emerges clearly only
after many repetitions. This remarkable fact is the basis for the idea of probability.

Example 7.1 Coin tossing

When you toss a coin, there are only two possible outcomes, heads or tails. Figure 7.1
shows the results of tossing a coin 1000 times. For each number of tosses from 1 to
1000, we have plotted the proportion of those tosses that gave a head. The first toss was a
head, so the proportion of heads starts at 1/1  1.0. The second toss was a tail, reducing
the proportion of heads to 1/2  0.5 after two tosses. The next three tosses produced a
tail followed by two heads, so the proportion of heads after five tosses is 3/5  0.6.

1.0

0.8
Proportion of heads

0.6
probability = 0.5

Figure 7.1 Toss a coin

0.4
many times. The propor-
tion of heads changes as
we make more tosses but 0.2
eventually gets very close
to 0.5. This is what we
0.0
mean when we say, “The
1 5 10 50 100 500 1000
probability of a head is
Number of tosses
one-half.”
314 CHAPTER 7 Probability: What Are the Chances?

The proportion of tosses that produce heads is quite variable at first, but it set-
tles down as we make more and more tosses. Eventually this proportion gets close
to 0.5 and stays there. We say that 0.5 is the probability of a head. The probability
0.5 appears as a horizontal line on the graph.

“Random” in statistics is not a synonym for “haphazard.” It is a description of

a kind of order that emerges only in the long run. We encounter the unpredictable
side of randomness in our everyday lives. But we rarely see enough repetitions of
the same random phenomenon to observe the long-term regularity that probability
describes. You can see that regularity emerging in Figure 7.1. In the very long run,
the proportion of tosses that give a head is 0.5. Probability 0.5 means “occurs half
the time in a very large number of trials.” This is the intuitive idea of probability.

ACTIVITY 7.1B
Tossing thumbtacks
Materials: Box of thumbtacks (enough for one per 2. Toss your thumbtack 50 times onto a flat surface.
student) Tally the number of point up (U) and point down
When you flip a fair coin, it is equally likely to land (D) landings. Calculate the proportion of point up
“heads” or “tails.” Do thumbtacks behave in the (U) landings you obtained.
same way? In this Activity, you will toss a thumbtack 3. What would you estimate as the probability of
several times and observe whether it comes to rest your thumbtack landing point up if you tossed it
point up (U) or point down (D). The question you one more time? Explain.
are trying to answer is: what is the probability that
the tossed thumbtack will land point up? 4. Compare your answer to Step 3 with those of
your classmates. Did everyone in the class come
1. Before you begin the Activity, make a guess about up with the same estimated probability? Give at
what will happen. If you could toss your thumbtack least two reasons why people’s probabilities might
over and over and over, what proportion of all tosses differ.
do you think would land point up (U)?

You might suspect that a coin has probability 0.5 of coming up heads just because
the coin has two sides. What about the thumbtacks of Activity 7.1B? They also have two
ways to land—point up or point down—but the chance that a tossed thumbtack lands
point up isn’t 0.5. How do we know that? From tossing a thumbtack over and over and
over again. Probability describes what happens in very many trials, and we must actu-
ally observe many tosses of a coin or thumbtack to pin down a probability. In the case of
tossing a coin, some diligent people have actually made thousands of tosses.

Example 7.2 Some remarkable coin tossers

The French naturalist Count Buffon (1707–1788) tossed a coin 4040 times. Result:
2048 heads, or proportion 2048/4040  0.5069 for heads.
Around 1900, the English statistician Karl Pearson heroically tossed a coin
24,000 times. Result: 12,012 heads, a proportion of 0.5005.
 7. 1 Randomness, Probability, and Simulation 315

While imprisoned by the Germans during World War II, the South African
mathematician John Kerrich tossed a coin 10,000 times. Result: 5067 heads, a pro-
portion of 0.5067.

Here’s a brief summary of what we have learned so far about randomness and
probability.

Randomness and probability

We call a phenomenon random if individual outcomes are uncertain

but there is nonetheless a regular distribution of outcomes in a large
number of repetitions.

The probability of any outcome of a random phenomenon is a number

between 0 and 1 that describes the proportion of times the outcome
would occur in a very long series of repetitions.

An outcome with probability 0 never occurs. An outcome with probability 1

happens on every repetition. An outcome with probability 0.5 happens half the time
in a very long series of trials. Of course, we can never observe a probability exactly.
We could always continue tossing the coin, for example. Mathematical probability
is based on imagining what would happen in an indefinitely long series of trials.
Probability gives us a language to describe the long-term regularity of random
behavior. The outcome of a coin toss and the sex of the next baby born in a local
hospital are both random. So is the outcome of a random sample or a randomized
experiment. The behavior of large groups of individuals is often as random as the
behavior of many coin tosses or many random samples. Life insurance, for example,
is based on the fact that deaths occur at random among many individuals.

Example 7.3 Probability and life insurance

We can’t predict whether a particular person will die in the next year. But if we
observe millions of people, deaths are random. The National Center for Health
Statistics says that the proportion of men aged 20 to 24 years who die in any one
year is 0.0015. This is the probability that a randomly selected young man will die
next year. For women that age, the probability of death is about 0.0005.
If an insurance company sells many policies to people aged 20 to 24, it knows
that it will have to pay off next year on about 0.15% of the policies sold to men and
on about 0.05% of the policies sold to women. It will charge more to insure a man
because the probability of having to pay is higher.

The history of chance

Randomness is most easily noticed in many repetitions of games of chance—rolling
dice, dealing shuffled cards, spinning a roulette wheel.1 The most common method of
randomization in ancient times was “rolling the bones,” which involved tossing several
316 CHAPTER 7 Probability: What Are the Chances?

astragali. The astragalus (Figure 7.2) is a solid bone from the heel of animals that, when
thrown, will come to rest on any of four sides. (The other two sides are rounded.)
Cubical dice, made of pottery or bone, came later, but even dice existed before 2000 b.c.

Figure 7.2 Animal heel

bones (astragali), shown
actual size. (From F. N.
David, Games, Gods, and
Gambling, Charles Griffin &
Company, 1962. Reproduced by
permission of the publishers.) Sheep Dog
Chance devices such as astragali have been used from the beginning of recorded
history. Yet none of the great ancient mathematicians studied the regular pattern
of many throws of bones or dice. Perhaps this is because astragali and most ancient
dice were so irregular that each had a different pattern of outcomes.
Professional gamblers, however, did notice the regular pattern of outcomes of
dice or cards and tried to adjust their bets to the odds of success. “How should I
bet?” is the question that launched mathematical probability. The systematic study
of randomness began when seventeenth-century French gamblers asked French
mathematicians for help in figuring out the “fair value” of bets on games of chance.
Probability theory, the mathematical study of randomness, originated with Pierre
de Fermat and Blaise Pascal in the seventeenth century and was well developed by
the time statisticians took it over in the twentieth century.

Exercises 7.1 Spinning a quarter With your forefinger, hold a new quarter (one with a
state featured on the reverse) upright, on its edge, on a hard surface. Then flick it
with your other forefinger so that it spins for some time before it falls and comes
to rest. Spin the coin a total of 25 times, and record the results.
(a) Based on 25 spins, estimate the probability of heads.
(b) Estimate the probability of tails.
7.2 How many tosses to get a head? When we toss a penny, experience shows
that the probability (long-term proportion) of a head is close to 1/2. Suppose that
we toss the penny repeatedly until we get a head. We want to know the probability
that the first head comes up in an odd number of tosses (1, 3, 5, and so on). To
find out, perform 20 trials, and keep a record of the number of tosses needed to
get a head on each of your 20 trials.
(a) From your results, estimate the probability of a head on the first toss. What
value should we expect this probability to have?
(b) Use your results to estimate the probability that the first head appears on an
odd-numbered toss.
7.3 Random digits The table of random digits (Table B) was produced by a
chance process that gives each digit a 1/10 chance of being a 0. Examine the first
200 digits in the table and determine what proportion are 0s. (This proportion is
 7. 1 Randomness, Probability, and Simulation 317

an estimate, based on 200 repetitions, of the true probability, which in this case is
known to be 0.1.)
If you want to find “true” random numbers, visit www.random.org. Check
www
out their coin flipper, die roller, playing card shuffler, and integer generator.
7.4 From words to probabilities Probability is a measure of how likely an event
is to occur. Match one of the probabilities that follow with each statement of
likelihood given. (The probability is usually a more exact measure of likelihood
than is the verbal statement.)

0 0.01 0.3 0.6 0.99 1

(a) This event is impossible. It can never occur.

(b) This event is certain. It will occur on every trial.
(c) This event is very unlikely, but it will occur once in a while in a long sequence of trials.
(d) This event will occur more often than not.
7.5 Rolling a die Imagine rolling a fair, six-sided die, like the kind used in many
board games. What is the probability of getting a 4? Justify your answer.
7.6 Toss three coins What is the probability of getting 2 heads and 1 tail when
you toss three coins? Give appropriate evidence to support your answer.

Myths about chance behavior

The idea of probability seems straightforward. It answers the question “What would
happen if we did this many times?” In fact, both the behavior of random phenomena
and the idea of probability are a bit subtle. We meet chance behavior constantly,
and psychologists tell us that we deal with it poorly.

ACTIVITY 7.1C
Investigating randomness
1. Pretend that you are flipping a fair coin. Without count the number of runs of size 2, 3, 4, etc. Record
actually flipping a coin, imagine the first toss. Write the number of runs of each size in a table like this:
down the result you see in your mind, heads (H) or
tails (T). Run length 2 3 4 5 6 7 8
Frequency
2. Imagine a second coin flip. Write down the result.
3. Keep doing this until you have 50 H’s or T’s 5. Use your calculator to generate a similar list of 50
written down. Write your results in groups of 5 coin flips. Let 1 represent a head and 0 represent a tail.
to make them easier to read, like this: HTHTH • On the TI-83/84, go to the home screen and
TTHHT, etc. execute the command randInt(0,1,50)S
4. A run is a repetition of the same result. In L1. The command randInt can be found under
the example in Step 3, there is a run of two tails MATH/PRB/5:randInt(.
followed by a run of two heads in the first 10 coin • On the TI-Nspire, open a new document
flips. Read through your 50 imagined coin flips, and and add a Lists & Spreadsheet page. Go to the
ACTIVITY 7.1C (continued)
318 CHAPTER 7 Probability: What Are the Chances?

ACTIVITY 7.1C (continued)

formula line in column A, enter the command Run length 2 3 4 5 6 7 8

=randint(0,1,50) and press b.
Frequency
Record the number of runs of size 2, 3, 4, and so
forth in a table like this: 6. Compare the two results. Did you or your
calculator have the longest run? How much longer?

The myth of short-run regularity The idea of probability is that randomness is

predictable in the long run. Unfortunately, our intuition about randomness tries to
tell us that random phenomena should also be predictable in the short run. When
they aren’t, we look for some explanation other than chance variation.2

Example 7.4 What looks random?

Toss a coin six times and record heads (H) or tails (T) on each toss. Which of the
following outcomes is more probable?

HTHTTH TTTHHH

Almost everyone says that HTHTTH is more probable because TTTHHH doesn’t
“look random.” But both are equally likely. That heads and tails are equally probable
says that about half a long sequence of tosses will be heads. It doesn’t say that heads
and tails must come close to alternating. The coin has no memory. It doesn’t know
what past outcomes were, and it can’t try to create a balanced sequence.

The outcome TTTHHH in tossing six coins looks unusual because of the runs
of 3 straight tails and 3 straight heads. Runs seem “not random” to our intuition but
are quite common. Here’s a more striking example than tossing coins.

Example 7.5 The hot hand in basketball

Belief that runs must result from something other than “just chance” is very common.
If a basketball player makes several consecutive shots, both the fans and his teammates
believe that he has a “hot hand” and is more likely to make the next shot. This is wrong.
Careful study has shown that runs of baskets made or missed are no more frequent in
basketball than would be expected if each shot were independent of the player’s previ-
ous shots. Players perform consistently, not in streaks. If a player makes half her shots
in the long run, her hits and misses behave just like tosses of a coin—and that means
that runs of hits and misses are more common than our intuition expects.3

The myth of the surprise meeting Mr. Starnes and his wife are on a two-week vaca-
tion in Hawaii. One day, at the Volcano National Park visitor center, they run into Her-
bie, one of Mr. Starnes’s former students. “How unusual! Maybe we were destined to
meet.” Well, maybe not. It is certainly unlikely that the Starnes family would run into
this particular acquaintance that day, but it is not at all unlikely that they would meet
 7. 1 Randomness, Probability, and Simulation 319

some acquaintance during their vacation. After all, a typical adult has about 1500 ca-
sual acquaintances. When something unusual happens, we look back and say, “Wasn’t
that unlikely?” We would have said the same if any of 1500 other unlikely things had
happened. Here’s an example where we can actually calculate the probabilities.

Example 7.6 Winning the lottery twice

Several years ago, Evelyn Marie Adams won the New Jersey state lottery for the sec-
ond time, adding $1.5 million to her previous $3.9 million jackpot. The New York Times
claimed that the odds of one person winning the big prize twice were about 1 in 17 tril-
lion. Nonsense, said two statistics professors in a letter that appeared in the Times two
weeks later. The chance that Evelyn Marie Adams would win twice in her lifetime is in-
deed tiny, but it is almost certain that someone among the millions of regular lottery play-
ers in the United States would win two jackpot prizes. The statisticians estimated even
odds of another double winner within seven years. Sure enough, Robert Humphries
won his second Pennsylvania lottery jackpot ($6.8 million total) two years later.

The myth of the law of averages While attending a convention in Las Vegas one of the
authors roamed the gambling floors, watching money disappear into the drop boxes un-
der the tables. You can see some interesting human behavior in a casino. When the shooter
in the dice game craps rolls several winners in a row, some gamblers think she has a “hot
hand” and bet that she will keep on winning. Others say that “the law of averages” means
that she must now lose so that wins and losses will balance out. Believers in the law of av-
erages think that, if you toss a coin six times and get TTTTTT, the next toss must be more
likely to give a head. It’s true that in the long run heads will appear half the time. What is a
myth is that future outcomes must make up for an imbalance like six straight tails.
Coins and dice have no memories. A coin doesn’t know that the first six outcomes
were tails, and it can’t try to get a head on the next toss to even things out. Of course,
things do even out in the long run. After 10,000 tosses, the results of the first six tosses
don’t matter. They are overwhelmed by the results of the next 9994 tosses.

“So the law of averages doesn’t guarantee me a girl after seven straight boys, but
can’t I at least get a group discount on the delivery fee?”
320 CHAPTER 7 Probability: What Are the Chances?

Example 7.7 We want a boy

Belief in this phony “law of averages” can lead to consequences close to disastrous. A
few years ago, “Dear Abby” published in her advice column a letter from a distraught
mother of eight girls. It seems that she and her husband had planned to limit their fam-
ily to four children. When all four were girls, they tried again—and again and again—
because they wanted a boy. After seven straight girls, even her doctor had assured her
that “the law of averages was in our favor 100 to 1.” Unfortunately for this couple, having
children is like tossing coins. Eight girls in a row is highly unlikely, but once seven girls
have been born, it is not at all unlikely that the next child will be a girl—and it was.

Probability and risk

People often use “probability” in a way that includes personal judgments of how likely it
is that some event will happen. They make decisions based on these judgments—some
people refuse to fly because they think the probability of an accident is too high. The
experts use probabilities from data to describe the risk of an unpleasant event. By
looking at what happens in many repetitions, the experts conclude that flying is actu-
ally safer than driving. Individuals and society seem to ignore data. We worry about
some risks that almost never occur while ignoring others that are much more likely.

Example 7.8 Asbestos in schools

High exposures to asbestos are dangerous. Low exposures, like those experienced by teach-
ers and students in schools where asbestos is present in the insulation around pipes, are not
very risky. The probability that a teacher who works for 30 years in a school with typical
asbestos levels will get cancer from the asbestos is around 15/1,000,000. The risk of dying
in a car accident during a lifetime of driving is about 15,000/1,000,000. That is, driving
regularly is 1000 times more risky than teaching in a school where asbestos is present.4
The probability of Risk does not stop us from driving. Yet the much smaller risk from asbestos
rain is . . . launched massive cleanup campaigns and a federal requirement that every school in-
You work all week. Then it spect for asbestos and make the findings public.
rains on the weekend. Can
there really be a statistical
truth behind our percep- Why do we take asbestos so much more seriously than driving? Why do we worry
tion that the weather is about very unlikely threats such as tornadoes more than we worry about heart attacks?
against us? At least on the
East Coast of the United • We feel safer when a risk seems under our control than when we cannot
States, the answer is “Yes.” control it. We are in control (or so we think) when we are driving, but we
Going back to 1946, it can’t control the risk from asbestos or tornadoes.
seems that Sundays receive
22% more precipitation • It is hard to comprehend very small probabilities. Probabilities of 15 per
than Mondays. The likely million and 15,000 per million are both so small that our intuition cannot
explanation is that the pollu- distinguish between them. Studies have shown that we generally overestimate
tion from all those workday very small risks and underestimate higher risks.
cars and trucks forms the
seeds for raindrops—with • The probabilities for risks like asbestos in the schools are not as certain as
just enough delay to cause probabilities for tossing coins. They must be estimated by experts from
rain on the weekend. complicated statistical studies.
 7. 1 Randomness, Probability, and Simulation 321

Our reactions to risk depend on more than probability. We are influenced by

social standards and by how we evaluate particular risks. As one writer noted, “Few
of us would leave a baby sleeping alone in a house while we drove off on a 10-minute
errand, even though car-crash risks are much greater than home risks.”5

Exercises 7.7 Personal random numbers? Ask several of your friends (at least 10 people)
to choose a four-digit number “at random.” How many of the numbers chosen
start with 1 or 2? How many start with 8 or 9? (There is strong evidence that
people in general tend to choose numbers that start with low digits.)
7.8 Playing “Pick 4” The Pick 4 games in many state lotteries announce a four-
digit winning number each day. The winning number is essentially a four-digit
group from a table of random digits. You win if your choice matches the winning
digits, in any order. The winnings are divided among all players who matched the
winning digits. That suggests a way to get an edge.
(a) The winning number might be, for example, either 2873 or 9999. Explain why
these two outcomes have exactly the same probability. (It is 1 in 10,000.)
(b) If you asked many people whether 2873 or 9999 is more likely to be the
randomly chosen winning number, most would favor one of them. Use the
information in this section to say which one and to explain why. (If you choose
a number that people think is unlikely, you have the same chance to win,
but you will win a larger amount because few other people will choose your
number.)
7.9 Surprising? You are getting to know your new roommate, assigned to
you by the college. In the course of a long conversation, you find that both
of you have sisters named Deborah. Should you be surprised? Explain your
answer.
7.10 Cold weather coming A TV weather man, predicting a colder-than-normal
winter, said, “First, in looking at the past few winters, there has been a lack of
really cold weather. Even though we are not supposed to use the law of averages,
we are due.” Do you think that “due by the law of averages” makes sense in talking
about the weather? Why or why not?
7.11 Reacting to risks National newspapers such as USA Today and the
New York Times carry many more stories about deaths from airplane crashes
than about deaths from automobile crashes. Auto accidents kill about 40,000
people in the United States each year. Crashes of all scheduled air carriers,
including commuter carriers, have killed between 0 and 575 people per
year in recent years, including the four hijacked planes that crashed on
September 11, 2001.
(a) Why do the news media give more attention to airplane crashes?
(b) How does news coverage help explain why many people consider flying more
dangerous than driving?
7.12 In the long run Suppose that the first six tosses of a coin give six tails and
that tosses after that are exactly half heads and half tails. What is the proportion of
heads after the first six tosses? What is the proportion of heads after 100 tosses if
322 CHAPTER 7 Probability: What Are the Chances?

the last 94 produce 47 heads? What is the proportion of heads after 1000 tosses if
half of the last 994 produce heads? What is the proportion of heads after 10,000
tosses if half of the last 9994 produce heads? What’s the point?

Simulation
The probabilities of heads and tails in tossing a coin are very close to 1/2. In principle,
these probabilities come from data on many coin tosses. What about the probability
that we get a run of three straight heads somewhere in 10 tosses of a coin? Do we re-
ally have to toss a coin 10 times over and over and over again? Of course not. In the
next chapter, we will learn how to find this probability by calculation from a model
that describes tossing coins. Unfortunately, the math needed to do probability cal-
culations is often tough. Technology comes to the rescue: we can use random digits
from a table, calculator, or computer to simulate many repetitions. This is easier than
math and much faster than actually running many repetitions in the real world.

Simulation

Using random digits from a table, a calculator, or computer software

to imitate chance behavior is called simulation.

Simulation is an effective tool for finding probabilities involving chance behav-

ior. We can use random digits to simulate many repetitions quickly. The proportion
of repetitions on which some outcome of interest occurs will eventually be close
to its probability, so simulation can give good estimates of probabilities. The art of
simulation is best learned by example.

Example 7.9 Doing a simulation: Shaquille O’Neal

at the free-throw line
For many years, Shaquille O’Neal was one of the NBA’s most dominant centers. He al-
ways had impressive statistics, and in 2000, he won all three MVP (Most Valuable Player)
awards. At 300 pounds and 7 feet 4 inches, he was a commanding presence on the
court. Still, he never became a “complete” player. Despite much practice and working
with special coaches, Shaq remained a 50% free-throw shooter for most of his career.
Let’s assume that every time Shaq steps up to the free-throw line, the probability that
he will make the shot is 0.5. Suppose we want to know how likely he is to make at least 3
free throws in a row out of 10 attempts. Three or more baskets in a row is called a “run.”
Step 1: State assumptions about the chance behavior.
• Each free throw has probabilities 0.5 for a basket (a “hit”) and 0.5 for a miss.
• Free throws are independent of each other. That is, knowing the outcome of
one free throw does not change the probabilities for the outcomes of any other
free throw. (Studies of consecutive free throws have shown this to be the case.)
Step 2: Assign digits to represent outcomes. Digits in Table B (in the back of the
book) will stand for the outcomes, in a way that matches the probabilities from Step 1.
We know that each digit in Table B has probability 0.1 of being any one of 0, 1, 2, 3,
 7. 1 Randomness, Probability, and Simulation 323

4, 5, 6, 7, 8, or 9, and that successive digits in the table are independent. Here is one
assignment of digits for free throws:
• One digit simulates one free throw.
• Digits 0 to 4 represent a made shot; digits 5 to 9 represent a missed shot.
Successive digits in the table simulate independent tosses.
Step 3: Simulate many repetitions. Ten digits simulate 10 tosses, so looking at 10
consecutive digits in Table B simulates one repetition. Read many groups of 10 digits
from the table to simulate many repetitions. Be sure to keep track of whether or not
the event we want (a run of 3 hits) occurs on each repetition.
Here are the first three repetitions, starting at line 101 in Table B. We have un-
derlined all runs of 3 or more hits.
Repetition 1 Repetition 2
Digits 1922395034 0575628713
Hit/miss HMHHHMMHHH HMMMMHMMHH
Run of 3? Yes No

Repetition 3
Digits 9640912531
Hit/miss MMHHMHHMHH
Run of 3? No

Continuing in Table B, we did 25 repetitions. In 11 of them, Shaq made at least 3

consecutive baskets. So we estimate the probability of a run by the proportion
11
estimated probability 5 5 0.44
25
Of course, 25 repetitions are not enough to be confident that our estimate is accu-
rate. But now that we understand how to do the simulation, we can tell a computer
to do many thousands of repetitions. A long simulation (or some tough calcula-
tion) finds that the true probability is about 0.508.

You can use your calculator’s random integer command instead of Table B to
perform the simulation of Example 7.9. The command randInt(0,9,10) will
simulate shooting one set of 10 free throws. Pressing b several more times will
generate additional sets of 10 free throws.
The chance behavior in Example 7.9 is typical of many probability problems
because it consists of independent trials (the free throws) that all have the same pos-
sible outcomes with the same probabilities. Shooting 10 free throws and observing
the number of heads in 10 tosses of a coin are simulated in much the same way.
Independence simplifies our work because it allows us to simulate each of the 10
free throws in exactly the same way.

Independence

Two random phenomena are independent if knowing the outcome of

one does not change the probabilities for the outcomes of the other.
324 CHAPTER 7 Probability: What Are the Chances?

Independence, like all aspects of probability, can be verified only by observing

many repetitions. It is plausible that repeated tosses of a coin are independent (the
coin has no memory), and observation shows that they are. It seems less plausible
that successive shots by a basketball player are independent, but observation shows
that they are at least very close to independent. Step 2 (assigning digits) rests on the
properties of the random digit table. Here are some examples of this step.

Example 7.10 Assigning digits for simulation

(a) Choose a family at random from a group of which 80% have an Internet connection.
One digit simulates one family:

Was he good or 0, 1, 2, 3, 4, 5, 6, 7  Internet connection

was he lucky? 8, 9  no connection
When a baseball player hits.
300, everyone applauds. A (b) Choose one family at random from a group of which 83% have an Internet connec-
.300 hitter gets a hit in 30% tion. Now two digits simulate one family:
of times at bat. Could a .300
year just be luck? Typical 00, 01, 02, . . . , 82  Internet connection
major leaguers bat about 83, 84, 85, . . . , 99  no connection
500 times a season and hit
about .260. A hitter’s suc- We assigned 83 of the 100 two-digit pairs to “Internet connection” to get probability
cessive tries seem to be in-
0.83. Representing “Internet connection” by 01, 02, . . . , 83 would also be correct.
dependent. From this model,
we can calculate or simulate (c) Choose one family at random from a group of which 50% have a dial-up connec-
the probability of hitting tion, 20% have no connection, and 30% have a cable or DSL connection. There are now
.300. It is about 0.025. Out three possible outcomes, but the principle is the same. One digit simulates one family:
of 100 run-of-the-mill major
league hitters, two or three 0, 1, 2, 3, 4  dial-up
each year will bat .300 be-
cause they were lucky. 5, 6 5 no connection
7, 8, 9 5 cable or DSL

The simulation of chance behavior is a powerful tool of contemporary science.

Doing a few simulations will help increase your understanding of probability.

Exercises 7.13 Which party does it better? An opinion poll selects adult Americans at
random and asks them, “Which political party, Democratic or Republican, do
you think is more concerned with providing health care for the poor?” Explain
carefully how you would assign digits from Table B to simulate the response of
one person in each of the following situations.
(a) Of all adult Americans, 50% would choose the Democrats and 50% the Republicans.
(b) Of all adult Americans, 60% would choose the Democrats and 40% the Republicans.
(c) Of all adult Americans, 40% would choose the Democrats, 40% would choose
the Republicans, and 20% would be undecided.
(d) Of all adult Americans, 53% would choose the Democrats and 47% the Republicans.
 7. 1 Randomness, Probability, and Simulation 325

7.14 Get rid of exams! Suppose that 80% of a school’s students favor abolishing
exams. You ask 10 students chosen at random. What is the probability that all 10
favor abolishing exams?
(a) State assumptions for the chance behavior involved in this setting.
(b) Assign digits to represent the answers “in favor” and “against.”
(c) Simulate 25 repetitions. Use your calculator or Table B. What is your estimate
of the probability?
7.15 Basic simulation Use Table B to simulate the responses of 10
independently chosen adults in each of the four situations of Exercise 7.13.
(a) For situation (a), use line 110.
(b) For situation (b), use line 111.
(c) For situation (c), use line 112.
(d) For situation (d), use line 113.
7.16 Organ donors A recent opinion poll showed that about 75% of Americans
would donate their organs upon death. Suppose that this is exactly true. Choosing
an American at random then has probability 0.75 of getting one who would
donate his or her organs. If we interview Americans separately, we can assume
that their responses are independent. We want to know the probability that a
simple random sample of 100 Americans will contain at least 80 who would
donate their organs. Explain carefully how to do this simulation and simulate
one repetition of the poll using your calculator or Table B. How many of the 100
would donate their organs? Explain how you would estimate the probability by
simulating many repetitions.
7.17 First ace Begin with a standard deck of playing cards. Shuffle the cards
thoroughly, and then draw a card. Replace the card in the deck, shuffle the deck,
and draw a card again. Continue until you draw an ace, or until you draw 10 cards,
whichever comes first. What is the probability of drawing an ace in 10 draws?
(a) State assumptions for the chance behavior involved in this setting.
(b) Assign digits to represent outcomes.
(c) Simulate 10 repetitions using your calculator or Table B. What is your estimate
of the probability?
7.18 Rock-paper-scissors Almost everyone has played the rock-paper-scissors
game when they were younger. Two players face each other and, at the count of
3, make a fist (rock), an extended hand, palm side down (paper), or a “V” with the
index and middle fingers (scissors). The winner is determined by these rules: rock
smashes scissors; paper covers rock; and scissors cut paper. If the two players choose
different objects, then there is a winner. But both could choose the same object
(for example, rock), in which case there is no winner. One of the questions that we
might explore is “What is the probability that a game will result in a winner?”
(a) State assumptions for the chance behavior involved in this setting.
(b) Assign digits to represent outcomes.
(c) Simulate 20 repetitions using your calculator or Table B. What is your estimate
of the probability?
326 CHAPTER 7 Probability: What Are the Chances?

ws
Ne
ws
Ne

APPLICATION 7.1 B]
RO
g
<
Se
a

Luck, skill, and randomness

In 2007, actress Danica McKellar authored her first
book, Math Doesn’t Suck: How to Survive Middle-
School Math without Losing Your Mind or Breaking
a Nail. A Newsweek article about the book included
a picture of McKellar standing by a chalkboard. The
following string of 60 digits was handwritten on the
chalkboard:

49894 84820 45868 34365 63811 77203 09179

80576 28621 35448 62270 52604

Is this a randomly generated set of digits or a

human-generated set of digits? Give appropriate
evidence to support your answer.
By permission of Penguin Group (USA) Inc.

Section 7.1 Summary

Some events in the world, both natural and of human design, are random. That is,
their outcomes have a clear pattern in very many repetitions even though the out-
come of any one trial is unpredictable. Probability describes the long-term regu-
larity of random phenomena. The probability of an outcome is the proportion of
very many repetitions on which that outcome occurs. A probability is a number
between 0 (never occurs) and 1 (always occurs).
Probabilities describe only what happens in the long run. Short runs of random
phenomena like tossing coins or shooting a basketball often don’t look random because
they do not show the regularity that in fact emerges only in very many repetitions.
We can use random digits to simulate chance behavior. A common situation
is several independent trials with the same possible outcomes and probabilities
on each trial. Think of several tosses of a coin or several rolls of a die. Successful
simulation requires that we state our assumptions about the chance behavior, as-
sign digits to represent outcomes, and perform many repetitions.

Section 7.1 7.19 An unenlightened gambler

Exercises (a) A gambler knows that red and black are equally likely to occur on each spin
of a roulette wheel. He observes five consecutive reds occur and bets heavily on
black at the next spin. Asked why, he explains that black is “due by the law of
averages.” Explain to the gambler what is wrong with this reasoning.
(b) After hearing you explain why red and black are still equally likely after five reds
on the roulette wheel, the gambler moves to a poker game. He is dealt five straight red
 7. 1 Randomness, Probability, and Simulation 327

cards. He remembers what you said and assumes that the next card dealt in the same
hand is equally likely to be red or black. Is the gambler right or wrong, and why?
7.20 Sports risks The probability of dying if you play high school football is about 10
per million each year you play. The risk of getting cancer from asbestos if you attend a
school in which asbestos is present for 10 years is about 5 per million. If we ban asbestos
from schools, should we also ban high school football? Briefly explain your position.
7.21 Rainy days The TV weatherman says, “There’s a 30% chance of rain
tomorrow.” Explain what this statement means.
7.22 Long runs Most people are surprised at the occurrence of long runs of
the same outcome in a random sequence. A science writer once said that, if you
flip a fair coin 250 times, then approximately 32 runs will have at least two heads;
about 16 runs will have at least three heads; 8 runs, at least four heads; 4 runs, at
least five heads; 2 runs, at least six heads; and 1 run, at least 7 heads.6 Carry out a
simulation to see if the writer’s claim is reasonable.
7.23 Is the Belgian euro coin “fair”? Two Polish math professors and their
students spun a Belgian euro coin 250 times. It landed “heads” 140 times. One of the
professors concluded that the coin was minted so that one side was heavier than the
other. A representative from the Belgian mint said the result was just chance.7 Design
and carry out a simulation using random digits to help you decide who is correct.
7.24 A game of chance I have a little bet to offer you. Toss a coin 10 times. If
there is no run of three or more straight heads or tails in the 10 outcomes, I’ll
pay you $2. If there is a run of three or more, you pay me just $1. Surely you will
want to take advantage of me and play this game? Is it to your advantage to play?
Design and carry out a simulation to help you answer this question.
7.25 Birth months If you choose a student from your school at random, what
is the probability that he or she was born in July? Justify your answer.
7.26 First ace again In Exercise 7.17 (page 325), you performed a simulation
to estimate the probability of drawing an ace in 10 draws. How would your results
change if you did not replace the card in the deck after each draw?
(a) Think about this new setting. Would the probability of drawing an ace in 10
draws be higher, lower, or the same as in Exercise 7.17? Explain.
(b) Design and carry out 10 repetitions of a simulation to imitate the chance
behavior in this new setting. Use Table B or your calculator. Describe the steps in
your simulation clearly enough so that a classmate could repeat your process.

CALCULATOR CORNER Prob Sim APP

The Prob Sim APP for the TI-83/84 allows you to simulate tossing
coins, rolling dice, spinning a spinner, drawing cards, and playing the
lottery. If you don’t have the APP, download it from your teacher. If you
are using the TI-Nspire, you’ll need to put on the TI-84 faceplate.
• To run the APP, press the Œ key. Then choose Prob Sim. You should see
the introductory screen on the left at the top of the next page. Press Í to
see the main menu (shown on the right at the top of the next page).
CALCULATOR CORNER (continued)
328 CHAPTER 7 Probability: What Are the Chances?

CALCULATOR CORNER (continued)

• Let’s use the Prob Sim APP to carry out the simulations described in
Exercises 7.17 (page 325) and 7.26, both of which involve drawing a card until
you get an ace. Choose 5:Draw Cards from the main menu. You should see
the screen shown at the left below. Press p to DRAW a card. (We got the
jack of spades, as shown at the right below.)

• Keep pressing p until you get an ace or until you have drawn a card
10 times.
• If you repeat this process many, many times, and keep track of the overall
proportion of trials in which you get an ace within 10 draws, you’ll have a good
estimate of the probability for Exercise 7.17.
• In Exercise 7.26, we asked about the probability of getting an ace within 10
draws if you don’t replace the card in the deck after each draw. To simulate this
new setting, press s to CLEAR your previous trials. Press o to confirm
the deletion.
• Now press q to adjust the APP’s settings. Change the Replace option to
No as shown in the screen shot below. Then press s to OK the changes.

• Draw cards until you get an ace or have drawn 10 cards, as before. If you do
this many times, you will get a good estimate of the probability for Exercise 7.26.
The results of our first six draws are shown in the screen shot above, right.
 7. 2 Probability Rules 329

7 . 2 Probability Rules

ACTIVITY 7.2
Monty’s dilemma
Materials: Several decks of cards • Designate one person to serve as game show
host. The host should shuffle and arrange the
In Parade magazine, a reader posed the following
cards so that he or she knows where the ace is
question to Marilyn vos Savant, author of the “Ask
located.
Marilyn” column:
• The contestant then picks a door.
Suppose you’re on a game show, and you’re given • The host will then reveal one of the doors that
the choice of three doors. Behind one door is a car, wasn’t picked and that has a goat behind it (so the
behind the others, goats. You pick a door, say #1, host will always show a 2).
and the host, who knows what’s behind the doors, • The contestant must then decide whether to
opens another door, say #3, which has a goat. He stick with the original choice or to switch to the
says to you, “Do you want to pick door #2?” Is it to door that hasn’t been opened.
your advantage to switch your choice of doors?
The game show in question was Let’s Make a • Finally, the host reveals what is located behind
Deal and the host was Monty Hall. the contestant’s chosen door.
• Play the game 10 times, and record the results
Here’s the first part of Marilyn’s response: in a table like the one below. The player should be
sure to stick sometimes and to switch sometimes.
Yes; you should switch. The first door has a 1/3
chance of winning, but the second door has a
2/3 chance. Trial Door chosen first Stick/switch Win/lose
1 2 Switch Lose
Thousands of readers wrote to Marilyn to disagree with 2 1 Stick Win
her answer. But she held her ground. In this Activity, 3 2 Stick Lose
you will use simulation to decide who was right.
1. Discuss with your classmates: should the contestant 3. Switch roles and repeat Step 2.
stay with the original choice or switch? Why?
4. What’s the probability that you picked the door
2. Working with a partner, simulate the game as with the nice prize behind it in the first place?
follows.
5. Intuition tells us that it shouldn’t make any
• Pull an ace and two 2s from the deck of cards. difference whether you stick or switch. There’s still a
These represent the three “doors” with prizes (the 1/3 chance that you’re right. Based on the results of
ace represents the nice prize). this Activity, would you agree or disagree?

Probability models
As Activity 7.2 illustrates, our intuition about probability is sometimes unreliable. In
such cases, simulation can help clarify our thinking about chance behavior. But do
we always need to repeat a chance process many times to determine the probability
of a particular outcome? Fortunately, the answer is no. What we need is a probability
model that describes the chance behavior. Our probability model should include a
list of all possible outcomes (the sample space) and a probability for each outcome.
330 CHAPTER 7 Probability: What Are the Chances?

Probability Model

A probability model is a description of some chance behavior that

consists of two parts: a list of all possible outcomes and the probability
for each outcome.
The list of all possible outcomes of some random phenomenon is
called the sample space.

Let’s see how a probability model works in a familiar setting—tossing a coin.

Example 7.11 Probability model: tossing a coin

three times
If you toss a fair coin three times, what’s the probability of getting two heads and one
tail? In this case, we’re trying to model the chance behavior involved in tossing a coin
three times. One possible outcome of this random phenomenon is heads on the first
toss, tails on the second toss, and tails on the third toss, which we can represent as
H-T-T. The sample space consists of all possible outcomes, which we can list as
follows:

H-H-H H-H-T H-T-H T-H-H T-T-H T-H-T H-T-T T-T-T

Since each toss is equally likely to land heads or tails, each of these 8 possible
outcomes has the same chance of occurring: 1/8. Three of these outcomes con-
sist of two heads and one tail. So the probability of getting two heads and one
tail is 3/8.

You can see from Example 7.11 that a probability model does more than just
assign a probability to each outcome. It allows us to find the probability of any
collection of outcomes, which we call an event. Events are usually designated by
capital letters, like A, B, C, etc.

Event

An event is any collection of outcomes from some chance process.

In Example 7.11, if we let event A  getting 2 heads and 1 tail, then we can
write P(A)  3/8 as a shorthand for saying “the probability that event A will hap-
pen is 3/8.” Consider two more events:

B 5 getting three heads and C 5 getting more heads than tails

Convince yourself that P(B)  1/8. How do we find P(C)? There are four outcomes
in the sample space with more heads than tails, so P(C)  4/8, or 1/2. What is the
relationship between P(A), P(B), and P(C)? Why does this make sense?
 7. 2 Probability Rules 331

Here’s one more event: D  getting at least one tail. What’s P(D)? (Recall that
“at least one” means one or more.) Confirm that P(D)  7/8. What’s the relation-
ship between P(D) and P(B)? Why does this make sense?
Basic rules of probability
Our coin-tossing example has revealed some basic rules that any probability model
must obey:
• The probability of any event is a number between 0 and 1. The probability
of an event is the long-run proportion of repetitions on which that event
occurs. Any proportion is a number between 0 and 1, so any probability is
also a number between 0 and 1. An event with probability 0 never occurs,
and an event with probability 1 occurs on every trial. An event with
probability 0.5 occurs in half the trials in the long run.
• All possible outcomes together must have probability 1. Because some
outcome must occur on every trial, the sum of the probabilities for all
possible outcomes must be exactly 1.
• The probability that an event does not occur is 1 minus the probability
that the event does occur. If an event occurs in (say) 70% of all trials, it
fails to occur in the other 30%. The probability that an event occurs and the
probability that it does not occur always add to 100%, or 1. (This explains
why P(D)  1  P(B) in the coin-tossing example, since event D occurring
is the same as event B not occurring.)
• If two events have no outcomes in common, the probability that one or the
other occurs is the sum of their individual probabilities. If one event occurs
in 40% of all trials, a different event occurs in 25% of all trials, and the two can
never occur together, then one or the other occurs on 65% of all trials because
40%  25%  65%. (This explains why P(A)  P(B)  P(C) in the coin-tossing
example. Event A was getting exactly two heads; event B was getting three
heads. These two events have no outcomes in common, so the probability that
one or the other occurs is P(A)  P(B). Note that event C—getting more heads
than tails—is equivalent to “event A occurs or event B occurs.”)
• If all outcomes in the sample space are equally likely, the probability that
event A occurs can be found using the formula

P1A2 5
number of outcomes corresponding to event A
total number of outcomes in sample space
Here’s an example that illustrates the use of these probability rules.

Example 7.12 Probability model: rolling two dice

Imagine rolling two fair, six-sided dice—one that’s red and one that’s green. Figure 7.3
(on the next page) shows the 36 possible outcomes in the sample space. How should
we assign probabilities to these outcomes?
If the dice are fair, then it is reasonable to assign the same probability to each
of the 36 outcomes in Figure 7.3. Because these 36 probabilities must have sum 1,
each outcome must have probability 1/36.
332 CHAPTER 7 Probability: What Are the Chances?

Figure 7.3 The 36 possible outcomes from rolling two dice, one red and one green.

We are interested in the sum of the spots on the up-faces of the dice. What is
the probability that this sum is 5? The event A  “roll a 5” contains four outcomes,
and its probability is the sum of these outcomes:

P(A)  P(roll a 5) = P ( +P ) ( ) + P( ) + P( )
1 1 1 1
5 1 1 1
36 36 36 36
4
5 5 0.111
36
What’s the probability that the sum is not 5?

P 1 sum is not 5 2 5 1 2 P 1 sum is 5 2 5 1 2

4 32
5 5 0.889
36 36

The rules tell us only what probability models make sense. They don’t tell us
whether the probabilities are correct, that is, whether they describe what actually
happens in the long run. The probabilities in Example 7.12 are correct for fair dice,
like those used in casinos. Inexpensive dice with hollowed-out spots are not bal-
anced, and this probability model does not describe their behavior.
The basic probability rules apply to more than just tossing coins or rolling dice—
they apply in any situation involving chance behavior, as the following example
illustrates.

Example 7.13 Probability model: marital status

Choose a woman aged 25 to 29 years old at random and record her marital status. “At
random” means that we give every such woman the same chance to be the one we
choose. That is, we choose a random sample of size 1. The probability of any marital sta-
tus is just the proportion of all women aged 25 to 29 who have that status—if we chose
many women, this is the proportion we would get. Here is the set of probabilities:

Marital status: Never married Married Widowed Divorced

Probability: 0.506 0.452 0.002 0.04
 7. 2 Probability Rules 333

This table gives a probability model for picking a young woman at random and
finding out her marital status. It tells us what the possible outcomes are (there are only
four) and it assigns probabilities to these outcomes. The probabilities here are the pro-
portions of all women who are in each marital class. That makes it clear that the
probability that a woman is not married is just the sum of the probabilities of the
three classes of unmarried women:

P 1 not married 2 5 P 1 never married 2 1 P 1 widowed 2 1 P 1 divorced 2

5 0.506 1 0.002 1 0.04 5 0.548
Alternatively, we can use one of our other basic probability rules:

P 1 not married 2 5 1 2 P 1 married 2

5 1 2 0.452 5 0.548

That is, if 45.2% are married, 54.8% are not married.

Exercises 7.27 Toss 4 times Imagine tossing a fair coin 4 times.

(a) List all possible outcomes in the sample space.
(b) What probability would you assign to each outcome? Why?
(c) Event A  getting three tails and one head. Find P(A). Show your method.
7.28 Use your head Refer to Exercise 7.27. In two different ways, find the
probability of getting at least one head. Show your work.
7.29 Pair-a-dice Imagine rolling two fair, six-sided dice—one red and one
green. Use the probability model from Example 7.12 and the basic probability
rules to help you find the probability of each of the following events. Show your
work.
D 5 doubles 1 the same number of spots showing on both dice 2
M 5 sum of the spots showing on the two dice is 10 or less
R 5 red die has a higher number of spots showing than green die
7.30 Four-sided dice A tetrahedron (see image) is a pyramid with four faces,
each a triangle with all sides equal in length. Label the four faces of a tetrahedral
die with 1, 2, 3, and 4 spots.
(a) Give a probability model for rolling two such dice—one blue and one yellow—
and recording the number of spots on the down-face. Explain why you think your
model is at least close to correct.
(b) What is the probability that the sum of the down-faces is 5? Show your work.
7.31 Causes of death Government data assign a single cause for each death
that occurs in the United States. The data show that the probability is 0.45 that
a randomly chosen death was due to cardiovascular (mainly heart) disease, and
0.23 that it was due to cancer.
334 CHAPTER 7 Probability: What Are the Chances?

(a) What is the probability that a death was due either to cardiovascular disease or
to cancer? Show your work.
(b) What is the probability that the death was due to some other cause? Show
your work.
7.32 Do husbands do their share? An opinion poll interviewed a random
sample of 1025 married women. The women were asked whether their husbands
did their fair share of household chores. Here are the results:

Outcome Probability
Does more than his fair share 0.12
Does his fair share 0.61
Does less than his fair share ?

These proportions are probabilities for the random phenomenon of choosing a

married woman at random and asking her opinion.
(a) What must be the probability that the woman chosen says that her husband
does less than his fair share? Why?
(b) What is the probability of the event “I think my husband does at least his fair
share”?

Two-way tables and probability

So far, we have learned basic tools for modeling chance behavior and for finding the
probability of an event. What if we’re interested in finding probabilities involving
two events?

Example 7.14 Shuffle up and deal!

A standard deck of playing cards (with jokers removed) consists of 52 cards in four
suits—clubs, diamonds, hearts, and spades. Each suit has 13 cards, with denomina-
tions ace, 2, 3, 4, 5, 6, 7, 8, 9, 10, jack, queen, and king. The jack, queen, and king are
referred to as “face cards.” Imagine that we shuffle the deck thoroughly and deal one
card. Let’s define two events: A 5 getting a face card and B 5 getting a heart.
We can build a two-way table that describes the possible outcomes in terms
of these two events. The rows of the table relate to event A and the columns of the
table relate to event B. Notice that there are only two possibilities for each event—
either it happens or it doesn’t. So we need two rows (A happens; A doesn’t happen)
and two columns (B happens; B doesn’t happen).

Heart Not a heart

Face card
Not a face card

Now we need to fill in the cells of the table. Let’s start with the rows, which relate
to event A  getting a face card. There are a total of 12 face cards—3 in each suit times
 7. 2 Probability Rules 335

4 suits. Since 3 of those face cards are hearts, we put a “3” in the top-left cell of the table.
That leaves 9 face cards that are not hearts. This value goes in the top-right cell of the
table. There are 13 hearts in all, so the bottom-left cell of the table must contain a 10.
Do you see how we got the “30” in the bottom-right cell? If not, look at the marginal
totals.
Heart Not a heart
Face card 3 9 12
Not a face card 10 30 40
13 39 52

What’s P(A)? Since each of the 52 cards is equally likely to be dealt, that means we
have a 1/52 chance of getting any particular card. Since there are 12 face cards, P(A) 5
12/52. Finding P(B) is just as easy. Since there are 13 hearts in the deck, P(B) 5 13/52.

The two-way table in Example 7.14 helped organize the outcomes in the sample
space in a way that allowed us to easily find P(A) and P(B). To be honest, we could
have determined these probabilities without using a two-way table. All we would have
to do is list the sample space in some other way and then use the same reasoning as in
Example 7.14 to obtain the probabilities of these two events. But what if we wanted to
find a probability involving both of these events, like the probability of getting a face
card and a heart? In that case, the two-way table would make the task much easier.

Example 7.15 Playing cards and two-way tables

Continuing with the scenario of Example 7.14, we’re interested in finding the prob-
ability of getting a face card and a heart. That is, we’re trying to determine the prob-
ability P(A and B). The event “face card and heart” can be seen at the intersection
of the “face card” row and the “heart” column in the completed two-way table of
Example 7.14. There are 3 cards in the deck that are both face cards and hearts
(namely, the jack, queen, and king of hearts). So the desired probability is

P(A and B)  P(face card and heart)  3/52

What about the probability of getting a face card or a heart? (Note the mathemati-
cal use of the word “or” here—the card could be a face card, or a heart, or both.) There
are 12 face cards in the deck, and 13 hearts. If you weren’t paying close attention, you
might be tempted to add these two numbers together and to give the (incorrect) an-
swer 25/52 for the desired probability. Do you see why this is wrong? There are 3 cards
in the deck that are both face cards and hearts. If we add the row total for “face card” to
the column total for “heart,” we are counting these three cards twice. How many cards
are there that are face cards or hearts (or both)? From the two-way table, we can see
that there are 10 1 3 1 9 5 22 such cards. So the desired probability is

P(A or B) 5 P(face card or heart) 5 22/52

 336 CHAPTER 7 Probability: What Are the Chances?

There’s another way to get the correct value of 22/52 for P(A or B) in Example
7.15. If we add P(A) 5 12/52 and P(B) 5 13/52, we get the incorrect answer (25/52)
mentioned in the example. Why? Because we are counting three outcomes—the
cards that are face cards and hearts—in both P(A) and P(B). If we subtract the
probability P(A and B) 5 3/52, we get

P(A) 1 P(B) 2 P(A and B) 5 12/52 1 13/52 2 3/52

5 22/52
5 P(A or B)

This intuitive result is known as the general addition rule.

General addition rule for two events

If A and B are any two events resulting from some chance process,
then
P(A or B) 5 P(A) 1 P(B) 2 P(A and B)

You might be wondering whether there is also a rule for P(A and B). There is,
but it’s not quite as intuitive. Stay tuned for that later. For now, let’s see what we can
do with two-way tables and the general addition rule.

Example 7.16 Who has pierced ears?

Students in a college statistics class wanted to find out how common it was for young
adults to have their ears pierced. They recorded data on two variables—gender and
whether the student had pierced ears—for all 178 individuals in the class. The two-
way table below displays the data.

Pierced ears?
Gender Yes No Total
Male 19 71 90
Female 84 4 88
Total 103 75 178
iStockphoto

• If we randomly select a student from the class, what’s the probability that
the student has pierced ears? Let event A 5 has pierced ears. Since each of
the 178 students in the class is equally likely to be chosen, and there are 103
students with pierced ears, P(A) 5 103/178.
• If we randomly select a student from the class, what’s the probability that
we choose a male with pierced ears? Let event B 5 is a male. We want to find
P(pierced ears and male), that is, P(A and B). Looking at the intersection
of the “male” row and “Yes” column, we see that there are 19 males with
pierced ears. So P(A and B) 5 19/178.
 7. 2 Probability Rules 337

• If we randomly select a student from the class, what’s the probability that
we choose someone with pierced ears or a male? This time, we’re interested
in P(A or B). Using the general addition rule,

P(A or B) 5 P(A) 1 P(B) 2 P(A and B)

5 103/178 1 90/178 2 19/178
5 174/178

Can you think of another (simpler) way to find this probability?

When we found the probability of getting a male with pierced ears in Example
7.16, we could have described this as either P(A and B) or P(B and A). Why? Because
“A and B” describes the same event as “B and A.” Likewise, P(A or B) is the same as
P(B or A). Don’t get so caught up in the notation that you lose sight of what’s really
happening! If you convert the symbols to words, it’s pretty obvious that P(pierced
ears and male) is the same as P(male and pierced ears). Likewise, P(pierced ears or
male) is clearly the same as P(male or pierced ears).

Exercises
7.33 Card tables and two-way tables Refer to Example 7.14. Let event C 5 getting
a card that is not a face card and event D 5 getting a spade, club, or diamond.
(a) Find P(C) and P(D). Show your method clearly.
(b) Describe the event “C and D” in words. Then find P(C and D). Show your work.
(c) Describe the event “C or D” in words. Then find P(C or D). Show your work.
7.34 More cards Shuffle a standard deck of playing cards and deal one card, as in
Example 7.14. Let event J 5 getting a jack and event R 5 getting a red card.
(a) Construct a two-way table that describes the sample space in terms of events
J and R.
(b) Find P(J), P(R), and P(J and R). Show your work.
(c) Explain why P(J or R) Z P(J) 1 P(R). Then use the general addition rule to
compute P(J or R).
7.35 Sampling senators, I The two-way table below describes the members of
the U.S. Senate in 2008.
Male Female
Democrats 40 11
Republicans 44 5

(a) Who are the individuals? What variables are being measured?
(b) If we select a U.S. senator at random, what’s the probability that we choose
• a Democrat? • a female Democrat?
• a female? • a female or a Democrat?
338 CHAPTER 7 Probability: What Are the Chances?

7.36 Who eats breakfast? Students in an urban school were curious about how
many children regularly eat breakfast. They conducted a survey, asking, “Do you eat
breakfast on a regular basis?” All 595 students in the school responded to the survey.
The resulting data are shown in the two-way table below.8

Male Female Total

Eat breakfast regularly 190 110 300
Don’t eat breakfast regularly 130 165 295
Total 320 275 595

(a) Who are the individuals? What variables are being measured?
(b) If we select a student from the school at random, what is the probability that
we choose

• a female?
• someone who eats breakfast regularly?
• a female who eats breakfast regularly?
• a female or someone who eats breakfast regularly?

7.37 Playing roulette An American roulette wheel has 38 slots with numbers 1
through 36, 0, and 00 (see Figure 7.4). On the wheel, 18 of the numbered slots are
red, 18 are black, and two—the 0 and 00—are green. When the wheel is spun, a
metal ball is dropped onto the middle of the wheel. If the wheel is balanced, the
ball is equally likely to settle in any of the numbered slots. Imagine spinning a fair
wheel once. Let event B 5 ball lands in a black slot and event E 5 ball lands in an
even-numbered slot. (Treat 0 and 00 as even numbers.)

1 1
3 3 00 10 2
7
24
6 25
29
12
3
15

8
9
34

21 6 18 31 1
22 5
17 32

33
20

16
7
11

30
26 23
5
9 28 4 3
Figure 7.4 Diagram of an 0 2 1
American roulette wheel.
 7. 2 Probability Rules 339

(a) Make a two-way table that displays the sample space in terms of events B and E.
(b) Find P(B) and P(E). Show your method clearly.
(c) Describe the event “B and E” in words. Then find P(B and E). Show your work.
(d) Describe the event “B or E” in words. Then find P(B or E). Show your work.
7.38 More roulette Refer to Exercise 7.37. Let event G 5 ball lands in a green slot.
(a) Find P(B and G). Explain why this makes sense.
(b) Find P(B or G) using the general addition rule. What do you notice?
(c) Find P(E and G). Explain why this makes sense.
(d) Find P(E or G) using the general addition rule.

Venn diagrams and probability

When we’re trying to find probabilities involving two events, two-way tables dis-
play the sample space in a way that makes probability calculations simple. Venn
diagrams provide another way to illustrate the set of all possible outcomes of a
chance process. A Venn diagram usually consists of two or more circles surrounded
by an outer rectangle, as in Figure 7.5. Each circle represents an event. The area
inside the rectangle represents the sample space of the random phenomenon
(sometimes called the “universe”). An example should help illustrate how Venn
diagrams work.

Sample space

Event Event

Figure 7.5 A typical Venn diagram shows the sample space and the relationship
between two or more events.

Example 7.17 Venn diagrams and dealing cards

In Examples 7.14 and 7.15 (pages 334 and 335), we modeled the process of dealing
a playing card from a well-shuffled deck. Our two events of interest were A 5 getting
a face card and B 5 getting a heart. Earlier, we used a two-way table to show the
sample space. Now we’ll use a Venn diagram.
Figure 7.6 (on the next page) is a Venn diagram showing all 52 possible outcomes
of this chance process. The face cards are all shown inside circle A; the hearts are all
shown inside circle B. Those cards that are both face cards and hearts are placed in the
intersection of the two circles. Which cards are outside the two circles? Those cards
that are neither hearts nor face cards. Recall that since each card is equally likely to be
the one dealt, each outcome in the sample space has a 1/52 chance of occurring.
340 CHAPTER 7 Probability: What Are the Chances?

A♦ 2♦ 3♦ 4♦ 5♦ 6♦ 7♦ 8♦ 9♦ 10 ♦
B = Hearts

J♦ Q♦ K♦ J♥ A♥ 2♥ 3♥ A♣
A = Face 2♣
cards J♣ Q♣ K♣ Q♥ 4♥ 5♥ 6♥ 3♣
4♣
J♠ Q♠ K♠ K♥ 7♥ 8♥ 9♥
5♣
10♥ 6♣
7♣
8♣
9♣
A♠ 2♠ 3♠ 4♠ 5♠ 6♠ 7♠ 8♠ 9♠ 10♠ 10 ♣

Figure 7.6 A Venn diagram showing the sample space when we deal a card from a
well-shuffled deck. Our two events of interest are A 5 getting a face card and B 5 getting
a heart.

How can we find probabilities from a Venn diagram? By counting the out-
comes in the appropriate region(s) of the diagram, as the following example
illustrates.

Example 7.18 Finding probabilities from

a Venn diagram
Let’s use Figure 7.6, the Venn diagram in Example 7.17, to find some probabilities.
• P(A): There are 12 face cards inside circle A. So the probability of getting a
face card is 12/52.
• P(B): There are 13 hearts inside circle B. So the probability of getting a
heart is 13/52.
• P(A and B): There are 3 cards in the intersection of circles A and B—these
are the cards that are both face cards and hearts. So the probability of getting
a heart that is a face card is 3/52.
• P(A or B): We are looking for the cards that are face cards or hearts (or
both). These are the cards anywhere inside the two circles. Since there are
22 such cards, P(A or B) 5 22/52.
• P(not A and not B): We want the cards that are not face cards and also not
hearts. Where are they located? In the region of the Venn diagram that falls
outside the two circles. Since there are 30 such cards, the desired probability
is 30/52. How is this probability related to P(A or B)? Why does this make
sense?

Venn diagrams have uses in other branches of mathematics. As a result,

some common vocabulary and notation have been developed that you ought to
know. The following chart shows the language that we have been using, along
with standard mathematical notation and terminology, and the corresponding
Venn diagrams.
 7. 2 Probability Rules 341

Our language Notation Terminology Venn diagram

not A Ac complement of A
A

A and B A>B intersection of A and B

A B

A or B A<B union of A and B

A B

Sometimes, instead of listing the actual outcomes of a chance process in the Venn
diagram, we summarize the sample space by writing the number of outcomes in
each region of the Venn diagram. Here’s an example that illustrates this approach.

Example 7.19 Another way to use Venn diagrams

In Example 7.16 (page 336), we looked at data from a survey on gender and ear pierc-
ings for a large group of college students. The chance process consisted of selecting a
student in the class at random. Our events of interest were A 5 has pierced ears and B 5
is male. Here, once again, is the two-way table that summarizes the sample space:
Pierced ears?

Gender Yes No
Male 19 71
Female 84 4

How can we construct a Venn diagram that displays this same information?
There are four distinct regions in the Venn diagram shown in Figure 7.7 (on the
next page). These regions correspond to the four cells in the two-way table. We can
describe this correspondence in tabular form as follows:
Region in Venn diagram In words In symbols Count
In the intersection of two circles pierced ears and male A>B 19
Inside circle A, outside circle B pierced ears and female A > Bc 84
Inside circle B, outside circle A no pierced ears and male Ac > B 71
Outside both circles no pierced ears and female Ac > Bc 4
342 CHAPTER 7 Probability: What Are the Chances?

We have added the appropriate counts of students to the four regions in Figure 7.7.
No pierced ears and
female

A = pierced ears B = male

84 19 71

Pierced ears and male

Figure 7.7 The completed Venn diagram for the large group of college students.
The circles represent the two events A 5 has pierced ears and B 5 is male.

Venn diagrams can also be used to solve probability problems, as the following
example illustrates.

Example 7.20 Who reads the paper?

In a large apartment complex, 35% of residents read USA Today. Only 20% read the New
York Times. Five percent of residents read both papers. If we select a resident of the apart-
ment complex at random, what’s the probability that he or she doesn’t read either paper?
Even though we don’t know how many residents there are in the apartment com-
plex, we can still construct a Venn diagram to model this situation. Our events of interest
are A 5 reads USA Today and B 5 reads New York Times. Figure 7.8(a) shows the empty
Venn diagram. In problems of this kind, it’s usually a good strategy to start in the “mid-
dle.” Since we know that 5% of residents read both papers, we can put this value in the
intersection of the two circles, A d B. Figure 7.8(b) shows the updated Venn diagram.

A = reads B = reads
USA Today New York Times A = reads Reads both B = reads
USA Today papers New York Times

5%

Figure 7.8(a) A Venn diagram with Figure 7.8(b) Updated Venn diagram show-
event A showing the residents who read ing the 5% of residents who read both papers.
USA Today and B showing those who read
the New York Times.

Now let’s work our way from the inside out. We know that 35% of residents read
USA Today. (That is, P(A) 5 0.35.) But we have already accounted for the 5% of
people in the group that read both papers. That leaves 35%  5% 5 30% who read
 7. 2 Probability Rules 343

USA Today but not the New York Times. We’ll fill in that value in the corresponding
region of Figure 7.8(c). Similarly, we can calculate that 20%  5% 5 15% read the
New York Times but not USA Today. This value has also been added to Figure 7.8(c).
How can we figure out the percent of residents who read neither paper? Those
are the people in the region of the Venn diagram that’s outside both circles. Since
all residents in the apartment complex fall somewhere in the diagram, we start with
100% of residents and subtract those who read either paper. That is,

percent who read neither paper 5 100% 2 30% 2 5% 2 15% 5 50%

Figure 7.8(d) shows the completed Venn diagram.

A = reads Reads both B = reads

USA Today papers New York Times
A = reads Reads both B = reads
USA Today papers New York Times

30% 5% 15%

30% 5% 15%
50%

Reads neither paper

Figure 7.8(c) The updated Venn diagram Figure 7.8(d) The completed Venn diagram
showing the percents of residents who read only showing that 50% of residents do not read either
one of the two newspapers. paper.

So far, we have tried to keep the wording of probability questions as consistent

and straightforward as possible. There are times, though, when you will encounter
more complicated language. One common example is found in questions like “What’s
the probability that at least one of the two events occurs?” and “What’s the probability
that at most one of the two events occurs?” Let’s consider the first of these questions
briefly in the setting of Example 7.20. We’ll save the other question for an exercise.
For events A and B as described in Example 7.20, at least one of the two events
occurs if we randomly select a resident who reads USA Today, the New York Times,
or both papers. The probability that this happens is 0.50, since 50% of the residents
read at least one of the two papers. Note that this is also P(A < B), since the event
A < B includes those who read one or both of the newspapers.

Exercises 7.39 Sampling senators, II The two-way table below describes the members of
the U.S. Senate in 2008. Construct a Venn diagram of these data using D 5 is a
Democrat and F 5 is female as the events of interest.
Male Female
Democrats 40 11
Republicans 44 5
344 CHAPTER 7 Probability: What Are the Chances?

7.40 Breakfast eaters, again The two-way table below describes the 595 students who
responded to a school survey about eating breakfast. Construct a Venn diagram of
these data using B 5 eats breakfast regularly and M 5 is male as the events of interest.
Male Female Total
Eat breakfast regularly 190 110 300
Don’t eat breakfast regularly 130 165 295
Total 320 275 595

7.41 Playing roulette again Refer to Exercise 7.37 (page 338). Construct a Venn
diagram that shows all possible outcomes of spinning the roulette wheel in terms of
event B 5 ball lands in black slot and event E 5 ball lands in an even-numbered slot.
7.42 At least, at most Refer to Example 7.20 (page 342). What’s the probability
that at most one of the two events occurs? Explain your method clearly.
7.43 MySpace versus Facebook, I A recent survey suggests that 85% of college
students have posted a profile on Facebook, 54% use MySpace regularly, and 42%
do both. Suppose we select a college student at random.
(a) Construct a Venn diagram to represent this setting.
(b) Consider the event that the randomly selected college student does not
use MySpace regularly or has posted a profile on Facebook. Write this event in
symbolic form using the two events of interest that you chose in (a).
Chris Jackson/Getty Images
(c) Find the probability of the event described in (b). Explain your method.
(d) Can you make a two-way table for this setting? If so, do it. If not, explain why not.
7.44 Computers at Princeton An October 2007 census revealed that 40%
of Princeton students primarily used Macintosh computers (MACs). The rest
primarily used PCs. At the time of the census, 67% of Princeton students were
undergraduates. The rest were graduate students. According to the census, 23% of
Princeton’s graduate students said that they used PCs as their primary computers.
Suppose that we select a Princeton student at random.
(a) Construct a Venn diagram to represent this setting.
(b) Consider the event that the randomly selected student is a graduate student
who uses a MAC. Write this event in symbolic form using the two events of
interest that you chose in (a).
(c) Find the probability of the event described in (b). Explain your method.
(d) Can you make a two-way table for this setting? If so, do it. If not, explain why not.

Mutually exclusive (disjoint) events

Thus far, we have drawn our Venn diagrams with two overlapping circles. That’s be-
cause in every example we’ve studied, there have been outcomes that fell in the inter-
section of the two events. In some situations, two events may not have any outcomes
in common. We say that the two events are mutually exclusive, or disjoint. When
that happens, a Venn diagram displaying the events should show two nonoverlapping
circles. Figure 7.9 shows a Venn diagram for mutually exclusive events A and B.
 7. 2 Probability Rules 345

A B

Figure 7.9 A Venn diagram

showing disjoint events A and B.

Let’s return to the card-dealing scenario of Examples 7.17 and 7.18. Recall
that we had previously defined event B 5 getting a heart. If we define a new event
C 5 getting a black card, then events B and C are mutually exclusive (disjoint).
Why? Because if we deal one card from a standard deck, that card cannot be both
a heart and a black card. What’s P(B > C) in this case? It’s 0, of course. What is
P(B < C)? By the general addition rule,

P(B < C) 5 P(B) 1 P(C) 2 P(B > C) 5 P(B) 1 P(C) since P(B > C) 5 0

These special results for disjoint events are worth summarizing.

Special rules for disjoint (mutually exclusive) events

If events A and B are disjoint (mutually exclusive), then

P(A > B) 5 0
P(A < B) 5 P(A) 1 P(B)

ws
Ne
ws
Ne

APPLICATION 7.2 B]
RO
g
<
Se
a

Genetics at work
You inherited certain physical characteristics from two events of interest. Use initials to represent each
your parents—gender, eye and hair color, and student.
the shape of your hairline and your earlobes. In 4. Now make a two-way table that summarizes the
this Application, we’ll investigate the relationship number of students in each of the four regions of
between two other traits that are believed to be the Venn diagram.
inherited—tongue curling and hand clasping.
5. Describe a method for selecting a student from
QUESTIONS the class at random.
1. Have everyone in your class try to curl their 6. Suppose we select a student from the class at
tongue (rolling the sides up to form a cup). random. Find each of the following probabilities.
2. Now have everyone interlock the fingers of their Describe in words what your result means.
two hands. Note which thumb (right or left) comes (a) P(Tc > R c)
out on top. (b) P(Tc < R)
Our two events of interest are T 5 can curl 7. Challenge: Consider the event F 5 is female. See
tongue and R 5 right thumb on top. if you can construct a Venn diagram with three
3. Make a Venn diagram that shows where each circles to classify the members of your class with
member of your class falls with respect to the respect to events T, R, and F.
346 CHAPTER 7 Probability: What Are the Chances?

Section 7.2 Summary

Probability is the study of chance behavior. A probability model describes chance
behavior by listing the possible outcomes in the sample space and giving the prob-
ability that each outcome occurs. An event is a subset of the possible outcomes in
a chance process. To find the probability that an event A happens, we can rely on
some basic probability rules:
• 0 # P1A2 # 1
• P(Ac) 5 1 2 P(A), where Ac is the complement of event A; that is, the event
that A does not happen
• P(S) 5 1, where S 5 the sample space
• If all outcomes in the sample space are equally likely,

P1A2 5
number of outcomes corresponding to event A
total number of outcomes in sample space
A two-way table or a Venn diagram can be used to display the sample space of
some chance process when we have two events of interest, say A and B. Two-way tables
and Venn diagrams can also be used to find probabilities involving events A and B, like
the union (A < B) and intersection (A > B). The event A < B (“A or B”) consists of all
outcomes in event A, event B, or both. The event A > B (“A and B”) consists of those
outcomes in both A and B. The general addition rule can be used to find P(A < B):
P(A or B) 5 P(A < B) 5 P(A) 1 P(B) 2 P(A > B)
If events A and B cannot happen together, we say that they are mutually exclusive, or
disjoint. For mutually exclusive events A and B, the general addition rule simplifies to
P(A or B) 5 P(A < B) 5 P(A) 1 P(B)

Section 7.2 7.45 Popular kids Researchers carried out a survey of fourth-, fifth-, and sixth-
Exercises grade students in Michigan. Students were asked whether good grades, athletic
ability, or being popular was most important to them. The two-way table below
summarizes the survey data.9
4th grade 5th grade 6th grade Total
Grades 49 50 69 168
Athletic 24 36 38 98
Popular 19 22 28 69
Total 92 108 135 335

(a) Who are the individuals? What variables are being measured?
(b) If we select a survey respondent at random, what’s the probability that we choose
• a sixth-grader?
• a student who rated good grades as most important?
• a sixth-grader who rated good grades as most important?
• a sixth-grader or a student who rated good grades as most important?
(c) How is the two-way table in this exercise different from the ones you saw in
earlier examples?
 7. 2 Probability Rules 347

7.46 TCNJ survey, I The 28 students in Mr. Starnes’s introductory statistics class
at The College of New Jersey (TCNJ) completed a brief survey. One of the questions
asked whether each student was right- or left-handed. The two-way table below
displays information on the gender and handedness of the students in the class.

Gender
Row
Handedness F M summary
L 3 1 4
R 18 6 24
Column summary 21 7 28

(a) Nationally, about 10% of people are left-handed. Are the students in Mr. Starnes’s
class similar to the nation as a whole? Justify your answer.
(b) What’s the probability that a randomly selected student from the class is

• male or right-handed?
• male and right-handed?

7.47 TCNJ survey, II Refer to Exercise 7.46. Construct a Venn diagram to display
the sample space for the chance process of choosing a student at random from
Mr. Starnes’s class. The events of interest are M 5 male and R 5 right-handed.
7.48 Mutually exclusive versus complementary For each part below, classify
the statement as true or false. Justify your answer.
(a) If one event is the complement of another event, then the two events are
mutually exclusive.
(b) If two events are mutually exclusive, then one event is the complement of the other.
7.49 Are you my (blood) type? Each of us has an ABO blood type, which
describes whether two characteristics called A and B are present. Every human
being has two blood type alleles (gene forms), one inherited from our mother and
one from our father. Each of these alleles can be A, B, or O. Which two we inherit
determines our blood type. Here is a table that shows what our blood type is for
each combination of two alleles:

Alleles inherited Blood type Alleles inherited Blood type

A and A A B and B B
A and B AB B and O B
A and O A O and O O

Children are equally likely to inherit either allele from their mother and also
equally likely to inherit either allele from their father.
(a) Mary and John both have alleles A and B. List the sample space that shows the
alleles that their next child could receive.
(b) Give the possible blood types that this child could have, along with the
probability for each blood type.
348 CHAPTER 7 Probability: What Are the Chances?

7.50 More on blood The table below shows the distribution of ABO blood types
in the United States.10

Blood type: O A B AB
Relative frequency: 44% 10% 4%

(a) The entry for type A blood is missing from the table. What is this value? Explain.
(b) Suppose we select one person from the United States at random. Find the prob-
ability that the person does not have type O blood. Show your method clearly.
7.51 Even more on blood A person’s blood type can be further classified
based on whether a certain substance known as a Rhesus antigen is present. For
example, a person with type B blood and the Rhesus antigen present is said to be
type B1 (B-positive). A person with type B blood but without the Rhesus antigen
present would be type B– (B-negative).
(a) People with blood type B1 can receive blood from anyone with blood type
O or B. What’s the probability that a randomly selected person from the United
States can donate blood to someone with blood type B1?
(b) In the United States, 84% of people have the Rhesus antigen present, 8.5% of
people have blood type B1 and 1.5% have blood type B–. Suppose we choose a U.S.
person at random and record his or her blood type. Construct a Venn diagram that
displays event B 5 has blood type B and event P 5 has the Rhesus antigen present.
7.52 Ask Marilyn again! In a Parade magazine column, Marilyn vos Savant was
asked the following question:
A woman and a man (who are unrelated) each have two children. At least one
of the woman’s children is a boy, and the man’s older child is a boy. Which is
more likely: that the man has 2 boys or that the woman has 2 boys?
What’s your answer? Explain your reasoning.

7.3 Conditional Probability and Independence

ACTIVITY 7.3
Weird dice
Materials: Four special six-sided dice provided by • Then the teacher must pick one of the three
your teacher remaining dice to roll.
In this Activity, a few lucky students will get the • In each round of the game, both players roll
chance to compete for a desirable reward from your their die. The player with the highest number
teacher. After explaining the rules of the game, your showing wins the round.
teacher will ask for volunteers. • The first player to win 3 rounds wins the
game.
Rules
• One student at a time may challenge the teacher When the competition is over, your teacher
in the “weird dice” game. will show you the number of spots on the six
• The student gets to pick any one of the four dice faces of each die from the last game that was
provided by the teacher to roll. played.
 7. 3 Conditional Probability and Independence 349

Questions 2. Find the probability that your teacher would win

1. Suppose your teacher’s chosen die and the final the round. Show your work.
player’s chosen die were rolled again. Use what
3. Working with your classmates, find the
you learned in the previous section to illustrate the
probability that your teacher would win if the two
sample space for this chance process.
competitors played another game.

What is conditional probability?

Let’s return to the college statistics class of Example 7.16 (page 336). Earlier, we used
Pierced the two-way table at left to find probabilities involving events A 5 has pierced ears and
ears? B 5 is male for a randomly selected student. Here is a summary of our previous results.
Gender Yes No Total • P(A) 5 103/178
• P(A > B) 5 P(A and B) 5 19/178
Male 19 71 90
Female 84 4 88 • P(A < B) 5 P(A or B) 5 174/178

Total 103 75 178

Since the probability that a randomly selected student is male is 90/178, we could
add P(B) 5 90/178 to the list. Now let’s turn our attention to some other interesting
probability questions.

Example 7.21 Getting acquainted with conditional

probability

(a) If we know that a randomly selected student has pierced ears, what is the
probability that the student is male? There are 103 students in the class with pierced
ears. We can restrict our attention to this group, since we are told that the chosen
student has pierced ears. Since there are 19 males among the 103 students with
pierced ears, the desired probability is
P(is male given has pierced ears) 5 19/103, or about 18.4%
(b) If we know that a randomly selected student is male, what’s the probability
that the student has pierced ears? This time, our attention is focused on the males
in the class. Since only 19 of the 90 males in the class have pierced ears,
P(has pierced ears given is male) 5 19/90, or about 21.1%
In this class, it’s slightly more likely for a male to have pierced ears than it is for
someone with pierced ears to be a male.

A probability like “the probability that a randomly selected student is male

given that the student has pierced ears” is known as a conditional probability. The
name comes from the fact that we are trying to find the probability that one event
will occur under the condition that some other event is already known to have oc-
curred. We often use the phrase “given that” to signal the condition. There’s even
a special notation to indicate a conditional probability. In the example above, we
would write P(is male | has pierced ears), where the | means “given that” or “under
the condition that.” Since we already defined the events A 5 has pierced ears and
B 5 is male, we could write the conditional probability as P(B|A).
350 CHAPTER 7 Probability: What Are the Chances?

Conditional probability

The probability that one event happens under the condition that an-
other event is already known to have happened is called a conditional
probability. Suppose we know that event A has happened. Then the
probability that event B happens given that event A has happened is
denoted by P(B|A).

With this new notation available, we can restate the results of Example 7.21 as

P(B|A) 5 19/103 and P(A|B) 5 19/90

Notice that the conditional probability P(B|A) 5 19/103 is very different from the
unconditional probability P(B) 5 90/178. About half of the students in the class are
male, but far fewer than half of the students with pierced ears are male. Also note that
P(A|B) 5 19/90 is much smaller than P(A) 5 103/178. Over half of the students in
the class have pierced ears, but only 19 out of 90 males in the class have pierced ears.

Conditional probability and independence

To get a better feel for how conditional probability works, let’s return to two famil-
iar settings involving chance behavior—tossing coins and rolling dice.

Example 7.22 Conditional probability: coins and dice

Suppose you toss a fair coin twice. Define events A 5 first toss is a head and B 5
second toss is a head. We have already established that P(A) 5 1/2 and P(B) 5 1/2.
What’s P(B|A)? It’s the conditional probability that the second toss is a head given
that the first toss was a head. The coin has no memory, so P(B|A) 5 1/2. In this
case, P(B|A) 5 P(B). The condition that the first toss was a head does not affect the
probability that the second toss is a head.
Let’s contrast the coin toss scenario with the chance process of rolling two fair,
six-sided dice—one red and one green. This time, we’ll define events R 5 red die
shows a 4 and S 5 sum of two dice is 11. Hopefully you’ll agree that P(R) 5 1/6.
What’s P(S)? From the sample space of Example 7.12 (page 331), it’s 3/36. What’s
the conditional probability P(S|R) that the sum of the two dice is 11 given that the
red die shows a 4? If the red die shows a 4, then it’s impossible to get a sum of 11 on
the two dice. So P(S|R) 5 0. This conditional probability is very different from the
unconditional probability P(S) of getting a sum of 11.

In Example 7.22, we observed that P(B|A) 5 P(B) for events A and B in the coin
toss setting. For the dice-rolling scenario, however, P(S|R) 2 P(S). When knowledge
that one event has happened does not change the likelihood that another event will
happen, we say that the two events are independent.
 7. 3 Conditional Probability and Independence 351

Independent events

Two events A and B are independent if the occurrence of one event has
no effect on the chance that the other event will occur. In other words,
events A and B are independent if P(A|B) 5 P(A) and P(B|A) 5 P(B).
If two events are not independent, we sometimes say that they are
dependent.

There’s something else interesting about events R and S in Example 7.22. See if you
can figure out what it is. (Here’s a hint: what’s the probability P(R and S) of both events
occurring?) By now, you’ve probably discovered that events R and S can’t happen to-
gether. If event R happens and the red die shows a 4, the sum can’t be 11. Likewise, if
event S happens and the sum is 11, then the red die can’t show a 4. That is, events R and
S are mutually exclusive. Two mutually exclusive events can never be independent,
because if one event happens, the other event is guaranteed not to happen.
We have one other important caution to offer. Don’t confuse independent trials
in a chance process with independent events. Both coin tosses and dice rolls result in
independent trials. But in either setting, you can define events A and B that are or
aren’t independent.

Exercises 7.53 Sampling senators, III The two-way table at left describes the members of
the U.S. Senate in 2008. Suppose we select a senator at random. Consider events
D 5 is a Democrat and F 5 is female.
Male Female (a) Find P(D|F). Explain what this value means.
Democrats 40 11 (b) Find P(F|D). Explain what this value means.
Republicans 44 5 (c) Are events D and F mutually exclusive? Justify your answer.
(d) Are events D and F independent? Justify your answer.
7.54 More breakfast eaters The two-way table below describes the 595 students
who responded to a school survey about eating breakfast. Suppose we select a
student at random. Consider events B 5 eats breakfast regularly and M 5 is male.
Male Female Total
Eat breakfast regularly 190 110 300
Don’t eat breakfast regularly 130 165 295
Total 320 275 595

(a) Find P(B|M). Explain what this value means.

(b) Find P(M|B). Explain what this value means.
(c) Are events B and M mutually exclusive? Justify your answer.
(d) Are events B and M independent? Justify your answer.
7.55 Rolling dice Suppose you roll two fair, six-sided dice—one red and one
green. Define two independent events A and B related to this chance process.
Show that your two events are, in fact, independent.
352 CHAPTER 7 Probability: What Are the Chances?

7.56 Tossing coins Suppose you toss a fair coin twice. Define two events A and B
related to this chance process that are not independent and are not disjoint.
7.57 Teachers and advanced degrees Select an adult at random. Let A 5 person
has earned an advanced degree (masters or PhD) and T 5 person’s career is teaching.
Rank the following probabilities from smallest to largest. Justify your answer.

P(A) P(T) P(A|T) P(T|A)

7.58 Mutually exclusive versus independent Decide whether the following
statement is true or false: two events that are not mutually exclusive must be
independent. Justify your answer.

Tree diagrams and conditional probability

If you toss a fair coin twice, what’s the probability that you get two heads? By the
reasoning of Section 7.1, it’s the proportion of times you would get H-H if you
tossed the coin two times over and over and over again. We could perform a simu-
lation to estimate the true probability. Using the methods of Section 7.2, we would
take a different approach to answering the question. We begin by listing the sample
space for this chance process:
H-H H-T T-H T-T
Since the result of one coin toss is independent from the result of any other toss, the
4 outcomes in the sample space are equally likely. So P(two heads) 5 P(H-H) 5
1/4. There’s another way to model chance behavior that involves a sequence of out-
comes, like tossing coins. It’s called a tree diagram.
Figure 7.10 shows a tree diagram for the coin-tossing scenario. After the first
toss, there are only two possible results—heads (H) or tails (T)—each with prob-
ability 1/2. These two possibilities are shown on the leftmost branches of the tree
diagram. From there, we toss a second time, which results in either an H or a T,
again with equal probability. This puts us on one of the four “nodes” beneath the
“Second toss” label on the tree diagram after two tosses.

Second toss

P 1 two heads 2 5
H 1#1 1
1 5
2 2 4
First toss 2
H
1
2 1 T
2
Toss a coin
1 H
1 2
2 T

1
2 T

Figure 7.10 A tree diagram displaying the sample space of tossing two coins.
 7. 3 Conditional Probability and Independence 353

What’s the probability of getting “heads, heads”? We already know it’s 1/4. Do
you see how we can obtain that value from the tree diagram? If we multiply prob-
abilities along the top branches, we get

1#1 1
P(two heads) 5 5
2 2 4

This same general strategy will work even when trials are not independent, as the
following example illustrates.

Example 7.23 Looking for hearts

If we shuffle a standard deck and deal the top two cards, what’s the probability that
we get two hearts? We know the probability that the first card dealt is a heart is
13/52. What about the probability that the second card is a heart? That depends on
whether the first card was a heart. In this case, the trials are not independent. This is
starting to sound complicated! It’s time to call on our new friend, the tree diagram.
Figure 7.11(a) shows a tree diagram for this setting with P(heart) 5 13/52 for
the first card. What’s the probability of not getting a heart on the first card? By the
complement rule, it’s

13 39
P(not heart) 5 1 2 P(heart) 5 1 2 5
52 52

We have added this probability on the bottom branch of Figure 7.11(a).

Second card
♥
First card

♥
13 not
52 ♥

Deal a card
♥
39 not
52 ♥

not
♥

Figure 7.11(a) A tree diagram for the chance process of dealing two cards and looking
for hearts. We have given the probabilities for the first card dealt.

If the first card we draw is a heart, then there are 12 hearts left among the 51 re-
maining cards. So the conditional probability of getting a heart on the second card
given that we got a heart on the first card is 12/51. Using the complement rule again,
354 CHAPTER 7 Probability: What Are the Chances?

the probability that the second card is not a heart given that the first card was a heart is
1 2 12/51 5 39/51.
What if the first card we draw isn’t a heart? Then there are still 13 hearts among
the 51 remaining cards. That is, the probability of getting a heart on the second card
given that we did not get a heart on the first card is 13/51. By the complement rule,

P(no heart on second card given no heart on first card) 5

13 38
12 5
51 51

Figure 7.11(b) shows the completed tree diagram. To get two hearts, we have to
get a heart on the first card and a heart on the second card. Following along the top
branches of the tree, we see that the probability is

P(two hearts) 5 P(heart on first card and heart on second card)

13 # 12
5 5 0.059
52 51

Second card
♥ P 1 two hearts 2 5
12
13 # 12
First card 51 5 0.059
52 51
♥
13 not
52 39 ♥
51
Deal a card
13 ♥
39 not 51
52 ♥

38 not
51 ♥

Figure 7.11(b) The completed tree diagram for the card-dealing scenario. We multiply
along the branches to get P(two hearts).

Being dealt two hearts is a pretty unusual occurrence!

The tree diagram method can be used to find the probability P(A and B)
that two events happen together. In Example 7.23, A 5 heart on first card and
B 5 heart on second card. To calculate P(A and B), we multiplied P(heart on first
card) by P(heart on second card given heart on first card). In symbols, we write

P(A > B) 5 P(A) ˙ P(B|A)

where P(B|A) is the conditional probability that event B will happen given that
event A has happened. This formula is known as the general multiplication rule.
 7. 3 Conditional Probability and Independence 355

General multiplication rule

The probability that events A and B both occur can be found using the
general multiplication rule
P(A > B) 5 P(A) ˙ P(B|A)
where P(B|A) is the conditional probability that event B occurs given
that event A has already occurred.

If we rearrange the terms in the general multiplication rule, we can get a

formula for the conditional probability P(B|A). Start with the general multipli-
cation rule

P(A > B) 5 P(A) ˙ P(B|A)

Divide both sides of the equation by P(A). This yields

P1A > B2
5 P1B0A2
P1A2
or

P 1 A >B 2
P1B0A2 5
P1A2

Conditional probability formula

To find the conditional probability P(B|A), use the formula

P 1 A >B 2
P1B0A2 5
P1A2

Let’s illustrate the use of this formula with a familiar example.

Example 7.24 Who reads the paper, continued

In Example 7.20 (page 342), we classified the residents of a large apartment complex based
on the events A 5 reads USA Today and B 5 reads the New York Times. The completed
Venn diagram is reproduced in Figure 7.12 (on the next page). What’s the probability that
a randomly selected resident who reads USA Today also reads the New York Times?
Since we’re given that the randomly chosen resident reads USA Today, we want
to find P(reads New York Times | reads USA Today), or P(B|A). By the conditional
probability formula,

P 1 A >B 2
P1B0A2 5
P1A2
356 CHAPTER 7 Probability: What Are the Chances?

Reads both
papers

A = reads B = reads
USA Today New York Times
30% 5% 15%

Figure 7.12 The com- 50%

pleted Venn diagram from
Example 7.20. Reads neither paper

Since P(A > B) 5 0.05 and P(A) 5 0.35, we have

P1B0A2 5
0.05
5 0.14
0.35
There’s a 14% chance that a randomly selected resident who reads USA Today also
reads the New York Times.

What about the special case when events A and B are independent? In the
toss two coins scenario of Example 7.22, A 5 head on first toss and B 5 head on
second toss. By the general multiplication rule,

P(A > B) 5 P(A) ˙ P(B|A)

5 P(head on first toss) ˙ P(head on second toss | head on first toss)
1 1 1
5 # 5
2 2 4
But for the coin-toss setting, the chance of getting a head on the second toss given
that we got a head on the first toss is just 1/2, because the coin has no memory.
That is, P(B|A) 5 P(B). In other words, A and B are independent events. We can
simplify the general multiplication rule in the case of independent events A and B
as follows:

P(A > B) 5 P(A) # P(B|A)

5 P(A) # P(B) since P(B|A) 5 P(B) for independent events

This result is known as the multiplication rule for independent events.

Multiplication rule for independent events

If A and B are independent events, then the probability that A and B

both occur is
P(A > B) 5 P(A) ˙ P(B)

Note that this rule applies only to independent events.

 7. 3 Conditional Probability and Independence 357

Exercises 7.59 Toss four more Imagine that you toss a fair coin 4 times.
(a) Draw a tree diagram to represent this chance process.
(b) Find the probability that you get at least one head. Explain your method.
7.60 Not just hearts Refer to Example 7.23 (page 353). Find the probability that
at least one of the two cards that are dealt is a heart. Show your method clearly.
7.61 Looking for hearts again Refer to Example 7.23 (page 353). Suppose we draw
one card from the deck, look at it, and then return it to the deck before we draw a
second card. Define events A 5 first card is a heart and B 5 second card is a heart.
(a) Draw a tree diagram to represent this chance process. Be sure to label all the
branches of the tree with the appropriate probabilities.
(b) Find the probability that both cards are hearts. Show your work.
(c) Are events A and B independent? Justify your answer.
7.62 Monopoly In the game of Monopoly, a player rolls two six-sided dice on
each turn. If a player is in jail, he or she must roll “doubles” (both dice show the
same number) in order to get out of jail. If a player in jail does not roll doubles on
three consecutive turns, he or she must pay $50 to get out of jail.
(a) Find the probability that a player rolls doubles on a turn. Show your method.
(b) John is in jail. Find the probability that John has to pay $50 to get out of jail
because he doesn’t roll doubles on three consecutive turns. Show your work.
7.63 MySpace versus Facebook, II A recent survey suggests that 85% of college
students have posted a profile on Facebook, 54% use MySpace regularly, and 42%
do both. Suppose we select a college student at random and learn that the student
has a profile on Facebook. Find the probability that the student uses MySpace
regularly. Show your work.
7.64 More computers at Princeton An October 2007 census revealed that 40% of
Princeton students primarily used Macintosh computers (MACs). The rest primarily
used PCs. At the time of the census, 67% of Princeton students were undergraduates.
The rest were graduate students. According to the census, 23% of Princeton’s
graduate students said that they used PCs as their primary computers. Suppose that
we select a Princeton student at random and learn that the student primarily uses a
MAC. Find the probability that this person is a graduate student. Show your work.

Applications of conditional probability

King Features Syndicate 2008

358 CHAPTER 7 Probability: What Are the Chances?

Conditional probability has important applications in the areas of risk manage-

ment, drug and disease testing, and quality control in manufacturing. Let’s look at
a couple of examples that show how an understanding of conditional probability
can help people make informed decisions.

Example 7.25 Medical decisions, risks, and probability

Morris’s kidneys are failing and he is awaiting a kidney transplant. His doctor gives
him the following information for patients in his condition: 90% survive the trans-
plant and 10% die. The transplant succeeds in 60% of those who survive, and the
other 40% must return to kidney dialysis. The proportions who survive five years are
70% for those with a new kidney and 50% for those who return to dialysis. Morris
wants to know the probability that he will survive for five years.
The tree diagram in Figure 7.13 organizes this information to give a probability
model in graphical form. The tree shows the three stages and the possible outcomes and
probabilities at each stage. Each path through the tree leads to either survival for five
years or to death in less than five years.

e
r viv
Su
0.7
y
ne Di
kid e
w
Ne 0.3
0.6

ive Di ive
rv aly rv
Su 0.4
sis Su
0.9 0.5

Di Di
e e
0.5
0.1
Figure 7.13 A tree dia-
gram showing the possible
outcomes and probabilities Stage 1: Stage 2: Stage 3:
for patients in Morris’s Transplant Transplant Survive
condition. succeeds? 5 years?

How can we calculate the probability that a patient in Morris’s condition will
survive for five years? In the tree diagram, there are two possible “paths” that lead to a
patient surviving for five years. The first such path would require a patient to survive
the transplant, have the new kidney function successfully, and then survive for five
years with the new kidney. What’s the probability that this sequence of events hap-
pens? There’s a 90% chance that the patient survives the transplant. For those who
survive the transplant, 60% have the new kidney function properly. Of those who
survive the transplant and have the new kidney function properly, 70% will survive.
So the probability of surviving for five years in this way is (0.9)(0.6)(0.7) 5 0.378.
The second path would require a patient to survive the transplant, be forced to
return to dialysis because the new kidney didn’t function properly, and then live for
 7. 3 Conditional Probability and Independence 359

five years. By multiplying along the branches of the tree, we can calculate this prob-
ability as (0.9)(0.4)(0.5) 5 0.18.
Since either path would allow a patient to survive for five years, the desired
probability is
0.378 1 0.18 5 0.558
That is, a patient in Morris’s condition has about a 56% chance of surviving for five years.

Lately, we’ve been hearing more in the news about athletes who have tested posi-
tive for performance-enhancing drugs. From Major League Baseball, to the Tour de
France, to the Olympics, we have learned about athletes who took banned substances.
Sports at all levels now have programs in place to test athletes’ urine or blood. There’s
an important question about drug testing that probability can help answer: if an ath-
lete tests positive for a banned substance, did the athlete necessarily attempt to cheat?

Example 7.26 Athletes and drug testing

Over 10,000 athletes competed in the 2008 Olympic Games in Beijing.
The International Olympic Committee wanted to ensure that the compe-
tition was as fair as possible. So they administered more than 5000 drug
tests to athletes. All medal winners were tested, as well as other randomly
selected competitors.
Suppose that 2% of athletes had actually taken (banned) drugs. No drug
test is perfect. Sometimes the test indicates that an athlete took drugs when
the athlete actually didn’t. We call this a false positive result. Other times,
© Paul J. Sutton/PCN/PCN/Corbis
the drug test indicates that an athlete is “clean” when the athlete actually
took drugs. This is called a false negative result. Suppose that the testing procedure
used at the Olympics has a false positive rate of 1% and a false negative rate of 0.5%.
What’s the probability that an athlete who tests positive actually took drugs?
The tree diagram in Figure 7.14 summarizes the situation. Since we assumed
that 2% of all Olympic athletes took drugs, that means 98% didn’t. Of those who
took drugs, 0.5%, or 0.005, would “get lucky” and test negative. The remaining

Test result


0.995
Took drugs?
Y

0.02 0.005 
Randomly choose
Figure 7.14 A tree Olympic athlete
diagram showing whether 
0.01
or not Olympic athletes 0.98 N
took drugs and the likeli-
hood of getting a positive
or negative test result from 0.99 
a drug test.
360 CHAPTER 7 Probability: What Are the Chances?

99.5%, or 0.995, would test positive. Among the athletes who didn’t take drugs,
an “unlucky” 1%, or 0.01, would test positive. The remaining 99%, or 0.99, would
(correctly) test negative.
Let’s define events B 5 took drugs and A 5 positive test result. We want to find
P(B|A). By the conditional probability formula,

P 1 A >B 2
P1B0A2 5
P1A2

To find P(A > B), we note that this is the probability that a randomly selected
athlete took drugs and got a positive test result. Multiplying along the Y and 
branches of the tree,

P 1 A >B 2 5 1 0.02 2 1 0.995 2 5 0.0199

To find P(A), we need to calculate the probability that a randomly selected ath-
lete gets a positive test result. There are two ways this can happen: (1) if the
athlete took drugs and the test result is positive, and (2) if the athlete didn’t
take drugs, but the test gives a false positive. From the tree diagram, the desired
probability is

P 1 A 2 5 1 0.02 2 1 0.995 2 1 1 0.98 2 1 0.01 2 5 0.0297

Using these two results, we can find the conditional probability:

P 1 A >B 2
P1B0A2 5
0.0199
5 5 0.67
P1A2 0.0297

That is, 67% of athletes who test positive actually took drugs.

Are you surprised by the final result of Example 7.26? Most people are. Some-
times a two-way table that includes counts is more convincing. To make calculations
simple, we’ll suppose that there were exactly 10,000 athletes at the 2008 Olympics,
and that exactly 200 took banned substances. How many of those 200 would test
positive? 99.5% of 200, or 199 of them. That leaves 1 who would test negative. How
many of the 9800 athletes who didn’t take drugs would get a false positive test result?
1% of them, or (0.01)(9800) 5 98 athletes. The remaining 9800 2 98 5 9702 would
test negative. In total, 199 1 98 5 297 athletes would have positive test results and
1 1 9702 5 9703 athletes would have negative test results. This information is sum-
marized in the two-way table below.

Took drugs?
Positive test? Yes No Total
Yes 199 98 297
No 1 9702 9,703
Total 200 9800 10,000
 7. 3 Conditional Probability and Independence 361

Given that a randomly selected athlete has tested positive, we can see from the
two-way table that the conditional probability P(took drugs | positive test) 5
199/297 5 0.67.
Now you can see why most drug-testing programs require an athlete to submit
two urine or drug samples at each test. If one sample tests positive, we can’t be sure
that the athlete took drugs. If both samples test positive, however, we can be pretty
sure that the athlete took a banned substance.

Exercises 7.65 Fill ’er up! In June 2008, 88% of automobile drivers filled their vehicles with
regular gasoline, 2% purchased midgrade gas, and 10% bought premium gas.11
Of those who bought regular gas, 28% paid with a credit card. Of customers who
bought midgrade and premium gas, 34% and 42%, respectively, paid with a credit
card. Suppose we select a customer at random.
(a) Construct a tree diagram to represent this situation.
(b) Find the probability that the customer paid with a credit card. Show your work.
(c) Given that the customer paid with a credit card, find the probability that she or
he bought premium gas.
7.66 Desktop or laptop? A computer company makes desktop and laptop
computers at factories in three states—California, Texas, and New York. The
California factory produces 40% of the company’s computers, the Texas factory
makes 25%, and the remaining 35% are manufactured in New York. Of the
computers made in California, 75% are laptops. Of those made in Texas and New
York, 70% and 50%, respectively, are laptops. All computers are first shipped to a
distribution center in Missouri before being sent out to stores. Suppose we select a
computer at random from the distribution center.12
(a) Construct a tree diagram to represent this situation.
(b) Find the probability that the computer is a laptop. Show your work.
(c) Given that the computer is a laptop, find the probability that it came from the
Texas factory.
7.67 Assessing risk Refer to Example 7.25 (page 358). Use the probabilities
provided to make a table that shows the status of 1000 randomly selected patients
with Morris’s condition.
7.68 False positives and negatives Which is a more serious error in each of the
following situations—a false positive result or a false negative result? Justify your
answer.
(a) Testing athletes for performance-enhancing drugs.
(b) Testing people for a life-threatening disease.
7.69 Testing for AIDS, I The ELISA test can help detect whether people
have the AIDS virus (HIV). As with any medical test, ELISA can give false
positive and false negative results. From experience, medical researchers
estimate that the ELISA test has a 0.2% false positive rate and a 0.1% false
negative rate. Suppose we use the ELISA test in an area where 5% of the
population has HIV.
362 CHAPTER 7 Probability: What Are the Chances?

(a) Draw a tree diagram that models this situation.

(b) Find the probability that a randomly selected person from this area would test
positive. Show your work.
(c) Given that a randomly chosen person has tested positive, what’s the probability
that the person actually has HIV? Show your work.
7.70 Testing for AIDS, II Refer to the previous exercise. Suppose that we use
the ELISA test in an area where only 0.1% of the population has HIV. Repeat parts
(a), (b), and (c). Are you surprised?

ws
Ne
ws
Ne

APPLICATION 7.3 B]
RO
g
<
Se
a

Baby hearing screening

According to the National Campaign for Hearing Baby’s hearing
Health, deafness is the most common birth defect. Test result Loss Normal
Experts estimate that 3 out of every 1000 babies
are born with some kind of hearing loss. Early Loss 54 6
Normal 4 36
identification of hearing loss through a screening
test shortly after birth and subsequent interven-
tion can have a significant impact on a child’s QUESTIONS
cognitive development as well as financial conse- 1. Estimate the probability that this new device
quences. Thus, universal newborn hearing screen- will show a hearing loss for a baby who actually has
ing (UNHS) calls for high-performing, accurate, hearing loss. Show your work.
and inexpensive testing equipment. The screening 2. Estimate the probability that this new device will
tests work by introducing a sound into a baby’s ear show normal hearing for a baby who actually has
and then measuring either the response of the ear’s normal hearing. Show your work.
internal mechanisms or the electrical activity of the
auditory portion of the brain. Just because a baby 3. Estimate the chance that a baby with a hearing
fails those tests, however, does not mean there is a loss will pass the hearing test (test normal) using
hearing problem. this newer miniature screening device. Show your
Studies conducted at the University of Utah work.
were undertaken to assess the adequacy of a new 4. Suppose we choose a baby at random from
miniature screening device called the Handtronix- all newborns and have the baby’s hearing tested
OtoScreener. The accuracy of this new device was using the Handtronix-OtoScreener. If the test
judged using as “truth” the results found using shows that the baby has a hearing loss, what’s the
standard equipment.13 Results are shown in the estimated probability that the baby really has a
following table: hearing loss?

Section 7.3 Summary

If one event has happened, the chance that another event will happen is a conditional
probability. The notation P(B|A) represents the probability that event B occurs
given that event A has occurred. When the chance that event B occurs is not af-
fected by whether event A occurs, we say that events A and B are independent. For
 7. 3 Conditional Probability and Independence 363

independent events A and B, P(B|A) 5 P(B) and P(A|B) 5 P(A). If two events A
and B are mutually exclusive, they cannot be independent.
When chance behavior involves a sequence of outcomes, a tree diagram can
be used to describe the sample space. Tree diagrams can also help in finding the
probability that two or more events occur together. We simply multiply along the
branches that correspond to the outcomes of interest.
The general multiplication rule states that the probability of events A and B
occurring together is

P 1 A >B 2 5 P 1 A 2 # P 1 B 0 A 2

Dividing both sides of the equation by P(A), we get the conditional probability
formula
P 1 A >B 2
P1B0A 2 5
P1A2

In the special case of independent events, the multiplication rule becomes

P1A >B2 5 P1A2 # P1B2

Section 7.3 7.71 Tall people and basketball players Select an adult at random. Let T 5 per-
Exercises son is over 6 feet tall and B 5 person is a professional basketball player. Rank the
following probabilities from smallest to largest. Justify your answer.

P(T) P(B) P(T|B) P(B|T)

7.72 TCNJ survey, III Exercise 7.46 (page 347) described the results of a class
survey. Here is the two-way table that shows the gender and handedness of the
students in the class:
Gender
Handedness F M Row summary
L 3 1 4
R 18 6 24
Column summary 21 7 28

(a) What percent of the males in the class are right-handed?

(b) What percent of the right-handed people in the class are male?
(c) Are the events “student is a male” and “student is right-handed” independent?
Justify your answer.
7.73 The chevalier’s problem In the early 1700s, French gamblers played a game
in which they bet on getting at least one “1” when a fair, six-sided die was rolled
four times.
(a) Draw a tree diagram that shows the sample space of this chance process.
(b) Find the probability of interest. Show your work.
7.74 The chevalier’s other problem To increase interest among French gamblers,
another game was devised in which players bet on getting at least 1 set of “double
364 CHAPTER 7 Probability: What Are the Chances?

1s” when rolling a pair of fair, six-sided dice 24 times. The chevalier de Mere, a
frequent gambler, believed that the probability of getting the desired outcome in
this game was the same as the probability in Exercise 7.73(b). He bet accordingly
and lost a lot of money. Show the chevalier that it’s less likely to get at least one set
of “double 1s” on 24 rolls of a pair of dice than it is to get at least one “1” on four
rolls of a single die.
7.75 Cats and dogs In an elementary school classroom, there are 40 students.
For every student, we record whether they have a cat and whether they have a dog.
Thirty students say that they have a dog, and 20 students say that they have a cat.
If possible, make a two-way table in which the events D  has a dog and C  has
a cat are independent if we select a student at random from the class.
7.76 Testing the test Are false positives too common in some medical tests?
Researchers conducted an experiment involving 250 patients with a medical
condition and 750 other patients who did not have the medical condition. The
medical technicians who were reading the test results were unaware that they were
subjects in an experiment.
(a) Technicians correctly identified 240 of the 250 patients with the condition.
They also identified 50 of the healthy patients as having the condition. What were
the false positive and false negative rates for the test?
(b) Given that a patient got a positive test result, what is the probability that the
patient actually had the medical condition? Show your work.
7.77 The birthday problem If 30 unrelated people are in a room at the same
time, what’s the probability that at least 2 of them have the same birthday (month
and day)? Make a guess before you perform any calculations.
(a) Let’s start with a simpler problem. If 2 unrelated people are in a room, what’s
the probability that they have different birthdays? What’s the probability that they
have the same birthday?
(b) If 3 unrelated people are in a room, what’s the probability that all 3 have different
birthdays? What’s the probability that at least 2 of them have the same birthday?
(c) If 30 unrelated people are in a room, what’s the probability that all 30 have different
birthdays? What’s the probability that at least 2 of them have the same birthday?
7.78 Activity 7.3 follow-up Return to Activity 7.3 (page 348). Assume that your
teacher has probability 2/3 of winning any individual round.
(a) List the sample space for the weird dice game. Find the probability for each
possible outcome.
(b) Find the probability that the teacher wins the next game. Show your work.

CHAPTER 7 REVIEW
Some phenomena are random. Although their individual outcomes are unpredict-
able, there is a regular pattern in the long run. Using gambling devices and taking
an SRS are examples of random phenomena. Probability gives us a language to de-
scribe randomness. To calculate the probability of a complicated event without using
complicated math, we use random digits to simulate many repetitions. Section 7.1
discusses randomness and simulation.
When randomness is present, probability answers the question “How often in
the long run?” Probability models assign probabilities to outcomes. Any such model
 Chapter 7 Review Exercises 365

must obey the rules of probability. Section 7.2 presents these rules, along with two
useful tools for describing outcomes of chance behavior—Venn diagrams and two-
way tables.
In some cases, we have partial information about the result of a chance process.
This leads to the study of conditional probability in Section 7.3. Tree diagrams help
organize our thinking about conditional probability in situations that involve a se-
quence of outcomes.

Here is a review list of the most important skills you should have developed
from your study of this chapter.
A. RANDOMNESS, PROBABILITY, AND SIMULATION
1. Recognize that some phenomena are random. Probability describes the long-
run regularity of random phenomena.
2. Understand the idea of the probability of an event as the proportion of times
the event occurs in very many repetitions of a random phenomenon. Use the idea
of probability as long-run proportion to think about probability.
3. Recognize that short runs of random phenomena do not display the regularity
described by probability. Accept that randomness is unpredictable in the short
run, and avoid seeking cause-and-effect explanations for random occurrences.

B. SIMULATION
1. Design a simulation using random digits to model chance behavior.
2. Estimate a probability by repeating a simulation many times.

C. PROBABILITY RULES
1. When probabilities are assigned to individual outcomes, find the probability of
an event by adding the probabilities of the outcomes that make it up.
2. Use probability rules to find the probabilities of events that are formed from
other events, including unions, intersections, complements, and conditional
probabilities.

D. PROBABILITY MODELS
1. Use Venn diagrams, two-way tables, and tree diagrams to model chance behavior.
2. Compute probabilities using information provided in Venn diagrams, two-way
tables, and tree diagrams.

CHAPTER 7 REVIEW EXERCISES

7.79 A dice game Your teacher has invented another “fair” dice game to play. Here’s
how it works. Your teacher will roll one fair eight-sided die, and you will roll a fair six-
sided die. Each player rolls once, and the winner is the person with a higher number.
In case of a tie, neither player wins.
(a) Diagram the sample space illustrating the possible outcomes on a single turn.
(b) Let A be the event “your teacher wins the first turn of the game.” Find P(A).
(c) Let B be the event “you get a 3 on your first roll.” Find P(A < B).
(d) Are events A and B independent? Justify your answer.
366 CHAPTER 7 Probability: What Are the Chances?

7.80 Sandblasters 2007 Eight teams of the world’s

best sand sculptors gathered near San Diego, California,
for Sandblasters 2007: The Extreme Sand Sculpting
Competition. At five points during the competition,
a randomly selected team’s sculpture was blown up.
That team was then forced to build a new sculpture
from scratch. No team could be selected for destruc-
tion more than once. Last year’s winning team did not
have a sculpture blown up either last year or this year.
Other teams were suspicious. Should they be? Use the
methods of this chapter to justify your answer. (Team
Sanding Ovation won the contest with their “zipper-
head” sculpture.)
Audrey Ustuzhanin/FeaturePics
7.81 Class rank Choose a college student at random
and ask his or her class rank in high school. Probabilities for the outcomes are

Class rank: Top 10% Top quarter but Top half but Bottom half
not top 10% not top quarter
Probability: 0.3 0.3 0.3 ?

(a) What must be the probability that a randomly chosen student was in the bottom
half of his or her high school class?
(b) To simulate the class standing of randomly chosen students, how would you as-
sign digits to represent the four possible outcomes listed?
7.82 The girls Suppose that about 48% of all infants are girls. The maternity ward of
a large hospital handles 1000 deliveries per year. A smaller hospital in the same city
records about 50 deliveries per year. At which hospital is it more likely that between
46% and 50% of the babies born there this year will be girls? Explain briefly.
7.83 Who smokes? The question “Do you smoke?” was asked of a random sample
of 100 people. Results are shown in the table below. Suppose we select one of these
people at random.
Smoke?
Gender Yes No Total
Male 19 41 60
Female 12 28 40
Total 31 69 100

(a) What’s the probability that the person smokes?

(b) Given that the person is male, what’s the probability that the person smokes?
(c) Are the two events “smokes” and “is male” independent? Why or why not?
7.84 More smokers Draw a Venn diagram to represent the sample space of
Exercise 7.83.
7.85 Challenger disaster On January 28, 1986, Space Shuttle Challenger exploded
on takeoff. All seven crew members were killed. Following the disaster, scientists and
statisticians helped analyze what went wrong. They determined that the failure of
 Chapter 7 Review Exercises 367

o-ring joints in the shuttle’s booster rockets was to blame. Under the cold conditions
that day, experts estimated that the probability that an individual o-ring joint would
function properly was 0.977. But there were six of these o-ring joints, and all six had
to function properly for the shuttle to launch safely. Assuming that o-ring joints suc-
ceed or fail independently, find the probability that the shuttle would launch safely
under similar conditions. Show your work.
7.86 Looking for metal A boy uses a homemade metal detector to look for valu-
able metal objects on a beach. The machine isn’t perfect—it identifies only 98% of
the metal objects over which it passes, and it identifies 4% of the nonmetallic objects
over which it passes. Suppose that 25% of the objects that the machine passes over are
metal. Find the probability that the boy has found a metal object when he receives a
signal from the machine. Show your work.
7.87 Random digits Suppose you instruct your calculator to choose a random digit
from 0 to 9. Define two events A and B related to this chance process that are mutu-
ally exclusive but not complements of each other.
7.88 Drug use in baseball On December 13, 2007, former senator George Mitchell
released an extensive report on the use of performance-enhancing drugs in Major
League Baseball. He even included a list of players who had been using such drugs.
One question of interest is whether performance-enhancing drugs really help. Let’s
examine the performance of 19 pitchers who were reported to have used drugs. In the
season when they first started using, 14 of the 19 showed improvement over the prior
year. Is this convincing evidence that using drugs helped improve their performance?
Design and carry out a simulation to help you answer this question. Assume that each
pitcher had probability 0.5 of showing improvement without taking drugs.
This page intentionally left blank
 CHAPTER 8

Probability Models
8.1 Probability Distributions
8.2 Counting and Probability
8.3 Binomial Distributions

The house edge, and the real

© Stewart Cohen/age fotostock

house edge
If you decide to gamble, you care about how often you’ll
win. The probability of winning tells you what proportion of
a large number of bets will be winners. You care even more
about how much you’ll win, because winning a lot is better
than winning a little. If you play a game with a 50% chance of winning $1, in the
long run you’ll win half the time, and average 50 cents in winnings per bet. A 10%
chance of winning $100 does better—now you win only 1 time in 10 in the long
run, but you average $10 in winnings per bet. Those “average winnings per bet”
numbers are expected values. Like probabilities, expected values tell us only what
will happen on the average in very many bets.
It’s no secret that the expected values of games of chance are set so that
the “house” wins on the average in the long run. In roulette, for example, the
expected value of a $1 bet is $0.947. The house keeps the other 5.3 cents per
dollar bet. Because the house plays hundreds of thousands of times each month,
it will keep almost exactly 5.3 cents for every dollar bet.
What is a secret, at least to naive gamblers, is that in the real world a casino
does much better than expected values suggest. In fact, casinos keep a bit over
20% of the money gamblers spend on roulette chips. That’s because players who
win keep on playing. Think of a player who gets back exactly 95% of each dollar
bet. After one bet, he has 95 cents; after two bets, he has 95% of that, or 90.25
cents; after three bets, he has 95% of that, or 85.7 cents. The longer he keeps
recycling his original dollar, the more of it the casino keeps. Real gamblers don’t
get a fixed percent back on each bet, but even the luckiest will lose his stake
if he plays long enough. The casino keeps 5.3 cents of every dollar bet but 20
cents of every dollar that walks in the door.1

369
370 CHAPTER 8 Probability Models

8.1 Probability Distributions

ACTIVITY 8.1A
Greed
Materials: One fair, six-sided die 1. Play the game once, recording your score for each
round in a table like this:
Are you more likely to take a risk or to stay with
a sure thing? In this Activity, you will get to make
decisions about whether to keep playing a game of Round 1 2 3 4 5 Total
chance in hopes of getting a higher score. Score
Rules: The game of Greed is played with a single
die, rolled by your teacher. Each game consists of 2. Your teacher will draw a stemplot on the board
5 rounds. Your total score for the game will be the with stems from 0 to 9 (or higher if a student scores
sum of your scores in the 5 rounds played. Here’s more than 99). After you get your total, go to the
how each round works. board and add your value to the stemplot.
• Everyone in the class stands up. Your teacher
3. Discuss the results with your classmates. Do
will throw a fair, six-sided die twice and add the
some students’ strategies sound better than the
numbers on the up-faces. This is your current score.
one you used? Think about how you would play
If you are happy with that score, sit down and re-
differently if you played Greed again.
cord this value as your score for the round.
• If someone is still standing, your teacher rolls 4. Play a second game of 5 rounds. Make a back-to-
the die again. Those still standing add the resulting back stemplot by adding your total to the left side of
number to their total unless the die shows a 5. If a 5 the stem.
is rolled, anyone standing loses all their points for 5. Compare shape, center, spread, and unusual
the round and sits down. values for the two distributions. Did the class
• Your teacher will continue to roll the die until all earn higher scores in the second game than in the
students have decided to sit down or until a 5 is rolled. first?

What is a probability distribution?

A probability model describes the possible outcomes of a chance process and the
likelihood that those outcomes will occur. For example, suppose we toss a fair coin
3 times. The sample space for this chance process is

H-H-H H-H-T H-T-H T-H-H H-T-T T-H-T T-T-H T-T-T

Since there are 8 equally likely outcomes, the probability is 1/8 for each possible
outcome. Define the variable X 5 the number of heads obtained. The value of X
will vary from one set of tosses to another but will always be one of the numbers
0, 1, 2, or 3. How likely is X to take each of those values? We can summarize the
probability distribution of X as follows:

Value: 0 1 2 3
Probability: 1/8 3/8 3/8 1/8
 8. 1 Probability Distributions 371

3/8

Probability
2/8

1/8

0
0 1 2 3
Value

Figure 8.1 A histogram of the probability distribution for X 5 number of heads in

three tosses of a fair coin.

Figure 8.1 shows the probability distribution of X in graphical form. Notice the
symmetric shape.
We can use the probability distribution to answer questions about the variable
X. What’s the probability that we get at least one head in three tosses of the coin? In
symbols, we want to find P(X $ 1). We could add P(X 5 1), P(X 5 2), and P(X 5 3)
to get the answer. Or we could use the complement rule from Chapter 7:
P(X $ 1) 5 1 2 P(X , 1) 5 1 2 P(X 5 0)
5 1 2 1/8 5 7/8
A numerical variable that describes the outcomes of a chance process (like X in
the coin-tossing scenario) is called a random variable.

Random variables and probability distributions

A random variable takes numerical values that describe the out-

comes of some random phenomenon. The probability distribution
of a random variable gives its possible values and associated prob-
abilities. The probability distribution can be displayed in a table or a
graph.

There is an obvious connection between probability distributions and the prob-

ability models of Chapter 7. We use the probability model for a chance process to
help determine the probability distribution of a random variable. The probability
model for tossing three coins tells us that P(T-T-T) 5 1/8. Once we define the ran-
dom variable X 5 the number of heads in three tosses, the probability distribution
for X says that P(X 5 0) 5 1/8. Why? Because the chance of getting no heads (all
tails) in three tosses of a coin is 1/8.
Here’s an example involving something a bit more serious than tossing coins.
372 CHAPTER 8 Probability Models

Example 8.1 Apgar scores: babies’ health at birth

In 1952, Dr. Virginia Apgar suggested five criteria for measuring a baby’s health at
birth: skin color, heart rate, muscle tone, breathing, and response when stimulated. She
developed a 0-1-2 scale to rate a newborn on each of the five criteria. Babies are usually
tested at one minute after birth and again at five minutes after birth. A baby’s Apgar
score is the sum of the ratings on each of the five scales, which gives a whole number
value from 0 to 10. Apgar scores are still used today to evaluate the health of newborns.
(Although this procedure was later named for Dr. Apgar, the acronym APGAR also
represents the five scales: Appearance, Pulse, Grimace, Activity, and Respiration.)
What Apgar scores are typical? To find out, researchers recorded the Apgar
scores of over 2 million newborn babies in a single year.2 Imagine selecting one of
these newborns at random. (That’s our chance process.) Define the random vari-
able Y 5 Apgar score of a randomly selected baby one minute after birth. The table
below gives the probability distribution for Y.

Value: 0 1 2 3 4 5 6 7 8 9 10
Probability: 0.001 0.006 0.007 0.008 0.012 0.020 0.038 0.099 0.319 0.437 0.053

Figure 8.2 displays a histogram showing the probability distribution of Y. Notice

the heavily left-skewed shape of the histogram.

0.5

0.4
Probability

0.3

0.2

0.1

0.0
0 2 4 6 8 10
Apgar score
Figure 8.2 A histogram showing the probability distribution of the random variable
Y 5 Apgar score of a randomly selected newborn at one minute after birth.

Doctors decided that Apgar scores from 7 to 10 indicate a healthy baby. What’s
the probability that a randomly selected baby is healthy? It’s P(7 # Y # 10). We can
calculate this probability as follows:

P(7 #Y # 10) 5 P(Y 5 7) 1 P(Y 5 8) 1 P(Y 5 9) 1 P(Y 5 10)

5 0.099 1 0.319 1 0.437 1 0.053 5 0.908

That is, we’d have about a 91% chance of randomly choosing a healthy baby.
 8. 1 Probability Distributions 373

Expected value
Gambling on chance outcomes goes back to ancient times. Both public and private
lotteries were common in the early years of the United States. After disappearing for a
century or so, government-run gambling reappeared in 1964, when New Hampshire
introduced a lottery to raise public revenue without raising taxes. After some initial
controversy, many larger states adopted the idea. Forty-two states and all Canadian
provinces now sponsor lotteries.
State lotteries made gambling acceptable as entertainment. Some form of
legal gambling is allowed in 48 of the 50 states. Over half of all adult Ameri-
cans have gambled legally. They spend more money betting than on spectator
sports, video games, theme parks, and movie tickets combined. If you are going
to bet, you should understand what makes a bet good or bad. As the opening
story says, we care about how much we win as well as about our probability of
winning.

Example 8.2 Winning (and losing) at roulette

1 1
00 27 On an American roulette wheel, there are 38 slots numbered 1 through 36, plus 0
3 3 10
24
6 25
29
and 00. Half of the slots from 1 to 36 are red; the other half are black. Both the
0 and 00 slots are green. A player can place a simple $1 bet on either red or
12
3
15

black. If the ball lands in a slot of that color, the player gets the original dol-
19
34

lar back, plus an additional dollar for winning the bet. If the ball lands in a
21 6 18 31
22

different-colored slot, the player loses the dollar bet to the casino.
5 17 32

Let’s define the random variable Z 5 amount gained on a single $1 bet.

The possible values of Z are 2$1 and $1. (The player either gains a dollar or
33
20

loses a dollar.) What are the corresponding probabilities? The chance that the
16
7

ball lands in a slot of the correct color is 18/38. The chance that the ball lands
11

30
23
26
9 28 4 3
5 in a different-colored slot is 20/38. Here is the probability distribution of Z:
0 2 1

Value: 2$1 $1
Probability: 20/38 18/38

What is the player’s average gain? The ordinary average of the two possible
outcomes 2$1 and $1 is $0. But $0 isn’t the average winnings because the player
is less likely to win $1 than to lose $1. In the long run, the player gains a dollar 18
times in every 38 games played and loses a dollar on the remaining 20 of 38 bets.
The player’s long-run average gain for this simple bet is

20 18
2$1 # 1$1 # 5 2$0.05
38 38
You see that in the long run the player loses (and the casino gains) 5 cents per bet.

Here is a general definition of the kind of “average outcome” we used to evalu-

ate the bets in Example 8.2.
374 CHAPTER 8 Probability Models

Expected value

The expected value of a random variable is found by multiplying each

possible value of the variable by its probability and then summing
over all possible outcomes.

In symbols, if the possible values of the variable are x1, x2, . . . , xk and
their probabilities are p1, p2, . . . , pk, the expected value is
expected value 5 x1 # p1 1 x2 # p2 1 c1 xk # pk

An expected value is an average of the possible outcomes, but it is not an ordinary

average in which all outcomes get the same weight. Instead, each outcome is weighted
by its probability, so outcomes that occur more often get higher weights. The idea of ex-
pected value as an average applies to random phenomena other than games of chance.

Example 8.3 Finding the average Apgar score

What’s the “average” Apgar score for a newborn baby? It’s the expected value of the
random variable Y that we defined in Example 8.1. From the probability distribution
for Y, we see that 1 in every 1000 babies would have an Apgar score of 0, 6 in every
1000 babies would have an Apgar score of 1, and so on. So the expected value of Y is
1 6 7 53
0# 11# 12# 1 c1 10 # 5 8.13
1000 1000 1000 1000
The mean Apgar score of a randomly selected newborn one minute after birth is 8.13.

Exercises 8.1 Random digits, I Imagine choosing a digit from 0 to 9 at random, either
from Table B or with your calculator. Let D 5 the digit selected.
(a) Explain why D is a random variable.
(b) Display the probability distribution of D in a table.
(c) Construct a histogram that shows the probability distribution of D.
8.2 Roll two dice, I Imagine that you roll two fair, six-sided dice—one red and
one green. Let T 5 the sum of the spots showing on the up-faces of the two dice.
(a) Explain why T is a random variable.
(b) Display the probability distribution of T in a table.
(c) Construct a histogram that shows the probability distribution of T.
8.3 Unhealthy babies Refer to Example 8.1. Newborns with Apgar scores of 3
and below may require medical attention. Write the probability that a randomly
chosen baby has an Apgar score of 3 or below in terms of the random variable Y.
Then find this probability.
 8. 1 Probability Distributions 375

8.4 Household size According to the Census Bureau, the number of people X in
a randomly selected U.S. household follows the probability distribution given in
the table below.
Number of people xi : 1 2 3 4 5 6 7
Probability pi : 0.25 0.32 0.17 0.15 0.07 0.03 0.01

(a) Sketch a histogram that displays the probability distribution of X.

(b) Calculate and interpret P(X . 2).
(c) Find the expected value of X and explain what it tells you.
8.5 Random digits, II Refer to Exercise 8.1. Find the expected value of D and
interpret the result.
8.6 Roll two dice, II Refer to Exercise 8.2. Find the expected value of T and
interpret the result.

The law of large numbers

The definition of “expected value” says that it is an average of the possible outcomes—
but an average in which outcomes with higher probability count more. We argued
that the expected value is also the average outcome in another sense: it represents
the long-run average we will actually see if we repeat a bet many times or choose
many newborn babies at random. This is more than just intuition. Mathematicians
can prove, starting from the basic rules of probability, that the expected value cal-
culated from a probability distribution really is the “long-run average.” This famous
fact is called the law of large numbers.

The law of large numbers

If a random phenomenon with numerical outcomes is repeated many

times independently, the mean of the observed outcomes approaches
the expected value.

The law of large numbers is closely related to the idea of probability. In many
independent repetitions, the proportion of times an outcome occurs will be close to
its probability. As a result, the average outcome obtained will be close to the expected
value. These facts express the long-run predictability of chance events. They are the
true version of the “law of averages.”
The law of large numbers explains why gambling, which is a recreation or an
addiction for individuals, is a business for a casino. The “house” in a gambling op-
eration is not gambling at all. The average winnings of a large number of customers
will be quite close to the expected value. The house has calculated the expected
value ahead of time and knows what its take will be in the long run. There is no
need to load the dice or stack the cards to guarantee a profit. If enough bets are
placed, the law of large numbers guarantees the house a profit.
Life insurance companies operate much like casinos—they bet that the people
who buy insurance will not die. Some do die, of course, but the insurance com-
pany knows the probabilities and relies on the law of large numbers to predict the
376 CHAPTER 8 Probability Models

average amount it will have to pay out. Then the company sets its premiums high
enough to guarantee a profit.
The following Activity lets you see the law of large numbers in action.

ACTIVITY 8.1B
The Expected Value applet
AP
PLET Go to the Statistics Through Applications Web site, at www.whfreeman.com/sta2e, and launch the
Expected Value applet.
Directions Use the “Fewer dice” and “More dice” buttons to fix the number of dice to be rolled with
each trial. Next, enter the number of rolls (100 maximum) and then click the “Roll dice” button to have the
applet roll the dice and plot the accumulating average of the roll totals.

18

Average

10.5

total - 9 3
1 10
Number of rolls

Fewer dice Roll dice Show mean

Rolls 1
More dice Reset Show roll totals

Think individual roll totals and the accumulating averages

1. When you roll just 1 die, what values can you behave like you thought they would?
get? Are they all equally likely? APPLET TIP: Click the “Show roll totals” button to see
the results of each individual roll (shown as blue dots).
2. When you roll 10 dice, what values can you get
for the roll total? Are they all equally likely? Experiment
4. The average of the numbers from 1 to 6 is (1 1
3. Which process is likely to get close to its mean
2 1 3 1 4 1 5 1 6)/6 5 3.5. Set the applet to roll
value in fewer trials: (a) the accumulating averages
1 die and click the “Show mean” box to see that the
of rolling 1 die at a time or (b) the accumulating
expected value for rolling a single die is 3.5. Now
averages of rolling 10 dice at a time?
alternately click the “More dice” button and the
Verify “Roll dice” button while you watch the behavior of
Test your answers to the three questions above the expected value for the sum of the rolls as more
through repeated use of the applet. Did the dice are added. What is the pattern?
 8. 1 Probability Distributions 377

Thinking about expected values

As with probability, it is worth exploring a few fine points about expected values
and the law of large numbers.
How large is a large number? The law of large numbers says that the actual aver-
age outcome of many trials gets closer to the expected value as more trials are made.
It doesn’t say how many trials are needed to guarantee an average outcome close to
the expected value. That depends on the variability of the random outcomes.
The more variable the outcomes, the more trials are needed to ensure that the
mean outcome is close to the expected value. Games of chance must be quite vari-
able if they are to hold the interest of gamblers. Even a long evening in a casino has
an unpredictable outcome. Gambles with extremely variable outcomes, like state
lottos with their very large but very unlikely jackpots, require impossibly large num-
bers of trials to ensure that the average outcome is close to the expected value.
Though most forms of gambling are less variable than lotto, there’s a practical
answer to how the law of large numbers applies: the casino plays often enough to
rely on it, but you don’t. Much of the appeal of gambling is its unpredictability for
the player. The business of gambling rests on the fact that the result is not unpre-
dictable for the casino.
Is there a winning system? Serious gamblers often follow a system of betting in
which the amount bet on each play depends on the outcome of previous plays. You
might, for example, double your bet on each spin of the roulette wheel until you
win—or until your money runs out. Such a system tries to take advantage of the
fact that you have a memory even though the roulette wheel does not. Can you beat
the odds with a system? No. Mathematicians have established a stronger version
of the law of large numbers that says that if you do not have an infinite fortune to
gamble with, your average winnings (the expected value) remain the same as long
as successive trials of the game (such as spins of the roulette wheel) are indepen-
High-tech gambling dent. Sorry.
There are more than
450,000 slot machines in
the United States. Once Finding expected values by simulation
upon a time, you put in a How can we calculate expected values in practice? You know the mathematical
coin and pulled the lever formula, but that requires that you start with the probability of each outcome.
to spin three wheels, each
with 20 symbols. No longer.
Expected values too difficult to compute in this way can be found by simulation.
Now the machines are The procedure is the same as before: give a probability model, use random digits to
video games with flashy imitate it, and simulate many repetitions. By the law of large numbers, the average
graphics and outcomes pro- outcome of these repetitions will be close to the expected value.
duced by random number
generators. Machines can
accept many coins at once, Example 8.4 Drawing an ace
can pay off on a bewilder-
ing variety of outcomes, You draw cards, one after the other, from a standard deck of 52 playing cards until you
and can be networked to draw an ace. On average, how many cards can you expect to turn over until you get
allow common jackpots. an ace? That is, what is the expected value for the number of cards drawn? This is
Gamblers still search for
not an easy question to answer. But with simulation methods, we can obtain a fairly
systems, but in the long
run the random number good estimate, although it is a bit tedious.
generator guarantees the Clearly, the cards drawn are not independent since the first card drawn is no
house its 5% profit. longer in the deck and so cannot be drawn again. If we use the random digit table,
378 CHAPTER 8 Probability Models

we would let the numbers 01 to 52 represent the 52 cards and arbitrarily let num-
bers 01, 02, 03, and 04 represent the four aces. Read pairs of digits from the table,
discarding numbers in the range 53 to 99 and 00.
Because about half of the pairs of numbers would be thrown out, we will use
the calculator rather than Table B. So that you can check our work, we’ll show you
how to seed the calculator’s random digit generator to match our results. Our goal
is to randomly generate integers between 1 and 52. As soon as a 1, 2, 3, or 4 is ob-
served, we’ll stop and record the number of cards drawn.
Seed the random number generator:
• TI-84: Type 1234, press XI, then choose PRB and 1:rand.
Complete the command 1234 ! rand and press ¸.
• TI-Nspire: On a Calculator page, press b, then choose
4:Probability, 4:Random, and 6:Seed. Complete the command
RandSeed 1234 and press ·.
Generate random integers from 1 to 52 until you get an ace (1 through 4). If
the calculator generates a duplicate number, ignore it (the same card can’t be drawn
twice), and draw again. Count the number of trials (cards) required to get an ace.
• TI-84: Press I, then choose PRB and 5:randInt(. Complete the
command randInt(1,52) and press ¸. Keep pressing ¸ until
you get a 1, 2, 3, or 4.
• TI-Nspire: Press b, then choose 4:Probability, 4:Random, and
2:Integer. Complete the command randInt(1,52)and press ·.
Keep pressing · until you get a 1, 2, 3, or 4.
Here are the results we got in the first six repetitions of the simulation:

Repetition Number to
number Cards drawn (* 5 skipped repeated number) get an ace
1 2 1
2 10, 5, 2 3
3 20, 42, *23, 52, 31, 19, 1 7
4 9, 8, 45, 46, 33, 43, 17, 47, 22, 14, 11, *5, *16, 32, 20, *31, 23
**30, 34, *38, *51, 49, 13, *3
5 8, 1 2
6 41, 18, 33, 28, 44, 24, 27, *47, 14, 31, 29, 34, 32, 16, 17, 7, 42, *3 18

The average number of cards drawn for these six repetitions is 9. But there’s a
lot of variability. We got lucky on the first repetition and drew an ace on the first
card. But it took 23 cards in the fourth repetition to draw an ace. Because of the
large variability, it will take a large number of repetitions until we have some con-
fidence in our estimate. Math or a long simulation shows that the actual expected
value is 10.6. We were lucky to get a result as close to 10.6 as we did after only six
repetitions.
Note: Before you continue, seed your calculator’s random number generator
with a unique number, such as the last four digits of your telephone number or
Social Security number. For example, 1089! rand (or RandSeed 1089 on
the TI-Nspire).
 8. 1 Probability Distributions 379

Exercises 8.7 16 boys! The Associated Press reported an unusual occurrence at the
Canton-Potsdam Hospital in Potsdam, New York: between December 15 and
December 21, all 16 newborns at the hospital were boys.
(a) Would the births of 16 consecutive boys at a different hospital in another small
town in the United States be just as newsworthy? Do you think this Associated
Press report is very unusual? Explain.
(b) It was also noted that on the two days before the report appeared, all 5
newborns at the Canton-Potsdam Hospital had been girls. Does this show the law
of averages in action? Explain.
8.8 Spin the wheel Joey has been watching people play roulette for a few
minutes. In the past 20 spins of the wheel, the ball has landed in a red slot 14
times. Joey wants to place a bet on “black” for the next spin, because he believes
black is “due.” He asks you for advice. What do you tell him?
8.9 Selecting a jury A county selects people for jury duty at random from a list
of one million registered voters, which happens to be evenly divided in terms of
gender (that is, exactly 50% of the county’s registered voters are women). Which
of the following is true and which is false? Explain.
(a) Out of the next 2000 names drawn, it is likely that exactly 1000 will be women.
(b) Out of the next 2000 names drawn, it is likely that the percent of women will
be around 50%, to within a percent or two.
8.10 Bringing down the house? You may have heard about a group of MIT
students who beat the casinos at blackjack. The 2008 movie 21 and the book
Bringing Down the House tell their story. Do some research: how did the students
use probability to get an edge on the casinos?
8.11 Duplicate birthdays There are 30 students in Austin’s precalculus class,
all unrelated. He wants to bet you $1 that at least 2 of the students were born
on the same day of the year (same month and day but not necessarily the same
year). Should you take the bet? Before you answer, you decide to conduct
a simulation to see how likely that event would be. Here is the probability
model:
• The birth date of a randomly chosen person is equally likely to be any of the
365 dates of the year (ignoring February 29).
• The birth dates of different people in the room are independent.
(a) Describe how you could use either Table B or your calculator to simulate 30
random birthdays from the 365 possible days.
(b) Carry out your simulation. Perform 10 repetitions. Based on your results,
should you take Austin’s bet?
(c) Combine your results with those from other members of the class. Are you
surprised?
8.12 Birthdays revisited Refer to the previous exercise. On average, how
many different people can you expect to ask before you find two with the same
birthday? Use Table B or your calculator to simulate birthdays of randomly
chosen people until you hit the same birthday a second time.
380 CHAPTER 8 Probability Models

(a) How many people did you ask before you found two with the same birthday?
(b) Combine your results with those from other members of the class and average
the results. This is an estimate of the expected value.

Probability distributions for sampling

Choosing a random sample from a population and calculating a statistic such as
the sample proportion is certainly a random phenomenon. The distribution of the
statistic tells us what values it can take and how often it takes those values. That
sounds a lot like a probability distribution.

Example 8.5 A sampling distribution

Suppose you take a simple random sample (SRS) of 1523 adults. Ask each person
whether they bought a lottery ticket in the last 12 months. The proportion who say
“Yes,”
number who say “Yes”
p^ 5
1523
is the sample proportion p^ . Imagine repeating this sampling process 1000 times.
You would collect 1000 sample proportions, one from each of the 1000 samples.
The histogram in Figure 8.3 shows the distribution of 1000 sample proportions
when the truth about the population is that 60% have bought lottery tickets.

0.54 0.56 0.58 0.60 0.62 0.64 0.66

Value of the sample proportion

Figure 8.3 The probability distribution of a sample proportion p^ from SRSs of size
1523 drawn from a population in which 60% of the members would give positive answers.
The histogram shows the distribution from 1000 samples. The Normal curve is the ideal
pattern that describes the results of a very large number of samples.
 8. 1 Probability Distributions 381

The results of random sampling are of course random: we can’t predict the
outcome of one sample, but the figure shows that the outcomes of many samples
have a regular pattern. The Normal curve in the figure is a good approximation to
the histogram. The histogram is the result of these particular 1000 SRSs. Think of
the Normal curve as the idealized pattern we would get if we kept on taking SRSs
from this population forever. That’s exactly the idea of probability—the pattern we
would see in the very long run. The Normal distribution assigns probabilities to the
outcomes of random sampling.
This Normal distribution has mean 0.6 and standard deviation about 0.0125.
The “95” part of the 68–95–99.7 rule says that 95% of all samples will give a p^ falling
within two standard deviations of the mean. That’s within 0.025 of 0.6, or between
0.575 and 0.625. We now have more concise language for this fact: the probability
is 0.95 that between 57.5% and 62.5% of the people in a sample will say “Yes.” The
word “probability” says we are talking about what would happen in the long run,
in very many samples.

A statistic from a large sample has many possible values. Assigning a prob-
ability to each possible value of a random variable worked well for tossing coins
or recording the Apgar score of a randomly selected baby, but it is awkward when
there are thousands of possible outcomes. Example 8.5 uses a different approach:
assign probabilities to intervals of outcomes by using areas under a Normal den-
sity curve. Density curves have area 1 underneath them, which lines up nicely
with total probability 1. The total area under the Normal curve in Figure 8.3 is 1,
and the area between 0.575 and 0.625 is 0.95, which is the probability that a sam-
ple gives a result in that interval. When a Normal curve assigns probabilities, you
can calculate probabilities from the 68–95–99.7 rule or from Table A of Normal
distribution areas. These probabilities satisfy the probability rules of Chapter 7.

Sampling distribution

The sampling distribution of a statistic tells us what values the statis-

tic takes in repeated samples from the same population and how often
it takes those values.

We think of a sampling distribution as assigning probabilities to the

values the statistic can take. Because there are usually many possible
values, sampling distributions are often described by a density curve
such as a Normal curve.

Here’s another example of sampling distributions in action.

Example 8.6 Do you approve of gambling?

An opinion poll asks an SRS of 501 teens, “Generally speaking, do you approve of
legal gambling or betting?” Suppose that in fact exactly 50% of all teens would say
“Yes” if asked. (This is close to what polls show to be true.) The poll’s statisticians tell
 382 CHAPTER 8 Probability Models

us that the sample proportion who say “Yes” will vary in repeated samples according
to a Normal distribution with mean 0.5 and standard deviation about 0.022. This is
the sampling distribution of the sample proportion p^ .
What is the probability that the poll will get a sample in which 52% or more say
“Yes”? Because 0.52 is not 1, 2, or 3 standard deviations away from the mean, we can’t
use the 68–95–99.7 rule. We will use Table A of standard Normal probabilities.
To use Table A, first turn the outcome p^ 5 0.52 into a standardized z score by
subtracting the mean of the distribution and dividing by its standard deviation:

0.52 2 0.5
z5 5 0.91
0.022
Now look in Table A. The area to the left of a z-score of 0.91 is 0.8186 under a
Normal curve. This means that the probability is 0.8186 that the poll gets a smaller
result. By the complement rule (or just the fact that the total area under the curve is 1),
this leaves probability 0.1814 for outcomes of 52% or more who say “Yes.” Figure 8.4
shows the probabilities as areas under the Normal curve.

Probability Probability
0.8186 0.1814
Figure 8.4 The Normal
sampling distribution for
the opinion poll of 501
teens. Table A tells us that
the area under the curve
to the left of 0.52 is 0.8186.
The area to the right of
0.5 0.52
0.52 is 0.1814.

Exercises 8.13 Do you jog? An opinion poll asks an SRS of 1500 adults, “Do you happen
to jog?” Suppose (as is approximately correct) that the population proportion who
jog is p 5 0.15. In a large number of samples, the proportion p^ who answer “Yes”
will be approximately Normally distributed with mean 0.15 and standard deviation
0.009. Sketch this Normal curve and use it to answer the following questions.
(a) What percent of many samples will have a sample proportion who jog that is
0.15 or less? Explain clearly why this percent is the probability that p^ is 0.15 or less.
(b) What is the probability that p^ will take a value between 0.141 and 0.159? (Use
the 68–95–99.7 rule.)
Stephen Coburn/FeaturePics (c) What is the probability that p^ does not lie between 0.141 and 0.159?
 8. 1 Probability Distributions 383

(d) Would you be surprised if the sample proportion who said they jogged was
p^ 5 0.17? Justify your answer.
8.14 Applying to college You ask an SRS of 1500 college students whether they
applied for admission to any other college. Suppose that in fact 75% of all college
students applied to colleges besides the one they are attending. (That’s close to the
truth.) The sampling distribution of the proportion of your sample who say “Yes”
is approximately Normal with mean 0.75 and standard deviation 0.01. Sketch this
Normal curve and use it to answer the following questions.
(a) Explain in simple language what the sampling distribution tells us about the
results of our sample.
(b) What percent of many samples would have a p^ larger than 0.77? (Use the
68–95–99.7 rule.) Explain in simple language why this percent is the probability
of an outcome larger than 0.77.
(c) What is the probability that your sample result will be either less than 0.73 or
greater than 0.75? Show your work.
(d) Would you be surprised if the sample proportion who said they applied to
another college was p^ 5 0.725? Justify your answer.
8.15 Generating a sampling distribution Let us illustrate the idea of a sampling
distribution in the case of a very small sample from a very small population. The
population consists of the numbers 1, 2, 4, and 9. The parameter of interest is the mean
of this population. The sample is an SRS of size n 5 2 drawn from the population.
(a) Find the mean of the four numbers in the population. This is the population
mean m.
(b) There are only six possible samples of size 2 that can be chosen. One such sample
is {1, 2}. For this sample, x 5 1.5. This statistic, x, is an estimate of the population
mean m. List the other five possible SRSs of size 2, and compute the value of x for each.
(c) Make a dotplot that shows the distribution of x-values from (b). This is the
sampling distribution of x.
(d) Find the mean of the six x-values from (b). How does this compare to the
population mean m?
Exercises 8.16 to 8.18 refer to the following setting. The Wechsler Adult Intelligence
Scale (WAIS) is a common “IQ test” for adults. The distribution of WAIS scores
for persons over 16 years of age is approximately Normal with mean 100 and
standard deviation 15.
8.16 An IQ test, I
(a) What is the probability that a randomly chosen individual has a WAIS score of
115 or higher?
(b) In what range do the scores of the middle 95% of the adult population lie?
8.17 An IQ test, II Use Table A or your calculator to find the probability that a
randomly chosen person has a WAIS score of 112 or higher. Show your work.
8.18 An IQ test, III How high must a person score on the WAIS test to be in
the top 10% of all scores? Use Table A or your calculator to answer this question.
Show your work.
384 CHAPTER 8 Probability Models

ws
Ne
ws
Ne

APPLICATION 8.1 B]
RO
g
<
Se
a

Benford’s Law
1. Using the newspaper, an almanac, the Statistical distributed. In an article that year in the Ameri-
Abstract of the United States, or similar text can Journal of Mathematics, Newcomb conjec-
material, locate a compact collection of numbers tured that, with naturally occurring data, you are
such as closing stock prices or sports statistics. most likely to observe a 1 as the leading digit, and
(Alternatively, your teacher may provide you with that each digit from 2 to 9 is less likely than its
data on closing stock prices from the New York predecessor to occur.3 In fact, he conjectured that
Stock Exchange.) Select a group of at least 100 many (but not all) such data have the following
separate numbers. Copy the table below onto your distribution:
paper. Then use tally marks to record the frequency
of each first digit, from 1 to 9. (Note: For decimals, Digit: 1 2 3 4 5 6 7 8 9
Probability: 0.301 0.176 0.125 0.097 0.079 0.067 0.058 0.051 0.046
like 0.00305, you would take the first nonzero digit.)
Calculate the relative frequency for each digit.
Newcomb got little recognition for this
remarkable result. Almost 60 years later, in 1938,
Digit Tally marks Counts Relative frequency Frank Benford rediscovered Newcomb’s result.
1 Benford was a physicist working for General Elec-
2 tric at the time. Apparently unaware of Newcomb’s
3 paper, Benford examined a large collection of data
4 that were not produced randomly. He published
5 his findings and his explanation, which he called
6 Benford’s Law, in another journal.4
7 Figure 8.5 is a histogram of the probabilities for
8
each first digit. Notice that we can’t locate the mean
9
of the right-skewed distribution by eye—calculation
You might think that each of the digits from 1 is needed. The mean is shown in Figure 8.5 as a
to 9 would occur equally often in the distribution of dotted line.
first digits. Were your results uniformly distributed? 2. Calculate the average first digit (expected
Scientist Simon Newcomb (1835–1909) dis- value) for data that obey Newcomb’s probability
covered in 1881 that the digits are not uniformly distribution.

0.4

0.3
Probability

0.2

0.1

0.0
0 1 2 3 4 5 6 7 8 9
Outcomes

Figure 8.5 Digits between 1 and 9 chosen from records that obey Benford’s Law.
 8. 1 Probability Distributions 385

3. Compare your results from step 1 with the This is a good way to visually compare how well
theoretical probability for each digit. Construct your results agree with the conjectured relative
a graph that shows two bars for each digit: one frequencies.
bar for the probability and a bar to the right of 4. Explain clearly why the distribution of first digits
it for the relative frequency from your data. Use in the blocks of 5 digits in Table B would not obey
color or shading to distinguish the two bars. Benford’s Law.

Section 8.1 Summary

A random variable is a numerical variable that describes the outcomes of a chance
process. A probability distribution tells us how often (in the very long run) a random
variable takes each of its possible values. We are often interested in the long-run aver-
age outcome, which we call the expected value of the random variable. The expected
value is found as an average of all the possible outcomes, each weighted by its probabil-
ity. If you don’t know the outcome probabilities, you can estimate the expected value
(along with the probabilities) by simulation. The law of large numbers says that the
mean outcome in many repetitions eventually gets close to the expected value.
A second kind of probability distribution assigns probabilities as areas under a
density curve, such as a Normal curve. The total probability is 1 because the total
area under the curve is 1. This kind of probability model is often used to describe
the sampling distribution of a statistic. This is the pattern of values of the statistic
in many samples from the same population.

Section 8.1 Exercises 8.19 to 8.22 refer to the following scenario. As a promotion gimmick, a
Exercises drug store decides to raffle off ice cream bars in the following way. They advertise,
“Ice cream bar: 0 to 43 cents. Take a chance.” To get one, the customer pops a bal-
loon and then reads the price he or she must pay on a slip of paper that was inside
the balloon. Of course, the customer doesn’t know what his or her chances are
of getting a fairly low price for his or her ice cream bar. The manager of the drug
store has marked prices on slips of paper as shown in the table. Suppose that the
usual price of an ice cream bar is 40 cents.

Price: $0.43 $0.42 $0.41 $0.40 $0.34 $0.20 $0.09 $0.02 $0.01 $0.00
Slips: 30 18 20 19 15 10 5 5 5 5

Mary decides to play the raffle and pops a balloon at random. Let X 5 the price
she must pay.
8.19 Pop a balloon, I
(a) Find P(X , $0.10) and interpret your result.
(b) What is the probability that Mary will pay less than the usual cost of an ice
cream bar? Write this probability in terms of the random variable X.
(c) How much money will the store take in if all of the balloons are popped? How
much would the store have made on these ice cream bars without the promotion?
Comment on the difference.
386 CHAPTER 8 Probability Models

8.20 Pop a balloon, II Sketch a graph of the probability distribution of X.

8.21 Pop a balloon, III Compute the expected price of an ice cream bar when
Mary pops a balloon. Is it to her benefit to play the raffle? Explain.
8.22 Pop a balloon, IV Make a new table of prices and numbers of slips so that
the expected price a customer will pay in the pop-the-balloon game is 40 cents.
Use a total of 100 slips.
8.23 Fire insurance Suppose a homeowner spends $300 for a home insur-
ance policy that will pay out $200,000 if the home is destroyed by fire. Let
Y 5 the profit made by the insurance company on a single policy. From
previous data, the probability that a home in this area will be destroyed by
fire is 0.0002.
(a) Make a table that shows the probability distribution of Y.
(b) Compute the expected value of Y. Explain what this result means for the
insurance company.
8.24 Polling women, I Suppose that 47% of all adult women think they do
not get enough time for themselves. An opinion poll interviews 1025 ran-
domly chosen women and records the sample proportion who feel they
o
hot

don’t get enough time for themselves. This statistic will vary from sample
ckp
iSto

to sample if the poll is repeated. The sampling distribution is approxi-

rt/
vae

mately Normal with mean 0.47 and standard deviation about 0.016.
e
cG
Eri

Sketch this Normal curve and use it to answer the following questions.
(a) If the true population proportion is 0.47, in what range will the middle
95% of all sample results fall?
(b) What is the probability that the poll gets a sample in which fewer than 45.4%
say they do not get enough time for themselves?
8.25 Polling women, II In the setting of Exercise 8.24, what is the probability
of getting a sample in which more than 51% of the women think they do not get
enough time for themselves? Show your work. (Use Table A or your calculator.)
8.26 Collecting cereal box prizes Several years ago, every box of Frosted
Mini-Wheats contained a NASCAR driver’s decal. There was a set of 6 decals, and
consumers were encouraged to “collect all 6.” The probability model is
• Every box of Frosted Mini-Wheats cereal contains one NASCAR driver’s
decal.
• A single decal is randomly selected and placed in the cereal box, one per box.
• There are 6 decals in the set.
The question we want to answer is “How many boxes of Frosted Mini-Wheats cere-
al can we expect to buy in order to collect the complete set of 6 NASCAR decals?”
(a) Begin with a simpler problem. Suppose there are only 2 decals. How would
you use the table of random digits to simulate this problem? How would you
use a single die to simulate the problem? How would you use your calculator to
simulate the problem? Choose one of these methods, and carry out 10 repetitions.
Then average the 10 results to estimate the expected number of boxes needed to
collect a set of 2 decals.
(b) Answer the questions in (a) if you assume a set of 3 decals, and then 4 decals.
(c) Now answer the original question for a set of 6 decals. Is there a pattern? If so,
describe it.
 8. 2 Counting and Probability 387

8 .2 Counting and Probability

ACTIVITY 8.2
Deal or no deal?
Materials: Computer with Internet connection 2. Before you play the game a second time, discuss
and projector the following questions with your classmates:
In the TV game show Deal or No Deal, a contestant • What’s the probability that you choose the
begins by choosing 1 of the 26 numbered cases $1,000,000 case to begin with?
shown on the game board below. Each case holds • If you played the game many, many times, what
a different amount of money ranging from $0.01 is the expected value of the amount that your case
to $1,000,000, as shown on the game board. The contains?
amounts are assigned to the cases at random before • Suppose you get lucky and pick the $1,000,000
the show. Neither the show’s host, Howie Mandel, case to begin with. (Of course, you don’t know what
nor the banker, who offers to buy the contestant’s case your case actually contains.) How likely do you think
after each round of the game, knows what any case you would be to keep turning down the banker’s of-
contains. fers and to actually leave the show with $1,000,000?
1. To get familiar with how the game works, go to 3. Play the game again as a class. This time, record
the online version of the game at www.nbc.com/ which cases you open in each round and the
Deal_or_No_Deal/game and play it once as a class. banker’s offer in a table like the one shown below.
Decline all of the banker’s offers and keep playing
until you get to see what’s in your case.
DEAL? Round Cases opened Banker’s offer
$
0.
01

$1
0
00
1,

00
1 $100,000; $0.01; $300; $500; $35,100
$

$5 ,0
22 23 24 25 26 $5 00
0,0 $10,000; $25
$1
0 $1
00
5,0
$ 25 17 18 19 20 21 $2 2 $
00
$ 50,0
$ 50
$ 75,000
12 13 14 15 16 4. Add an extra column on the right side of your
$ 75 $ 100,000
table for the average amount of money in the cases
$ 100 7 8 9 10 11 $ 200,000

$ 300
,000
that still remain on the board. Compute this value
$ 200
1 2 3 4 5 6 $4
00,
000
after each round. Is the banker’s offer higher than,
00
$3 $5
00 $7
00
,00
0
the same as, or lower than this average amount?
$4 50
,00
0
5. Play the game one last time as a class. This time,
$

50 0
1,

$
00

Choose
one of the briefcases
0,
0
75

00

This will try to finish with as much money as you can. Did
0
$

be your briefcase
you make a good deal?

Popular TV game shows like Deal or No Deal and The Price Is Right rely on
probability, expected values, and the law of large numbers to tell them how
much contestants will win—not in a single show, but on average in the long
run. The same is true for state governments who sponsor lotteries, where the
payouts can reach tens or even hundreds of millions. Casinos invite “high roll-
ers” to place large bets on complicated games of chance whose results are not
as predictable as those of roulette or slot machines. As the games get more
388 CHAPTER 8 Probability Models

complex, we need more advanced methods to determine all possible outcomes

and their probabilities.
In the early days of the Internet, many companies and individuals scrambled
to get the domain name of their choosing. American Airlines landed aa.com, In-
ternational Business Machines (IBM) got ibm.com, and Kentucky Fried Chicken
snagged kfc.com. Not everyone was so lucky. A small business in Canton, Ohio,
had to settle for a domain name that was 15 letters long. The company is The
Karcher Group, which specializes in Web design. Until recently, their domain
name was thekarchergroup.com. After waiting 10 years and paying $15,000, the
company has now acquired the domain name tkg.com. Imagine how much easier
it will be now for people to get to The Karcher Group’s Web site and to send emails
to company employees!

Example 8.7 Choosing a domain name

Internet domain names must have more than one character. That rules out names
like i.com. How many such names are there? If we consider letters only, there are
26 possibilities. What if we allowed a digit or a letter? Since there are 10 digits, we’d
have 26 1 10 5 36 possible one-character domain names.
Two-character domain names are allowed. How many two-letter domain
names are there? There are 26 possible choices for the first letter. Suppose we choose
“a.” How many choices are there for the second letter? 26, right? So there are 26 two-
letter domain names that begin with “a.” Similarly, there are 26 names that begin
with “b,” “c,” “d,” . . . , “z.” For any choice of first letter, we get 26 possibilities for the
second letter. So the total number of two-letter domain names is

(# of choices for first letter) (# of choices for second letter)

5 26 26 5 676

What if we allow letters or numbers for either character? Then there are a total of

(# of choices for first character) (# of choices for second character)

5 36 36 5 1296

possible two-character domain names. All 1296 have been claimed.

Let’s head to the ice cream parlor for some more counting.

Example 8.8 Ice cream sundaes

The 28 Choices ice cream shop offers 28 flavors of ice cream, 11 different toppings,
and 3 sizes of bowl for their ice cream sundaes. In how many ways can you order a
sundae with one scoop of ice cream and one topping?
Let’s start with the ice cream. You have 28 flavor choices. For each choice of
flavor, you can select 11 possible toppings. So there are 28 11 5 308 different ice
 8. 2 Counting and Probability 389

cream plus topping selections. For each of those 308 concoctions, you have three
possible sizes of bowl. In all, that gives you

(# of ice cream flavors) (# of topping choices) (# of size options)

5 28 11 3 5 924

ways to order a one-scoop sundae.

Examples 8.7 and 8.8 illustrate the multiplication counting principle.

Multiplication counting principle

For a process involving multiple (say k) steps, if there are n1 ways to

do Step 1, n2 ways to do Step 2, . . . , and nk ways to do Step k, then the
total number of different ways to complete the process is
n1 # n2 # c # nk
This result is called the multiplication counting principle.

The multiplication counting principle can also help us determine how many
ways there are to arrange a group of people, animals, or things. For example, sup-
pose you have 5 framed photographs of different family members that you want to
arrange in a line on top of your dresser. In how many ways can you do this? Since
you are arranging the photographs in a line, let’s count the options moving from left
to right across the dresser. There are 5 choices for the leftmost photo. For each of
these 5 possibilities, there are 4 options for the next photo to the right. By the mul-
tiplication counting principle, there are 5 4 5 20 different possibilities for these
first two photos. For each of these 20 options, there are 3 choices for the next photo
to the right, then 2 choices for the photo to the right of that one, and finally, only 1
choice left for the final photo. Again by the multiplication counting principle, that
gives a total of 5 4 3 2 1 5 120 different photo arrangements.
Expressions like 5 4 3 2 1 occur often enough in counting problems that
mathematicians invented a special name and notation for them. We write 5 4 3
2 1 5 5!, read as “5 factorial.”

Factorial

For any positive integer n, we define n! (read “n factorial”) as

n!  n 1 n 2 1 2 1 n 2 2 2 c # 3 # 2 # 1
That is, n factorial is the product of the numbers starting with n and
going down to 1.

The following example shows how quickly the number of different arrange-
ments of n individuals grows.
390 CHAPTER 8 Probability Models

Example 8.9 A special offer from the teacher

At the beginning of class, your teacher makes a special offer. She will dismiss class
early if you can make all possible arrangements with every student in the class in
a single-file line. To try to tempt you, she lets you try with one student, then two
students, and finally with three students. Should you accept the offer?
How many possible arrangements are there? If a small class has 15 students,
then there are 15 choices for which student may be first in line. For each of these
possibilities, there are 14 choices for who is second in line, 13 choices for who is
third in line, and so on. By the multiplication counting principle, there are

15 # 14 # 13 # c # 3 # 2 # 1 5 15!

possible arrangements. Instead of multiplying all 15 numbers together, we can use

the factorial command on the calculator to evaluate 15!
• TI-84: Type 15. Then press I, arrow to PRB, choose 4:!, and
press ¸.
• TI-Nspire: On a Calculator page, type 15, press b, then choose
4:Probability, 1:Factorial(!), and press ·.
There are about 1.31 # 1012, or just over 1 trillion possible arrangements. (Note that
the TI-Nspire shows all 13 digits of the result.)
Our advice: don’t take the offer!

Exercises 8.27 More domain names Refer to Example 8.7. How many three-character
Internet domain names are possible that consist of
(a) letters only?
(b) numbers or letters?
Note: All three-character domain names have been taken.
8.28 More ice cream Refer to Example 8.8. Mary stops by 28 Choices on the way
home from school. She only has enough money for a sundae with one scoop of ice
cream and one topping in a small bowl. How many different choices does Mary
have for her sundae?
8.29 New Jersey plates, I The illustration shows what New Jersey license plates
looked like in two different years—1959 and 1999. Assume that letters and
numbers must be in the positions shown on each license plate: for 1959, three
letters followed by three numbers; for 1999, three letters, then two numbers,
and then a letter. If there are no restrictions on the letters and numbers used
in any position, how many different license plates were possible in 1959? 1999?
Show your work.
8.30 New Jersey plates, II Refer to the previous exercise. In 1999, New Jersey
license plates were actually not allowed to have the letters D, T, or X in the first
position, or the letters I, O, or Q in any position. With these restrictions, how
many different license plates were actually possible in 1999? Show your work.
 8. 2 Counting and Probability 391

8.31 Organizing homework Suppose you have six homework assignments to

complete one night.
(a) In how many different orders can you complete all of the assignments?
(b) In how many different orders can you complete all six assignments if you
simply must do your statistics assignment first?
8.32 Assigned seating Mrs. Random decides to randomly assign seats for the 25
students in her statistics class.
(a) If there are 25 seats in the classroom, how many different seating assignments
are possible? Show your work.
(b) Repeat part (a) for a classroom with 30 seats.

Permutations
Mr. Paradise likes to get the students in his class involved in the action. But he
doesn’t want to play favorites. Each day, Mr. Paradise puts the names of all 28 of
his students in a hat and mixes them up. He then draws out 3 names, one at a time.
The student whose name is chosen first gets to operate the display calculator for
the day. The second student chosen is in charge of reading the answers to the even-
numbered homework problems. The third student selected writes class notes on the
Smart board. In how many different ways can Mr. Paradise randomly assign these
three jobs?
There are 28 possibilities for the first name chosen, 27 possibilities for the sec-
ond name, and 26 possibilities for the third name. By the multiplication counting
principle, there are

28 # 27 # 26 5 19,656

different ways for Mr. Paradise to assign the three jobs. We call arrangements like
this, where the order of selection matters, permutations. In this case, we calculated
the number of permutations of 28 people taken 3 at a time. In symbols, we’ll write
this as 28P3.
With a little clever math, we can rewrite as 28P3 as follows:

28P3 5 28 # 27 # 26
28 # 27 # 26 # 25 # 24 # c # 3 # 2 # 1
5
25 # 24 # c # 3 # 2 # 1
28!
5
25!
28!
5
1 28 2 3 2 !

In general, the number of permutations of n things taken k at a time is given by

n!
5
nPk
1n 2 k2!
392 CHAPTER 8 Probability Models

Permutation

A permutation is a distinct arrangement of some collection of indi-

viduals. If there are n individuals, then the notation n Pk represents the
number of different permutations of k individuals chosen from the en-
tire group of n. We can calculate nPk using the multiplication counting
principle or with the formula
n!
5
nPk
1n 2 k2!
By definition, 0! 5 1.

To make the formula for n P k work properly for all possible values of k, we must
define 0! 5 1. Why? Consider 28 P28. This is the number of different arrangements of
28 individuals, taken 28 at a time. In other words, it’s 28 # 27 # 26 # c # 3 # 2 # 1 5 28!
Using the permutation formula for 28P28, we get
28! 28!
5 5
28 P28
1 28 2 28 2 ! 0!
If we define 0! 5 1, then the values obtained by the two methods will agree.

Combinations
With permutations, order matters. Sometimes, we’re just interested in the number
of ways to choose some number of individuals from a group, but we don’t care about
the order in which the individuals are selected. For example, suppose Mr. Paradise
decides to randomly select 3 students’ homework papers to grade each day. He once
again puts all 28 names in a hat, mixes them up, and draws out 3 names, one at a
time. In how many different ways can he choose 3 students’ papers to grade?
It’s tempting to say that there are 28 # 27 # 26 5 19,656 ways for Mr. Para-
dise to do this. That’s not correct, however. Suppose he picks Leucretia, Tim, and
Kiran—in that order. That’s really no different from getting Tim, then Leucretia,
then Kiran. Or Kiran, then Leucretia, then Tim. How many selections consist of
these same 3 students? Use the multiplication counting principle: there are 3 pos-
sibilities for the first pick, 2 possibilities for the second pick, and only 1 option for
the last pick. So there are 3 # 2 # 1 5 3! 5 6 arrangements that consist of these same
3 students. This same argument applies for any 3 students that Mr. Paradise selects.
So to get the correct number of ways for him to choose 3 students’ papers to grade,
we need to divide our original (wrong) answer by 6:

28 # 27 # 26
number of ways to choose 3 homework papers to grade out of 28 5
3#2#1

With a little fancy math, we can rewrite this answer as

28 # 27 # 26 28P3
# # 5
3 2 1 3!
 8. 2 Counting and Probability 393

When the order in which we select individuals from a group doesn’t matter, we
call the number of possible selections a combination. In the previous paragraph,
we computed the number of combinations of 28 students taken 3 at a time, which
we’ll write as 28C3. We also saw an important connection between the number of
permutations and the number of combinations in this setting:

28P3
28C3 5
3!

In general, the number of combinations of n things taken k at a time can be found

using the formula

nPk
nCk 5
k!

Substituting the formula we found earlier for nPk yields

n!
¢ ≤
n Pk 1n 2 k2! n!
5 5 5
k! 1 n 2 k 2 !
nCk
k! k!

Combination

A combination is a selection of individuals from some group where the

order of selection doesn’t matter. If there are n individuals, then the nota-
tion nCk represents the number of different combinations of k individuals
chosen from the entire group of n. We can calculate nCk using the formula

n Pk n!
5 5
n Ck
k! k! 1 n 2 k 2 !

Finding probabilities using counting methods

With these new counting tools—the multiplication counting principle, permuta-
tions, and combinations—to help us, we are ready to tackle some more challenging
probability applications.

Example 8.10 Pick 3 lottery

Here is a simple lottery wager, the “Straight” from the Pick 3 game of the Tri-State Daily
Numbers offered by New Hampshire, Maine, and Vermont. You pay $1 and choose a
three-digit number. The state chooses a three-digit winning number at random and
pays you $500 if your number is chosen. What’s the probability that you win?
By the multiplication counting principle, there are 10 # 10 # 10 5 1000 possible
winning numbers. Your chance of matching the winning number picked randomly by
the computer is 1/1000. Of course, the probability that you don’t match the winning
394 CHAPTER 8 Probability Models

number is 1 2 1/1000 5 999/1000. If we let X 5 the amount of your winnings, then

the probability distribution of X is

Value: $0 $500
Probability: 999/1000 1/1000

Your long-run average winnings from a ticket are

999 1
$0 # 1 $500 # 5 $0.50
1000 1000
or 50 cents. You see that in the long run the state pays out half the money bet and
keeps the other half.

In the popular card game Texas Hold ’Em, players are initially dealt two cards.
The “best” hand a player can start with is a pair of aces (known as “pocket rockets”).
What’s the probability that a particular player is dealt a pair of aces? Since Texas
Hold ’Em often involves several players, and thus several starting hands that must
be dealt, let’s begin with a simpler version.

Example 8.11 Looking for aces

Suppose we deal two cards from a well-shuffled, standard deck. What’s the probabil-
ity that both are aces? In Chapter 7, we used a tree diagram and the general multipli-
cation rule to answer a similar question. Let’s follow that strategy again here.
Method 1—Tree diagram and general multiplication rule Let’s define the events
A 5 ace on first card and B 5 ace on second card. The probability that the first card dealt
is an ace is 4/52. Given that the first card was an
Second card ace, there’s a 3/51 chance that the second card is an
Ace ace. Then the probability of getting two aces can be
3 found using the general multiplication rule:
First card
51
Ace
4
P(A d B) 5 P(A) ˙ P(B|A)
52 48
No ace 5 (4/52)(3/51) 5 0.0045
51
Deal 2 Figure 8.6 shows a tree diagram of this chance
cards 4
51 Ace process. Getting two aces isn’t very likely!
48 No ace There are two completely different ways to
52 approach this problem using the methods of
Chapter 8.
No ace
47 Method 2—Permutations If we think of
51 the two cards that are dealt as Card 1 and Card 2,
Figure 8.6 A tree diagram then there are 52 possibilities for Card 1. Once
showing the sample space for the Card 1 is dealt, there are only 51 possibilities for
chance process of dealing two cards Card 2. So there are 52 51, or 52 P 2, possible ar-
from a deck and looking for aces. rangements of the first two cards in the deck, each
 8. 2 Counting and Probability 395

of which is equally likely after shuffling. To get two aces, we have four choices for Card 1
and then three choices remaining for Card 2. That is, there are 4 3, or 4 P 2, arrange-
ments in which the first two cards are aces. The probability of getting two aces is

number of arrangements with two aces

P(two aces) 5
number of arrangements of first two cards
4 P2 4#3
5 5 5 0.0045
52 P2 52 # 51

Method 3—Combinations We can also use combinations to get this result.

First, we count the number of possibilities for dealing two cards from the deck. Note
that the order in which the cards are dealt really doesn’t matter here. All we care about
is which two cards are dealt. Since we are “choosing” two cards out of 52 at random,
there are 52C2 different ways to do this. Each of these possible two-card sets is equally
likely to be dealt. How many ways are there to get two aces? With four aces to choose
from, there are 4C2 ways to obtain two aces. So the probability of getting two aces is

number of ways to get two aces

P(two aces) 5
number of ways to deal two cards
4! 4 # 3 # 2! 4# 3
4C2 2! 1 4 2 2 2 ! 2!2! 2# 1 6
5 5 5 5 5 5 0.0045
52C2 52! 52 # 51 # 50! 52 # 51 1326
2! 1 52 2 2 2 ! 2!50! 2# 1

In Example 8.11, we calculated the probability of getting two aces when we deal
two cards from a standard deck. In other words, we found the probability that two
aces will end up as the top two cards in the deck after shuffling. But in card games
involving several players, like Texas Hold ’Em, the top card on the deck is dealt to
Player 1, the second card to Player 2, and so on until each player has one card. Then
a second card is dealt to Player 1, Player 2, and so forth until all the players have two
cards. What’s the probability that Player 1 gets two aces? It’s the chance that there
are aces in Positions 1 and n  1 in the deck, where n is the number of players. Let’s
think about this. The probability that the top card and second card in the deck are
aces is the same as the probability that the top card and the (n  1)st card are aces.
What is that chance? By Example 8.11, it’s 0.0045. Every player at the table has the
same chance (before the cards are dealt) of getting “pocket rockets.”

Exercises 8.33 Two scoops On the way home, you decide to stop by 28 Choices ice cream parlor
for a snack. With 28 flavors to choose from, how many different ways are there to order:
(a) a cone with two scoops if you care about the order of flavors?
(b) a cone with two scoops of different flavors if you care about the order of flavors?
(c) a cup with two scoops of different flavors (order doesn’t matter)?
(d) a cup with two scoops of ice cream (order doesn’t matter)?
396 CHAPTER 8 Probability Models

8.34 Three scoops Refer to Exercise 8.33. How many different ways are there to order
(a) a cone with three scoops if you care about the order of flavors?
(b) a cone with three scoops of different flavors if you care about the order of flavors?
(c) a cup with three scoops of different flavors (order doesn’t matter)?
(d) a cup with three scoops of ice cream (order doesn’t matter)?
8.35 Pick 3 more Refer to Example 8.10.
(a) Find the number of ways in which the computer can pick a winning number
with three different digits. Show your work.
(b) What’s the probability that the computer picks a winning number with two or
more matching digits? Show your work.

© 2009 Darby Conley, distributed by UFS, Inc.

8.36 Pick 4 Refer to Example 8.10. In the Pick 4 game of the Tri-State Daily
Numbers, you pay $1 and choose a four-digit number. The state chooses a four-
digit winning number at random and pays you $5000 if your number is chosen.
(a) What’s the probability that you win? Show your method clearly.
(b) Find the expected value of your winnings from playing Pick 4.
8.37 Permutations and combinations Six friends—Aaron, Betty, Caleb, Deanne,
Ernie, and Flo—are having a study session on permutations and combinations.
(a) Find the value of 6 P0. Explain why this value makes sense.
(b) Compute the value of 6 P3 and 6 C 3. Create an example involving the six friends
to show how these values are related.
(c) Calculate 6 C 2 and 6 C 4. Create an example involving the six friends to show
how these values are related.
8.38 Big slick In Texas Hold ’Em, players are initially dealt two cards. One very
strong starting hand is called “big slick,” which consists of an ace and a king (not
necessarily of the same suit).
(a) If you deal two cards from a well-shuffled deck, what’s the probability of
getting “big slick”? Show your method clearly.
(b) In a Texas Hold ’Em game with six players, what’s the probability that Player 1
is dealt “big slick”? Justify your answer.
 8. 2 Counting and Probability 397

CALCULATOR CORNER Permutations and combinations

on the calculator
The TI-84 and TI-Nspire have built-in commands for calculating the
number of permutations and combinations.
TI-84
• Permutations: To find 28 P 3, type 28, then press 1, choose PRB, and 2:nPr.
Complete the command 28 nPr 3 and press b.
• Combinations: To find 28C3, type 28, then press Ì, choose PRB, and 3:nCr.
Complete the command 28 nCr 3 and press b.

TI-Nspire
• Permutations: To find 28 P 3, on a Calculator page press b, choose 4:
Probability, 2:Permutations. Complete the command nPr(28,3)
and press ·.
• Combinations: To find 28C3, press b, choose 4:Probability, 3:
Combinations. Complete the command nCr(28,3) and press ·.
The last line in both calculator screen shots reminds you of the relationship
between permutations and combinations.

ws
Ne
ws
Ne

APPLICATION 8.2 B]
RO
g
<
Se
a

iPod play list

Janine wants to set up a play list with 8 songs on her Suppose Janine
iPod. She has 75 songs to choose from, including 15 decides to let her iPod
by the Jonas Brothers, 12 by Miley Cyrus, and 10 by select an 8-song play
Bow Wow. list at random.
4. What’s the
QUESTIONS probability that all 8
1. How many different sets of 8 songs are possible songs are by Miley
for Janine’s play list? Assume that the order of the Cyrus?
songs doesn’t matter. 5. What’s the
2. How many different 8-song play lists can Janine probability that all 8
create? This time, assume the order of the songs songs are by Miley
does matter. Cyrus, the Jonas Andreanna Seymore/Getty

3. How many 8-song play lists contain only songs Brothers, or Bow Wow?
by the Jonas Brothers? No songs by the Jonas 6. What’s the probability that at least 1 of the 8
Brothers? songs is by Bow Wow?
398 CHAPTER 8 Probability Models

Section 8.2 Summary

Most probability problems involve counting the number of possible outcomes of
a chance process. When that process involves multiple steps, the multiplication
counting principle can be used to determine the total number of possible out-
comes. The multiplication counting principle also helps us count the number of
distinct arrangements of some group of individuals. Such arrangements are called
permutations. The number of permutations of n individuals is n factorial:

n! 5 n # 1 n 2 1 2 # 1 n 2 2 2 # . . . # 3 # 2 # 1

The number of permutations of n individuals taken k at a time is given by

n!
5
n Pk
1n 2 k2!

With permutations, the order in which individuals are selected matters. When
order doesn’t matter, we use combinations. The number of combinations of n in-
dividuals taken k at a time is

n Pk n!
5 5
k! 1 n 2 k 2 !
nC k
k!

To solve a complicated probability problem, begin by counting all possible out-

comes. Use a tree diagram, the multiplication counting principle, permutations, or
combinations as needed.

Section 8.2 8.39 Radio station call signs, I In 1912, the U.S. government began issuing li-
Exercises censes to radio stations. Each station was given a unique three-letter “call sign.” By
international agreement, the U.S. eventually received rights to all call signs begin-
ning with the letters W, N, and K. Radio stations in the western United States were
given call signs starting with K. Stations in the east were given call signs starting
with W. (N was reserved for use by the U.S. Navy.)
(a) How many three-letter call signs start with the letter W? Show your method.
(b) How many three-letter call signs starting with W or K were available for U.S.
radio stations? Show your work.
8.40 Radio station call signs, II Refer to Exercise 8.39. By 1922, there were more
applications for radio station licenses than the number of three-letter call signs avail-
able. A radio station in New Orleans applied for and was granted the call sign WAAB.
(a) How many four-letter call signs start with the letter W? Show your method.
(b) In fact, no call sign was permitted in which the same letter appeared three times in
a row. (That’s why the New Orleans station got WAAB, not WAAA.) How many four-
letter call signs starting with W are possible with this restriction? Show your work.
(c) How many four-letter call signs start with W or K and meet the restriction in
(b)? Show your method.
8.41 Who gets picked? Mr. Paradise uses a calculator program to select students
at random to answer questions during class. Each of the 28 students in his class
 8. 3 Binomial Distributions 399

is equally likely to be chosen every time he runs the program. Suppose Mr. Paradise
runs the program twice. The first student selected gets to answer Question 1,
and the second student selected (who might be the same student) gets to answer
Question 2.
(a) Use the multiplication counting principle to find the number of different ways
that the program can choose who answers these two questions.
(b) Now find the number of ways in which two different students are chosen to
answer the two questions.
(c) What’s the probability that the program selects two different students in the
class for the first two questions?
8.42 Who else gets picked? Refer to Exercise 8.41. This time, Mr. Paradise uses
his program three times in a row. Find the probability that three different students
in the class are chosen.
8.43 Pick Six In the New Jersey “Pick Six” lotto game, a player chooses six different
numbers from 1 to 49. The six winning numbers for the lottery are chosen at ran-
dom. If the player matches all six numbers, she wins the jackpot, which starts at $2
million. Find the probability of picking all six winning numbers. Show your work.
8.44 Pick Six again Refer to Exercise 8.43. Find the probability that an unlucky
player does not match any of the six winning numbers. Show your work.
8.45 Roll 5 dice Suppose you roll 5 six-sided dice at one time.
(a) Find the probability that all 5 dice show the same number of spots on the up-
faces. Show your work.
(b) What’s the probability that all 5 dice show a different number of spots on the
up-faces? Show your work.
8.46 Deal or No Deal? In Activity 8.2 (page 387), you explored the TV game
show Deal or No Deal. Now, you should be ready to analyze some of the probabili-
ties involved in the game. Suppose you get very lucky and choose the case with
$1 million in it as your case. In the first round of the game, you must choose 6 of
the remaining 25 cases to open.
(a) In how many different ways can you select 6 cases containing dollar
amounts on the left side of the game board? What’s the probability that this
happens?
(b) If you happen to choose the 6 cases with the lowest amounts of money in
them ($0.01 to $50), how much should the banker offer you? Justify your answer.

8 .3 Binomial Distributions

ACTIVITY 8.3A
Rock-paper-scissors-art?
In 2005, Takashi Hashiyama, president of a successful or Sotheby’s—to offer the collection for sale. Since he
Japanese electronics company, wanted to sell his felt that the two companies were equally qualified, Mr.
company’s $20 million art collection. He decided to Hashiyama invited representatives from Christie’s and
use one of two famous auction companies—Christie’s Sotheby’s to compete in a game of rock-paper-scissors.
ACTIVITY 8.3A (continued)
400 CHAPTER 8 Probability Models

ACTIVITY 8.3A (continued)

(Using games of 2. Count the number of rounds won by each player.

chance to make such Is there a winner?
decisions is common
3. Based on your results, does it appear that both
in Japanese business.)
players were selecting among the three possible
The winner would get
throws at random in each round? Or do you
to sell the collection.
think some more sophisticated “strategy” was
Who won, and what
being used by one or both players? Justify your
was their strategy?
answer.
In this Activity,
you will learn more 4. Pool your results with those of your classmates
about the game of to determine the highest number of rounds won by
© 2007 Vic Lee, distributed by King Features
Syndicate, Inc. rock-paper-scissors any student in the class. Do you think it’s possible
via simulation and the that a player could win this many rounds if both
rules of probability. Let’s review the basic rules of players were selecting throws at random in each
the game before you begin. A “round” of the game round? Explain.
consists of a “throw” by each of the two opposing
5. Work with your partner to design and carry out
players. There are three possible throws that a player
a simulation to help answer the question posed in
may choose: rock, paper, or scissors. If both play-
Step 4. You may use your calculator or Table B of
ers make the same “throw” (for example, rock and
random digits. Perform 10 trials of your simulation.
rock), then the round is a tie. Otherwise, the winner
What conclusion do you draw? Why?
of the round is determined as follows: rock breaks
scissors, scissors cuts paper, and paper covers rock. 6. Now pool results with your classmates. Does
your conclusion change? Why or why not?
1. Play 10 rounds of rock-paper-scissors against
your opponent. Record the result of each round in a What happened in the $20 million game of
table like the one shown. rock-paper-scissors described earlier? Christie’s got
Round My throw Opponent’s throw Outcome
a little help from the 11-year-old twin daughters of
a company executive. From their experience playing
1 Rock Paper Lose the game many times at school, the girls recom-
2 Scissors Scissors Tie mended “scissors.” The Christie’s executive took
3 Scissors Paper Win their advice. Sotheby’s chose “paper,” which gave
… Christie’s the win and the art collection.

Binomial settings and the binomial theorem

What do each of the following scenarios have in common?
• Toss a coin 5 times. Count the number of heads.
• Spin a roulette wheel 8 times. Record how many times the ball lands in a red slot.
• Play 10 games of rock-paper-scissors with a friend in which you both use a
chance process (like rolling a die) to determine what you throw each time.
Count how many games you win.
• Take a random sample of 100 babies born in U.S. hospitals today. Record
the number of females.
In each case, we’re performing repeated trials of the same chance process. The
number of trials is fixed in advance. In addition, the outcome of one trial has no
 8. 3 Binomial Distributions 401

effect on the outcome of any other trial. That is, the trials are independent. We’re
interested in the number of times that a specific outcome (we’ll call it a “success”)
occurs. Our chances of getting a “success” are the same on each trial. When these
conditions are met, we have a binomial setting.

Binomial setting

A binomial setting arises when we perform several independent trials

of the same chance process and record the number of times that a par-
ticular outcome occurs. The four conditions for a binomial setting are

• Binary? The possible outcomes of each trial can be classified as “suc-

cess” or “failure.” Note that there can be more than two possible out-
comes per trial—a roulette wheel has numbered slots of three colors: red,
black, and green. If we define “success” as having the ball land in a red
slot, then “failure” occurs when the ball lands in a black or a green slot.
• Independent? Trials must be independent; that is, the outcome of
one trial must not have any effect on the outcome of any other trial.
• Number? The number of trials n of the chance process must be fixed
in advance.
• Success? On each trial, the probability p of success must be the same.

The boldface letters in the previous box give you a helpful way to remember the
conditions for a binomial setting: just check the BINS!

Example 8.12 Is it a binomial setting?

Here are three scenarios involving chance behavior. Which describe binomial settings?
(a) Shuffle a deck of cards. Turn over the top card. Put the card back in the deck, and
shuffle again. Repeat this process 10 times. Count the number of aces you observe.
• Binary? “Success” 5 get an ace; “Failure” 5 don’t get an ace.
• Independent? Since you are replacing the card in the deck and shuffling each
time, the result of one trial does not affect the outcome of any other trial.
• Number? You’re performing n 5 10 trials of this chance process.
• Success? The probability of a success is 4/52 on each trial.
So this is a binomial setting.
(b) Shuffle a deck of cards. Turn over the first 10 cards, one at a time. Count the
number of aces you observe.
• Binary? “Success” 5 get an ace; “Failure” 5 don’t get an ace.
• Independent? In this case, the trials are not independent. If the first card you
turn over is an ace, then the next card is less likely to be an ace. That is, P(first
card is an ace) 5 4/52, but P(second card is an ace | first card is an ace) 5 3/51.
402 CHAPTER 8 Probability Models

Since the trials of this chance process are not independent, this is not a binomial
setting.
(c) Shuffle a deck of cards. Turn over the top card. Put the card back in the deck,
and shuffle again. Repeat this process until you get an ace. Count the number of
trials required.
• As in (a), the Binary and Independent conditions are satisfied.
• Number? The number of trials is not set in advance. You could get an ace
on the first card you turn over, or it may take many cards to get an ace.
Since there is no fixed number of trials, this is not a binomial setting.

Part (c) of Example 8.12 raises another important point about binomial settings.
In addition to checking the BINS, make sure that you’re being asked to count the num-
ber of successes in a certain number of trials. In Part (c), for example, you’re asked to
count the number of trials until you get a success. That can’t be a binomial setting.
In a binomial setting, we can define a random variable (say X) as the number of
successes in n independent trials. What’s the probability distribution of X? Let’s see
if an example can help shed some light on this question.

Example 8.13 1 in 6 wins a prize

As a special promotion for its 20-ounce bottles of soda, a soft drink company printed
a message on the inside of each cap. Some of the caps said “Please try again,” while
others said “You’re a winner!” The company advertised the promotion with the slogan
“1 in 6 wins a prize.” Assume that the company is telling the truth and that each bottle
of soda has a 1/6 chance of having a cap that says “You’re a winner!” Suppose that
Alan and five of his friends—Bart, Chloe, Delores, Ed, and Frankie—each buy one 20-
ounce bottle. What’s the probability that none of them wins a prize?
First, let’s be sure that this is a binomial setting. We’ll check the BINS. Note that
a trial consists of buying a soda and looking under the cap. The chance is involved
in which soda bottles receive caps that say “You’re a winner!”
• Binary? “Success” 5 cap says “You’re a winner!”; “Failure” 5 cap says
“Please try again.”
• Independent? If the company is telling the truth, then the message printed
on one bottle’s cap should have no effect on which message appears on
another bottle’s cap.
• Number? Between Alan and his friends, there are n 5 6 trials of this
chance process.
• Success? Each bottle has a p 5 1/6 chance of having a cap that says “You’re
a winner!”
Since the conditions are met and we’re counting the number of successes in a fixed
number of trials, this is a binomial setting.
Let X 5 the number of prizes won by Alan and his friends. We want to find
P(X 5 0). The complement rule tells us that the probability that a particular bottle
is not a winner is 1 2 1/6 5 5/6. If X 5 0, that means that none of the six wins a
 8. 3 Binomial Distributions 403

prize. In other words, all six bottles must have caps that say “Please try again.” Since
the trials are independent, the chance that this happens is
1 5 /6 2 1 5 /6 2 1 5 /6 2 1 5 /6 2 1 5 /6 2 1 5 /6 2 5 0.335

There’s about a 1/3 chance that none of the six wins a prize.

In Example 8.13, we found the probability that Alan and his five friends won
nothing in the soft drink company’s “1 in 6 wins a prize” contest. What’s the prob-
ability that exactly one of them wins a prize? One way that can happen is if Alan’s
bottle of soda says “You’re a winner!” while all five friends’ bottles say “Please try
again.” The chance that this happens is

1 1/6 2 3 1 5/6 2 3 1 5/6 2 3 1 5/6 2 3 1 5/6 2 3 1 5/6 2

Alan Bart Chloe Delores Ed Frankie

since the trials are independent. Another way this can happen is if Bart wins a
prize, but no one else does. The probability of this happening is

1 5/6 2 3 1 1/6 2 3 1 5/6 2 3 1 5/6 2 3 1 5/6 2 3 1 5/6 2

Alan Bart Chloe Delores Ed Frankie

Note that the probability that Bart is the only winner in the group is the same as the
probability that Alan is the only winner. There are four more possibilities to con-
sider—the ones in which Chloe, Delores, Ed, and Frankie are the only winners. Of
course, the probability will be the same for each of those cases. So the chance that
exactly one of the six wins a prize is

6 3 1 1/6 2 3 1 5/6 2 3 1 5/6 2 3 1 5/6 2 3 1 5/6 2 3 1 5/6 2

1 1 5 5
5 6 # a b # a b 5 0.402
6 6

Where did the 6 come from? It’s the number of ways to choose which one of the six
friends wins a prize. In Section 8.2, we wrote this as 6C1, the number of combina-
tions of 6 things taken 1 at a time. Using this notation, we can rewrite the above
probability as

P 1 X 5 1 2 5 6C1 # a b # a b 5 0.402
1 1 5 5
6 6

Can you see where we’re headed? What’s the probability P(X 5 2) that exactly
two of the six friends win a prize? One way this can happen is for Alan and Bart to
win a prize, but the other four to get a soda that says “Please try again.” The chance
that this occurs is

1 1/6 2 3 1 1/6 2 3 1 5/6 2 3 1 5/6 2 3 1 5/6 2 3 1 5/6 2

Alan Bart Chloe Delores Ed Frankie
404 CHAPTER 8 Probability Models

Since there are 6C2 ways to choose which two of the six friends win prizes, the prob-
ability we want is

P 1 X 5 2 2 5 6C2 # a b # a b 5 0.201
1 2 5 4
6 6

Our work on this problem suggests a more general result, known as the bino-
mial theorem.

Binomial theorem

In a binomial setting involving n independent trials of a chance

process with probability p of success on each trial, the probability of
getting exactly k successes is given by

P 1 X 5 k 2 5 nCk # pk # 1 1 2 p 2 n2k

Exercises In Exercises 8.47 to 8.50, determine whether the given scenario describes a
binomial setting. If so, confirm that all of the required conditions are met. If not,
explain why not.
8.47 Sowing seeds Seed Depot advertises that 85% of its flower seeds will
germinate (grow). Suppose that the company’s claim is true. Judy buys a packet
with 20 flower seeds from Seed Depot and plants them in her garden. She counts
how many of the seeds germinate.
8.48 Long or short? Put the names of all the students in your class in a hat.
Mix them up, and draw four names without looking. Count the number of these
students whose last names have more than six letters.
8.49 Lefties, I About 10% of people are left-handed. Suppose you select students
at random from your school, one at a time, until you find one who is left-handed.
Record the number of students you selected.
8.50 Lefties, II About 10% of people are left-handed. Select 15 students at
random from your school and count the number who are left-handed.
8.51 All 6 win Refer to Example 8.13 (page 402). Find the probability that Alan
and all five of his friends win a prize. Show your work.
8.52 Three winners Refer to Example 8.13 (page 402). Find the probability that
exactly three of the six friends win a prize. Show your work.

Binomial distributions
In Example 8.13, we examined the chance process of Alan and his five friends playing
the “1 in 6 wins a prize” soft drink contest. We defined the random variable X 5 the
number of prizes won by Alan and his friends. Earlier, we found P(X 5 0), P(X 5 1),
 8. 3 Binomial Distributions 405

and P(X 5 2). If you completed Exercises 8.51 and 8.52, then you found P(X 5 6) and
P(X 5 3). By the binomial theorem,

P 1 X 5 4 2 5 6C4 # a b # a b 5 0.008
1 4 5 2
6 6
and

P 1 X 5 5 2 5 6C5 # a b # a b 5 0.0006
1 5 5 1
6 6

Here is the completed probability distribution for X:

Value: 0 1 2 3 4 5 6
Probability: 0.335 0.402 0.201 0.054 0.008 0.0006 0.00002

Figure 8.7 shows a histogram of this probability distribution, which we call

a binomial distribution. It has a clear right-skewed shape. Why does this make
sense? Each 20-ounce bottle of soda has only a 1/6 chance of being a prize win-
ner. If Alan and his five friends buy a total of six bottles, then it’s much more
likely that 0, 1, or 2 of them win a prize than that 3 or more of them win prizes.

0.5

0.4
Probability

0.3

0.2

0.1

0.0
0 1 2 3 4 5 6
Successes
Figure 8.7 A histogram showing the probability distribution of X 5 number of prizes
won by Alan and his friends in the “1 in 6 wins” soft drink contest.

How many prizes would we expect Alan and his friends to win? The expected
value of X is

expected value 5 0 # 0.335 1 1 # 0.402 1 2 # 0.201 1 3 # 0.054

1 4 # 0.008 1 5 # 0.0006 1 6 # 0.00002
5 1.00

This should make sense, too! If six people play a game that gives each person a 1/6
chance of winning, then we’d expect one of them to win, on average, if the process
were repeated many times.
406 CHAPTER 8 Probability Models

We can generalize this result for any binomial setting.

Expected value for a binomial distribution

If a chance process involves n independent trials, with success prob-

ability p on each trial, the expected number of successes is
expected value  np

ACTIVITY 8.3B
Galton’s Board applet
Sir Frances Galton invented the “bean machine” in AP
PLET
1. Go to the book’s Web site,
the late 1880s to demonstrate what happens when www.whfreeman.com/sta2e, and launch the
a large number of beans or balls are dropped down Galton’s Board applet.
a chute and allowed to bounce off a series of pegs.
2. Set the number of trials at 1. Click “Start” to have
Figure 8.8 shows a small version of such a “bean
the applet drop one ball. Watch as the ball bounces
machine,” which we now tend to call a quincunx or
off the pegs on its way down. What do you think is
Galton’s Board. In this Activity, you will use a Java
the probability that the ball falls to the left when it
applet to see how Galton’s Board works.
hits a peg? To the right?
= wooden pegs 3. Reset the applet, and set the number of trials
= ball in chute at 100. Click “Start.” After watching the 100 balls
bounce down the board, what do you notice
about where they have landed? Which slot(s)
seem most likely? Least likely? Does that make
sense to you?

A B C D E F G

Figure 8.8 An example of a “bean machine,” or

Galton’s Board, named for its inventor, Sir Francis Galton.

Let’s look more closely at Figure 8.8. When a ball is released, it hits the first peg, and
then has an equal chance of falling to the left or to the right. The ball then proceeds down
to the next row of pegs, hits one of them, and falls left or right with equal probability. This
continues until the ball settles in one of the “bins” at the bottom of the board.
What is the probability that a ball lands in Slot A? For this to happen, the ball
must fall to the left all 6 times it hits a peg. The probability that this will happen is

1 6
a b 5 0.015625
2

Can you figure out the probability that a ball lands in Slot G?
 8. 3 Binomial Distributions 407

What’s the probability that a ball lands in Slot B? For this to happen, the ball
must fall to the right once and to the left five times, in some order. If we designate
L to mean “falls to the left” and R to mean “falls to the right,” then there are six pos-
sible “routes” for the ball to travel to land in Slot B:

RLLLLL LRLLLL LLRLLL LLLRLL LLLLRL LLLLLR

The probability that the ball follows the route RLLLLL is

1 1 1 5
a b # a b
2 2

Likewise, the probability that the ball follows the route LRLLLL is

1 1 1 5
a b # a b
2 2

In fact, this is the probability that it follows any of the six specific routes. So the
chance that the ball will land in Slot B is given by

1 1 1 5
6 # a b # a b 5 0.09375
2 2

Does this calculation look familiar? It uses the binomial theorem with n 5 6,
p 5 1/2, and k 5 1. But why is this a binomial setting? Consider the chance process
of a ball bouncing its way down Galton’s Board. At each row of the board, the ball
encounters a peg and falls either left or right. If we think of each peg as a “trial,”
then we can define a “success” as falling to the right. There are a fixed number of trials:
n 5 6. The trials are independent, and the probability of success on each trial is 1/2.
If we define the variable X 5 the number of times the ball falls to the right on its
way down the board, then X has a binomial distribution. Let’s find the probability
distribution of X.

Example 8.14 Galton’s Board: finding probabilities

Earlier, we found the probability that the ball falls to the left all six times: 0.015625.
That’s P 1 X 5 0 2 . We also determined that P 1 X 5 1 2 5 0.09375. Now let’s find the
probability that a ball lands in Slot C. For this to happen, the ball must fall to the
right at two pegs and to the left at four pegs. So we want P 1 X 5 2 2 . By the binomial
theorem,

P 1 X 5 2 2 5 6C2 # a b # a b 5 0.234375
1 2 1 4
2 2

Similarly, the probability that a ball lands in Slot D is

P 1 X 5 3 2 5 6C3 # a b # a b 5 0.3125
1 3 1 3
2 2
408 CHAPTER 8 Probability Models

What’s the probability that the ball lands in Slot E? It’s

P 1 X 5 4 2 5 6C4 # a b # a b 5 0.234375
1 4 1 2
2 2
That’s the same as P(X 5 2). Does it make sense to you that a ball is equally likely to
land in Slot C or Slot E? By similar logic, the chance that a ball lands in Slot F is

P 1 X 5 5 2 5 6C5 # a b # a b 5 0.09375 5 P 1 X 5 1 2
1 5 1 1
2 2
and the chance that a ball lands in Slot G is

P 1 X 5 6 2 5 6C6 # a b # a b 5 0.015625 5 P 1 X 5 0 2
1 6 1 0
2 2
Putting it all together, the probability distribution of X is

Value xi : 0 1 2 3 4 5 6
Probability pi : 0.015625 0.09375 0.234375 0.3125 0.234375 0.09375 0.015625

Figure 8.9 displays this probability distribution. Notice the perfectly symmetric shape!

0.35

0.30

0.25
Probability

0.20

0.15

0.10

0.05

0.00
0 1 2 3 4 5 6
Value
Figure 8.9 A histogram of the probability distribution for X 5 the number of times the
ball falls to the right in Galton’s Board. This is a binomial distribution with n 5 6 and p 5 1/2.

Exercises 8.53 Blood types, I In the United States, 44% of adults have type O blood.
Suppose we choose seven U.S. adults at random.
(a) What’s the probability that all seven people have type O blood? Show your
method.
(b) What’s the probability that exactly four of those selected have type O blood?
Show your method.
8.54 Blood types, II Refer to Exercise 8.53. Let X 5 the number of people
chosen with type O blood.
 8. 3 Binomial Distributions 409

(a) Find the probability distribution of X. Display it in a table.

(b) Graph the probability distribution of X. Describe what you see.
8.55 Sowing seeds again Seed Depot advertises that 85% of its flower seeds will
germinate (grow). Suppose that the company’s claim is true. Judy buys a packet
with 20 flower seeds from Seed Depot and plants them in her garden.
(a) How many seeds should Judy expect to germinate? Explain.
(b) Find the probability that exactly the expected number of seeds from (a)
actually grow. Show your work.
(c) Find the probability that fewer than the expected number of seeds from (a)
actually grow. Show your work.
8.56 Sowing more seeds Refer to Exercise 8.55.
(a) If only 15 seeds actually germinate, should Judy be suspicious that the company’s
claim is not true? Compute an appropriate probability to support your answer.
(b) If only 12 seeds actually germinate, should Judy be suspicious that the
company’s claim is not true? Compute an appropriate probability to support your
answer.
8.57 Coin tosses Imagine that you toss a fair coin six times.
(a) Find the probability that you get three heads and three tails. Show your work.
(b) Find the probability that you get four heads and two tails. Show your work.
8.58 Bigger bean machine Draw a Galton’s Board with two more rows of pegs
than the one in Figure 8.8. Label the slots at the bottom of the board A through I
from left to right.
(a) Find the probability that a ball lands in Slot C. Show your work.
(b) Find the probability that a ball lands in Slot F. Show your work.

CALCULATOR CORNER Binomial distributions

Both the TI-84 and the TI-Nspire have built-in functions that can help
you find probabilities involving binomial distributions. In a binomial
setting involving n independent trials with success probability p on each
trial, we can find the probability of getting exactly k successes using the
binomial theorem:

P 1 X 5 k 2 5 nCk # pk # 1 1 2 p 2 n2k

The equivalent command on the calculator is binompdf(n,p,k). There is

also a command (binomcdf) for finding the probability of getting k or fewer
successes in n trials.
Let’s use the “1 in 6 wins a prize” binomial setting from Example 8.13
(page 402) to illustrate how these commands can be used.
CALCULATOR CORNER (continued)
410 CHAPTER 8 Probability Models

CALCULATOR CORNER (continued)

1. To find the probability that Alan and his friends win a total of two prizes,
P(X 5 2):
TI-84
• From the home screen, press 2L (DISTR) and choose binompdf(.
• Complete the command binompdf(6,1/6,2)and press b.

TI-Nspire
• On a Calculator page, press b, choose 4:Probability,
5:Distributions..., and D:Binomial Pdf.
• Enter the number of trials (6), success probability (1/6), and X Value (2) in the
dialog box as shown. Then choose OK.

2. To find the probability that Alan and his friends win two prizes or fewer,
P(X # 2):
TI-84
• Press 2L (DISTR) and choose binomcdf(.
• Complete the command binomcdf(6,1/6,2)and press b.

TI-Nspire
• Press b, choose 4:Probability, 5:Distributions..., and E:
Binomial Cdf.
• Enter the number of trials (6), success probability (1/6), Lower Bound (0), and
Upper Bound (2) in the dialog box as shown.
3. Can you figure out how to use the binomcdf command to find the
probability that Alan and his friends win two or more prizes, P(X $ 2)?
 8. 3 Binomial Distributions 411

ws
Ne
ws
Ne

APPLICATION 8.3 B]
RO
g
<
Se
a

The lady tasting tea

One summer af- tea to test the lady’s claim? Explain the reasoning
ternoon in the late behind your design.
1920s, a tea party was Assume for a moment that the lady can’t
held in Cambridge, actually tell the difference between the two types
England. A lady at the of beverages—tea with milk added and milk with
party claimed that she tea added. Then for each of the 10 cups presented
could tell the differ- in the experiment, she would just be guessing.
ence between tea to Suppose the experiment is conducted in such a
which milk had been way that each cup presented to the lady is equally
added and milk to likely to contain either beverage. Let X be the
which tea had been number of cups of tea that the lady identifies
added. Many of those correctly.
at the party were 2. Show that this is a binomial setting. (Check the
doubtful. A few of the BINS.)
© Pietus/Dreamstime.com
more open-minded
gentlemen thought about designing an experiment 3. Find and interpret P(X 5 5).
to test the lady’s claim. One of those gentlemen was 4. Find and interpret P(X . 5).
Ronald A. Fisher, widely considered to be the father 5. How many correct identifications would you
of experimental design. With his guidance, the tea- have required before concluding that the lady
tasting experiment was carried out successfully us- was not just guessing? Justify your answer using
ing 10 cups of tea.5 Want to know what happened? probability.
Not so fast! You’ll appreciate the ending of the story
better if you learn about the experiment first. 6. In the actual experiment, the lady identified
all 10 cups of tea correctly. Fisher and his friends
QUESTIONS concluded that the lady could actually tell how
1. If you had been at the party that day, how would the tea had been prepared. Do you agree? Why or
you have designed an experiment using 10 cups of why not?

Section 8.3 Summary

A process that involves counting the number of successful outcomes in a set num-
ber of independent trials of the same chance phenomenon is a binomial setting. To
confirm that a chance process describes a binomial setting, check the four required
conditions—Binary, Independent, Number, and Success. Binary means that the
outcomes can be classified as either “success” or “failure.” The trials of the chance
process must be independent, and the number of trials must be fixed in advance.
On each trial, the probability of success p must be the same.
In a binomial setting, we often define a numerical variable (say X) to represent the
number of successes. The binomial theorem tells us how to calculate the probability of
getting exactly k successes in n independent trials, each with probability p of success:
P 1 X 5 k 2 5 nCk # pk # 1 1 2 p 2 n2k
The probability distribution of X is called a binomial distribution.
412 CHAPTER 8 Probability Models

Galton’s Board can be used to show the possible routes a ball can take as it
bounces down through several rows of pegs. To find the probability that a ball lands
in a particular slot on Galton’s Board, use the binomial theorem.

Section 8.3 8.59 Going to the dogs, I Ladies Home Journal magazine reported that 66% of
Exercises all dog owners greet their dog before greeting their spouse or children when they
return home at the end of the workday. Suppose that 12 dog owners are selected at
random.
(a) Show that the four requirements for a binomial setting are met.
(b) Find the probability that exactly 7 of the 12 dog owners greet their dog first
when they arrive home. Show your work.
(c) Find the probability that at least 5 of the 12 dog owners greet their dog first
when they arrive home. Show your work.
8.60 Going to the dogs, II Refer to the previous exercise.
(a) What is the expected number of dog owners who greet their dog first when
they arrive home? Justify your answer.
(b) Make a table that shows the probability distribution of the variable X 5 the
number of owners who greet their dog first.
(c) Construct a histogram that shows the probability distribution for X. Describe
what you see.
8.61 More lefties Suppose that exactly 10% of the students at your school
are left-handed. Imagine selecting an SRS of 15 students from the school
population.
(a) Find the probability that exactly 3 students in the sample are left-handed.
Show your work.
(b) Find the probability that 3 or fewer students in the sample are left-handed.
Show your work.
(c) Would you be surprised if the SRS contained 4 left-handed students? Justify
your answer using probability.
8.62 Righties and lefties Refer to the previous exercise. Let X 5 the number of
right-handed students in your sample.
(a) Find and interpret P(X 5 12). How does this compare with your answer to
8.61(a)?
(b) Write a probability expression involving X that is equivalent to the question in
8.61(b).
(c) Find the probability that the number of right-handed students in your sample
is more than the expected value of X. Show your work.
8.63 Orange M&M’s, I According to the Mars candy company, 20% of its
plain M&M’s candies are orange. Assume the company’s claim is true. Suppose
you reach into a large bag of plain M&M’s (without looking) and pull out eight
candies.
(a) Would you be surprised if none of the candies were orange? Compute an
appropriate probability to support your answer.
(b) Would you be surprised if exactly two of the candies were orange? Compute
an appropriate probability to support your answer.
 Chapter 8 Review 413

(c) Would you be surprised if four of the candies were orange? Compute an ap-
propriate probability to support your answer.
8.64 Orange M&M’s, II Refer to Exercise 8.63. Let X 5 the number of orange
M&M’s chosen.
(a) Find the probability distribution of X. Display it in a table.
(b) Graph the probability distribution of X. Describe what you see.
8.65 Galton’s Board, I Refer to Example 8.14 (page 407). A clever person has
redesigned the “bean machine” of Figure 8.8. The pegs have been adjusted so that
a ball is more likely to fall to the right (0.6) than to the left (0.4).
(a) Find the probability that a ball lands in Slot A. Show your work.
(b) Find the probability that a ball lands in Slot C. Show your work.
(c) In which slot is the ball most likely to land? Justify your answer with probability.
8.66 Galton’s Board, II Refer to the previous exercise. Let X 5 the number of
times that the ball falls to the right.
(a) Make a histogram that shows the probability distribution of X. How is this
graph different from the one in Figure 8.9 (page 408)?
(b) Suppose 100 balls are dropped down the chute, one at a time, onto this modi-
fied Galton’s Board. How many of the balls would you expect to land in Slot D?
Explain.

CHAPTER 8 REVIEW
Sometimes it is helpful to define a random variable that summarizes the outcomes
of a chance process. A probability distribution specifies the values that the random
variable can take and how often it takes those values. We can use the probability
distribution to determine the expected value of the random variable, a number
that describes the long-run average outcome. If the random variable can take all
values in an interval, we find probabilities using areas under a density curve (like
the Normal curve).
To solve a probability problem, we often need to determine the number of
possible outcomes of a chance process. We can use the multiplication counting
principle to do this when the chance behavior occurs in several stages. Both per-
mutations and combinations involve choosing individuals from an existing popu-
lation. With permutations, we’re interested in the order of selection; with combi-
nations, we’re not.
A binomial setting occurs when several independent trials of the same chance
process are performed. If we define a variable to represent the number of trials on
which a successful outcome occurs, that variable follows a binomial distribution.
We can use the binomial theorem to find probabilities associated with a binomial
distribution.

Here is a review list of the most important skills you should have developed
from your study of this chapter.
A. EXPECTED VALUES AND PROBABILITY DISTRIBUTIONS
1. Understand the idea of expected value as the average of numerical outcomes in
a great many repetitions of a random phenomenon.
2. Find the expected value of a random variable from a probability distribution.
414 CHAPTER 8 Probability Models

3. Estimate an expected value by repeating a simulation many times.

4. When probabilities are assigned by a Normal curve, find the probability of an
event by finding an area under the curve.
B. COUNTING TECHNIQUES AND PROBABILITY
1. Use the multiplication counting principle to determine the number of possible
outcomes of a chance process involving multiple steps.
2. Count the number of distinct arrangements of a group of individuals using
permutations.
3. Count the number of distinct selections of a group of individuals using
combinations.
4. Distinguish counting situations in which order matters from those where it
doesn’t. Explain the relationship between nPk and nCk.
5. Solve probability problems using the multiplication counting principle, permu-
tations, and combinations.

C. BINOMIAL DISTRIBUTIONS
1. Determine whether a given chance process describes a binomial setting.
2. Use the binomial theorem to calculate probabilities involving a binomial
distribution.
3. Compute the expected value of a binomial distribution.

CHAPTER 8 REVIEW EXERCISES

8.67 Pastabilities A popular Italian restaurant chain claimed that you could get “42
different pasta-sauce combinations.” Explain what you think they meant by this claim.
8.68 What are the chances? Open your local telephone directory to any page in
the residential listing. Look at the last four digits of each telephone number, the
digits that specify an individual number within an exchange given by the first three
digits. Note the first of these four digits in each of the first 100 telephone numbers
on the page.
(a) How many of the digits were 1, 2, or 3? What is the approximate probability that
the first of the four “individual digits” in a telephone number is 1, 2, or 3?
(b) If all 10 possible digits had the same probability, what would be the probability
of getting a 1, 2, or 3? Based on your work in (a), do you think the first of the four
“individual digits” in telephone numbers is equally likely to be any of the 10 possible
digits? Justify your answer.
8.69 Phone numbers, I In North America, phone numbers have the form XXX-
XXX-XXXX. The first three digits give the area code, and the second three digits
indicate the exchange.
(a) Nick lives in western North Carolina, where the area code is 828. If there were no
restrictions on the remaining seven digits, how many phone numbers would be pos-
sible in the 828 area code? Show your method.
(b) In fact, exchanges cannot begin with 0 or 1. How many possible numbers are there
in the 828 area code, subject to this restriction? Show your method.
 Chapter 8 Review Exercises 415

(c) How many of the possible phone numbers in (b) include at least one 9? Show
your work.
8.70 Phone numbers, II Refer to the previous exercise. Suppose that a computer
assigns the seven digits in Nick’s phone number at random, subject to the restriction in
Exercise 8.69(b). Find the probability that Nick’s phone number contains no repeated
digits. Show your work.
8.71 Free throws A basketball player claims to make 80% of her free throws.
Suppose this probability is the same for each free throw she attempts. In one game,
she shoots 7 free throws. Let X 5 the number of free throws on which she makes a
basket.
(a) Show that this is a binomial setting.
(b) Find the probability that she makes 4 or fewer baskets on 7 free throws. Show
your work.
(c) If you were watching this game and saw the player get a basket on 4 out of 7 free
throws, would you be suspicious about her claim of being an 80% free-throw shooter?
Justify your answer.
8.72 Getting a flush If you deal five cards from a well-shuffled deck, what’s the
probability that they’re all the same suit? (Card players call this kind of hand a “flush.”)
(a) Start by calculating the probability that all five cards are diamonds. Show your work.
(b) Now find the probability of getting a flush. Show your work.
8.73 We like opinion polls Are Americans interested in opinion polls about
the major issues of the day? Suppose that 40% of all adults are very interested
in such polls. (According to sample surveys that ask this question, 40% is about
right.) A polling firm chooses an SRS of 1015 people. If they do this many
times, the percent of the sample who say they are very interested will vary from
sample to sample following a Normal distribution with mean 40% and standard
deviation 1.5%.
(a) What is the probability that one such sample gives a result within 63% of the
truth about the population? Show your method.
(b) Find the probability that one sample misses the truth about the population by
4% or more. (This is the probability that the sample result is either less than 36% or
greater than 44%.) Show your work.
8.74 It pays to eat cereal Some time ago, Cheerios cereal boxes displayed a dollar
bill on the front of the box and a cartoon character who said, “Free $1 bill in every
20th box.” Mae Lee wants to know how many boxes of Cheerios she can expect to buy
in order to get one of the “free” dollar bills.
(a) Describe how you would establish a correspondence between boxes of cereal that
do and do not contain a dollar bill and numbers from Table B.
(b) Describe how you could use your calculator to simulate buying boxes of Cheerios
and checking for dollar bills.
(c) Carry out a simulation to answer Mae Lee’s question. Perform enough repetitions
so that when you pool your results with your classmates, you will have at least 25 rep-
etitions (don’t forget that the more repetitions, the better the results).
8.75 Keno, I The game of Keno is played with 80 balls numbered 1 through 80.
During each play of the game, the casino selects 20 of the balls at random. In a
simple version of the game, a player pays $1 and chooses one number between
416 CHAPTER 8 Probability Models

1 and 80. If the player’s number matches one of the 20 randomly selected balls,
the player wins $3 (and gets the original $1 back). If not, the player loses the
original $1 bet.
(a) Find the probability that the player’s chosen number is a winner. Show your work.
(b) Let X 5 the amount gained by the player in a single game of Keno. Give the prob-
ability distribution of X in tabular form.
(c) Find the expected value of X.
8.76 Keno, II Refer to the previous exercise. Suppose a player decides to play
25 games of this simple version of Keno. Let Y 5 the number of games won by the
player.
(a) Show that this is a binomial setting.
(b) Find and interpret P(Y 5 5). Show your work.
(c) Find and interpret P(Y . 5). Show your work.
(d) How much money should the player expect to win or lose in 25 games? Explain.
 PA RT D

INFERENCE
HOWARD HUGE*
To infer is
to draw a
conclusion
from evidence.

©2003 WM. Hoest Enterprises, Inc.

Statistical
inference draws a
conclusion about
a population
from evidence
provided by a “Howard's been like this ever since he heard that having a dog
sample. Drawing lowers your blood pressure.”

conclusions in mathematics is a matter of starting from a hypothesis and using

logical argument to prove without doubt that the conclusion follows. Statistics
isn’t like that.
Statistical conclusions are uncertain, because the sample isn’t the entire
population. So statistical inference has to state conclusions and also say how
uncertain they are. We use the language of probability to express uncertainty.
Because inference must both give conclusions and say how uncertain they are, it
is the most technical part of statistics.
Texts and courses intended to train people to do statistics spend most of
their time on inference. Our aim in this book is to help you understand statistics,
which takes less technique but often more thought. We will look only at a few
basic techniques of inference. The techniques are simple, but the ideas are subtle,
so prepare to think. To start, think about what you already know and don’t be
too impressed by elaborate statistical techniques: even the fanciest inference
cannot remedy fundamental flaws such as voluntary response samples or
uncontrolled experiments.

417
This page intentionally left blank
 CHAPTER 9

Introduction to
Inference
9.1 What Is a Confidence Interval?
9.2 What Is a Significance Test?
9.3 Use and Abuse of Statistical Inference

Don’t get mad

Know someone who is prone to anger? Nature has a
way to calm such people: they get heart disease more

© Timnichols/Dreamstime.com
often. Several observational studies have discovered a link
between anger and heart disease. The best study looked
at 12,986 people, both black and white, chosen at random
from four communities.1 When first examined, all subjects
were between the ages of 45 and 64 and were free of heart
disease. Let’s focus on the 8474 people in this sample who
had normal blood pressure.
A short psychological test, the Spielberger Trait Anger Scale, measured
how easily each person became angry. There were 633 people in the high
range of the anger scale, 4731 in the middle, and 3110 in the low range. Now
follow these people forward in time for almost six years and compare the rate
of heart disease in the high and low groups. There are some lurking variables:
people in the high-anger group are somewhat more likely to be male, to have
less than a high school education, and to be smokers and drinkers. After
adjusting for these differences, high-anger people were 2.2 times as likely to
get heart disease and 2.7 times as likely to have an acute heart attack as low-
anger people.
We know that the numbers 2.2 and 2.7 won’t be exactly right for the
population of all people aged 45 to 64 years with normal blood pressure. News
reports of the study cited those numbers, but the full report in the medical journal
Circulation gave confidence intervals.With 95% confidence, high-anger people are
between 1.36 and 3.55 times as likely to get heart disease as low-anger people.
They are between 1.48 and 4.90 times as likely to have an acute heart attack. The
intervals remind us that any statement we make about the population is uncertain
because we have data only from a sample. For the sample, we can say “exactly 2.2 419
420 CHAPTER 9 Introduction to Inference

times as likely.” For the whole population, the sample data allow us to say only
“between 1.36 and 3.55 times as likely,” and only with 95% confidence.To go behind
the news to the real thing, in medicine and in other areas, we must speak the
language of inference.

9.1 What Is a Confidence Interval?

ACTIVITY 9.1A
The candy machine
Imagine a very large candy machine filled with
www orange, brown, and yellow candies. Suppose
that 45% of the candies in the machine are
orange. When you insert money, the machine dispenses
a sample of 25 candies. Would you be surprised if
your sample contained 10 orange candies (that’s 40%
orange)? How about 5 orange candies (20% orange)? In
this Activity, you will use a Java applet to investigate the
sample-to-sample variability in the proportion of orange
candies dispensed by the machine.
1. Launch the Reese’s Pieces applet at
www.rossmanchance.com/applets/Reeses/
ReesesPieces.html. Click on the “Draw Samples”
button. An animated simple random sample of
25 candies should be dispensed. Figure 9.1 shows
the results of one such sample. Was your sample
proportion of orange candies (look at the value of p^
in the applet window) close to the actual population Figure 9.1 The result of taking one simple random
sample of 25 candies from a large candy machine in
proportion, p 5 0.45?
which 45% of the candies are orange.
2. Click “Draw Samples” 9 more times, so that you
have a total of 10 sample results. Look at the dotplot (that’s 40% orange)? How about 5 orange candies
of your p^ -values. What is the mean of your 10 sample (20% orange)? Explain.
proportions? What is their standard deviation?
5. How would the distribution of the sample
3. To take many more samples quickly, enter 190 proportion p^ change if the machine dispensed 50
in the “num samples” box. Click on the “Animate” candies each time instead of 25? “Reset” the applet.
box to turn the animation off. Then click “Draw Take 200 samples of 50 candies. Describe the shape,
Samples.” You have now taken a total of 200 samples center, and spread of the distribution of sample
of 25 candies from the machine. Describe the shape, proportions (p^ ).
center, and spread of the distribution of sample
6. Would you be surprised if a sample of 50 candies
proportions (p^ ) shown in the dotplot.
from the machine contained 20 orange candies
4. Would you be surprised if a sample of 25 candies (that’s 40% orange)? How about 10 orange candies
from the machine contained 10 orange candies (20% orange)? Explain.
 9. 1 What Is a Confidence Interval? 421

Estimating
Statistical inference draws conclusions about a population based on data from
a sample. One kind of conclusion answers questions like “What percent of teens
smoke?” or “What is the mean survival time for patients with a particular type
of cancer?” These questions ask about a number (a percent, a mean) that de-
scribes a population. Numbers that describe a population are parameters. To
estimate a population parameter, choose a sample from the population and use
a statistic, a number calculated from the sample, as your estimate. Here’s an
example.

Example 9.1 Teen smoking

How common is behavior that puts people at risk of lung cancer? The 2007 Youth
Risk Behavior Survey questioned a nationally representative sample of 14,041 stu-
dents in grades 9 to 12. Of these, 2808 said they had smoked cigarettes at least one
day in the past month. That’s 20.0% of the sample.2 This result may be biased by
reluctance to tell the truth about any kind of risky behavior. For now, assume that
the people in the sample told the truth. Based on these data, what can we say about
the percent of all high schoolers who smoked?
Our population is high school students in the United States. The parameter is
the proportion that have smoked cigarettes in the past month. Call this unknown
parameter p, for “proportion.” The statistic that estimates the parameter p is the
sample proportion

count in the sample

p^ 5
sample size
2808
5 5 0.200
14,041

© Reprinted with special permission of King Feature Syndicate.

A basic strategy in statistical inference is to use a sample statistic to estimate a

population parameter. Once we have the sample data, we estimate that the propor-
tion of all high school students who have smoked cigarettes in the past month p is
“about 20.0%” because the proportion in the sample was exactly 0.200. We can only
estimate that the truth about the population is “about” 20.0% because we know that
422 CHAPTER 9 Introduction to Inference

the sample result is unlikely to be exactly the same as the true population propor-
tion. A 95% confidence interval makes that “about” precise.

95% confidence interval

A 95% confidence interval is an interval calculated from sample data

that is guaranteed to capture the true population parameter in 95% of
all samples.

Let’s look at how to calculate a confidence interval for a population proportion.

Then we’ll reflect on what we have done and generalize a bit.

Estimating with confidence

We want to estimate the proportion p of the individuals in a population who have
some characteristic—they are orange (candies), or they smoke (teens), for example.
Let’s call the characteristic we are looking for a “success.” We use the proportion
p^ (read as “p-hat”) of successes in a simple random sample (SRS) to estimate the
proportion p of successes in the population. How good is the statistic p^ as an esti-
mate of the parameter p? To find out, we ask, “What would happen if we took many
samples?” Well, we know that p^ would vary from sample to sample. If you did
Activity 9.1A, you saw that this sampling variability isn’t haphazard. It has a clear
pattern in the long run, a pattern that is pretty well described by a Normal curve.
Here are the facts.

Sampling distribution of a sample proportion

The sampling distribution of a statistic is the distribution of values

taken by the statistic in all possible samples of the same size from the
same population.

Take an SRS of size n from a large population that contains proportion

p of successes. Let p^ be the sample proportion of successes,
count of successes in sample
p^ 5
n
Then, if the sample is large enough:

• The sampling distribution of p^ is approximately Normal.

• The mean of the sampling distribution is p.
• The standard deviation of the sampling distribution is
p11 2 p2
Å n
 9. 1 What Is a Confidence Interval? 423

These facts can be proved by mathematics, so they are a solid starting point.
Figure 9.2 summarizes them in a form that also reminds us that a sampling distri-
bution describes the results of lots of samples from the same population.

SRS size n
p
SRS size n p(1 – p)
p n
SRS s
ize n
p
Mean p
Population
proportion p
Values of p

Figure 9.2 Repeat many times the process of selecting an SRS of size n from a popula-
tion in which the proportion p are successes. The values of the sample proportion of
successes p^ have this Normal sampling distribution.

Example 9.2 Risky behavior

Suppose the truth is that 21% of high school students smoke. Then in the setting of
Example 9.1, p 5 0.21. The Youth Risk Behavioral Survey sample of size n 5 14,041
would, if repeated many times, produce sample proportions that closely follow the
Normal distribution with

mean 5 p 5 0.21

and

p11 2 p2
standard deviation 5
Å n
0.21 1 0.79 2
5 5 0.0034
Å 14,041

The center of this Normal distribution is at the truth about the population. That’s
the absence of bias in random sampling once again. The standard deviation is small
because the sample is quite large. So almost all samples will produce a statistic that
is close to the true p. In fact, the 95 part of the 68–95–99.7 rule says that 95% of all
sample outcomes will fall between

mean 2 2 standard deviations 5 0.21 2 2(0.0034) 5 0.21 2 0.0068 5 0.2032

and

mean 1 2 standard deviations 5 0.21 1 2(0.0034) 5 0.21 1 0.0068 5 0.2168

Figure 9.3 (on the next page) displays these facts.

424 CHAPTER 9 Introduction to Inference

0.2032 0.2168

95% of all
values of p

0.1998 0.2032 0.2066 0.21 0.2134 0.2168 0.2202

Values of the sample proportion

Figure 9.3 Repeat many times the process of selecting an SRS of size 14,041 from a
population in which the proportion p 5 0.21 are successes. The middle 95% of the values
of the sample proportion will lie between 0.2032 and 0.2168.

So far, we have just put numbers on what we already knew. We can trust the re-
sults of large random samples because almost all such samples give results that are
close to the truth about the population. The numbers say that in 95% of all samples
of size 14,041, the statistic p^ and the parameter p are within 0.0068 of each other.
We can put this another way: 95% of all samples give an outcome p^ such that the
population truth p is captured by the interval from p^ 2 0.0068 to p^ 1 0.0068. The
0.0068 came from substituting p 5 0.21 into the formula for the standard deviation
of p^ . For any value of p, the general fact is: When the population proportion has
the value p, about 95% of all samples catch p in the interval extending 2 stan-
dard deviations on either side of p^ . That’s the interval

p11 2 p2
p^ 6 2
Å n

Is this the 95% confidence interval we want? Not quite. The interval can’t be
found just from the data because the standard deviation involves the population
proportion p, and in practice we don’t know p. In Example 9.2, we applied the for-
mula using p 5 0.21, but this may not be the true p.
What should we do? Well, the standard deviation of the statistic does depend
on the parameter p, but it doesn’t change a lot when p changes. Let’s go back to
Example 9.2 and redo the calculation for other values of p. Here’s the result:

Value of p: 0.19 0.20 0.21 0.22 0.23

Standard deviation: 0.0033 0.00338 0.00344 0.0035 0.00355
 9. 1 What Is a Confidence Interval? 425

We see that if we guess a value of p reasonably close to the true value, the stan-
dard deviation found from the guessed value will be about right. We know that,
when we take a large random sample, the statistic p^ is almost always close to the
parameter p. So we will use p^ as the guessed value of the unknown p. Now we have
an interval that we can calculate from the sample data.

95% confidence interval for a proportion

Choose an SRS of size n from a large population that contains an

unknown proportion p of successes. Call the proportion of successes
in this sample p^ . An approximate 95% confidence interval for the
parameter p is
p^ 11 2 p^ 2
p^ 6 2
Å n

Example 9.3 A confidence interval for youthful smokers

The Youth Risk Behavioral Survey random sample of 14,041 high school students
found that 2808 had smoked cigarettes at least one day in the past month, a sample
proportion p^ 5 0.200. The 95% confidence interval for the proportion of all high
school students who had smoked in the past month is

p^ 11 2 p^ 2 0.200 1 0.800 2
p^ 6 2 5 0.200 6 2
Å n Å 14,041
5 0.200 6 2 1 0.0034 2
5 0.200 6 0.0068

or the interval

1 0.1932, 0.2068 2

Interpret this result as follows: we got this interval by using a method that
catches the true unknown population proportion in 95% of all samples. The short-
hand is: we are 95% confident that the true proportion of high school students who
smoked in the past month lies between 19.32% and 20.68%.

Exercises 9.1 A student survey Tonya wants to estimate what proportion of the seniors in
her school plan to attend the prom. She interviews an SRS of 50 of the 750 seniors
in her school. She finds that 36 plan to attend the prom.
(a) What population does Tonya want to draw conclusions about?
(b) What does the population proportion p represent in this setting?
(c) What is the numerical value of the sample proportion p^ from Tonya’s sample?
426 CHAPTER 9 Introduction to Inference

9.2 Our dirty little secret More than 90% of people surveyed say they always
wash their hands after using public restrooms. But observers posted in public
restrooms found that only 4679 of 6076 observed individuals actually washed
their hands.3
(a) What population does inference concern here?
(b) Explain clearly what the population proportion p represents in this setting.
(c) What is the numerical value of the sample proportion p^ ?
9.3 Marijuana use A 2007–2008 survey of teens conducted by the Partnership
for a Drug-Free America found that 33% of 6511 U.S. teenagers surveyed
reported use of marijuana. The report states, “With 95% confidence, we can say
that between 31.4% and 34.6% of all American teenagers have used marijuana.”4
Explain to someone who knows no statistics what the phrase “95% confidence”
means in this report.
9.4 Do you drink the cereal milk? A USA Today poll asked a random sample of
1012 U.S. adults what they do with the milk in the bowl after they have eaten the
cereal. Suppose that 70% of all U.S. adults actually drink the cereal milk.
(a) Find the mean and standard deviation of the proportion p^ of the sample who
say they drink the cereal milk.
(b) Draw a Normal curve like the one in Figure 9.3 that shows the sampling
distribution of p^ in this case.
(c) Of the 1012 adults surveyed, 67% said they drink the cereal milk. Plot this
sample result on the graph from (b).
(d) Using your Normal curve, find the probability of obtaining a sample in which
67% or fewer say they drink the cereal milk. Do you think it’s plausible that the
true population proportion is p  0.70? Explain.
9.5 Polling women A New York Times Poll on women’s issues interviewed 1025
women randomly selected from the United States, excluding Alaska and Hawaii.
Of the women in the sample, 482 said they do not get enough time for themselves.
Although the samples in national polls are not SRSs, they are close enough that
our method gives approximately correct confidence intervals.
(a) Explain in words what the parameter p is in this setting.
(b) Use the poll results to give a 95% confidence interval for p.
(c) Write a short explanation of your findings in (b) for someone who knows no
statistics.
9.6 Gun violence The Harris Poll asked a sample of 1009 adults which causes
of death they thought would become more common in the future. Gun violence
topped the list: 706 members of the sample thought deaths from guns would
increase. Although the samples in national polls are not SRSs, they are close
enough that our method gives approximately correct confidence intervals.
(a) Say in words what the population proportion p is for this poll.
(b) Find a 95% confidence interval for p.
(c) Harris announced a margin of error of plus or minus 3 percentage points for
this poll result. How well does your work in (b) agree with this margin of error?
 9. 1 What Is a Confidence Interval? 427

Understanding confidence intervals

Our 95% confidence interval for a population proportion has the familiar form

estimate 6 margin of error

We know that news reports of sample surveys, for example, usually give the estimate
and the margin of error separately: “A new Gallup Poll shows that 66% of women favor
new laws restricting guns. The margin of error is plus or minus four percentage points.”
This result suggests that between 62% and 70% of the population of women favor these
laws. We also know that news reports usually leave out the level of confidence.
Not all confidence intervals have this form. Here’s a complete description of a
confidence interval.

Confidence interval

A level C confidence interval for a parameter has two parts:

• An interval calculated from the data.

• A confidence level C, which gives the probability that the interval
will capture the true parameter value in repeated samples.

ACTIVITY 9.1B
The Confidence Intervals applet
AP Go to the Statistics Through Applications Web
PLET Directions Click the “Sample” button to have the
site www.whfreeman.com/sta2e and launch applet take a simple random sample and construct
the Confidence Intervals applet. Although a confidence interval. Click the “Sample 50” button
this applet involves constructing confidence intervals to have the applet take 50 different simple random
for the population mean, m, the ideas of this Activity samples and make 50 different confidence intervals
apply to both means and proportions. from the resulting data.
Think
1. If you make just one 90% confidence
interval, will it necessarily capture the
parameter value?
2. If you make 10 different 90%
confidence intervals, how many of
them do you think will capture the
parameter?
3. If you make a thousand 90% confidence
intervals, how many of them do you think
will capture the parameter?

ACTIVITY 9.1B (continued)

428 CHAPTER 9 Introduction to Inference

ACTIVITY 9.1B (continued)

APPLET TIP: The cumulative totals are cleared when happens to the number of intervals that fail to
you click the “Reset” button. When you change the capture the parameter?
confidence level, the totals pertain to the displayed 7. Have the applet make 50 new confidence
intervals. intervals. The dots in the center of the intervals
mark the sample mean. Are the intervals produced
Test
by the applet symmetric around the sample mean?
4. Test your answers to the preceding three
The tick marks on the horizontal axis below the
questions through repeated use of the applet. How
Normal curve are each separated by one standard
far did the results vary from what you expected in
deviation of the sample mean. According to the
Questions 2 and 3? How far did the results vary
68–95–99.7 rule, how many of the 50 sample
from what you expected on a percentage basis?
means would you expect to fall within one standard
deviation of the parameter? How many of the 50
Explore
sample means did fall within one standard deviation?
5. Set the confidence level to 80%, and sample 50
confidence intervals. How many of these intervals 8. How is the margin of error related to the length
fail to capture the parameter (they are drawn in red of the interval? Explain.
on the computer screen)? 9. Examine the 50 intervals at the 95% confidence
6. Increase the confidence level in succession to level. What is the typical margin of error for
90%, then to 95%, and then to 99% while watching these intervals measured in units of the standard
the changing behavior of the 50 intervals. What deviation of the mean? Explain how this is related to
happens to the lengths of the intervals? What the Normal curve.

There are many formulas for confidence intervals that apply in different situ-
ations. Be sure you understand how to interpret a confidence interval. The inter-
pretation is the same for any formula, and you can’t use a calculator or a com-
puter to do the interpretation for you. As Activity 9.1B illustrates, confidence
intervals use the central idea of probability: ask what would happen if we
repeated the sampling many times. The 95% in a 95% confidence interval is a
probability, the probability that the method produces an interval that does cap-
ture the true parameter.

Example 9.4 How confidence intervals behave

The Youth Risk Behavior Survey sample of 14,041 high school seniors found 2808
smokers, so the sample proportion was

2808
p^ 5 5 0.200
14,041

and the 95% confidence interval was

p^ 1 1 2 p^ 2 0.200 1 0.800 2
p^ 6 2 5 0.200 6 2 5 0.200 6 0.0068
Å n Å 14,041
 9. 1 What Is a Confidence Interval? 429

Draw a second sample from the same population. It finds 2877 of its 14,041
respondents who smoke. For this sample,

2877
p^ 5 5 0.205
14,041
p^ 1 1 2 p^ 2 0.205 1 0.795 2
p^ 6 2 5 0.205 6 2 5 0.205 6 0.0068
Å n Å 14,041

Draw another sample. Now the count is 2719 and the sample proportion and
confidence interval are

2719
p^ 5 5 0.194
14,041
p^ 1 1 2 p^ 2 0.194 1 0.806 2
p^ 6 2 5 0.194 6 2 5 0.194 6 0.0067
Å n Å 14,041

Keep sampling. Each sample yields a new estimate and a new confidence inter-
val. If we sample forever, 95% of these intervals will capture the true parameter. This
is true no matter what the true value is. Figure 9.4 summarizes the behavior of the
confidence interval in graphical form.

SRS size 14,041

0.200 
 0.0068
SRS size 14,041 0.205 
 0.0068
95% of these
SRS size 14,041 0.194 
 0.0067 intervals cover
the true p

Population
proportion p
Figure 9.4 Repeated samples from the same population give different 95% confidence
intervals, but 95% of these intervals capture the true population proportion p.

Figures 9.4 and 9.5 (on the next page) show how confidence intervals behave.
Example 9.4 and Figure 9.4 remind us that repeated samples give different results
and that we are guaranteed only that 95% of the samples give a correct result.
Figure 9.5 goes behind the scenes. The vertical line is the true value of the
population proportion p. The Normal curve at the top of the figure is the sampling
distribution of the sample statistic, which is centered at the true p. We are behind
the scenes because in real-world statistics we don’t know p. The 95% confidence
intervals from 25 SRSs appear one after the other. The central dots are the values
of p^ , the centers of the intervals. The arrows on either side span the confidence in-
terval. In the long run, 95% of the intervals will cover the true p and 5% will miss.
Of the 25 intervals in Figure 9.5, 24 hit and 1 misses. (Remember that probability
430 CHAPTER 9 Introduction to Inference

Sampling
distribution of p

Values of p

Figure 9.5 Twenty-five samples from the same population give these 95% confidence
intervals. In the long run, 95% of all such intervals cover the true population proportion,
marked by the vertical line.

describes only what happens in the long

run—we don’t expect exactly 95% of 25
intervals to capture the true parameter.)
Don’t forget that our interval is only
approximately a 95% confidence interval.
It isn’t exact for two reasons. The sampling
distribution of the sample proportion p^
isn’t exactly Normal. And we don’t get the
standard deviation of p^ exactly right, be-
cause we used p^ in place of the unknown p.
Both of these difficulties become less seri-
ous as the sample size n gets larger. So our
formula is good only for large samples.
What’s more, our method assumes
that the population is really big—at least
10 times the size of the sample. Profes-
sional statisticians use more elaborate
methods that take the size of the popula-
© The New Yorker Collection 1992 Dana Fradon from cartoonbank.com. All rights reserved.
tion into account and work even for small
samples.5 But our method works well enough for many practical uses. More impor-
tant, it shows how we get a confidence interval from the sampling distribution of a
statistic. That’s the reasoning behind any confidence interval.
 9. 1 What Is a Confidence Interval? 431

Confidence intervals for a population proportion

We used the 95 part of the 68–95–99.7 rule to get a 95% confidence interval for the pop-
ulation proportion. Perhaps you think that a method that works 95% of the time isn’t
good enough. You want to be 99% confident. For that, we need to mark off the central
99% of a Normal distribution. For any probability C between 0 and 1, there is a number
z* such that any Normal distribution has probability C within z* standard deviations of
the mean. Figure 9.6 shows how the probability C and the number z* are related.

Shaded area represents

probability C
z* standard deviations

3 2 1 0 1 2 3

–z* z*
Standard deviations above and below the mean

Figure 9.6 Critical values z* of the Normal distributions. In any Normal distribution,
there is area (probability) C under the curve between z* and z* standard deviations
away from the mean.

Table 9.1 gives the numbers z* for various choices of C. For convenience, the table
gives C as a confidence level in percent. The numbers z* are called critical values of
the Normal distributions. Table 9.1 shows that any Normal distribution has prob-
ability 99% within 62.58 standard deviations of its mean. The table also shows that
any Normal distribution has probability 95% within 61.96 standard deviations of
its mean. The 68–95–99.7 rule uses 2 in place of the critical value z* 5 1.96. That is
good enough for practical purposes, but the table gives the more exact value.

Table 9.1 Critical Values of the Normal Distribution

Confidence level C Critical value z* Confidence level C Critical value z*

50% 0.67 90% 1.64

60% 0.84 95% 1.96
70% 1.04 99% 2.58
80% 1.28 99.9% 3.29
432 CHAPTER 9 Introduction to Inference

From Figure 9.6 we see that, with probability C, the sample proportion takes
a value within z* standard deviations of p. That is just to say that, with probability
C, the interval extending z* standard deviations on either side of the observed p^
captures the unknown p. Using the estimated standard deviation of p^ produces the
following result.

Confidence interval for a population proportion

Choose an SRS of size n from a population of individuals of which

proportion p are successes. The proportion of successes in the
sample is p^ . When n is large, an approximate level C confidence
interval for p is
p^ 1 1 2 p^ 2
p^ 6 z*
Å n
where z* is the critical value for probability C from Table 9.1.

Example 9.5 A 99% confidence interval

The Youth Risk Behavior Survey random sample of 14,041 high school students
found that 2808 were smokers. We want a 99% confidence interval for the propor-
tion p of all high school students who have smoked cigarettes in the past month.
Table 9.1 says that for 99% confidence, we must go out z*  2.58 standard devia-
tions. Here are our calculations:

2808
p^ 5 5 0.200
14,041
p^ 1 1 2 p^ 2 0.200 1 0.800 2
p^ 6 z* 5 0.200 6 2.58
Å n Å 14,041
5 0.200 6 2.58 0.00338 2
1
5 0.200 6 0.0087

So our 99% confidence interval is (0.1913, 0.2087).

Who is a smoker? We are 99% confident that the true population proportion is between 19.13%
When estimating a propor- and 20.87%. That is, we got this range of percents by using a method that gives a
tion p, be sure you know
correct answer 99% of the time.
what counts as a “success.”
The news says that 20%
of adolescents smoke.
Shocking. It turns out that Compare Example 9.5 with the calculation of the 95% confidence interval in
this is the percent who Example 9.3. The only difference is the use of the critical value 2.58 for 99% con-
smoked at least once in the fidence in place of 2 for 95% confidence. That makes the margin of error for 99%
past month. If we say that confidence larger and the confidence interval wider. Higher confidence isn’t free—
a smoker is someone who
smoked on at least 20 of we pay for it with a wider interval. Figure 9.6 reminds us why this is true. To cover
the past 30 days, only 8% of a higher percent of the area under a Normal curve, we must go farther out from
adolescents qualify. the center.
 9. 1 What Is a Confidence Interval? 433

Exercises 9.7 Interpreting poll results Here’s a quote from a news report about a sample
survey: “A new Gallup Poll shows that 66% of women favor new laws restricting
guns. The margin of error is plus or minus four percentage points.” The news
report omitted the confidence level, which was 95%. Explain what this poll result
means to someone who knows little statistics. Be sure to discuss the meaning of
both the interval and the confidence level.
9.8 Confidence level visual Figure 9.5 (page 430) shows the meaning of 95%
confidence. Trace the sampling distribution in Figure 9.5. Then draw 10 intervals
that illustrate the meaning of 80% confidence. Be sure to consider the lengths of
your intervals.
9.9 98% confidence Table 9.1 (page 431) gives critical values z* for several
confidence levels. Find z* for a 98% confidence interval using Table A. Show your
method clearly.
9.10 93% confidence Table 9.1 (page 431) gives critical values z* for several
confidence levels. Find z* for a 93% confidence interval using Table A. Show your
method clearly.
9.11 Going to the prom The prom committee of Exercise 9.1 (page 425)
wants to know how many seniors they can expect to attend the prom so that
they can develop a budget. An SRS of 50 of the 750 seniors in the school
yielded 36 who said they plan to attend the prom.
(a) Construct and interpret a 95% confidence interval for the true
proportion of seniors who will attend the prom.
(b) The committee asks if you can provide a “more precise” interval.
Should you use a 90% confidence level or a 99% confidence level? Explain.
(c) Construct a new interval using the confidence level you chose in (b).
Interpret the resulting interval for the prom committee in simple language.
9.12 Random digits We know that the proportion of 0s among a large
set of random digits is p  0.1 because all 10 possible digits are equally
probable. The entries in a table of random digits are a sample from the
population of all random digits. To get an SRS of 200 random digits, look
at the first digit in each of the 200 five-digit groups in lines 101 to 125 of
Table B in the back of the book.
ckphoto
uss/iSto (a) How many of these 200 digits are 0s? Give a 95% confidence interval for the
Kevin R

proportion of 0s in the population from which these digits are a sample.

(b) Does your interval include the true parameter value, p  0.1? What does this tell you?

CALCULATOR CORNER Confidence interval for a

population proportion
You can use your TI-84/TI-Nspire to construct a confidence interval
for a population proportion. We will use the teen smoking data from
Examples 9.1 to 9.5 to illustrate.
CALCULATOR CORNER (continued)
434 CHAPTER 9 Introduction to Inference

CALCULATOR CORNER (continued)

TI-84
• Press A, choose TESTS, and then choose 1-PropZInt.
• Enter the number of teen smokers, 2808, as the count x and the number of
teens in the sample, 14041, as the sample size n. Specify the confidence level as .95.
Then highlight “Calculate” and press b.

TI-Nspire
You can calculate confidence intervals in a Calculator page or a Lists &
Spreadsheet page. We’ll show the steps for a Calculator page here.
• On a Calculator page, press b, and choose 5:Statistics,
6:Confidence Intervals…, and 5:1-Prop z Interval.
• In the dialog box, enter the number of teen smokers, 2808, as the number of
“Successes” x and the number of teens in the sample, 14,041, as the sample size
n. Specify the confidence level as 0.95. Then click OK.

DATA EXPLORATION
No Child Left Behind
In 2002, the Bush administration sponsored would be classified as failing schools. Students at
legislation known as the “No Child Left Behind” those schools would be able to transfer to other,
act, which attempted to improve public education supposedly better, schools. Many states established
by mandating a testing program and minimum 40% as the minimum passing rate.
standards. The act left it up to each state to develop Principals and school system administrators
a statewide testing program in reading and math don’t want their schools to be declared failing
and a minimum proficiency level. Schools that schools, so many states have adopted a controversial
didn’t meet the minimum performance level procedure to buy some “wiggle room” on test results.
 9. 1 What Is a Confidence Interval? 435

Here is their position. Instead of considering test 2. What if the Oregon school had tested 75
results for one year to be the entire population of students this year instead of 50? Calculate a new
interest, they argue that test results should be viewed 95% confidence interval, still assuming a 28% pass
as a sample of school performance from a population rate in the sample. Explain why the school would
of multiple years. Each year’s results come from a not meet the state requirement now.
sample of students gathered at one point in time,
3. Here’s another idea: the school could use a 99%
they say. Consequently, it makes sense to construct
confidence level instead of 95%. Recalculate the
confidence intervals for the population passing rate.
confidence interval from Question 2, this time
At this writing, 35 states use the confidence interval
using 99% confidence. What would you conclude?
approach or a variation of it to interpret test scores.
In fact, if the Oregon school had up to 92
Critics argue that using the confidence interval
students take the test and 28% passed, then the
approach to determine which schools pass distorts
state’s 40% requirement would be within the 99%
the truth. For example, a small high school in Oregon
confidence interval and the school would be
met that state’s requirement of a 40% pass rate on the
considered passing. Life is good. Thirteen states use
state reading test even though only 28% of the school’s
a 99% confidence interval.
students passed the test that year. And in Maryland,
an elementary school met that state’s standards even 4. Show how it is possible for the Maryland
though only 31% of its pupils passed a state math test elementary school to meet the state’s 41.5%
when a 41.5% pass rate was required.6 requirement even though only 31% of the school’s
students passed the state math test that year.
1. Suppose the Oregon school tested 50 students
this year, and that 28% of this sample passed. 5. Do you think a confidence interval approach to
Calculate the resulting 95% confidence interval. interpreting test scores is appropriate? What do you
Explain why the school could be considered to have think is the most compelling reason for or against
met Oregon’s requirement. this practice?

Determining sample size

National survey organizations like the Gallup Poll typically sample between 1000
and 1500 American adults who are interviewed by telephone. Why do they choose
such sample sizes? The following example shows you why.

Example 9.6 Sample size for a desired margin of error

National polls like to use the 95% confidence level because it has become a standard
and people are used to it. Similarly, pollsters like to report results with a margin of
error of 3 percentage points. Media reports of polls frequently have information
about how the poll was conducted and how to interpret the results. How many
people do the pollsters need to interview in order to give a 3% margin of error at
the 95% confidence level? The details require some algebra.
The form for the confidence interval is

p^ 1 1 2 p^ 2
p^ 6 z*
Å n
where p^ is the proportion of people who respond in a certain way and the term fol-
lowing the sign is the margin of error. For the 95% confidence interval, z*  1.960,
correct to three decimal places.
436 CHAPTER 9 Introduction to Inference

Since we want the margin of error to be 3 percentage points, we have to solve

the equation

p^ 1 1 2 p^ 2
5 0.03
Å
z*
n

Notice that we use the decimal form for 3%. Substituting, we have

p^ 1 1 2 p^ 2
5 0.03
Å
1.96
n

Multiplying both sides by !n and then dividing both sides by 0.03 yields

"p^ 1 1 2 p^ 2 5 !n
1.96
0.03

Squaring both sides of the equation gives

b p^ 1 1 2 p^ 2 5 n
1.96 2
a
0.03

The largest value of n will be required if p^ 5 0.5, since that will make the product
p^ 1 1 2 p^ 2 as large as possible. With p^ 5 0.5, we have

b 1 0.5 2 1 1 2 0.5 2 5 n
1.96 2
a
0.03
1067.111 5 n

Since you can’t interview 0.111 of a person, you have to round up to the next whole
number of people: 1068. Why not round to the nearest whole number—in this case,
1067? Because a smaller sample size will result in a larger margin of error, possibly
more than the desired 3% for the poll.

ws
Ne
ws
Ne

APPLICATION 9.1 B]
RO
g
<
Se
a

The beads
Before class, your teacher will prepare a large popula- class should use a small cup to scoop out a sample
tion of different-colored beads and put them into a of beads. Explain why this sampling method
container that you cannot see inside. Your goal is to es- should result in a simple random sample from the
timate the actual proportion of beads in the population population.
that have a particular color (specified by your teacher).
2. Count the total number of beads selected and
QUESTIONS the number having the specified color. What
1. Your teacher will mix the beads in the container proportion of the beads in the sample have the
thoroughly. Without looking, a member of your specified color?
 9. 1 What Is a Confidence Interval? 437

3. Calculate and interpret a 95% confidence 4. Is it possible that your confidence interval from
interval for the proportion of beads in the Question 3 does not capture the actual population
population that have the specified color. proportion? Explain your answer.

Section 9.1 Summary

Statistical inference draws conclusions about a population on the basis of data
from a sample. Because we don’t have data for the entire population, our conclu-
sions are uncertain. A confidence interval estimates an unknown parameter in a
way that tells us how uncertain the estimate is. The interval itself says how closely
we can pin down the unknown parameter. The confidence level is a probability
that says how often in many samples the method produces an interval that does
contain the parameter. Confidence intervals are based on the sampling distribu-
tion of a statistic, which shows how the statistic varies in repeated sampling.
In this section we found one specific confidence interval, for the proportion p
of “successes” in a population, based on an SRS from the population. To calculate
such an interval, we use critical values z* from a Normal distribution. You will find
more advice on interpreting confidence intervals in Section 9.3.

Section 9.1 9.13 Determining sample size, I Suppose a polling organization wants to report their
Exercises results using a 99% confidence level with a margin of error of 3 percentage points. Use
the method of Example 9.6 to find the number of people they would need to interview.
9.14 Determining sample size, II Suppose a polling organization wants to
report their results at the 95% confidence level with a margin of error of 4 per-
centage points. Use the method of Example 9.6 to find the number of people they
would need to interview.
9.15 The good life Roper ASW, a market research and consulting firm,
conducted in-person interviews with 2004 adults. When asked if they felt they
had achieved the “good life,” only 180 of the respondents said “Yes.” This was
despite the fact that a majority of respondents had the things they said consti-
tuted the good life: a house, good health, a car, and children. What is the mar-
gin of error for this poll? (Assume a 95% confidence level.) Show your work.
9.16 Losing weight A Gallup Poll in November 2008 found that 59% of the peo-
ple in its sample said “Yes” when asked, “Would you like to lose weight?” Gallup
announced: “For results based on the total sample of national adults, one can say
with 95% confidence that the margin of (sampling) error is 3 percentage points.”7
(a) What is the 95% confidence interval for the percent of all adults who want to
lose weight?
(b) Explain clearly to someone who knows no statistics why we can’t just say that
59% of all adults want to lose weight.
(c) Then explain clearly what “95% confidence” means.
(d) As Gallup indicates, the 3% margin of error for this poll includes only sampling
variability (what they call “sampling error”). What other potential sources of error
(Gallup calls these “nonsampling errors”) could affect the accuracy of the 59% estimate?
438 CHAPTER 9 Introduction to Inference

9.17 Harley motorcycles Harley-Davidson motorcycles make up 14% of all the

motorcycles registered in the United States. You plan to interview an SRS of 600
motorcycle owners.
(a) Describe the shape, center, and spread of the sampling distribution of the
sample proportion p^ who own Harleys.
(b) How likely is your sample to contain 18.2% or more who own Harleys? Show
your work.
(c) How likely is your sample to contain less than 12% Harley owners? Show your work.
9.18 The quick method The quick method of Section 5.2 uses 1y !n as a rough
estimate for the margin of error for a 95% confidence interval for a population
proportion. The margin of error from the quick method is actually a bit larger
than needed. It differs most from the more accurate method of this chapter when
p^ is close to 0 or 1. An SRS of 600 motorcycle registrations finds that 68 of the
A baseball poll motorcycles are Harley-Davidsons. Calculate a 95% confidence interval for the
Polls can sometimes yield proportion of all motorcycles that are Harleys by the quick method and then by the
surprising and entertaining method of this chapter. How much larger is the quick-method margin of error?
results. A Sports Illustrated 9.19 Do college students pray? Social scientists asked 127 undergraduate stu-
poll conducted in summer dents “from courses in psychology and communications” about prayer and found
2003 asked 550 Major
League Baseball players to
that 107 prayed at least a few times a year.8
name the greatest living (a) To what population is it reasonable to generalize the results of this survey?
player. Barry Bonds was Justify your answer.
named most often (39.9%). (b) Calculate and interpret a 99% confidence interval for the population proportion p.
Next was Alex Rodriguez 9.20 Wildlife management Wildlife biologists know that males of most big-
(12.8%), followed by Willie game species can be “harvested” without affecting the total population because
Mays (12.1%), Nolan Ryan
(7.1%), Hank Aaron (6.7%), they tend to mate with several females a season. It is the female that requires
and Pete Rose (3.6%). Babe the most protection. In a recent year, Virginia hunters killed 928 black bears
Ruth was named by 8 of the during the fall hunting season. Of these, 328 were females.9 Consider the 928
players (1.5%). Babe Ruth bears a random sample of the population of all bears killed over many seasons.
died in 1948. (a) What is the sample proportion of females harvested?
(b) Find a 95% confidence interval for the true proportion of female bears killed
in Virginia over many seasons.
(c) What is the margin of error for your confidence interval in (b)? What two fac-
tors determine the margin of error?

9.2 What Is a Significance Test?

ACTIVITY 9.2A
Is this deck fair?
Materials: Three decks of playing cards, provided by 1. If a deck is fair, then the actual proportion of red
your teacher cards in the deck is p  _______. For each of the
three decks that your teacher will present, start out
Your teacher has what appear to be three standard decks
assuming that the deck is fair.
of playing cards. In this Activity, your task is to try to
determine whether each deck is “fair” by collecting data 2. The first deck: Your teacher will have one student
about the distribution of black and red cards in the deck. draw a card, note the color, and replace the card in the
 9. 2 What Is a Significance Test? 439

deck. Write RED and BLACK on the board and use a until the class decides whether the deck is fair. How
tally mark to record the result of the first draw. After many draws did it take?
the deck has been shuffled, a second student should
4. The third deck: Your teacher will introduce a
draw a card and tally the result. Continue until your
third deck. Repeat the process described in Step 2
class reaches a consensus on the proportion p of red
until the class decides whether the deck is fair. How
cards in the deck. Do you think this deck is fair?
many draws did it take? Once the class has made a
3. The second deck: Your teacher will introduce a decision about all three decks, your teacher will
second deck. Repeat the process described in Step 2 reveal how you did.

The reasoning of statistical tests

Confidence intervals are one of the two most common types of statistical inference.
Use a confidence interval when your goal is to estimate a population parameter.
The second common type of inference, called significance tests, has a different
goal: to assess the evidence provided by data about some claim concerning a popu-
lation. Here’s a simple example.

Example 9.7 Shooting free throws

The local hot-shot playground basketball player claims to make 80% of his free
throws. “Show me,” you say. He shoots 20 free throws and makes 8 of them. “Aha,”
you conclude, “if he makes 80%, he would almost never make as few as 8 of 20. So
I don’t believe his claim.”
Your reasoning is based on asking what would happen if the shooter’s claim
were true and we repeated the sample of 20 free throws many times—he would
almost never make as few as 8. This outcome is so unlikely that it gives strong evi-
dence that his claim is not true.
You can say how strong the evidence against the shooter’s claim is by giving the
probability that he would make as few as 8 out of 20 free throws if he really makes
80% in the long run. This probability is 0.0001. The shooter would make as few as 8
of 20 only once in 10,000 tries in the long run if his claim to make 80% is true. Th e
small probability convinces you that his claim is false.

Example 9.7 shows you the reasoning of statistical tests at the playground level:
an outcome that is very unlikely if a claim is true is good evidence that the claim is
not true.
Statistical inference uses data from a sample to draw conclusions about a
population. So once we leave the playground, statistical tests deal with claims
about a population. Tests ask if sample data give good evidence against a claim. A
test says, “If we took many samples and the claim were true, we would rarely get a
result like this.” To get a numerical measure of how strong the sample evidence is,
replace the vague term “rarely” by a probability. Here is an example of this reason-
ing at work.
440 CHAPTER 9 Introduction to Inference

Example 9.8 Is the coffee fresh?

People of taste are supposed to prefer fresh-brewed coffee to the instant variety. On
the other hand, perhaps many coffee drinkers just want their caffeine fix. A skeptic
claims that only half of all coffee drinkers prefer fresh-brewed coffee. Let’s do an
experiment to test this claim.
Each of 50 subjects tastes two unmarked cups of coffee and says which he or
she prefers. One cup in each pair contains instant coffee; the other, fresh-brewed
coffee. The statistic that records the result of our experiment is the proportion of
the sample who say they like the fresh-brewed coffee better. We find that 36 of our
50 subjects choose the fresh coffee. That is,

36
p^ 5 5 0.72 5 72%
50

To make a point, let’s compare our outcome p^ 5 0.72 with another possible
result. If only 28 of the 50 subjects like the fresh coffee better than the instant coffee,
the sample proportion is

28
p^ 5 5 0.56 5 56%
50

Surely 72% is stronger evidence against the skeptic’s claim than 56%. But how
much stronger? Is even 72% in favor in a sample convincing evidence that a major-
ity of the population prefer fresh coffee? Statistical tests answer these questions.
Here’s the answer in outline form:
• The claim. The skeptic claims that only half of all coffee drinkers prefer
fresh-brewed coffee. That is, he claims that the population proportion p is
only 0.5. Suppose for the sake of argument that this claim is true.
• The sampling distribution (page 422). If the claim p  0.5 were true and
we tested many random samples of 50 coffee drinkers, the sample proportion
would vary from sample to sample according to (approximately) the Normal
distribution with

mean 5 p 5 0.5
and

p11 2 p2
standard deviation 5
Å n
0.5 1 0.5 2
5
Å 50
5 0.0707

Figure 9.7 displays this Normal curve.

• The data. Place the sample proportion on the sampling distribution. You
see in Figure 9.7 that p^  0.56 isn’t an unusual value but that p^  0.72 is
 9. 2 What Is a Significance Test? 441

Sampling
distribution
of p if p = 0.5

Area = 0.2

Area = 0.001

0.5 p = 0.56 p = 0.72

Figure 9.7 The sampling distribution of the proportion of 50 coffee drinkers who prefer
fresh-brewed coffee. This distribution would hold if the truth about all coffee drinkers is
that 50% prefer fresh coffee. The shaded area is the probability that the sample proportion
is 56% or greater.

unusual. We would rarely get 72% of a sample of 50 coffee drinkers preferring

fresh-brewed coffee if only 50% of all coffee drinkers felt that way. So the
sample data do give evidence against the claim.
• The probability. We can measure the strength of the evidence against
the claim by a probability. What is the probability that a sample gives a p^ this
large or larger if the truth about the population is that p  0.5? If p^ 5 0.56,
this probability is the shaded area under the Normal curve in Figure 9.7.
This area is 0.20. But our sample actually gave p^ 5 0.72. The probability of
getting a sample outcome this large is only 0.001, an area too small to see
in Figure 9.7. An outcome that would occur just by chance in 20% of all
samples is not strong evidence against the claim. But an outcome that would
happen only 1 in 1000 times is good evidence.

Be sure you understand why this evidence is convincing. There are two possible
explanations of the fact that 72% of our subjects prefer fresh to instant coffee:
1. The skeptic is correct (p  0.5) and by bad luck a very unlikely outcome
occurred.
2. The population proportion favoring fresh coffee is greater than 0.5, so the
sample outcome is about what would be expected.
We cannot be certain that Explanation 1 is untrue. Our taste test results could be
due to chance alone. But the probability that such a result would occur by chance is
so small (0.001) that we are quite confident that Explanation 2 is right.

Stating hypotheses
A significance test starts with a careful statement of the claims we want to compare. In
most studies, we hope to show that some definite effect is present in the population.
442 CHAPTER 9 Introduction to Inference

In Example 9.8, we suspect that a majority of coffee drinkers prefer fresh-brewed

coffee. A statistical test begins by supposing that the effect we seek is not present.
We then look for evidence against this possibility and in favor of the effect we
hope to find. The first step in a significance test is to state a claim that we will try
to find evidence against. We call this claim our null hypothesis. The statement
we hope or suspect is true instead of the null hypothesis is called the alternative
hypothesis.

Null and alternative hypotheses

The claim being tested in a statistical test is called the null hypothesis.
The test is designed to assess the strength of the evidence against the
null hypothesis. Usually the null hypothesis is a statement of “no
effect” or “no difference.”

The claim about the population that we are trying to find evidence for
is the alternative hypothesis.

We abbreviate the null hypothesis as H0 (read as “H-naught”) and the alterna-

tive hypothesis as Ha. Both of these hypotheses are statements about the population
and so must be stated in terms of a population parameter.
In Example 9.8, the parameter is the proportion p of all coffee drinkers who
prefer fresh to instant coffee. The null hypothesis is

H0 : p 5 0.5

The alternative hypothesis is that a majority of the population favor fresh coffee. In
terms of the population parameter, this is

Ha : p . 0.5

ACTIVITY 9.2B
The Significance Test applet
AP The Significance Test applet on the book’s
PLET The “null hypothesis” is always the same, that
Web site (www.whfreeman.com/sta2e) the player makes 80% of shots taken: we write this
animates Example 9.7. That example asks if a as H 0 : p 5 0.80. The applet does not do a formal
basketball player’s actual performance gives evidence statistical test. Instead, it allows you to ask the
against the claim that he makes 80% of his free player to shoot until you are reasonably confident
throws. The parameter in question is the proportion that the true percent of shots made is or is not very
p of free throws that the player will make if he shoots close to 80%. The shooter claims that he makes 80%
free throws forever. The population is all free throws of his free throws. To test his claim, we have him
the player will ever shoot. shoot 20 free throws.
 9. 2 What Is a Significance Test? 443

1. Set the applet to take 20 shots. Check “Show null 3. If you are not convinced, click “Shoot” again
hypothesis” so that the shooter’s claim is visible in for 20 more shots. Keep going until either you are
the graph. convinced that he doesn’t make 80% of his shots
or it appears that his true percent made is pretty
close to 80%. How many shots did you watch him
shoot? How many did he make? What did you
conclude?
4. Click “Show true %” to reveal the truth. Was your
conclusion correct?

Comment: You see why statistical tests say how

strong the evidence is against some claim. If
the shooter makes only 10 of 40 shots, you are
pretty sure he can’t make 80% in the long run.
But even if he makes exactly 80 of 100, his true
2. Click “Shoot.” How many of the 20 shots did he long-term percent might be 78% or 81% instead
make? Are you convinced that the shooter really of 80%. It’s hard to be convinced that he makes
makes less than 80%? exactly 80%.

Exercises 9.21 A fair deck? Activity 9.2A (page 438) asks you to determine whether a deck of
cards is fair by looking at the proportion of red cards you obtain in a simple random
sample. Suppose that the deck of cards is fair and that you take an SRS of 25 cards.
(a) What are the mean and standard deviation of the sampling distribution of p^ ,
the sample proportion of red cards?
(b) If your sample contains 16 red cards, what is the value of p^ ?
(c) Sketch a graph like Figure 9.7 (page 441) that shows the sampling distribution. Mark
the location of the sample proportion p^ from (b) on the horizontal axis of your sketch.
(d) How likely is it that you would get a value of p^ as unusual as the one you did if the
deck is fair? What do you conclude about whether the deck is fair?
AP
PLET 9.22 Free throws Activity 9.2B shows you how to use the Significance Test applet
at the book’s Web site, www.whfreeman.com/sta2e, to test a basketball player’s
claim that he makes 80% of his free throws. Suppose that the player’s claim is true
and that you use the applet to shoot 20 shots.
(a) What are the mean and standard deviation of the sampling distribution of p^ , the
sample proportion of made free throws?
(b) If the player actually makes 8 shots, what is the value of p^ ?
(c) Sketch a graph like Figure 9.7 (page 441) that shows the sampling distribution. Mark
the location of the sample proportion p^ from (b) on the horizontal axis of your sketch.
(d) How likely is it that you would get a value of p^ as unusual as the one you did
if the player’s claim is true? What do you conclude about whether the player is an
80% free-throw shooter?
444 CHAPTER 9 Introduction to Inference

9.23 Fresh coffee, I Show how to calculate the two probabilities given in
Example 9.8.
9.24 Fresh coffee, II In Example 9.8, 72% of the subjects in the
experiment preferred fresh coffee. Explain why this result does
not prove that more than half of all coffee drinkers prefer fresh to
instant coffee.
The situations in Exercises 9.25 and 9.26 call for a significance
test. State the appropriate null hypothesis H0 and
alternative hypothesis Ha in each case. Be sure to define
Ukap

your parameter.
ala/
Dr

9.25 Do you argue? A May 2005 Gallup Poll report on a national

eam
sti

survey of 1028 teenagers revealed that 72% of teens said they rarely or
m

.c o
e

m
never argue with their friends.10 You wonder whether this national result
would be true in your school. You conduct your own survey of a simple
random sample of students at your school.
9.26 One potato, two potato A potato chip producer and a supplier of potatoes
agree that each shipment of potatoes must meet certain quality standards. If the
producer is convinced that more than 8% of the potatoes in the shipment have
blemishes, the truck will be sent away to get another load of potatoes from the
supplier. Otherwise, the entire truckload will be accepted.

P-values
A significance test looks for evidence against the null hypothesis and in favor of
the alternative hypothesis. The evidence is strong if the outcome we observe would
rarely happen if the null hypothesis is true but is more probable if the alternative
hypothesis is true. For example, it would be surprising to find 36 of 50 subjects
favoring fresh coffee if in fact only half of the population feels this way. How sur-
prising? A significance test answers this question by giving a probability: the prob-
ability of getting an outcome at least as far as the actually observed outcome is from
what we would expect when H0 is true. We call this probability the P-value.What
counts as “far from what we would expect” depends on Ha as well as H0. In the taste
test, the probability we want is the probability that 36 or more of 50 subjects favor
fresh coffee. If the null hypothesis p  0.5 is true, this probability is very small
(0.001). That’s good evidence that the null hypothesis is not true.

P-value

The probability, computed assuming that H0 is true, that the sample

outcome would be as extreme or more extreme than the actually
observed outcome is called the P-value of the test. The smaller the
P-value is, the stronger is the evidence against H0 provided by the data.

In practice, most statistical tests are carried out by computer software that cal-
culates the P-value for us. It is usual to report the P-value in describing the results
 9. 2 What Is a Significance Test? 445

of studies in many fields. You should therefore understand what P-values say even
if you don’t do statistical tests yourself, just as you should understand what “95%
confidence” means even if you don’t calculate your own confidence intervals.

Example 9.9 Count Buffon’s coin

The French naturalist Count Buffon (1707–1788) tossed a coin 4040 times. He got
2048 heads. The sample proportion of heads is
2048
p^ 5 5 0.507
4040

That’s a bit more than one-half. Is this evidence that Buffon’s coin was not balanced?
This is a job for a significance test.

The truth about Count Buffon’s coin-tossing experiment.

The hypotheses. The null hypothesis says that the coin is balanced (p 5 0.5).
We did not suspect a bias in a specific direction before we saw the data, so the alter-
native hypothesis is just “the coin is not balanced.” The two hypotheses are
H0 : p 5 0.5
Ha : p 2 0.5
The sampling distribution. If the null hypothesis is true, the sample proportion
of heads has approximately the Normal distribution with
mean 5 p 5 0.5
p11 2 p2
standard deviation 5
Å n
0.5 1 0.5 2
5
Å 4040
5 0.00787
446 CHAPTER 9 Introduction to Inference

Figure 9.8 shows this sampling distribution with Buffon’s sample outcome p^ 5 0.507
marked. The picture already suggests that this is not an unlikely outcome that would
give strong evidence against the claim that p 5 0.5.

Sampling
distribution
of p if p = 0.5

0.5
p = 0.507

Figure 9.8 The sampling distribution of the proportion of heads in 4040 tosses of a
balanced coin. Count Buffon’s result, proportion 0.507 heads, is marked.

The P-value. How unlikely is an outcome as far from 0.5 as Buffon’s p^ 5 0.507?
Because the alternative hypothesis allows p to lie on either side of 0.5, values of p^
far from 0.5 in either direction provide evidence against H0 and in favor of Ha. The
P-value is therefore the probability that the observed p^ lies as far from 0.5 in either
direction as the observed p^ 5 0.507. Figure 9.9 shows this probability as an area
under the Normal curve. It is P 5 0.37.

P-value = 0.37

0.493 0.5 0.507

Figure 9.9 The P-value for testing whether Count Buffon’s coin was balanced. This
is the probability, calculated assuming a balanced coin, of a sample proportion as far or
farther from 0.5 as Buffon’s result of 0.507.

Conclusion. A truly balanced coin would give a result this far or farther from
0.5 in 37% of all repetitions of Buffon’s trial. His result gives no reason to think that
his coin was not balanced.
 9. 2 What Is a Significance Test? 447

The alternative Ha : p . 0.5 in Example 9.8 is a one-sided alternative because

the effect we seek evidence for says that the population proportion is greater than
one-half. The alternative Ha : p 2 0.5 in Example 9.9 is a two-sided alternative
because we ask only whether or not the coin is balanced. Whether the alternative
is one-sided or two-sided determines whether sample results, extreme in one or in
both directions, count as evidence against H0 and in favor of Ha.
Try the Java applet at www.math.usu.edu/~schneit/CTIS/PValue/. It allows
www
you to test whether a coin is fair by simulating repeated tosses.

Statistical significance
We can decide in advance how much evidence against H0 we will insist on. The
way to do this is to say how small a P-value we require. The decisive value of P
is called the significance level. It is usual to write it as a, the Greek letter alpha.
If we choose a 5 0.05, we are requiring that the data give evidence against H0
so strong that it would happen no more than 5% of the time (1 time in 20) when
H0 is true. If we choose a 5 0.01, we are insisting on stronger evidence against
H0, evidence so strong that it would appear only 1% of the time (1 time in 100)
if H0 is in fact true.

Statistical significance

If the P-value is smaller than a, we say that the data are statistically
significant at level a.

“Significant” in the statistical sense does not mean “important.” It means

simply “not likely to happen just by chance.” We used these words in Section 6.1
(page 267). Now we have attached a number to statistical significance to say
what “not likely” means. You will often see significance at level 0.01 expressed
by the statement “The results were significant 1P , 0.012 .” Here P stands for the
P-value.
We don’t need to make use of traditional levels of significance such as 5% and
1%. The P-value is more informative because it allows us to assess significance at any
level we choose. For example, a result with P  0.03 is significant at the a 5 0.05
level but not significant at the a 5 0.01 level. Nonetheless, the traditional signifi-
cance levels are widely accepted guidelines for “how much evidence is enough.”
We might say that P , 0.10 indicates “some evidence” against the null hypothesis,
P , 0.05 is “moderate evidence,” and P , 0.01 is “strong evidence.” Don’t take
these guidelines too literally, however. We will say more about interpreting tests in
the next section.

Calculating P-values
Finding the P-values we gave in Examples 9.8 and 9.9 requires doing Normal dis-
tribution calculations using Table A. That was addressed in Section 3.1. In practice,
software or a calculator does the calculation for us, but here is an example that
shows how to use Table A.
448 CHAPTER 9 Introduction to Inference

Example 9.10 Tasting coffee

The hypotheses. In Example 9.8, we want to test the hypotheses

H0 : p 5 0.5
Ha : p . 0.5
Here p is the proportion of the population of all coffee drinkers who prefer fresh
coffee to instant coffee.
The sampling distribution. If the null hypothesis is true, then p  0.5. For
samples of size 50, p^ follows a Normal distribution with mean 0.5 and standard
deviation 0.0707.
The data. A sample of 50 people found that 36 preferred fresh coffee. The sam-
ple proportion is p^ 5 0.72.
The P-value. The alternative hypothesis is one-sided on the high side. So the
P-value is the probability of getting an outcome at least as large as 0.72. Figure 9.7
(page 441) displays this probability as an area under the Normal sampling distribu-
tion curve. To find any Normal curve probability, move to the standard scale. The
standard score for the outcome p^ 5 0.72 is
observation 2 mean
standardized score 5
standard deviation
0.72 2 0.5
z5 5 3.11
0.0707

Table A says that the area under a Normal curve to the left of 3.11 (in the standard scale)
is 0.9991. The area to the right is therefore 1  0.9991  0.0009, and that is our P-value.
Conclusion. The small P-value means that the data provide very strong evi-
dence that a majority of the population prefer fresh coffee.

Exercises 9.27 Baggage check! Thousands of travelers pass through Guadalajara airport
each day. Before leaving the airport, each passenger must pass through Customs.
Customs officials want to be sure that passengers do not bring illegal items into
the country. But they do not have time to search every traveler’s luggage. Instead,
they require each person to press a button. Either a red or a green bulb lights
up. If the red light shows, the passenger will be searched by Customs agents. A
green light means “go ahead.” Customs officers claim that the probability that the
light turns red on any press of the button is 0.30. A weary traveler waiting for her
luggage watches the next 100 people press the button. The light turns red 20 times.
(a) State null and alternative hypotheses for testing the claim made by Customs
officials.
(b) Describe the sampling distribution assuming the null hypothesis from (a) is true.
(c) How extreme is the sample outcome? Find the P-value.
(d) Explain your conclusions in nontechnical language.
 9. 2 What Is a Significance Test? 449

9.28 Affirmative action The Chronicle of Higher Education commissioned a

comprehensive survey of American attitudes toward colleges and universities.
Of the 1000 American adults aged 25 to 64 surveyed, 640 said that colleges
should not admit minority students who have lower grades than other qualified
candidates.11 Is this sufficient evidence to conclude that a majority of adults have
this view of affirmative action?
(a) State null and alternative hypotheses.
(b) Describe the sampling distribution.
(c) How extreme is the sample outcome? Find the P-value.
(d) Explain your conclusions in nontechnical language.
9.29 Bullies in middle school If one study is any indication,
many middle school students are not shy about admitting
aggressive behavior in school. A University of Illinois study
on aggressive behavior surveyed 558 students in a midwestern
middle school. When asked to describe their behavior in the
last 30 days, 445 students said their behavior included physical
aggression, social ridicule, teasing, name-calling, and issuing
threats. This behavior was not defined as bullying in the
questionnaire.12
© Godfer/Dreamstime.com (a) Is this evidence that more than three-quarters of the students at
that middle school engage in bullying behavior? Carry out a significance test to
answer this question. State hypotheses, describe the sampling distribution, and
find the P-value. State your conclusions.
(b) Does this study provide strong evidence of frequent bullying in America’s
middle schools? Explain.
9.30 Count Buffon’s coin again Refer to Example 9.9 (page 445).
(a) Construct and interpret a 95% confidence interval for the overall proportion
of times that Buffon’s coin would land heads up.
(b) Is p  0.5 a plausible value for the population proportion? Justify your
answer.
9.31 P-values and significance, I A test of the null hypothesis H0 : p 5 0.3
yields z 5 21.77.
(a) If the alternative hypothesis is Ha : p , 0.3, find the P-value for this test. Is this
result significant at the a 5 0.05 level? At the a 5 0.01 level? Explain.
(b) If the alternative hypothesis is Ha : p 2 0.3, find the P-value for this test. Is this
result significant at the a 5 0.05 level? At the a 5 0.01 level? Explain.
9.32 P-values and significance, II A test of the null hypothesis H0 : p 5 0.68
yields z 5 2.35.
(a) If the alternative hypothesis is Ha : p . 0.68, find the P-value for this test. Is
this result significant at the a 5 0.05 level? At the a 5 0.01 level? Explain.
(b) If the alternative hypothesis is Ha : p 2 0.68, find the P-value for this test. Is
this result significant at the a 5 0.05 level? At the a 5 0.01 level? Explain.
450 CHAPTER 9 Introduction to Inference

CALCULATOR CORNER Significance test for a

population proportion
The TI-84/TI-Nspire can be used to test a claim about a population
proportion. Consider Example 9.9 about Count Buffon’s coin-tossing.
In n  4040 tosses, Buffon observed X  2048 heads. Recall that our
hypotheses were

H 0 : p 5 0.5
H a : p 2 0.5

To perform a significance test:

TI-84
• Press A, choose TESTS, and then choose 1-PropZTest.
• On the 1–PropZTest screen, enter these values: p0 5 0.5, x 5 2048, and
n 5 4040. Specify the alternative hypothesis as prop2p0 .
• If you select the “Calculate” option and press b, you will see that the
z statistic is 0.88 and the P-value is 0.3783.
• If you select the “Draw” option, you will see the screen shown at the right, below.

TI-Nspire
You can perform significance tests in a Calculator page or a Lists & Spreadsheet
page. There are some slight differences, which we’ll illustrate now.
• On a Calculator page, press b, and choose 5:Statistics,7:Stat
Tests…, and 5:1-Prop z Test.
• In the dialog box, enter these values: p0  0.5, x  2048, and n  4040. Specify
the alternative hypothesis as H a : prop 2 p0. Then click OK.
 9. 2 What Is a Significance Test? 451

• You can also carry out the hypothesis test in a Lists & Spreadsheet page. Press
b, and choose 4:Statistics, 4:Stat Tests…, and 5:1-Prop z
Test.
• In the dialog box, enter these values: p0  0.5, x  2048, and n  4040. Specify
the alternative hypothesis as H a : prop 2 p0. Choose any empty column for the
“1st Result Column.” If you click the box next to “Draw,” you will get a picture of
the standard Normal curve with appropriate shading, along with the values of
z and P.

ws
Ne
ws
Ne

APPLICATION 9.2 B]
RO
g
<
Se
a

Pair-a-dice!
Materials: Pair of dice (one red, one white) • Combine your results with the other teams to
provided by your teacher obtain a grand total of red wins and white wins.
Students pair off, Player 1 with the red die and QUESTIONS
Player 2 with the white die. Each player rolls the 1. Our assumption is that each die should win
die, and the higher number wins that round. A tie about half the time in the long run. State null and
is ignored and the dice are rolled again. The pur- alternative hypotheses in terms of the proportion of
pose is to look at the proportion of wins by the red times the red die should win.
die over the long term.
2. Draw a graph of the sampling distribution of p^
• Roll the dice 10 times and use tally marks to assuming that H 0 is true. Mark the location of the
record the winner. class’s p^ -value on the sketch.
Red wins: 3. Calculate the P-value. Is there sufficient evidence
White wins: to reject the null hypothesis?
4. State your conclusion clearly.
To make sure that no one has an unfair advantage,
exchange the dice and roll another 10 times. 5. Construct and interpret a 95% confidence
Exchange the dice again and roll 10 more times, interval for the parameter p. What additional
and then exchange again and roll 10 more times. information does this confidence interval provide?
Continue to record your results. You should have 6. Carefully inspect the red die. Can you explain
a total of 40 rolls. the results that you observed?
452 CHAPTER 9 Introduction to Inference

Section 9.2 Summary

A confidence interval estimates an unknown parameter. A significance test as-
sesses the evidence for some claim about the value of an unknown parameter. In
practice, the purpose of a statistical test is to answer the question “Could the effect
we see in the sample just be an accident due to chance, or is it good evidence that
the effect is really there in the population?”
Significance tests answer this question by giving the probability that a sample
effect as large as the one we see in this sample would arise just by chance. This prob-
ability is the P-value. A small P-value says that our outcome is unlikely to happen
just by chance. To set up a test, state a null hypothesis that says the effect you seek
is not present in the population. The alternative hypothesis says that the effect
is present. The P-value is the probability, calculated taking the null hypothesis to
be true, of an outcome as extreme in the direction specified by the alternative hy-
pothesis (either a one-sided alternative or a two-sided alternative) as the actually
observed outcome. A sample result is statistically significant at the 5% level if it
would occur just by chance no more than 5% of the time in repeated samples.
This section concerns the basic reasoning of tests and the details of tests for hy-
potheses about a population proportion. There is more discussion of the practical
interpretation of statistical tests in the next section.

Section 9.2 The situations in Exercises 9.33 and 9.34 call for a significance test. State the
Exercises appropriate null hypothesis H0 and alternative hypothesis Ha in each case. Be sure
to define your parameter.
9.33 No homework?! Mr. Starnes believes that fewer than 75% of the students at
his school completed their math homework last night. The math teachers inspect
the homework assignments from a random sample of students at the school to
help Mr. Starnes test his claim.
9.34 Lefties Sally reads a newspaper report claiming that 12% of all adults in the
United States are left-handed. She wonders whether 12% of the students at her
large public high school are left-handed.
9.35 Where’s the blunder? Write a sentence or two to explain the blunder in
each of the following statements.
(a) Austin says that a result that is significant at the a 5 0.05 level must also be
significant at the a 5 0.01 level.
(b) Asked to explain the meaning of “statistically significant at the a 5 0.05 level,”
Carla says: “This means that the probability that the null hypothesis is true is less
than 0.05.”
9.36 Is the Belgian euro coin fair? Exercise 7.23 (page 327) asked you to use
simulation to determine whether a Belgian euro coin that landed “heads” 140
times out of 250 spins was minted fairly. One of the Polish math professors who
spun the coin concluded that the coin was not fair. A representative from the
Belgian mint argued that the result was due to chance.
(a) Carry out a significance test to determine whether the coin was fair. Follow the
steps in Example 9.10 (page 448). What do you conclude?
 9. 3 Use and Abuse of Statistical Inference 453

(b) Construct and interpret a 95% confidence interval. What additional information
does the confidence interval provide?
9.37 P-values and significance, III A test of the null hypothesis H0 : p 5 0.81
versus the one-sided alternative Ha : p . 0.81 yields z 5 2.36.
(a) Find the P-value for this test.
(b) Is this result significant at the a 5 0.05 level? At the a 5 0.01 level? Explain.
9.38 P-values and significance, IV A test of the null hypothesis H0 : p 5 0.81
versus the two-sided alternative Ha : p 2 0.81 yields z 5 2.36.
(a) Find the P-value for this test.
(b) Is this result significant at the a 5 0.05 level? At the a 5 0.01 level? Explain.
9.39 Are cell phones distracting? The American Automobile Association (AAA)
commissioned a study by researchers at the University of North Carolina on sources
of distraction for drivers. The study tracked 70 drivers aged 18 to 80 for a week. Min-
iature cameras were placed in the drivers’ cars, with their knowledge, and their driv-
ing behavior was randomly viewed for a week. The first three hours of each tape were
discarded in the hope that drivers would act more naturally later in the week. The re-
searchers considered a wide range of activities to be distracting, such as leaning over
to reach for something, fiddling with radio controls, attending to babies, talking to
passengers, applying makeup, and opening and reading their mail. The study found
that 21 of the subjects used cell phones while their vehicles were moving.13
(a) Is there significant evidence that fewer than half of all drivers use cell phones
while their vehicles are moving? Carry out appropriate inference, and report your
results and conclusions.
(b) Do you believe that this study is subject to bias? In what regard?
9.40 Hack-a-Shaq Any NBA fan has heard about the free-throw shooting woes
of the NBA’s most dominant center, Shaquille O’Neal. Over his NBA career, Shaq
has made 53.3% of his free throws. One off-season, Shaq worked with an assistant
coach on his free-throw technique. During the first two games of the next season,
Shaq made 26 out of 39 free throws. Do these results provide evidence that Shaq
has significantly improved his free-throw shooting? Carry out a significance test
to help answer this question.

9 . 3 Use and Abuse of Statistical Inference

ACTIVITY 9.3
Did you get enough blues?
Materials: One 1.69-ounce bag of milk chocolate extensive national survey, they chose blue as the new
M&M’s per student color. The company mixes the colors for each type
of M&M’s candy (plain, peanut, etc.) according to a
The M&M/Mars Company, headquartered in specific percent distribution. At the time we’re writing,
Hackettstown, New Jersey, makes milk chocolate the company claims that the distribution of colors in
candies. In 1995, they decided to replace the tan- bags of plain M&M’s candies is 24% blue, 20% orange,
colored M&M’s with a new color. After conducting an 16% green, 14% yellow, 13% red, and 13% brown. You
ACTIVITY 9.3 (continued)
454 CHAPTER 9 Introduction to Inference

ACTIVITY 9.3 (continued)

may want to check this information at the M&M’s 4. Construct and interpret a 95% confidence
Web site, www.mms.com, before you continue. In interval for the population proportion of blue
this Activity, we’ll investigate the company’s claim M&M’s based on your sample. Is your result
about the proportion of M&M’s that are blue. consistent with your significance test in Step 2?
What additional information does your confidence
1. Your teacher will supply each student with a 1.69-
interval give?
ounce bag of M&M’s candies. Open your bag and
carefully count the number of M&M’s of each color 5. Some students may reject the company’s claim
as well as the total number of M&M’s in the bag. based on their sample data even though the
Record your counts in a table like this: company’s claim is actually true. Discuss with your
classmates how this might happen.
Color: Blue Orange Green Yellow Red Brown
Total Count: 6. Your teacher will draw a horizontal axis at the top
of the chalkboard with a scale that goes from 0 to
2. Carry out a significance test of the company’s claim 0.6. Each student should draw his or her confidence
about the actual proportion p of blue M&M’s based on interval below the axis. What proportion of the
the data in your sample (bag). Use a 5% significance class’s confidence intervals contained 0.24? Are you
level 1 a 5 0.05 2 . What conclusion do you draw? surprised?

3. Pool results with your classmates. What percent of the 7. Keep your data from this Activity handy. You will
class rejected the company’s claim? Are you surprised? need it again in the next chapter.

Using inference wisely

We have met the two major types of statistical inference: confidence intervals and
significance tests. We have, however, seen only one inference method of each type,
designed for inference about a population proportion p. There are libraries of both
books and software filled with methods for inference about various parameters in
various settings. The reasoning of confidence intervals and significance tests re-
mains the same, but the details can seem overwhelming. The first step in using
inference wisely is to understand your data and the questions you want to answer
and to fit the method to its setting. Here are some tips on inference, adapted to the
one setting we are familiar with.
The design of the data production matters “Where do the data come from?”
remains the first question to ask in any statistical study. Any inference method is
intended for use in a specific setting. For our confidence interval and test for a
proportion p:
• The data must be a simple random sample (SRS) from the population of
interest. When you use these methods, you are acting as if the data are an
SRS. In practice, it is often not possible to actually choose an SRS from the
population. Your conclusions may then be open to challenge. In Activity 9.3, for
example, each individual bag of plain M&M’s candies may not be equivalent
to an SRS. As a result, several students in your class may reject the null
hypothesis H0 : p 5 0.24 even though the company is telling the truth
about the population proportion of blue M&M’s. Likewise, your class’s 95%
 9. 3 Use and Abuse of Statistical Inference 455

confidence intervals for p may not have a 95% capture rate if each student’s
data do not come from an SRS.
• These methods are not correct for sample designs more complex than
an SRS, such as stratified samples. There are other methods that fit these
settings.
• There is no correct method for inference from data haphazardly collected
with bias of unknown size. Fancy formulas cannot rescue badly produced
data.
• Other sources of error, such as dropouts and nonresponse, are important.
Remember that confidence intervals and tests use the data you give them
and ignore these practical difficulties.

Know how confidence intervals behave A confidence interval estimates the un-
known value of a parameter and also tells us how uncertain the estimate is. All
Dropping out confidence intervals share these behaviors:
An experiment found
that weight loss is signifi- • The confidence level says how often the method catches the true parameter
cantly more effective than in very many uses. We never know whether a specific data set gives us an
exercise for reducing high interval that contains the true parameter. All we can say is that “we got this
cholesterol and high blood result from a method that works 95% of the time.” Our data set might be
pressure. The 170 subjects one of the 5% that produce an interval that misses the parameter. If that risk
were randomly assigned
to a weight-loss program,
bothers you, use a 99% confidence interval.
an exercise program, or a • High confidence is not free. A 99% confidence interval will be wider than
control group. Only 111 of a 95% confidence interval based on the same data. There is a trade-off
the 170 subjects completed between how closely we can pin down the parameter and how confident we
their assigned treatment,
and the analysis used data
are that we have caught the parameter.
from these 111. Did the • Larger samples give shorter intervals. If we want high confidence and a
dropouts create bias? short interval, we must take a larger sample. The length of our confidence
Always ask about details interval for p goes down in proportion to the square root of the sample size.
of the data before trusting
inference.
To cut the interval in half, we must take four times as many observations.
This is typical of many types of confidence intervals.
Know what statistical significance says Many statistical studies hope to show
that some claim is true. A clinical trial compares a new drug with a standard drug
because the doctors hope that patients given the new drug will do better. A psy-
chologist studying gender differences suspects that women will do better than men
(on the average) on a test that measures social-networking skills. The purpose of
significance tests is to weigh the evidence that the data give in favor of such claims.
That is, a test helps us know if we have found what we were looking for.
To do this, we ask what would happen if the claim were not true. That’s the null
hypothesis—no difference between the two drugs, no difference between women
and men. A significance test answers only one question: “How strong is the evi-
dence that the null hypothesis is not true?” A test answers this question by giving a
P-value. The P-value tells us how unlikely our data would be if the null hypothesis
were true. Data that are very unlikely are good evidence that the null hypothesis is
not true. We never know whether the hypothesis is true for this specific population.
All we can say is, “Data like these would occur only 5% of the time if the hypothesis
were true.”
456 CHAPTER 9 Introduction to Inference

This kind of indirect evidence against the null hypothesis (and for the effect we
hope to find) is less straightforward than a confidence interval.
Know what your methods require Our test and confidence interval for a pro-
portion p require that the population be much larger than the sample. They also
require that the sample itself be reasonably large so that the sampling distribution
of the sample proportion is close to Normal. We have said little about the specifics
of these requirements because the reasoning of inference is more important. Just
as there are inference methods that fit stratified samples, there are methods that
fit small samples and small populations. If you plan to use statistical inference in
practice, you will need help from a statistician (or need to learn lots more statistics)
to manage the details.
Most of us read about statistical studies more often than we actually work with
data ourselves. Concentrate on the big issues, not on the details of whether the
authors used exactly the right inference methods. Does the study ask the right ques-
tions? Where did the data come from? Do the results make sense? Does the study
report confidence intervals so you can see both the estimated values of important
parameters and how uncertain the estimates are? Does it report P-values to help
convince you that findings are not just good luck?

Exercises 9.41 Activity 9.3 follow-up Refer to Activity 9.3 (page 453). Statistics classes
like yours have examined the M&M’s color distribution stated by the M&M/Mars
Company for many years. In response to consumer questions about the
accuracy of their claims, the company wrote:
While we mix the colors as thoroughly as possible, the above ratios may
vary somewhat, especially in the smaller bags. This is because we combine
the various colors in large quantities for the last production stage (printing).
The bags are then filled on high-speed packaging machines by weight, not by
count.
Which of the four big ideas of this section is the company addressing?
mediablitzimages Limited/Alamy
9.42 A television poll A television news program conducts a call-in poll about
a proposed city ban on handgun ownership. Of the 2372 calls, 1921 oppose the
ban. The station, following recommended practice, makes a confidence statement:
“81% of the Channel 13 Pulse Poll sample opposed the ban. We can be 95%
confident that the true proportion of citizens opposing a handgun ban is within
1.6% of the sample result.”
(a) Show that the station’s confidence interval calculation is correct.
(b) Nonetheless, the station’s conclusion is not justified. Why?
9.43 Only-child presidents Joe is writing a report on the backgrounds of
American presidents. He looks up information on how many of the 44 presidents
were only children. Because Joe took a statistics course, he uses these results to
get a 95% confidence interval for the true proportion p of these 44 U.S. presidents
who were only children. This makes no sense. Why not?
 9. 3 Use and Abuse of Statistical Inference 457

9.44 Who will win? A poll taken shortly before an election finds that 52% of
the voters favor candidate Shrub over candidate Snort. The poll has a margin of
sampling error of plus or minus 3 percentage points at 95% confidence. The poll
press release says the election is too close to call. Why?
9.45 Dropping out The marginal nugget on page 455 describes an experiment
comparing the effectiveness of weight loss and exercise in reducing cholesterol
level and blood pressure. Explain how the 59 people who dropped out may have
created bias in the results of the study.
9.46 Inference conditions Explain why each of the two conditions for
performing inference about a population proportion p—population much larger
than sample size and sample size large enough—is important.

The woes of significance tests

The purpose of a significance test is usually to give evidence for the presence of
some effect in the population. The effect might be a probability of heads different
from one-half for a coin or a longer mean survival time for patients given a new can-
cer treatment. If the effect is large, it will show up in most samples—the proportion
of heads among our tosses will be far from one-half, or the patients who get the new
treatment will live much longer than those in the control group. Small effects, such
as a probability of heads only slightly different from one-half, will often be hidden
behind the chance variation in a sample. This is as it should be: big effects are easier
to detect. That is, the P-value will usually be small when the population truth is far
from the null hypothesis.
The “woes” of testing start with the fact that a test just measures the strength
of evidence against the null hypothesis. It says nothing about how big or how im-
portant the effect we seek in the population really is. For example, our hypothesis
might be “This coin is balanced.” We express this hypothesis in terms of the prob-
ability p of getting a head as H0 : p 5 0.5. No real coin is exactly balanced, so we
know that this hypothesis is not exactly true. If this coin has probability p 5 0.502
of a head, we might say that for practical purposes it is balanced. A statistical test
doesn’t think about “practical purposes.” It just asks if there is evidence that p is not
exactly equal to 0.5. The focus of tests on the strength of the evidence against an
exact null hypothesis is the source of much confusion in using tests.
Pay particular attention to the size of the sample when you read the result of a
significance test. Here’s why:
• Larger samples make tests of significance more sensitive. If we toss a coin
hundreds of thousands of times, a test of H0 : p 5 0.5 will often give a very
low P-value when the truth for this coin is p 5 0.502. The test is right—it
found good evidence that p really is not exactly equal to 0.5—but it has
picked up a difference so small that it is of no practical interest. A finding
can be statistically significant without being practically important.
• On the other hand, tests of significance based on small samples are often
not sensitive. If you toss a coin only 10 times, a test of H0 : p 5 0.5 will often
give a large P-value even if the truth for this coin is p 5 0.7. Again the test
is right—10 tosses are not enough to give good evidence against the null
hypothesis. Lack of significance does not mean that there is no effect,
458 CHAPTER 9 Introduction to Inference

only that we do not have good evidence for an effect. Small samples often
miss effects that are really present in the population.
Whatever the truth about the population—whether p 5 0.7 or p 5 0.502—more
observations allow us to pin down p more closely. If p is not 0.5, more observations
will give more evidence of this, that is, a smaller P-value. Because significance de-
pends strongly on the sample size as well as on the truth about the population, sta-
tistical significance tells us nothing about how large or how practically important
an effect is. Large effects (like p 5 0.7 when the null hypothesis is p 5 0.5) often
give data that are insignificant if we take only a small sample. Small effects (like
p 5 0.502) often give data that are highly significant if we take a large sample. Let’s
return to a favorite example to see how significance changes with sample size.

Example 9.11 Count Buffon’s coin again

Count Buffon tossed a coin 4040 times and got 2048 heads. His sample proportion
of heads was

2048
p^ 5 5 0.507
4040

Is Buffon’s coin balanced? The hypotheses are

H0 : p 5 0.5
Ha : p 2 0.5

The test of significance works by locating the sample outcome p^ 5 0.507 on the
sampling distribution that describes how p^ would vary if the null hypothesis were
true. Figure 9.10 repeats Figure 9.8 (page 446). It shows that the observed p^ 5 0.507
is not surprisingly far from 0.5 and therefore is not good evidence against the hy-
pothesis that the true p is 0.5. The P-value, which is 0.37, just makes this precise.

Sampling
distribution
of pˆ if p = 0.5

0.5 pˆ = 0.507

Figure 9.10 The sampling distribution of the proportion of heads in 4040 tosses of a
coin if in fact the coin is balanced. Sample proportion 0.507 is not an unusual outcome.
 9. 3 Use and Abuse of Statistical Inference 459

Suppose that Count Buffon got the same result, p^ 5 0.507, from tossing a coin
1000 times and also from tossing a coin 100,000 times. The sampling distribution of
p^ when the null hypothesis is true always has mean 0.5, but its standard deviation
gets smaller as the sample size n gets larger. Figure 9.11 displays the three sampling
distributions, for n 51000, n 5 4040, and n 5 100,000. The middle curve in this
figure is the same Normal curve as in Figure 9.10, but drawn on a scale that allows
us to show the very tall and narrow curve for n 5 100,000. Locating the sample
outcome p^ 5 0.507 on the three curves, you see that the same outcome is more or
less surprising depending on the size of the sample.

n = 100,000

Is this outcome
surprising?
n = 4040

n = 1000

0.5 0.507

Figure 9.11 The three sampling distributions of the proportion of heads in 1000, 4040,
and 100,000 tosses of a balanced coin. Sample proportion 0.507 is not unusual in 1000 or
4040 tosses but is very unusual in 100,000 tosses.

The P-values are P 5 0.66 for n 5 1000, P 5 0.37 for n 5 4040, and P 5 0.000009
for n 5 100,000. Imagine tossing a balanced coin 1000 times repeatedly. You will get
a proportion of heads at least as far from one-half as Buffon’s 0.507 in about two-
thirds of your repetitions. If you toss a balanced coin 100,000 times, however, you
will almost never (9 times in a million repeats) get an outcome this unbalanced.
The outcome p^ 5 0.507 is not evidence against the hypothesis that the coin is
balanced if it comes up in 1000 tosses or in 4040 tosses. It is completely convincing
evidence if it comes up in 100,000 tosses.

The moral of Example 9.11 is clear: sample size matters!

Beware the naked P-value

The P-value of a significance test depends strongly on the size of the

sample, as well as on the truth about the population.

It is bad practice to report a P-value without also giving the sample

size and a statistic or statistics that describe the sample outcome.
460 CHAPTER 9 Introduction to Inference

The advantages of confidence intervals

Example 9.11 suggests that we not rely on significance alone in understanding a
statistical study. Just knowing that the sample proportion was p^ 5 0.507 helps a
lot. You can decide whether this deviation from one-half is large enough to interest
you. Of course, p^ 5 0.507 isn’t the exact truth about the coin, just the chance result
of Count Buffon’s tosses. So a confidence interval, whose width shows how closely
we can pin down the truth about the coin, is even more helpful. Here are the 95%
confidence intervals for the true probability of a head p, based on the three sample
sizes in Example 9.11. You can check that the method of Section 9.1 gives these
answers.

Number of tosses 95% confidence interval

n 5 1000 0.507 6 0.031, or 0.476 to 0.538
n 5 4040 0.507 6 0.015, or 0.492 to 0.522
n 5 100,000 0.507 6 0.003, or 0.504 to 0.510

The confidence intervals make clear what we know (with 95% confidence) about the
true p. The intervals for 1000 and 4040 tosses include 0.5, so we are not confident
the coin is unbalanced. For 100,000 tosses, however, we are confident that the true
p lies between 0.504 and 0.510. In particular, we are confident that it is not 0.5.

Give a confidence interval

Confidence intervals are more informative than tests because they

actually estimate a population parameter. They are also easier to inter-
pret. It is good practice to give confidence intervals whenever possible.

Significance at the 5% level isn’t magical

The purpose of a significance test is to describe the degree of evidence provided
by the sample against the null hypothesis. The P-value does this. But how small a
P-value is convincing evidence against the null hypothesis? This depends mainly
on two circumstances:
• How plausible is H0? If H0 represents an assumption that the people you
must convince have believed for years, strong evidence (small P) will be
needed to persuade them.
• What are the consequences of rejecting H0? If rejecting H0 in favor of Ha means
making an expensive changeover from one type of product packaging to
another, you need strong evidence that the new packaging will boost sales.
These criteria are a bit subjective. Different people will often insist on different
levels of significance. Giving the P-value allows each of us to decide individually if
the evidence is sufficiently strong.
Users of statistics have often emphasized standard levels of significance such
as 10%, 5%, and 1%. For example, courts have tended to accept 5% as a standard
in discrimination cases. This emphasis reflects the time when tables of critical
values rather than computer software dominated statistical practice. The 5% level
 9. 3 Use and Abuse of Statistical Inference 461

(a 5 0.05) is particularly common. There is no sharp border between “signifi-

cant” and “insignificant,” only increasingly strong evidence as the P-value de-
creases. There is no practical distinction between the P-values 0.049 and 0.051. It
makes no sense to treat P , 0.05 as a universal rule for what is significant.

Exercises Exercises 9.47 to 9.49 refer to the following brief newspaper article.14

Depression Pill Seems to

Help Smokers Quit
Boston—Taking an antidepression medicine appears to double smokers’ chances
of kicking the habit, a study found. The Food and Drug Administration approved
the marketing of this medicine, called Zyban or bupropion, to help smokers in May.
The results of several studies with the drug, including one published in today’s issue
of the New England Journal of Medicine, were made public then.
The newly published study was conducted on 615 volunteers who wanted to
give up smoking and were not outwardly depressed. They took either Zyban or
dummy pills for 6 weeks. A year later, 23 percent of those getting Zyban were still
off cigarettes, compared with 12 percent in the comparison group.

9.47 Significant, but important? The results of this experiment were significant
at the a 5 0.05 level. In your opinion, are the results practically important? Justify
your position.
9.48 Generalizability To what population can the results of this study be
generalized? Explain.
9.49 Causation? Can we conclude that taking Zyban causes people to quit
smoking? Justify your answer.
9.50 P-value or significance level? In performing a significance test, the
researcher can choose between adopting a fixed significance level or calculating a
P-value. Explain how a P-value gives more information.
9.51 Searching for significance You perform 1000 significance tests using a 5 0.05.
Assuming that all null hypotheses are true, about how many of the test results would
you expect to be statistically significant? Explain how you obtained your answer.
9.52 Plagiarizing An online poll posed the following question:
It is now possible for school students to log on to Internet sites and download
homework. Everything from book reports to doctoral dissertations can be down-
loaded free or for a fee. Do you believe giving a student who is caught plagiariz-
ing an “F” for their assignment is the right punishment?
Of the 20,125 people who responded, 14,793 clicked “Yes.” That’s 73.5% of the sample.
Based on this sample, a 95% confidence interval for the percent of the population
who would say “Yes” is 73.5% 6 0.61%. Why is this confidence interval worthless?
462 CHAPTER 9 Introduction to Inference

ws
Ne
ws
Ne

APPLICATION 9.3 B]
RO
g
<
Se
a

Return of the free-throw shooter

PLET
AP In Activity 9.2B (page 442), we introduced the shooter can hit 80% of his shots overall?
the Significance Test applet at the book’s Explain why or why not.
Web site, www.whfreeman.com/sta2e. 2. Add another five shots to the total. Can you
Now that you have studied some of the cautions for decide at this point?
performing inference about a population propor-
tion, let’s return to the applet once again. 3. Continue to add five shots at a time until you
are confident that you know whether
100 the null hypothesis is true or false.
How many shots did it take? Perform
% Hits
a significance test using the sample of
free throws that the player has shot.
Do the results of the significance test
50 confirm your belief?
4. Check the “Show True %” box to
see the true value of p. Record this
value.
Hits  12/25  48% 0 5. Remove the true value of p from the
1 40 plot and pick a new shooter. Repeat
Misses  13/25  52% Shots
the process of using the applet to
Show True % add five shots at a time until you feel
Shots 25 Shoot New Shooter
Show null hypothesis confident that you know whether the
null hypothesis is true or false. Did it
APPLET TIP: The null hypothesis is always p 5 0.8, take more shots or fewer shots this time? View and
and the alternative hypothesis is p , 0.8 since the record the true value of p.
applet never fixes p at a value higher than 0.8. 6. Repeat Step 5 several more times. Did it take
longer for you to determine whether the null
QUESTIONS hypothesis was true or false when p was closer to
1. Use the applet to have the shooter make five 0.8 or when it was farther from 0.8? Explain why
shots. Based on these data, can you decide whether this makes sense.

Section 9.3 Summary

Statistical inference is less widely applicable than exploratory analysis of data.
Any inference method requires the right setting, in particular the right design
for a random sample or randomized experiment. Understanding the meaning of
confidence levels and statistical significance helps avoid improper conclusions.
Increasing the number of observations has a straightforward effect on confidence
intervals: the interval gets shorter for the same level of confidence. More observa-
tions usually drive down the P-value of a test when the truth about the popula-
tion stays the same, making tests harder to interpret than confidence intervals.
A finding with a small P-value may not be practically interesting if the sample is
 9. 3 Use and Abuse of Statistical Inference 463

large, and an important truth about the population may fail to be significant if
the sample is small. Avoid depending on fixed significance levels such as 5% to
make decisions.

Section 9.3 9.53 Cholesterol and breast cancer The Journal of Women’s Health reported
Exercises on a study of 7528 women aged 65 and older. The women were all participants
in a study of fractures due to osteoporosis, a disease that weakens the bones.
At one clinic visit during the study, researchers asked the women whether
they took any cholesterol-lowering drugs (such as the popular statins). None
of the women had breast cancer at the time the researchers collected informa-
tion about their use of cholesterol-lowering medication. The scientists then
followed the women, for nearly seven years on average, to see who developed
breast cancer. After accounting for age and weight (two factors related to breast
cancer risk), the researchers found that statin users were about 75% less likely to
develop breast cancer than women who weren’t taking any cholesterol-lowering
medication.15 A news report said the study suggested that cholesterol-lowering
medications might reduce women’s risk of breast cancer. Can we be confident in
this conclusion? Explain.
9.54 What is significance good for? Which of the following questions does a
significance test answer?
(a) Is the sample or experiment properly designed?
(b) Is the observed effect due to chance?
(c) Is the observed effect important?
9.55 Searching for ESP A researcher looking for evidence of extrasensory per-
ception (ESP) tests 500 subjects. Four of these subjects do significantly better
(P , 0.01) than random guessing.
(a) Is it proper to conclude that these four people have ESP? Explain your answer.
(b) What should the researcher now do to see whether any of these four subjects
have ESP?
9.56 Is this convincing? You are planning to test a new vaccine for a virus that
now has no vaccine. Since the disease is usually not serious, you will expose 100
volunteers to the virus. After some time, you will record whether or not each
volunteer has been infected.
(a) Explain how you would use these 100 volunteers in a designed experiment to
test the vaccine. Include all important details of designing the experiment (but
don’t actually do any random allocation).
(b) You hope to show that the vaccine is more effective than a placebo. The ex-
periment gave a P-value of 0.25. Explain carefully what this means.
(c) Your fellow researchers do not consider this evidence strong enough to recom-
mend regular use of the vaccine. Do you agree?
9.57 Why are larger samples better? Statisticians prefer large samples. Describe
briefly the effect of increasing the size of a sample (or the number of subjects in an
experiment) on each of the following.
(a) The margin of error of a 95% confidence interval.
(b) The P-value of a test when H0 is false and all facts about the population remain
unchanged as n increases.
464 CHAPTER 9 Introduction to Inference

9.58 Low pressure The Automobile Association of America (AAA) reports that
85% of all cars on the road are driving on underinflated tires. Two researchers,
Geraldine and Patti, collect data on tire pressure from a random sample of cars
that were brought in for oil changes.
(a) Geraldine wants to perform a significance test of AAA’s claim. Patti would
rather calculate a confidence interval. Which would you recommend? Why?
(b) Will Geraldine and Patti be able to generalize their results to the population of
all cars on the road? Why or why not?
9.59 Why we seek significance Asked why statistical significance appears so of-
ten in research reports, a student says, “Because saying that results are significant
tells us that they cannot easily be explained by chance variation alone.” Do you
think that this statement is essentially correct? Explain your answer.
9.60 Online directory While surfing the Internet one day, Raul finds this claim
about the site www.whitepages.com: “We cover 80% of U.S. adults.” Raul thinks
the site is exaggerating.
(a) Describe the details of a study that Raul could realistically perform to investi-
gate this claim.
(b) If Raul performs a significance test using the data from the study in (a), is it
possible that the conclusion he draws about the claim is wrong? Explain.
(c) If Raul calculates a confidence interval using the data from the study in (a), is
it possible that the conclusion he draws about the claim is wrong? Explain.

CHAPTER 9 REVIEW
Statistical inference draws conclusions about a population on the basis of sample
data and uses probability to indicate how reliable the conclusions are. A confidence
interval estimates an unknown parameter. A significance test shows how strong
the evidence is for some claim about a parameter. Sections 9.1 and 9.2 present the
reasoning of confidence intervals and tests and give details for inference about a
population proportion p.
The probabilities in both confidence intervals and tests tell us what would happen
if we used the formula for the interval or test very many times. A confidence level is
the probability that the formula for a confidence interval actually produces an interval
that contains the unknown parameter. A 95% confidence interval gives a correct result
95% of the time when we use it repeatedly. Figure 9.12 illustrates the reasoning using
the approximate 95% confidence interval for a population proportion p.

SRS size n

SRS size n
About 95% of these
intervals cover
SRS size n the true p

Population
p=?
Figure 9.12 The idea of a confidence interval.
 Chapter 9 Review 465

A P-value is the probability that the test would produce a result at least as ex-
treme as the observed result if the null hypothesis really were true. Figure 9.13 il-
lustrates the reasoning, placing the sample proportion p^ from our one sample on the
Normal curve that shows how p^ would vary in all possible samples if the null hy-
pothesis were true. A P-value tells us how surprising the observed outcome is. Very
surprising outcomes (small P-values) are good evidence that the null hypothesis is
not true.

SRS size n
p
SRS size n
p
SRS size n
p
P-value

Population
1
H0 : p = 2 1 p
2
Ha: p > 1
2

Figure 9.13 The idea of a significance test.

To detect wise and unwise uses of inference, you must know the basic reason-
ing and also be aware of some potential pitfalls. Section 9.3 will help.

Here is a review list of the most important skills you should have developed
from your study of this chapter.
A. SAMPLING DISTRIBUTIONS
1. Explain the idea of a sampling distribution. See Figure 9.2 (page 423).
2. Use the Normal sampling distribution of a sample proportion and the 68–95–99.7
rule to find probabilities involving p^ .

B. CONFIDENCE INTERVALS
1. Explain the idea of a confidence interval. See Figure 9.12.
2. Explain in nontechnical language what is meant by “95% confidence” and other
statements of confidence in statistical reports.
3. Use the basic formula p^ 6 2 !p^ 1 1 2 p^ 2yn to obtain an approximate 95% confi-
dence interval for a population proportion p.
4. Understand how the margin of error of a confidence interval changes with the
sample size and the level of confidence.
5. Detect major mistakes in applying inference, such as improper data production,
selecting the best of many outcomes, ignoring high nonresponse and outliers.

C. SIGNIFICANCE TESTS
1. Explain the idea of a significance test. See Figure 9.13.
2. State the null and alternative hypotheses in a testing situation when the param-
eter in question is a population proportion p.
466 CHAPTER 9 Introduction to Inference

3. Explain in nontechnical language the meaning of the P-value when you are
given the numerical value of P for a test.
4. Explain the meaning of “statistically significant at the 5% level” and other state-
ments of significance. Explain why significance at a specific level such as 5% is less
informative than a P-value.
5. Recognize that significance testing does not measure the size or importance of
an effect.
6. Recognize and explain the effect of small and large samples on the significance
of an outcome.

CHAPTER 9 REVIEW EXERCISES

9.61 Body temperature We have all heard that 98.6 degrees Fahrenheit (or 37
degrees Celsius) is “normal body temperature.” In fact, there is evidence that most
people have a slightly lower body temperature. You plan to measure the body temper-
ature of a random sample of people very accurately. You hope to show that a majority
have temperatures lower than 98.6 degrees.
(a) Say clearly what the population proportion p stands for in this setting.
(b) In terms of p, what are your null and alternative hypotheses?
9.62 Significant versus important Explain how a result can be statistically signifi-
cant but have little practical importance.
9.63 We love football! A recent Gallup Poll conducted telephone interviews with a
random sample of adults aged 18 and older. Data were obtained for 1000 people. Of
these, 37% said that football is their favorite sport to watch on television.
(a) Explain to someone who knows no statistics why we can’t just say that 37% of
adults would say that football is their favorite sport to watch on television.
(b) Construct and interpret a 95% confidence interval for the population proportion p.
(c) The poll announced a margin of error of 3 percentage points. Do you agree?
(d) This poll was taken in December, an exciting part of the football season. Do you
think a similar poll conducted in June might produce different results? Explain why
or why not.
9.64 Roulette A roulette wheel has 18 red slots among its 38 slots. You observe many
spins and record the number of times that red occurs. Now you want to use these data
to test whether the probability p of a red has the value that is correct for a fair roulette
wheel.
(a) State the hypotheses H 0 and H a that you will test.
(b) In 50 spins, the ball landed in a red slot 31 times. Carry out a significance test and
report your conclusion.
(c) The casino manager uses your data to produce a 99% confidence interval for p
and gets (0.44, 0.80). He says that this interval provides convincing evidence that the
wheel is fair. How do you respond?
9.65 Senior citizens Table 2.4 (page 44) records the percent of residents aged 65 or
older in each of the 50 states. You can check that this percent is 14% or higher in 10 of
the states. So the sample proportion of states with at least 14% of elderly residents is
p^ 5 10y50 5 0.20. Explain why it does not make sense to calculate a 95% confidence
interval for the population proportion p.
 Chapter 9 Review Exercises 467

9.66 Comparing package designs A company compares two package designs for
a laundry detergent by placing bottles with both designs on the shelves of several
markets. Checkout scanner data on more than 5000 bottles bought show that more
shoppers bought Design A than Design B. The difference is statistically significant
(P 5 0.02). Can we conclude that consumers strongly prefer Design A? Explain your
answer.
9.67 Online gaming, I A random sample of 1100 teenagers (aged 12 to 17) were
asked whether they played games online; 775 said that they did.
(a) Construct and interpret a 99% confidence interval for the population proportion p.
(b) Suppose that the results of the survey were used to construct separate 99%
confidence intervals for boys and girls. Would the margins of error for those two
confidence intervals be the same as, larger than, or smaller than the margin of error
for the interval you constructed in (a)? Justify your answer.
(c) In the context of this exercise, describe a real-world issue that isn’t included in the
margin of error.
9.68 Online gaming, II Refer to the previous exercise. How large a sample would
you need to take to estimate p within 2% at a 99% confidence level? Show your work.
9.69 Abstinence? The Gallup Youth Survey asked a random sample of 439 U.S. teens
aged 13 to 17 whether they thought that young people should wait to have sex until
marriage.16 The number who said “Yes” was 246.
(a) Do these results give convincing evidence that a majority of U.S. teens would say
that young people should wait to have sex until they’re married? Carry out a signifi-
cance test to help answer this question.
(b) Construct and interpret a 95% confidence interval for the proportion of all teens
who would say “Yes” if asked this question.
(c) Do the inference methods in (a) and (b) tell us whether the teens who responded
to the survey were telling the truth? Explain.
9.70 Alternative medicine A nationwide random survey of 1500 adults asked about
attitudes toward “alternative medicine” such as acupuncture, massage therapy, and
herbal therapy. Among the respondents, 660 said they would use alternative medi-
cine if traditional medicine was not producing the results they wanted. Do these data
provide good evidence that more than 1/3 of all adults would use alternative medicine
if traditional medicine did not produce the results they wanted?
(a) State the hypotheses to be tested.
(b) If your null hypothesis is true, what is the sampling distribution of the sample
proportion p^ ? Sketch this distribution.
(c) Mark the actual value of p^ on the curve. Does it appear surprising enough to give
good evidence against the null hypothesis?
(d) Carry out the significance test called for in detail. Show the steps of the inference
procedure (hypotheses, sampling distribution, data, P-value, conclusion) clearly.
This page intentionally left blank
 C H A P T E R 10

Inference in Practice
1 0.1 Chi-Square Tests: Goodness of Fit
and Two-Way Tables
1 0.2 Inference about a Population Mean

Jason Lugo/iStockphoto.com
Do dogs resemble their owners?
Some people look a lot like their pets. Maybe they deliberately choose
animals that match their appearance. Or maybe we’re perceiving
similarities that aren’t really there. Researchers at the University of California,
San Diego, decided to investigate. They designed an experiment to test whether
dogs and their owners resembled one another. The researchers believed that the
resemblance might differ for owners of purebred and nonpurebred dogs.
A total of 45 dogs and their owners were photographed separately at three
dog parks. Then researchers “constructed triads of pictures, each consisting
of one owner, that owner’s dog, and one other dog photographed at the same
park.” The subjects in the experiment were 28 undergraduate psychology
students. Each subject was presented with the individual sets of photographs and
asked to identify which dog belonged to the pictured owner. A dog was classified
as resembling its owner if more than half of the 28 undergraduate students
matched dog to owner.1
Here are some results. Of the 45 dogs photographed, 23 were classified
as resembling their owners. That’s just over half of the dogs in the study. If we
assume that these dogs are representative of the larger canine population, we
could use the methods of Chapter 9 to construct a confidence interval for the
proportion of all dogs that resemble their owners.
What about the researchers’ belief that the resemblance between dog
and owner might differ for purebred and nonpurebred dogs? Of the 45
dogs in the study, 25 were purebreds and 20 were not. For the purebred
dogs, 16 resembled their owners. As a proportion, that’s 16/25  0.64. For
the nonpurebred dogs, 7 resembled their owners. That’s 7/25  0.28 as a
proportion. Does such a large difference in proportions provide convincing
evidence in favor of the researchers’ belief?

469
470 CHAPTER 10 Inference in Practice

To answer this question, we need a new kind of significance test—one that

allows us to compare proportions for two groups or populations. This new
test, called a chi-square test, is the subject of Section 10.1. In Section 10.2, we’ll
introduce confidence intervals and significance tests for a population mean.

1 0 . 1 Chi-Square Tests: Goodness of Fit and Two-Way Tables

ACTIVITY 10.1
Return of the M&M’s!
Materials: One 1.69-ounce bag of milk chocolate distribution of M&M’s in your individual bag with
M&M’s per student or data from Activity 9.3 the advertised distribution. We want to see if there is
sufficient evidence to dispute the company’s claim.
The M&M/Mars Company says that the color
distribution for its plain M&M’s candies is 24% blue, 1. Make a table on your paper like the one shown
20% orange, 16% green, 14% yellow, 13% red, and below. Fill in the counts, by color, and the total
13% brown. If you completed Activity 9.3 (page 453), number of M&M’s in your bag in the “Observed”
then you have already tested the company’s claim column. We have entered values from a bag that we
about the proportion of blue candies, p 5 0.24. The purchased for demonstration purposes. Leave the
purpose of this Activity is to compare the entire color other columns blank for now.

Color Observed (O) Expected (E) OE

Blue 9 (0.24)(61)  14.64 9  14.64  5.64
Orange 10 (0.20)(61)  12.20 10  12.20  2.20
Green 15 (0.16)(61)  9.76 15  9.76  5.24
Yellow 12 (0.14)(61)  8.54 12  8.54  3.46
Red 8 (0.13)(61)  7.93 8  7.93  0.07
Brown 7 (0.13)(61)  7.93 7  7.93  0.93
Total 61 61 0

2. Assuming that the company’s claim is true, how 3. How close are your observed counts to the expected
many of each color would you expect in your bag? values? To answer this question, calculate the difference
We had a total of 61 M&M’s in our bag. How many for each color: Observed  Expected. Record these
blue M&M’s should we expect in a bag with 61 differences in a column labeled O  E. Find the total of
candies? If 24% of all plain M&M’s are blue, then we’d your six O  E values. What do you notice?
expect (0.24)(61)  14.64 blue M&M’s, on average,
in a bag with 61 M&M’s. In a similar way, we can 4. In Step 3, the sum of the difference values O  E
calculate the expected values for the other five colors. should be 0 (up to roundoff error). Does that mean
Follow our model, and calculate these expected your individual bag matches the color distribution
values (to two decimal places) for your bag. Record claimed by M&M/Mars perfectly? Of course not.
your results in the “Expected” column of your table. You get a total difference of 0 because the positive
 10. 1 Chi-Square Tests: Goodness of Fit and Two-Way Tables 471

and negative values of O  E cancel each other. values (does that remind you of anything earlier in
How can we fix this problem? We could use absolute the text?). Compute the values of (O  E)2 and fill in
value, but instead, we choose to square the difference the next column of your table. What’s your total?

Color Observed (O) Expected (E) OE (O  E)2 (O  E)2 / E

Blue 9 (0.24)(61)  14.64 9  14.64  5.64 (5.64)2  31.81 31.81 / 14.64  2.173
Orange 10 (0.20)(61)  12.20 10  12.20  2.20 (2.20)  4.84
2
4.84 / 12.20  0.397
Green 15 (0.16)(61)  9.76 15  9.76  5.24 (5.24)  27.46
2
27.46 / 9.76  2.81
Yellow 12 (0.14)(61)  8.54 12  8.54  3.46 (3.46)  11.97
2
11.97 / 8.54  1.402
Red 8 (0.13)(61)  7.93 8  7.93  0.07 (0.07)  0.005
2
0.005 / 7.93  0.0006
Brown 7 (0.13)(61)  7.93 7  7.93  0.93 (0.93)  0.86
2
0.86 / 7.93  0.108
Total 61 61 0 76.945 6.8906

5. In the last column of your table, divide the values x 2 should be very small. Are the entries in the last
in the (O  E)2 column by the corresponding column all about the same, or do any values
expected values (E). Then find the total. This final stand out because they are much larger than the
result is called the chi-square statistic and is denoted others? Did you get many more or many fewer of a
by x 2. Keep this number handy—you will use it particular color than expected?
later in the chapter.
7. Compare your value of x 2 with the results
6. If your sample reflects the distribution obtained by your classmates. How do we know
advertised by the M&M/Mars Company, whether a particular x 2 statistic is large enough to
there should be very little difference between provide convincing evidence against the company’s
the observed counts and the expected values. claim? We need to learn more about chi-square
Therefore, the calculated values making up the sum before we can answer this question.

Goodness of fit and the chi-square statistic

Suppose you open a 1.69-ounce bag of plain M&M’s and discover that of the 61
M&M’s in the bag, only 9 are blue. In your sample of size 61, the proportion of
blue M&M’s is p^ 5 9y61 5 0.148. Since the company claims that 24% of all plain
M&M’s are blue, you might feel that you didn’t get your fair share of blues. You
could use the z test described in Chapter 9 to test the hypotheses

H0 : p 5 0.24
Ha : p 2 0.24

where p is the population proportion of blue M&M’s. Here are the results: z  1.69;
P-value  0.09. Remember: the P-value tells us the probability of getting a sample
result as “surprising” as the one we did if the null hypothesis is true. In this case,
P  0.09 says that roughly 1 in every 11 bags of 61 M&M’s would have a propor-
tion of blue candies that differed from p  0.24 by at least as much as our bag did.
The P-value gives marginal evidence against the company’s claim about the overall
proportion of blue M&M’s.
472 CHAPTER 10 Inference in Practice

You could then perform additional significance tests for each of the remaining
colors. But this would be inefficient. More important, performing six individual tests
wouldn’t tell you how likely it is that six sample proportions would differ from the val-
ues stated by M&M/Mars as much as your sample does. There is a single test that can
be applied to see if the observed distribution of sample data is significantly different in
some way from the stated population distribution. It is called the chi-square 1 x2 2 test
for goodness of fit. Like any other test, we must begin by stating hypotheses.

Example 10.1 Hypotheses for a goodness of fit test

Let’s continue with the M&M’s scenario. The null hypothesis in a chi-square good-
ness of fit test should state a claim about the distribution of a single categorical
variable in the population of interest. In the case of M&M’s, the categorical variable
we’re measuring is color. The appropriate null hypothesis is

H0 : The company’s stated color distribution for plain M&M’s candies is correct;
that is,
pblue 5 0.24, porange 5 0.20, pgreen 5 0.16, pyellow 5 0.14, pred 5 0.13, pbrown 5 0.13.

The corresponding alternative hypothesis is

Ha: The company’s stated color distribution for plain M&M’s candies is not
correct; that is, at least two of the stated proportions are incorrect.

The idea of the goodness of fit test is this: we compare the observed counts for
our sample with the counts that would be expected if the color distribution claimed
by M&M/Mars is correct. The more the observed counts differ from the expected
counts, the more evidence we have against the company’s claim. In general, the
expected counts for a categorical variable can be obtained by multiplying the pro-
portion of the distribution for each category by the sample size.

Example 10.2 Computing expected counts for a

goodness of fit test
If the company’s claim is true, 24% of all plain M&M’s produced are blue. Does that
mean that every 1.69-ounce bag of plain M&M’s will contain exactly 24% blue can-
dies? Of course not. Some bags will have more than 24% blue candies; other bags
will have less than 24%. On average, the bags will contain 24% blue M&M’s. For bags
with 61 M&M’s, the average number of blue M&M’s should be (0.24)(61) 5 14.64.
This is our expected count of blue M&M’s.
Using this same method, we can find the expected counts for the other color
categories:

Orange: (0.20)(61) 5 12.20 Green: (0.16)(61) 5 9.76 Yellow: (0.14)(61) 5 8.54

Red: (0.13)(61) 5 7.93 Brown: (0.13)(61) 5 7.93
 10. 1 Chi-Square Tests: Goodness of Fit and Two-Way Tables 473

Did you notice that the expected count in Example 10.2 sounds a lot like the
expected value of a random variable from Chapter 8? That’s no coincidence. The
number of M&M’s of a specific color in a bag of 61 candies is a random variable. As
a result, we can find its expected value.
To see if the data give evidence against the null hypothesis, we need to
compare the observed counts from our sample with the expected counts. If
the observed counts are far from the expected counts, that’s the evidence we
were seeking. The chi-square test of goodness of fit uses a statistic (called the
chi-square statistic) that measures how far apart the observed and expected
counts are.

Chi-square statistic

The chi-square statistic x2 is a measure of how far the observed

counts for a categorical variable are from the expected counts. The
formula for the statistic is
1 observed count 2 expected count 2 2
x 5 a
2
expected count
The symbol  means “sum over all the categories.”

The chi-square statistic is a sum of terms, one for each value of the categori-
cal variable. In the M&M’s example, 9 of the candies in our bag were blue. The
expected count of blues is 14.64. So the term in the chi-square statistic from this
category is

1 observed count 2 expected count 2 2 1 9 2 14.64 2 2

5
expected count 14.64
31.81
5 5 2.173
14.64

Example 10.3 Calculating the chi-square statistic

Here are the observed and expected counts for the M&M’s data of Activity 10.1 side
by side:

Color Observed Expected

Blue 9 14.64
Orange 10 12.20
Green 15 9.76
Yellow 12 8.54
Red 8 7.93
Brown 7 7.93
474 CHAPTER 10 Inference in Practice

We can now find the chi-square statistic, adding six terms for the six color categories:

1 9 2 14.64 2 2 1 10 2 12.20 2 2 1 15 2 9.76 2 2 1 12 2 8.54 2 2

x2 5 1 1 1
14.64 12.20 9.76 8.54
1 8 2 7.93 2 2 1 7 2 7.93 2 2
1 1
7.93 7.93
5 2.173 1 0.397 1 2.813 1 1.402 1 0.0006 1 0.108
5 6.89

Because x2 measures how far the observed counts are from what would be
expected if H0 were true, large values are evidence against H0. Is x2 5 6.89 a
large value? You know the drill: compare the observed value 6.89 against the
sampling distribution that shows how x2 would vary if the null hypothesis were
true. This sampling distribution is not a Normal distribution. It is a right-skewed
distribution that allows only positive values because x2 can never be negative.
Moreover, the sampling distribution differs depending on the number of pos-
sible values for the categorical variable (that is, on the number of categories).
Here are the facts.

The chi-square distributions

The sampling distribution of the chi-square statistic x2 when the null

hypothesis is true is called a chi-square distribution.

The chi-square distributions are a family of distributions that take

only positive values and are skewed to the right. A specific chi-square
distribution is specified by giving its degrees of freedom.

The chi-square test for goodness of fit uses critical values from the
chi-square distribution with degrees of freedom 5 the number of
categories 2 1.

Figure 10.1 shows the density curves for three members of the chi-square family
of distributions. As the degrees of freedom (df) increase, the density curves become
less skewed and larger values become more probable. We can’t find P-values as
areas under a chi-square curve by hand, though a computer or calculator can do it
for us.
Table 10.1 is a shortcut to avoid finding the P-value. It shows how large the
chi-square statistic x2 must be in order to be significant at various levels. This isn’t
as good as an actual P-value, but it is often good enough. Each number of degrees
of freedom has a separate row in the table. We see, for example, that a chi-square
statistic with 3 degrees of freedom is significant at the 5% level if it is greater than
7.81 and is significant at the 1% level if it is greater than 11.34.
 10. 1 Chi-Square Tests: Goodness of Fit and Two-Way Tables 475

Figure 10.1 The density

curves for three members
of the chi-square family
of distributions. The
sampling distributions of df = 1
chi-square statistics belong
to this family. df = 4

df = 8

Table 10.1 To Be Significant at Level a, a Chi-Square Statistic Must Be

Larger Than the Table Entry for a

Significance level a

df 0.25 0.20 0.15 0.10 0.05 0.01 0.001

1 1.32 1.64 2.07 2.71 3.84 6.63 10.83

2 2.77 3.22 3.79 4.61 5.99 9.21 13.82
3 4.11 4.64 5.32 6.25 7.81 11.34 16.27
4 5.39 5.99 6.74 7.78 9.49 13.28 18.47
5 6.63 7.29 8.12 9.24 11.07 15.09 20.51
6 7.84 8.56 9.45 10.64 12.59 16.81 22.46
7 9.04 9.80 10.75 12.02 14.07 18.48 24.32
8 10.22 11.03 12.03 13.36 15.51 20.09 26.12
9 11.39 12.24 13.29 14.68 16.92 21.67 27.88

We have reproduced Table 10.1 in the back of the book (Table E) for easier
reference.

Example 10.4 The M&M’s study: conclusion

Our bag of 61 plain M&M’s had fewer blues and more greens than expected if
the company’s claimed color distribution is correct. Comparing observed and
expected counts gave the chi-square statistic x2 5 6.89. The last step is to assess
significance.
476 CHAPTER 10 Inference in Practice

The variable “color” has 6 possible values. The chi-square statistic

therefore has degrees of freedom

number of categories 2 1 5 6 2 1 5 5

Look in the df 5 5 row of Table 10.1. We see that x2 5 6.89 is larger than the
critical value 6.63 required for significance at the a 5 0.25 level but smaller
than the critical value 7.29 for a 5 0.20. So the P-value is between 0.20 and
Marta Rostek/stock.xchng
0.25. That’s not enough evidence to refute the color distribution proposed by
M&M/Mars.

Like our test for a population proportion, the chi-square test uses some ap-
proximations that become more accurate as we take more observations. Here is a
rough rule for when it is safe to use this test.2

Expected counts required for the chi-square test

You can safely use the chi-square test when no more than 20% of the
expected counts are less than 5 and all individual expected counts are
1 or greater.

The M&M’s study easily passes this test: the smallest expected count is 7.93.

Exercises Exercises 10.1 to 10.5 refer to the following setting. The M&M/Mars Company reports
that their M&M Peanut Chocolate Candies have the following distribution: 23%
each of blue and orange, 15% each of green and yellow, and 12% each of red and
brown. Joey bought a bag of peanut M&M’s and counted out the following distribu-
tion of colors: 12 blue, 7 orange, 13 green, 4 yellow, 8 red, and 2 brown.
10.1 Hypotheses State an appropriate pair of hypotheses for testing the
company’s claim about the color distribution of peanut M&M’s.
10.2 Expected counts Joey’s bag contained 46 peanut M&M’s. Calculate the
expected count for each color assuming that the company’s claim is true. Show
your work.
10.3 Test statistic Calculate the chi-square statistic for Joey’s sample. Show your work.
10.4 P-value Which chi-square distribution should we use? (Specify the degrees of
freedom.) What’s the P-value for this test? Refer to Table E (in the back of the book).
10.5 Conclusion What conclusion do you draw based on the results of this test?
Justify your answer.
10.6 Activity 10.1 follow-up Use your M&M’s data from Activity 10.1 (page 470)
to perform a chi-square goodness of fit test for the company’s claimed color
distribution. What do you conclude?
 10. 1 Chi-Square Tests: Goodness of Fit and Two-Way Tables 477

Using the chi-square test for goodness of fit

One of the most common applications of the chi-square goodness of fit test is in
the field of genetics. Scientists want to investigate the genetic characteristics of off-
spring that result from mating (also called “crossing”) parents with known genetic
makeups. Scientists use rules about dominant and recessive genes to predict the
proportion of offspring that will fall in each possible genetic category. Then, the
researchers mate the parents and classify the resulting offspring. The chi-square
goodness of fit test helps the scientists assess the validity of their hypothesized dis-
tribution of proportions.

Example 10.5 Genetics and the chi-square

goodness of fit test
Biologists wish to cross two tobacco plants having genetic makeup Gg, indicating
that each has one dominant gene (G) and one recessive gene (g) for color. Each off-
spring plant will receive one gene for color from each parent. The following table,
often called a Punnett square, shows the possible combinations of genes received
by the offspring:

Parent 2 passes on:

Parent 1 passes on: G g

G GG Gg
g Gg gg

The Punnett square suggests that the expected ratio of green (GG) to yellow-green
(Gg) to albino (gg) tobacco plants should be 1:2:1. In other words, the biologists
predict that 25% of the offspring will be green, 50% will be yellow-green, and 25%
will be albino. To test their hypothesis about the distribution of offspring, the bi-
ologists mate 84 pairs of yellow-green parent plants. Of 84 offspring, 23 plants are
green, 50 are yellow-green, and 11 are albino. Do these data differ significantly
from what the biologists have predicted?
We can now follow the steps for a significance test, familiar from Section 9.2.
The hypotheses. The chi-square goodness of fit test starts with these hypotheses:

H0 : The biologists’ predicted color distribution for tobacco plant offspring

is correct; that is, pgreen 5 0.25, pyellow-green 5 0.5, palbino 5 0.25.
Ha : The biologists’ predicted color distribution is not correct; that is, at least
two of the proposed proportions are incorrect.

The sampling distribution. There are three possible values for the categorical
variable “offspring color”: green, yellow-green, and albino. We will use critical val-
ues from the chi-square distribution with degrees of freedom

df 5 number of categories 2 1 5 3 2 1 5 2
478 CHAPTER 10 Inference in Practice

The data. First find the expected cell counts. For example, the expected count
of green offspring is (0.25)(84) 5 21. Here is the complete table of observed and
expected counts side by side:
Offspring color Observed Expected
Green 23 (0.25)(84) 5 21
Yellow-green 50 (0.50)(84) 5 42
Albino 11 (0.25)(84) 5 21

Looking at these counts, we see that there were many more yellow-green tobacco
offspring than expected and many fewer albino plants than expected.
The P-value. The chi-square statistic is

1 23 2 21 2 2 1 50 2 42 2 2 1 11 2 21 2 2
x2 5 1 1
21 42 21
5 0.190 1 1.524 1 4.762 5 6.476

To find the P-value, look at the df 5 2 line of Table E (in the back of the book). The
observed chi-square x2 5 6.476 is larger than the critical value 5.99 for a 5 0.05 but
smaller than the critical value 9.21 for a 5 0.01. So the P-value is between 0.01 and
0.05. (Technology can give the actual P-value. It is P 5 0.0392.)
Conclusion. We have significant evidence (P , 0.05) that the biologists’ hy-
pothesized distribution of color for tobacco plant offspring is incorrect.

CALCULATOR CORNER Goodness of fit tests

You can use the TI-84/TI-Nspire to perform the calculations for a chi-
square goodness of fit test. We’ll use the M&M’s data from Activity 10.1
(page 470) to illustrate the steps.
1. Enter the observed counts and expected counts in two separate lists.
TI-84
• Clear lists L1 and L2.
• Enter the observed counts in list L1. Calculate the expected counts separately
and enter them in list L2.
TI-Nspire
• On a new Lists & Spreadsheet page, name column A “obsd” and column B “expd.”
• Enter the observed counts in the “obsd” list. Calculate the expected counts
separately and enter them in list “expd.”
2. Perform a chi-square goodness of fit test.
TI-84
Press A, arrow over to TESTS, and choose D:x 2 GOF-Test. Enter the in-
puts shown. If you choose Calculate, you’ll get a screen with the test statistic,
 10. 1 Chi-Square Tests: Goodness of Fit and Two-Way Tables 479

P-value, and df. If you choose the Draw option, you’ll get a picture of
the appropriate chi-square distribution with the test statistic marked
and with the area corresponding to the P-value shaded.

TI-Nspire
Press b, then choose 4:Statistics, 4:Stat Tests…, and
7:x 2GOF. Enter the inputs shown. If you choose OK, you’ll get the
results screen shown. If you click the “Plot data” button, you’ll also get a
picture of the appropriate chi-square distribution with the test statistic
marked and with the area corresponding to the P-value shaded.

Exercises Use Table E (in the back of the book) to help you with Exercises 10.7 to 10.12.
10.7 Mendel and the peas Gregor Mendel (1822–1884), an Austrian monk, is
considered the father of genetics. Mendel studied the inheritance of various traits
in pea plants. One such trait is whether the pea is smooth or wrinkled. Mendel
predicted a ratio of 3 smooth peas for every 1 wrinkled pea. In one experiment, he
observed 423 smooth and 133 wrinkled peas. Carry out a chi-square goodness of
fit test for Mendel’s prediction. What do you conclude?
10.8 Seagulls by the seashore Do seagulls show a preference for where they
stand? To answer this question, biologists examined a small piece of shore whose
area was made up of 56% sand, 29% mud, and 15% rocks. They recorded where
each of 200 seagulls was standing on a particular afternoon. In all, 128 seagulls
stood on the sand, 61 stood in the mud, and 11 stood on the rocks. Carry out a
chi-square goodness of fit test. What do you conclude?
10.9 Skittles Statistics teacher Jason Molesky contacted the M&M/Mars
Company to ask about the color distribution for Skittles. Here is an excerpt of the
response he received:
480 CHAPTER 10 Inference in Practice

The original flavor blend for the SKITTLES BITE SIZE CANDIES is lemon,
lime, orange, strawberry and grape. They were chosen as a result of consumer
preference tests we conducted. The flavor blend is 20 percent of each flavor.

(a) State appropriate hypotheses for a significance test of the company’s claim.
(b) Find the expected counts for a bag of Skittles with 60 candies.
(c) How large a x2 statistic would you need to get in order to have significant
evidence against the company’s claim at the a 5 0.05 level? Refer to Table E (in
the back of the book).
(d) How large a x2 statistic would you need to get in order to have significant
evidence against the company’s claim at the a 5 0.01 level?
10.10 More Skittles Refer to the previous exercise.
(a) Create a set of observed counts for a bag with 60 candies that results in a
P-value between 0.01 and 0.05. Show the calculation of your chi-square
statistic.
(b) What conclusion would you draw in this case? Explain.
10.11 Is your random number generator working? Use your calculator’s
randInt function to generate 200 digits from 0 to 9 and store them in a list.
(a) State appropriate hypotheses for a significance test to determine whether your
calculator’s random number generator is working properly.
(b) Complete a goodness of fit test. Report your expected counts, chi-square
statistic, P-value, and your conclusion.
10.12 Prob Sim The Prob Sim APP for the TI-84 allows you to simulate
tossing coins, rolling dice, picking marbles, spinning a spinner, drawing
cards, and playing the lottery. (If you have a TI-Nspire, just switch to the
TI-84 faceplate.)
• To run the APP, press the Œ key, and then choose Prob Sim. You
should see the introductory screen shown in the margin. Press b to see the
main menu.
• Choose 4. Spin Spinner. Spin the spinner a total of 200 times with 4
sets of 50 spins. Record the number of times that the spinner lands in each of
the four numbered sections.
• Perform a significance test of the hypothesis that this program yields an
equal proportion of 1s, 2s, 3s, and 4s. (If you do not have the Prob Sim APP, use
the following sample data: 51 1s, 39 2s, 53 3s, 57 4s.)

Relationships between categorical variables:

two-way tables
In Chapter 7, we used two-way tables to display the sample space of a random
phenomenon involving two events. Two-way tables can also be used to describe
the relationship between two categorical variables, as the following example
illustrates.
 10. 1 Chi-Square Tests: Goodness of Fit and Two-Way Tables 481

Example 10.6 Do angry people have more heart

disease?
People who get angry easily tend to have more heart disease. That’s the conclusion
of a study that followed a random sample of 12,986 people from three locations
for about four years.3 All subjects were free of heart disease at the beginning of the
study. The subjects took the Spielberger Trait Anger Scale, which measures how
prone a person is to sudden anger. Here is a two-way table that summarizes the data
for the 8474 people in the sample who had normal blood pressure. CHD stands for
“coronary heart disease.” This includes people who had heart attacks and those who
needed medical treatment for heart disease.

Low anger Moderate anger High anger Total

CHD 53 110 27 190
No CHD 3057 4621 606 8284
Total 3110 4731 633 8474

Anger rating on the Spielberger scale is a categorical variable that takes three
possible values: low, medium, and high. Whether someone gets heart disease is also
a categorical variable. The two-way table of Example 10.6 shows the relationship
between anger rating and heart disease for a random sample of 8474 people. How
can we make sense of the information contained in this table?
First, look at the distribution of each variable separately. The distribution of a
categorical variable says how often each outcome occurred. The “Total” column at
the right of the table contains the totals for each of the rows. These row totals give
the distribution of heart disease for all people in the study, all anger levels com-
bined. The “Total” row at the bottom of the table gives the distribution of anger rat-
ing for all 8474 people, CHD and non-CHD combined. It is often clearer to present
these distributions using percents.
We might report the distribution of heart disease as

190
percent CHD 5 5 0.0224 5 2.24%
8474
8284
percent no CHD 5 5 0.9776 5 97.76%
8474

Likewise, we could give the distribution of Spielberger anger rating as

3110
percent Low 5 5 0.3670 5 36.70%
8474
4731
percent Moderate 5 5 0.5583 5 55.83%
8474
633
percent High 5 5 0.0747 5 7.47%
8474
482 CHAPTER 10 Inference in Practice

The two-way table contains more information than the two distributions of
heart disease and anger rating alone. The nature of the relationship between these
two variables cannot be deduced from the separate distributions but requires
the full table. To describe relationships among categorical variables, calculate
appropriate percents from the counts given.

Example 10.7 Anger and heart disease

Since we’re interested in whether angrier people tend to get heart disease more
often, we can compare the percents of people who got heart disease (CHD) in each
of the three anger categories:
Low anger Moderate anger High anger
53 110 27
5 0.0170 5 1.70% 5 0.0233 5 2.33% 5 0.0427 5 4.27%
3110 4731 633
There is a clear trend: as the anger score increases, so does the percent who suffer
heart disease. The bar graph in Figure 10.2 displays these results visually. Is this
relationship between anger and heart disease statistically significant?

5%
Percent with CHD

4%
3%
2%
1%
0%
Low Moderate High
Anger rating
Figure 10.2 Bar graph comparing the percent of people in each anger category who
got coronary heart disease (CHD).

In working with two-way tables, you must calculate lots of percents. Here’s a tip to
help decide what fraction gives the percent you want. Ask, “What group represents the
total that I want a percent of?” The count for that group is the denominator of the frac-
tion that leads to the percent. In Example 10.7, we wanted the percent of each anger level
who got CHD, so the total counts in the three anger categories are the denominators.

Inference for a two-way table: hypotheses

and expected counts
We often gather data and arrange them in a two-way table to see if two categori-
cal variables are related to each other. The sample data are easy to investigate: turn
them into percents and look for an association between the row and column vari-
ables. Is the association in the sample evidence of an association between these
variables in the entire population? Or could the sample association easily arise just
from the luck of random sampling? This is a question for a significance test.
 10. 1 Chi-Square Tests: Goodness of Fit and Two-Way Tables 483

Our null hypothesis is that there is no association between the two categorical vari-
ables. The alternative hypothesis is that there is an association between the variables.
For the observational study of Examples 10.6 and 10.7, we want to test the hypotheses

H0 : There is no association between anger and CHD.

Ha : There is an association between anger and CHD.

To test H0 , we compare the observed counts in a two-way table with the expected
counts, the counts we would expect—except for random variation—if H0 were true.

Example 10.8 Anger and heart disease: finding

expected counts
Here again is the two-way table that summarizes the data from the heart disease
(CHD) and anger study.
Low anger Moderate anger High anger Total
CHD 53 110 27 190
No CHD 3057 4621 606 8284
Total 3110 4731 633 8474

Earlier, we determined that 190/8474 5 0.0224, or 2.24%, of the individuals

in this observational study got CHD. If the null hypothesis is true, then there is no
association between CHD and anger level. In that case, we would expect 2.24% of
the individuals in each anger category to get CHD. For the 3110 people classified as
“Low anger,” we would expect 0.0224(3110) 5 69.73 to get CHD.
Another way to write this calculation is
190 #
3110 5 69.73
8474
This is equivalent to

190 # 3110
5 69.73
8474

Do you see where the numbers in this calculation are coming from in the two-way
table? The numerator is the row total for CHD times the column total for “Low anger.”
The denominator is the table total, 8474. We can find the expected counts for the
other cells in the table in a similar way:

CHD, Low CHD, Moderate CHD, High

190 # 3110 190 # 4731 190 # 633
5 69.73 5 106.08 5 14.19
8474 8474 8474
no CHD, Low no CHD, Moderate no CHD, High
8284 # 3110 8284 # 4731 8284 # 633
5 3040.27 5 4624.92 5 618.81
8474 8474 8474
484 CHAPTER 10 Inference in Practice

As we saw in Example 10.8, there is a rule that makes it easy to find expected
counts for a two-way table. Here it is.

Expected counts

The expected count in any cell of a two-way table when H0 is true is

row total 3 column total
expected count 5
table total

Exercises 10.13 Extracurricular activities and grades, I North Carolina State University
studied student performance in a course required by its chemical engineering major.
One question of interest was the relationship between time spent in extracurricular
activities and whether a student earned a C or better in the course. Here are the data
for the 119 students who answered a question about extracurricular activities:4
Extracurricular activities
(hours per week)

,2 2–12 .12

C or better 11 68 3
D or F 9 23 5

(a) Calculate percents that describe the nature of the relationship between time
spent on extracurricular activities and performance in the course.
(b) Draw a bar graph that shows these percents graphically. Be sure to label your
graph appropriately. Give a brief summary in words.
eaturePics

10.14 Extracurricular activities and grades, II Refer to Exercise 10.13.

hoto/F

(a) Write the null and alternative hypotheses for the question of interest.
mpgp

(b) Calculate the expected number of C’s or better for those who spend 2 to 12
hours per week in extracurricular activities.
10.15 Trying to quit, I An observational study of 177 people who were trying
to quit smoking looked at whether or not alcohol consumption was a factor in
relapse (resuming smoking).5 Here are the results:
Alcohol consumption Smoked Did not smoke
Yes 20 13
No 48 96

The entry 20 means that 20 of the study participants consumed alcohol and
smoked during the cessation period.
(a) Find the row and column totals.
(b) Among the drinkers, what percent smoked? Among the teetotalers, what
percent smoked? What is your preliminary conclusion?
 10. 1 Chi-Square Tests: Goodness of Fit and Two-Way Tables 485

(c) Draw a bar graph that shows these percents graphically. Be sure to label your
graph appropriately.
10.16 Trying to quit, II Refer to Exercise 10.15.
(a) Write the null and alternative hypotheses for the question of interest.
(b) Find the expected counts.
10.17 Python eggs, I How is the hatching of water python eggs influenced by
the temperature of the snake’s nest? Researchers assigned newly laid eggs to one
of three water temperatures: hot, neutral, or cold. Hot duplicates the extra warmth
provided by the mother python, and cold duplicates the absence of the mother.
Here are the data on the number of eggs and the number that hatched:6
Eggs Hatched
Cold 27 16
Neutral 56 38
Hot 104 75

(a) Make a two-way table of temperature by outcome (hatched or not).

(b) Calculate the percent of eggs in each group that hatched. The researchers anticipated
that eggs would not hatch in cold water. Do the data support that anticipation?
10.18 Python eggs, II Refer to Exercise 10.17.
(a) Write the null and alternative hypotheses for the question of interest.
(b) What is the expected count of hatched pythons in the hot group?

Chi-square test for two-way tables

To see if the data give evidence against the null hypothesis of “no relationship,”
compare the counts in the two-way table with the counts we would expect if there
really were no relationship. To measure how far apart the observed and expected
counts are, use the chi-square statistic.

Chi-square test for two-way tables

To test the null hypothesis

H0: There is no association between two categorical variables.
against the alternative hypothesis
Ha: There is an association between two categorical variables.
compute the chi-square statistic
1 observed count 2 expected count 2 2
x2 5 a
expected count
This time, the symbol S means “sum over all the cells in the table.”

The chi-square test for a two-way table with r rows and c columns
uses critical values from the chi-square distribution with (r 2 1)(c 2 1)
degrees of freedom.
486 CHAPTER 10 Inference in Practice

Example 10.9 Anger and heart disease: chi-square test

Here are the observed and expected counts for the CHD and anger study side
by side:
Observed Expected
Low Moderate High Low Moderate High

CHD 53 110 27 69.73 106.08 14.19

No CHD 3057 4621 606 3040.27 4624.92 618.81

We can now follow the steps for a significance test, familiar from Section 9.2.
The hypotheses. The chi-square method tests these hypotheses:

H0 : There is no association between anger and heart disease.

Ha : There is an association between anger and heart disease.

The sampling distribution. All the expected cell counts are larger than 5, so
we can safely apply the chi-square test. The two-way table of anger versus CHD has
2 rows and 3 columns. We will use critical values from the chi-square distribution
with degrees of freedom df 5 (2 2 1)(3 2 1) 5 2.
The data. Looking at the observed and expected counts, we see that the high-
anger group has more CHD than expected and the low-anger group has less CHD
than expected. This is consistent with what the percents in Example 10.7 (page 482)
show.
The P-value. The chi-square statistic is

1 53 2 69.732 2 1110 2 106.08 2 2 1 27 2 14.192 2

x2 5 1 1
69.73 106.08 14.19
1 3057 2 3040.272 2
1 4621 2 4624.92 2 2 1 606 2 618.812 2
1 1 1
3040.27 4624.92 618.81
5 4.014 1 0.145 1 11.564 1 0.092 1 0.003 1 0.265 5 16.08

Look at the six terms that we sum to get x2. Most of the total comes from just one
cell: high-anger people have more CHD than expected.
Look at the df 5 2 line of Table E (in the back of the book). The observed chi-
square x2 5 16.08 is larger than the critical value 13.82 for a 5 0.001. So the P-value
is less than 0.001. Statistical software can give the actual P-value. It is P 5 0.0003.
Conclusion. We have highly significant evidence (P , 0.001) that anger and
CHD are associated.

Can we conclude that proneness to anger causes heart disease? The anger and
heart disease study is an observational study, not an experiment. It isn’t surprising
that some lurking variables are confounded with anger. For example, people prone
to anger are more likely than others to be men who drink and smoke. The study
report used advanced statistics to adjust for many differences among the three
 10. 1 Chi-Square Tests: Goodness of Fit and Two-Way Tables 487

anger groups. The adjustments raised the P-value from P 5 0.0003 to P 5 0.002
because the lurking variables explain some of the heart disease. But this is still good
evidence for a relationship. Because the study started with a random sample of
people who had no heart disease and followed them forward in time, and because
many lurking variables were measured and accounted for, it does give some evi-
dence for causation. The next step might be an experiment that involves showing
anger-prone people how to change. Will this reduce their risk of heart disease?

CALCULATOR CORNER Chi-square tests for

two-way tables
We will perform a chi-square test of the anger and heart disease data in
Examples 10.6 through 10.9 on the TI-84/TI-Nspire.
1. Enter the observed counts in a matrix.
TI-84
• Press yV (MATRIX), cursor over to EDIT, and choose 1:[A].
• Specify a 2  3 matrix. Enter the observed counts from the two-way table in
the appropriate cells of the matrix.

TI-Nspire
• Start with a blank Calculator page.
• Go into the Catalog (press k) and use the NavPad to move over to the
templates tab. Use the NavPad to move to the m-by-n matrix… template
and press a or · to select it.
• Enter the number of rows (2) and the number of columns (3) and choose OK.
• Enter the observed counts from the two-way table in the appropriate cells of
the matrix. Use the e key to move to the next cell.
• Store the matrix as a variable: press /h (STO) A·.

CALCULATOR CORNER (continued)

488 CHAPTER 10 Inference in Practice

CALCULATOR CORNER (continued)

2. Perform the chi-square test.

TI-84
• Specify the chi-square test, the matrix where the observed counts are found
(default is [A]), and the matrix where the expected counts will be stored (default
is [B]).
• Press A, arrow over to TESTS, and choose C:x 2-Test.
• Choose “Calculate” or “Draw” to carry out the test. If you choose “Calculate,”
you should get the results shown below. If you choose “Draw,” the chi-square
curve with 2 degrees of freedom will be drawn, the area in the tail will be
shaded, and the P-value will be displayed.

TI-Nspire
• Press b, choose 5:Statistics, 7:Stat Tests…, and 8:x 2
2-way Test.
• In the dialog box, enter a as the “Observed Matrix” and choose OK.
3. To see the expected counts, simply ask for a display of the matrix [B].
TI-84
• Press yV (MATRIX), cursor over to EDIT, and choose 2:[B].
• Verify that these calculator results agree with the results in Example 10.9.
TI-Nspire
• Press h, choose stat.ExpMatrix, and press ·.
 10. 1 Chi-Square Tests: Goodness of Fit and Two-Way Tables 489

Exercises 10.19 Extracurricular activities and grades, III In Exercises 10.13 and 10.14
(page 484), you described the relationship between extracurricular activities and
success in a required course. Here once again is the two-way table describing the
119 students who responded to a survey:

Extracurricular activities
(hours per week)
,2 2–12 .12

C or better 11 68 3
D or F 9 23 5

Is the observed association between these variables statistically significant? To

find out, proceed as follows.
(a) Find the chi-square statistic. Which cells contribute most to this statistic?
(b) What are the degrees of freedom? Use Table E (in the back of the book) to say
how significant the chi-square test is.
(c) Write a brief conclusion for your study.
10.20 Trying to quit, III In Exercise 10.15 (page 484), you saw an apparent
association between alcohol consumption and smoking. In Exercise 10.16 (page
485) you wrote hypotheses and calculated expected counts. Here again is the two-
way table that summarizes the data from the observational study:

Alcohol consumption Smoked Did not smoke

Yes 29 13
No 48 96

(a) Calculate a x2 statistic and determine the P-value.

(b) Write your conclusion from this study in plain language.
10.21 Python eggs, III Exercise 10.17 (page 485) presented data on the hatching
of python eggs in water at three different temperatures. Here again are the data
from that experiment. (Note: This is not a two-way table!)

Eggs Hatched
Cold 27 16
Neutral 56 38
Hot 104 75

Does temperature have a significant effect on hatching? Write a clear summary of

your work and your conclusion.
10.22 Smoking by students and their parents How are the smoking habits of
students related to their parents’ smoking? Here is a two-way table from a survey
of students in eight Arizona high schools:7
490 CHAPTER 10 Inference in Practice

Student Student
smokes does not smoke
Both parents smoke 400 1380
One parent smokes 416 1823
Neither parent smokes 188 1168

(a) Write the null and alternative hypotheses for the question of interest.
(b) Find the expected cell counts. Write a sentence that explains in simple
language what “expected counts” are.
(c) Find the chi-square statistic, its degrees of freedom, and the P-value.
(d) What is your conclusion about significance?
10.23 Cocaine addiction is hard to break, I Cocaine addicts need the drug to
feel pleasure. Perhaps giving them a medication that fights depression will help
them stay off cocaine. A three-year study compared an antidepressant called
desipramine with lithium (a standard treatment for cocaine addiction) and a
placebo. The subjects were 72 chronic users of cocaine who wanted to break
their drug habit. Twenty-four of the subjects were randomly assigned to each
treatment. Here are the counts and percents of the subjects who succeeded in
staying off cocaine during the study:

Group Treatment Subjects Successes Percent

1 Desipramine 24 14 58.3
2 Lithium 24 6 25.0
3 Placebo 24 4 16.7

(a) Compare the effectiveness of the three treatments. Use percents and draw a
bar graph.
(b) Construct a two-way table that shows the relationship between treatment and
success in staying off cocaine.
10.24 Cocaine addiction is hard to break, II Refer to the previous exercise. Is
the relationship between treatment and success statistically significant? Carry out
a significance test. What do you conclude?

ws
Ne
ws
Ne

APPLICATION 10.1 B]
RO
g
<
Se
a

Do dogs resemble their owners?

In the chapter-opening story (page 469), we dis- Resembles Doesn’t
cussed the details of an experiment to test whether owner resemble owner
dogs and their owners resemble one another. Here Purebred dogs 16 9
is a two-way table that summarizes the results: Nonpurebred dogs 7 13
 10. 1 Chi-Square Tests: Goodness of Fit and Two-Way Tables 491

QUESTIONS 4. Compute the chi-square statistic. Show your work.

1. Compute appropriate percents and draw a well- 5. Determine the P-value using Table E (in the
labeled bar graph to compare the results. What do back of the book). Be sure to state the degrees of
the data suggest? freedom.
2. State an appropriate pair of hypotheses for testing 6. What conclusion do you draw?
the research question.
7. Follow the steps in the Calculator Corner (page
3. Calculate the expected counts. Are we safe 487) to perform a chi-square test. The results should
performing a chi-square test? confirm your answers to Questions 3, 4, and 5.

Section 10.1 Summary

Categorical variables group individuals into classes. The chi-square test for good-
ness of fit tests the null hypothesis that a single categorical variable has a specified
distribution. The expected count for any category is obtained by multiplying the
proportion of the distribution in each category times the sample size.
To display the relationship between two categorical variables, make a two-way
table of counts in the classes. We describe the nature of an association between cat-
egorical variables by comparing selected percents. The chi-square test for two-way
tables tells us whether an observed association in a two-way table is statistically
significant. The expected count in any cell of a two-way table when H0 is true is

row total 3 column total

expected count 5
table total

To compare the observed and expected counts, calculate the chi-square statistic:

1 observed count 2 expected count 2 2

x2 5 a
expected count

The sampling distribution of the x2 statistic is not Normal. It is a new, right-

skewed distribution called a chi-square distribution. There is a whole family of
chi-square distributions having different degrees of freedom (df). For a goodness
of fit test, df 5 the number of categories minus 1. For a two-way table with r rows
and c columns, df 5 (r 2 1)(c 2 1).
The P-value for a chi-square test is the area under the density curve to the right
of x2. Large values of x2 are evidence against H0.

Section 10.1 10.25 Finding P-values Use Table E (in the back of the book) to find the
Exercises P-value for a chi-square test in each of the following situations.
(a) Goodness of fit test for a categorical variable with 10 possible values; x2 5 19.62
(b) Two-way table with 3 rows and 4 columns; x2 5 7.04
(c) Two-way table with 2 rows and 2 columns; x2 5 1.41
 492 CHAPTER 10 Inference in Practice

10.26 Munching Froot Loops Kellogg’s Froot Loops cereal comes in six fruit
flavors: orange, lemon, cherry, raspberry, blueberry, and lime. Charise poured out
her morning bowl of cereal and methodically counted the number of cereal pieces
of each flavor. Here are her data:

Flavor: Orange Lemon Cherry Raspberry Blueberry Lime

Count: 28 21 16 25 14 16

Test the null hypothesis that the population of Froot Loops produced by Kellogg’s
contains an equal proportion of each flavor.
10.27 Majors for men and women in business, I A study of the career plans
of young women and men sent questionnaires to all 722 members of the senior
class in the College of Business Administration at the University of Illinois. One
question asked which major within the business program the student had chosen.
Here are the data from the students who responded:8
Female Male
Accounting 68 56
Administration 91 40
Economics 5 6
Finance 61 59

Describe the differences between the distributions of majors for women and men
with percents, with a graph, and in words.
10.28 Majors for men and women in business, II Exercise 10.27 gives the responses
to questionnaires sent to all 722 members of a business school graduating class.
(a) Two of the observed cell counts are small. Do these data satisfy our guidelines
for safe use of the chi-square test?
(b) Is there a statistically significant relationship between the gender and major of
business students? Carry out a test and state your conclusion.
(c) What percent of the students did not respond to the questionnaire? How does
this nonresponse weaken the conclusions that can be drawn from these data?
10.29 Birds in the trees Researchers studied the behavior of birds that were
searching for seeds and insects in an Oregon forest. In this forest, 54% of the
trees were Douglas firs, 40% were ponderosa pines, and 6% were other types of
trees. The researchers made 156 observations of red-breasted nuthatches: 70 were
seen in Douglas firs, 79 in ponderosa pines, and 7 in other types of trees.9 Do
these data suggest that nuthatches prefer particular types of trees when they’re
searching for seeds and insects? Carry out a chi-square goodness of fit test to help
Alain Turgeon/FeaturePics
answer this question.
10.30 Preventing domestic violence A study conducted in Charlotte, North
Carolina, tested the effectiveness of three police responses to spousal abuse:
(1) advise and possibly separate the couple, (2) issue a citation to the offender,
and (3) arrest the offender. Police officers were trained to recognize eligible cases.
When presented with an eligible case, a police officer called the dispatcher, who
would randomly select one of the three available treatments to be administered.
There were 650 cases in the study. Each case was classified according to whether
the abuser was subsequently arrested within 6 months of the original incident.10
 10. 2 Inference about a Population Mean 493

No subsequent arrest Subsequent arrest

Advise and separate 187 25
Citation 181 43
Arrest 175 39

(a) Why was randomization important in this study? Be specific.

(b) Is there sufficient evidence from this study to conclude that the treatment
imposed on an abuser is related to the likelihood of the abuser’s subsequent
arrest? Give appropriate evidence to support your position.
10.31 Is this coin fair? A statistics student suspected that his penny was not a
fair coin, so he held it upright on a tabletop with a finger of one hand and spun
the penny repeatedly by flicking it with the index finger of the other hand. In 200
spins of the coin, it landed with tails side up 122 times.
(a) Perform a goodness of fit test to see if there is sufficient evidence to conclude
that spinning the coin does not produce an equal proportion of heads and tails.
(b) Use a one-proportion inference procedure to determine whether spinning the
coin is equally likely to result in heads or tails.
(c) Compare your results for parts (a) and (b).
10.32 Totals aren’t enough Here are the row and column totals for a two-way
table with two rows and two columns:
a b 50
c d 50
60 40 100
Find two different sets of counts a, b, c, and d for the body of the table that give
these same totals. This shows that the relationship between two variables cannot be
obtained from the two individual distributions of the variables.

1 0.2 Inference about a Population Mean

ACTIVITY 10.2
Exploring sampling distributions
Professor David Lane of Rice University has
www 2. There are choices for the population distribution:
developed a wonderful applet that probably
Normal, uniform, skewed, and custom. The default is
ranks number 1 among high school and college
Normal. Click the “Animated” button. What happens?
students studying statistics. It provides a dynamic
Click the button several more times. What do the
way for students to understand a fundamental
black boxes represent? What is the blue square that
concept in statistics: sampling distributions. And it’s
drops down onto the plot below? What does the red
just plain fun to play with. Let’s try it.
horizontal band under the population histogram tell us?
1. Go to the Web site www.onlinestatbook.com/ Notice the left panel. Important numbers are
stat_sim/sampling_dist/index.html/. When the displayed there. Did you notice that the colors of
BEGIN button appears on the left side of the screen, the numbers match up with the objects to the right?
click on it. You will then see a yellow page entitled As you make things happen, the numbers change
“Sampling Distributions.” accordingly, like an automatic scorekeeper.
ACTIVITY 10.2 (continued)
494 CHAPTER 10 Inference in Practice

ACTIVITY 10.2 (continued)

6. Clear the page, and select “Skewed” population.

Repeat the preceding steps for the skewed
distribution as you increase the sample size.

3. Click on “Clear lower 3” to start clean. Then

click on the “5” button under “Sample:” repeatedly
until you have simulated 100 repetitions (look for
“Reps 5 100” on the left panel in black letters).
7. Clear the page, and select “Custom” distribution.
Does the sampling distribution (blue bars) have
Click on a point on the population histogram to
a recognizable shape? Click the box next to “Fit
insert a bar of that height. Or click on a point on the
normal.” Compare the center (mean) of the
horizontal axis, and drag up to define a bar. Make
population distribution with the center of the
a distribution as strange as you can. (Note: You can
sampling distribution.
shorten a bar or get rid of it completely by clicking
4. Click “Clear lower 3.” Click on 1000 samples. on the top of the bar and dragging down to the axis.)
Click four more times until you have 5000 samples. Then repeat the preceding steps for your custom
What is the range of the data? How well does the distribution as you increase the sample size. Cool,
Normal curve fit your histogram now? huh?
5. Click “Clear lower 3.” Select sample size n 5 16. 8. What did you discover about the shape and
Repeat Steps 3 and 4, but with the bigger sample spread of the sampling distribution? Does
size. Then “Clear lower 3” and do the same for something special happen as you gradually increase
sample size n 5 25. Complete the statements: The the sample size? Fill in the blank to describe this
center of the sampling distribution is ________ [how phenomenon: As the sample size increases, the
related to] the center of the population distribution. sampling distribution of the sample mean gets closer
As the sample size increases, the spread of the and closer to ________.
sampling distribution ________. There is a name for this important result: the
There is a very explicit relationship between central limit theorem. We will learn more about the
the sample size and the spread of the sampling central limit theorem shortly.
distribution of the sample mean. You will learn what
that relationship is later in this section.
 10. 2 Inference about a Population Mean 495

Although the reasoning of confidence intervals and significance tests is always the
same, specific methods vary greatly. The form of an inference procedure depends
first on the parameter you want information about—a population proportion or
mean or median or whatever. The second influence is the design of the sample or
experiment. Estimating a population proportion from a stratified sample requires a
different method than if the data come from an SRS. This section presents a confi-
dence interval and a significance test for inference about a population mean when
the data are an SRS from the population.
The sampling distribution of a sample mean
What is the mean number of hours your school’s seniors study each week? What is
their mean grade point average? We often want to estimate the mean of a popula-
tion. To distinguish the population mean (a parameter) from the sample mean x,
we write the population mean as m (the Greek letter mu). We use the mean x of an
SRS to estimate the unknown mean m of the population.
Like the sample proportion p^ , the sample mean from a large SRS has a sam-
pling distribution that is close to Normal. The sampling distribution of x has m
as its mean. The standard deviation of x depends on the standard deviation of the
population, which is usually written as s (the Greek letter sigma). Using math-
ematics we can discover the following facts.

Sampling distribution of a sample mean

Choose an SRS of size n from a population in which individuals have

mean m and standard deviation s. Let x be the mean of the sample.
Then:

• The mean of the sampling distribution of x is equal to m. In symbols,

m x 5 m.
• The standard deviation of the sampling distribution is sy !n. In
symbols, sx 5 sy !n.

It isn’t surprising that the values that x takes in many samples are centered at the
true mean m of the population. That’s the lack of bias in random sampling once again.
The other fact about the sampling distribution makes precise a very important
property of the sample mean: the mean of several observations is less variable than
individual observations. Figure 10.3 (on the next page) illustrates this property. It com-
pares the distribution of a single observation with the sampling distribution of the
mean x of 10 observations. Both have the same center, but the distribution of the mean
of 10 observations is less spread out.
In Figure 10.3, the distribution of individual observations is Normal. If that
is true, then the sampling distribution of x is exactly Normal for any size sample,
not just approximately Normal for large samples. What if the population distribu-
tion is not Normal? As Activity 10.2 illustrates, if we take large enough samples,
the sampling distribution of the sample mean will be approximately Normal. Th is
remarkable statistical fact is known as the central limit theorem. The central limit
theorem lies behind the use of Normal sampling distributions for sample means.
496 CHAPTER 10 Inference in Practice

x– for 10 observations Standard deviation σ / 10

Standard deviation σ
1 observation

Figure 10.3 The sampling distribution of the sample mean x of 10 observations

compared with the distribution of a single observation.

Example 10.10 The central limit theorem in action

Figure 10.4 shows the central limit theorem in action. The top-left density curve
describes individual observations from a population. It is strongly right-skewed.
Distributions like this describe the time it takes to repair a household appliance, for
example. Most repairs are quickly done, but some are lengthy.

0 1 0 1
n=1 n=2

0 1 0 1
n = 10 n = 25

Figure 10.4 The distribution of sample means becomes more Normal as the size of the
sample increases. The distribution of individual observations (n 5 1) is far from Normal.
The distributions of means for samples of size n 5 2, n 5 10, and n 5 25 observations
move closer to the Normal shape.
 10. 2 Inference about a Population Mean 497

The other three density curves in Figure 10.4 show the sampling distributions of
the sample mean for samples of size n 5 2, n 5 10, and n 5 25 observations from
this population. As the sample size n increases, the shape becomes more Normal.
The mean remains fixed and the standard deviation decreases, following the pattern
sy !n. The distribution for n 5 10 observations is still somewhat skewed to the right
but is beginning to resemble a Normal curve. The density curve for n 5 25 is yet more
Normal. The contrast between the shapes of the population distribution and of the
sampling distribution of the mean for n 5 10 or n 5 25 observations is striking.

Exercises 10.33 The idea of a sampling distribution Figure 9.2 (page 423) shows the idea
of the sampling distribution of a sample proportion in picture form. Draw a similar
picture that shows the idea of the sampling distribution of a sample mean x.
10.34 Averages versus individuals Scores on the American College Testing
(ACT) college admissions examination vary Normally with mean m 5 18 and
standard deviation s 5 6. The range of reported scores is 1 to 36.
(a) What range contains the middle 95% of all individual scores? Explain.
(b) If the ACT scores of 25 randomly selected students are averaged, what range
contains the middle 95% of the averages x? Justify your answer.
10.35 A sampling distribution, I We used our calculator’s rand function to
generate samples of 100 random numbers between 0 and 1 repeatedly. Here are
the sample means for 50 samples of size 100:

0.532 0.450 0.481 0.508 0.510 0.530 0.499 0.461 0.543 0.490
0.497 0.552 0.473 0.425 0.449 0.507 0.472 0.438 0.527 0.536
0.492 0.484 0.498 0.536 0.492 0.483 0.529 0.490 0.548 0.439
0.473 0.516 0.534 0.540 0.525 0.540 0.464 0.507 0.483 0.436
0.497 0.493 0.458 0.527 0.458 0.510 0.498 0.480 0.479 0.499

The sampling distribution of x is the distribution of the means from all possible
samples. We actually have the means from 50 samples.
(a) Make a histogram of these 50 observations. Does the distribution appear to be
roughly Normal, as the central limit theorem says will happen for large enough samples?
(b) Sketch a graph of the density curve for the population distribution. What is
the value of the population mean?
10.36 A sampling distribution, II Exercise 10.35 presents 50 sample means x
from 50 random samples of size 100. Find the mean and standard deviation of
these 50 values. Then answer these questions.
(a) The mean of the population from which the 50 samples were drawn is m 5 0.5 if
the random number generator is working properly. What do you expect the mean
of the distribution of x’s from all possible samples to be? Is the mean of these 50
samples close to this value?
(b) The standard deviation of the distribution of x from samples of size n 5 100
is supposed to be sy !100 5 sy10, where s is the standard deviation of
498 CHAPTER 10 Inference in Practice

individuals in the population. Use this fact with the standard deviation you
calculated for the 50 x’s to estimate s. (In fact, s 5 0.289.)
10.37 Student attitudes The Survey of Study Habits and Attitudes (SSHA) is
a psychological test that measures students’ study habits and attitude toward
school. Scores range from 0 to 200. The mean score for U.S. college students is
about 115, and the standard deviation is about 30. A teacher suspects that older
students have better attitudes toward school. She gives the SSHA to 25 students
who are at least 30 years old. Assume that scores in the population of older
students are Normally distributed with standard deviation s 5 6. The teacher
wants to test the hypotheses

H0 : m 5 115
Ha : m . 115
(a) What is the sampling distribution of the mean score x of a sample of 25
older students if the null hypothesis is true? Sketch the density curve of this
distribution. (Hint: Sketch a Normal curve first, then mark the axis using what
you know about locating m and s on a Normal curve.)
(b) Suppose that the sample data give x 5 118.6. Mark this point on the axis of
your sketch. In fact, the outcome was x 5 125.7. Mark this point on your sketch.
Using your sketch, explain in simple language why one outcome is good evidence
that the mean score of all older students is greater than 115 and why the other
outcome is not.
(c) Shade the area under the curve that is the P-value for the sample result
x 5 118.6.
AP
PLET 10.38 The CLT applet Go to the textbook Web site, www.whfreeman.com/sta2e,
and click on “Statistical Applets.” Launch the Central Limit Theorem applet.
You should see a screen like the one shown in the margin. Click and drag the
slider to change the sample size, and watch how the idealized curve for the
sampling distribution changes with it. Write a few sentences describing what is
happening.

Confidence intervals for a population mean

If we want to estimate an unknown population mean m, we start by taking a simple
random sample from the population of interest and then calculate the sample mean
x. We use x as our estimate for m because the values of x in repeated samples center
on m. That is, the sampling distribution of x has mean m. The standard deviation
of the sampling distribution is sy !n. If the population distribution is Normal or
the sample size n is sufficiently large, the sampling distribution will have a Normal
shape.
Now we can find confidence intervals for m following the same reasoning that
led us to confidence intervals for a proportion p in Section 9.1. The big idea is that,
to cover the central area C under a Normal curve, we must go out a distance z* on
either side of the mean. Figure 10.5 reminds you how C and z* are related.
Our confidence interval has the same basic form as before:
estimate 6 margin of error
 10. 2 Inference about a Population Mean 499

Shaded area represents

probability C
z* standard deviations

Figure 10.5 Critical

values z* of the Normal
distributions. In any Nor-
mal distribution, there is
area (probability) C under 3 2 1 0 1 2 3
the curve between 2z* –z* z*
and z* standard deviations Standard deviations above and below the mean
away from the mean.

However, the details are a little different now that we are working with means in-
stead of proportions. The correct interval is
s
x 1 z*
!n
There’s just one problem. The standard deviation of x depends on both the sam-
ple size n and the standard deviation s of individuals in the population. In practice,
we know n but not s. When n is large, the sample standard deviation s is close to
s and can be used to estimate it, just as we use the sample mean x to estimate the
population mean m. The estimated standard deviation of x is therefore sy !n. For
large sample sizes (say n $ 50), an approximate level C confidence interval for an
unknown population mean m is therefore
s
x 6 z*
!n

Confidence interval for a population mean

Choose an SRS of size n from a large population of individuals having

mean m. The mean of the sample observations is x. When n is large, an
approximate level C confidence interval for m is
s
x 6 z*
!n
where z* is the critical value for confidence level C from Table D (in
the back of the book).

The cautions we noted in estimating p apply here as well. The formula is valid
only when an SRS is drawn and the sample size n is reasonably large. The margin of
500 CHAPTER 10 Inference in Practice

error again decreases only at a rate proportional to !n as the sample size n increases.
One additional caution: remember that x and s are strongly influenced by outliers.
Inference using x and s is suspect when outliers are present. Always look at your data.

Example 10.11 NAEP quantitative scores

The National Assessment of Educational Progress (NAEP) includes a short test of
quantitative skills, covering mainly basic arithmetic and the ability to apply it to
realistic problems. Scores on the test range from 0 to 500. For example, a person
who scores 233 can add the amounts of two checks appearing on a bank deposit
slip; someone scoring 325 can determine the price of a meal from a menu; a person
scoring 375 can transform a price in cents per ounce into dollars per pound.
In a recent year, 840 men 21 to 25 years of age were in the NAEP sample. Their
mean quantitative score was x  272, and the standard deviation of their scores
was s 5 59. These 840 men are a simple random sample from the population of all
young men. On the basis of this sample, what can we say about the mean score in
the population of all 9.5 million young men of these ages?
The 95% confidence interval for m uses the critical value z* 5 1.96 from Table D
(in the back of the book). The interval is
s 59
x 6 z* 5 272 6 1.96
!n !840
5 272 6 1 1.96 2 1 2.036 2 5 272 6 4.0

We are 95% confident that the mean score for all young men lies between 268
and 276.11

Tests for a population mean

As with confidence intervals, the reasoning that leads to significance tests for hy-
potheses about a population mean m follows the reasoning that led to tests about
a population proportion p. The big idea is to use the sampling distribution that
the sample mean x would have if the null hypothesis were true. Locate the x from
your data on this distribution and see if it is unlikely. A value of x that would rarely
appear if H0 were true is evidence that H0 is not true. The steps in performing a
significance test about m are the same as those in tests for a proportion.

Example 10.12 Executives’ blood pressures

The National Center for Health Statistics reports that the mean systolic blood pres-
sure for males 35 to 44 years of age is 128. The medical director of a large company
looks at the medical records of a sample of 72 executives in this age group and finds
that the mean systolic blood pressure is x 5 126.1 and that the standard deviation is
s 5 15.2. Is this evidence that the company’s executives have a different mean blood
pressure from the general population?
The hypotheses. The null hypothesis is “no difference” from the national mean.
The alternative is two-sided, because the medical director did not have a particular
 10. 2 Inference about a Population Mean 501

direction in mind before examining the data. So the hypotheses about the unknown
mean m of the executive population are
H0 : m 5 128
Ha : m 2 128
The sampling distribution. If the null hypothesis is true, the sample mean x has
approximately the Normal distribution with mean m and standard deviation
s 15.2
5 5 1.79
!n !72
The data. The sample mean is x 5 126.1. The standardized score for this out-
come is
Catching cheaters observation 2 mean
standardized score 5
Lots of students take a long standard deviation
multiple-choice exam. Can 126.12128
the computer that scores z5 5 21.06
the exam also identify 1.79
papers that are suspiciously
similar? Clever people have
We know that an outcome a little more than 1 standard deviation away from the
created a measure that mean of a Normal distribution is not very surprising. The last step is to make this
takes into account not just formal.
identical answers but the The P-value. Figure 10.6 locates the sample outcome 21.06 (in the standard
popularity of those answers scale) on the Normal curve that represents the sampling distribution if H0 is true.
and the total score on the The two-sided P-value is the probability of an outcome at least this far out in either
similar papers. The mea-
sure has close to a Normal direction. This is the shaded area under the curve. According to Table A, the area to
distribution, and the com-
puter flags pairs of papers
with a measure outside
4 standard deviations as
significant.

Figure 10.6 The 1.06 0 1.06

P-value for a two-sided _
Values of x (standard scale)
test when the standardized
score for the sample mean
is 21.06.
502 CHAPTER 10 Inference in Practice

the left of z 5 21.06 is 0.1446. The area to the left of 21.06 and to the right of 1.06
is double this, or 0.2892. This is our P-value.
Conclusion. The large P-value gives us no reason to think that the mean blood
pressure of the company’s executive population differs from the national average.

The test in Example 10.12 assumes that the 72 executives in the sample are an
SRS from the population of all middle-aged male executives in the company. We
should check this assumption by asking how the data were produced. If medical
records are available only for executives with recent medical problems, for example,
the data are of little value for our purpose. It turns out that all executives are given
a free annual medical exam, and that the medical director selected 72 exam results
at random.
The data in Example 10.12 do not establish that the mean blood pressure m for
this company’s executives is 128. We sought evidence that m differed from 128 and
failed to find convincing evidence. That is all we can say. No doubt the mean blood
pressure of the entire executive population of this company is not exactly equal to
128. A large enough sample would give evidence of the difference, even if it were very
small.

Exercises 10.39 Confidence level and margin of error The NAEP test (Example 10.11,
page 500) was also given to a sample of 1077 women aged 21 to 25 years. Their
mean quantitative score was 275, and the standard deviation was 58.
(a) Construct and interpret a 95% confidence interval for the mean score m in the
population of all young women.
(b) Calculate the 90% and 99% confidence intervals for m.
(c) What are the margins of error for 90%, 95%, and 99% confidence? How
does increasing the confidence level affect the margin of error of a confidence
interval?
10.40 Executives’ blood pressure Example 10.12 (page 500) found that the
mean blood pressure of a sample of executives did not differ significantly at the
10% level from the national mean, which is 128. Use the data in that example
to give a 90% confidence interval for the mean of the company’s executive
population. Why do you expect 128 to be inside this interval?
Exercises 10.41 to 10.43 refer to the following setting. A bank wonders whether
omitting the annual credit card fee for customers who charge at least $2400 in
a year will increase the amount charged on its credit cards. The bank makes
this offer to an SRS of 200 of its credit card customers. It then compares how
much these customers charge this year with the amount that they charged
last year. The mean increase in the sample is $332, and the standard deviation
is $108.
10.41 Charge more, I Is there significant evidence at the 1% level that the mean
amount charged increases under the no-fee offer? State H0 and Ha and carry out
a test.
 10. 2 Inference about a Population Mean 503

10.42 Charge more, II Calculate and interpret a 99% confidence interval for the
mean amount charges would have increased if this benefit had been extended to
all such customers.
10.43 Charge more, III In Exercises 10.41 and 10.42, you carried out the
calculations for a test and confidence interval based on a bank’s experiment in
changing the rules for its credit cards. You ought to ask some questions about this
study.
(a) The distribution of the amount charged is skewed to the right, but outliers
are prevented by the credit limit that the bank enforces on each card. Why can
we use a test and confidence interval based on a Normal sampling distribution
for the sample mean x?
(b) The bank’s experiment was not comparative. The increase in amount
charged over last year may be explained by lurking variables rather than by the
rule change. What are some plausible reasons why charges might go up? Outline
the design of a comparative randomized experiment to answer the bank’s
question.
10.44 Normal body temperature In Application 1.2 (page 25), we described
an observational study that investigated whether 98.6°F is really
“normal” body temperature. Allen Shoemaker, from Calvin College,
produced a data set with many of the same characteristics as the
original temperature readings. His data set consists of one oral
temperature reading for each of 130 individuals. The mean and
standard deviation of the 130 temperature readings are x 5 98.25
and s 5 0.73.
(a) Perform a significance test to determine whether these data
provide convincing evidence that “normal” body temperature is
not 98.6°F. Show each step of your test clearly.
iSt
oc

ph
k

oto
(b) Calculate a 95% confidence interval for the mean
body temperature m in the population of healthy adults.
What additional information does the confidence interval provide?

Inference in practice: the t distribution

What would cause the head brewer of the famous Guinness brewery in Dublin,
Ireland, not only to use statistics but also to invent new statistical methods? The
search for a better beer, of course.
William S. Gosset (1876–1937) joined Guinness as a brewer in 1899. He soon
became involved in experiments and in statistics to understand the data from those
experiments. What are the best varieties of barley and hops for brewing? How
should they be grown, dried, and stored? The results of the field experiments, as
you can guess, varied. Gosset faced the problem we noted earlier: he didn’t know
the population standard deviation s. He also observed that when the sample size
was small, replacing s by s and using a Normal distribution to estimate or test a
claim about a population mean m wasn’t accurate enough.
After much work, Gosset developed what we now call the t distributions.
Unlike the standard Normal distribution, there is a different t distribution for
504 CHAPTER 10 Inference in Practice

each sample size n. We specify a particular t distribution by giving its degrees

of freedom (df). When we perform inference about m using a t distribution, the
appropriate degrees of freedom is df 5 n  1.
Figure 10.7 compares the density curves of the standard Normal distribution
and the t distributions with 2 and 9 degrees of freedom.

t, 2 degrees of freedom
t, 9 degrees of freedom
standard Normal

Figure 10.7 Density curves for the t distributions with 2 and 9 degrees of freedom and
for the standard Normal distribution. All are symmetric with center 0. The t distributions
have more probability in the tails than does the standard Normal.

The figure illustrates these facts about the t distributions:

• The density curves of the t distributions are similar in shape to the standard
Normal curve. They are symmetric about zero, single-peaked, and bell-
shaped.
• The spread of the t distributions is a bit greater than that of the standard
Normal distribution. The t distributions in Figure 10.7 have more area
(probability) in the tails and less in the center than does the standard
Normal.
• As the degrees of freedom increase, the t density curves approach the
standard Normal distribution. This happens because s estimates s more
accurately as the sample size increases. So using s in place of s causes
little extra variation when the sample is large. (That explains why it was
acceptable to use z instead of t to perform inference about m in our earlier
examples—the sample sizes were all very large.)
Table C in the back of the book gives critical values t* for the t distributions.
Each row in the table contains critical values for one of the t distributions; the
degrees of freedom appear at the left of the row. For confidence intervals, several
of the more common confidence levels C (in percent) are given at the bottom of
the table. By looking down any column, you can check that the t critical values ap-
proach the Normal values as the degrees of freedom increase.
 10. 2 Inference about a Population Mean 505

Example 10.13 Finding t* using Table C

Upper-tail probability p Suppose you want to construct a 95% confidence interval for the mean m of a
df .05 .025 .02 .01 Normal population based on an SRS of size n 5 12. What critical value t* should
10 1.812 2.228 2.359 2.764 you use?
11 1.796 2.201 2.328 2.718 In Table C, an excerpt of which is shown, we consult the row corresponding
12 1.782 2.179 2.303 2.681 to df 5 n  1 5 11. We move across that row to the entry that is directly above
z* 1.645 1.960 2.054 2.326 95% confidence level at the bottom of the chart. The desired critical value is
90% 95% 96% 98% t* 5 2.201. Notice that the corresponding critical value from the z distribution
Confidence level C is z* 5 1.96.

Confidence intervals in practice: the t interval

Earlier, we used the formula
s
x 6 z*
!n
to construct a confidence interval for m when the sample size n is large. What if n is
small? In that case, we have to use critical values from the t distribution with n  1
degrees of freedom in place of the z critical values. Our formula becomes
s
x 6 t*
!n
In fact, even for large sample sizes, this t interval “works better” than the z interval
we showed you earlier. That is, 95% confidence intervals constructed using t criti-
cal values will tend to capture m closer to 95% of the time than intervals based on z
critical values. Our advice: use t for inference about means and z for inference about
proportions.

The t interval

Draw an SRS of size n from a population having unknown mean m. A

level C confidence interval for m is
s
x 6 t*
!n
where t* is the critical value for the t distribution with n 2 1 degrees
of freedom. This interval is exactly correct when the population distri-
bution is Normal. It is approximately correct as long as no outliers or
strong skewness are present.

The t interval is similar in both reasoning and computational detail to the z in-
terval we used earlier. So we will now pay more attention to questions about using
these methods in practice.
506 CHAPTER 10 Inference in Practice

Example 10.14 Measuring stream health

The level of dissolved oxygen in a river is an important indicator of the water’s ability
to support aquatic life. A researcher collects water samples at 15 randomly chosen
locations along a stream and measures the dissolved oxygen. Here are the results in
milligrams per liter:

4.53 5.04 3.29 5.23 4.13 5.50 4.83 4.40 5.42 6.38 4.01 4.66 2.87 5.73 5.55

We will construct and interpret a 95% confidence interval for the mean dissolved
oxygen level in this stream.
What is m in this setting? It’s the mean dissolved oxygen level in the stream. In
order to use a t interval to estimate m, we need to know that the population distri-
bution is Normal or that there are no outliers or strong skewness in the data.
Figure 10.8 shows a dotplot and boxplot of the dissolved oxygen (DO) mea-
surements. Both graphs are slightly skewed to the left. Since there are no outliers or
strong skewness present, we should be safe calculating a t interval.

2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5
DO (mg/l) DO (mg/l)
(a) (b)
Figure 10.8 (a) A dotplot and (b) a boxplot of the dissolved oxygen level at 15 randomly
selected locations along a stream.

For these data, x 5 4.77 mg/l and s 5 0.94 mg/l. To find the critical value t* for
a 95% confidence interval, we look at the df 5 15 2 1 5 14 row in Table C. We find
that t* 5 2.145. So our desired confidence interval is
s 0.94
x 6 t* 5 4.77 6 2.145
!n !15
Upper-tail probability p
df .05 .025 .02
14 1.761 2.145 2.264
15 1.753 2.131 2.249 5 4.77 6 0.52
16 1.746 2.120 2.235
90% 95% 96% We are 95% confident that the mean DO level in the stream is between 4.25 and
Confidence level C 5.29 mg/l.

Why did the researcher in Example 10.14 collect water samples at randomly
chosen locations along the stream? Because random sampling allows us to gen-
eralize from the sample results to the population of interest. If the researcher had
measured the DO level at 15 locations downstream from a farm, for example, the
 10. 2 Inference about a Population Mean 507

sample mean DO level x might not be a very good estimate for the true mean
DO level m in the stream. Except in the case of very small samples, the assumption
that the data are an SRS from the population of interest is more important than the
assumption that the population distribution is Normal.

Significance tests in practice: the t test

In tests as in confidence intervals, we allow for unknown s by using the standard
error and replacing z by t. Now we can do a realistic analysis of data produced to
test a claim about an unknown population mean.

Example 10.15 Is caffeine dependence real?

An experiment was conducted with 11 people diagnosed as being dependent on
caffeine. Each subject was barred from coffee, colas, and other substances contain-
ing caffeine. Instead, they took capsules containing their normal caffeine intake.
During a different time period, they took placebo capsules. The order in which
subjects took caffeine and the placebo was randomized. Table 10.2 contains data
on one of several tests given to the subjects.12 “Depression” is the score on the Beck
Depression Inventory. Higher scores show more symptoms of depression. Do these
data provide convincing evidence that depriving caffeine-dependent people of caf-
feine causes them to become more depressed?

Table 10.2 Results of a Caffeine-Deprivation Study

Depression score Depression score Difference

Subject (caffeine) (placebo) (placebo  caffeine)

1 5 16 11
2 5 23 18
3 4 5 1
4 3 7 4
5 8 14 6
6 5 24 19
7 0 6 6
8 0 3 3
9 2 15 13
10 11 12 1
11 1 0 1

The hypotheses. The population of interest is all people who are dependent
on caffeine. Our null hypothesis is that caffeine deprivation doesn’t affect de-
pression for these individuals. The alternative is one-sided, because researchers
believed that depriving people of caffeine would increase their level of depression.
508 CHAPTER 10 Inference in Practice

Our parameter m is the mean difference (placebo  caffeine) in depression scores

that would be reported if all individuals in the population took both the caffeine
capsule and the placebo. So the hypotheses about the unknown mean m are
H0 : m 5 0
Ha : m . 0
The sampling distribution. Since s is unknown and the sample size is small
(n 5 11), we have to use the t distribution with 11  1 5 10 degrees of free-
dom in place of the standard Normal curve to carry out the test. Is the popula-
tion distribution of differences Normal? We don’t know this, so we examine the
sample data. Figure 10.9 shows a dotplot and boxplot of the differences (placebo 
caffeine) in depression scores for our 11 subjects. There are no obvious outliers
or strong skewness in the sample data. We should be safe using t procedures.

2 0 2 4 6 8 10 12 14 16 18 20 2 0 2 4 6 8 10 12 14 16 18 20
Difference Difference
Figure 10.9 Dotplot and boxplot of the differences in Beck Depression Inventory
scores for the subjects in a caffeine-dependence experiment.

The data. For our 11 subjects, x 5 7.364 and s 5 6.918. The corresponding
standard error is

s 6.918
5 5 2.086
!n !11

If the null hypothesis is true, m 5 0 and our standardized score is

observation 2 mean
standardized score 5
standard deviation
7.364 2 0
t5 5 3.53
2.086

The P-value. Figure 10.10 locates the standardized value t 5 3.53 on the t
distribution with 10 degrees of freedom. The P-value for our one-sided test is the
shaded area to the right of 3.53. Table C says that this area is between 0.005 and
0.0025. (Technology gives a P-value of 0.0027.)
 10. 2 Inference about a Population Mean 509

t distribution with Area is between

Figure 10.10 The 10 degrees of 0.005 and 0.0025
P-value for a one-sided freedom
test when the standardized
score for the sample mean
is t 5 3.53. t  3.53
Upper-tail probability p
df .005 .0025 .001
9 3.250 3.690 4.297
Conclusion. A P-value this low (between 0.005 and 0.0025) gives quite strong
10 3.169 3.581 4.144 evidence against the null hypothesis. We reject H0 and conclude that withholding
11 3.106 3.497 4.025 caffeine from caffeine-dependent individuals may lead to depression.

Many experiments involve individuals who are not chosen at random from the
population of interest, like the caffeine-dependent subjects in Example 10.15. In
such cases, we may not be able to generalize our findings to the population of inter-
est. By randomly assigning the treatments, however, the researchers could say that
the large mean difference in depression scores was due to the different treatments.
Did you notice that the data in Example 10.15 came from a matched pairs ex-
periment? Comparative studies like this one are quite common. When we are deal-
ing with paired data, such as the two depression scores for each subject in this
experiment, we can use a t interval or t test on the observed differences. More
advanced methods known as two-sample t procedures can be used to perform infer-
ence about the means of two groups or populations.

Exercises 10.45 Finding critical values What critical value t* from Table C should be used for
a confidence interval for the population mean m in each of the following situations?
(a) A 99% confidence interval based on n 5 12 observations
(b) A 90% confidence interval from an SRS of 30 observations
(c) A 95% confidence interval from a sample of size 18
10.46 Blood pressure A randomized comparative experiment studied the effect
of diet on blood pressure. Researchers divided 54 healthy white males at random
into two groups. One group received a calcium supplement; the other, a placebo.
At the beginning of the study, the researchers measured many variables on the
subjects. The paper reporting the study gives x 5 114.9 and s 5 9.3 for the seated
systolic blood pressure of the 27 members of the placebo group.
(a) Calculate and interpret a 95% confidence interval for the mean blood pressure
of the population from which the subjects were recruited.
(b) The formula you used in part (a) requires an important assumption about the
27 men who provided the data. What is this assumption?
10.47 Caffeine and depression Refer to Example 10.15 (page 507). The results of
our significance test suggest that caffeine deprivation leads to increased depression
for caffeine-dependent people. Calculate a 95% confidence interval for the population
mean m. What additional information does the confidence interval provide?
510 CHAPTER 10 Inference in Practice

10.48 Healthy streams Refer to Example 10.14 (page 506). Do the data provide
strong evidence that the stream has a mean dissolved oxygen content of less than
5 mg per liter?
(a) Use the confidence interval from Example 10.14 to answer this question.
(b) Carry out a significance test to confirm your answer to (a).
10.49 Sleepless nights An experiment was carried out with 10 patients to investigate
the effectiveness of a drug that was designed to increase sleep time. The data below
show the number of additional hours of sleep gained by each subject after taking the
drug.13 (A negative value indicates that the subject got less sleep after taking the drug.)
1.9 0.8 1.1 0.1 20.1 4.4 5.5 1.6 4.6 3.4
Is there convincing evidence that the drug is effective? Carry out a significance
test, and state your conclusion clearly.
10.50 Darwin’s plants Charles Darwin, author of The Origin of Species (1859),
designed an experiment to compare the effects of cross-fertilization and self-
fertilization on the size of plants. He planted pairs of very similar seedling plants, one
self-fertilized and one cross-fertilized, in the same pot at the same time. After a period
of time, Darwin measured the heights (in inches) of all the plants. Here are his data:14
Pair Cross Self Pair Cross Self
1 23.5 17.4 9 18.3 16.5
2 12.0 20.4 10 21.6 18.0
3 21.0 20.0 11 23.3 16.3
4 22.0 20.0 12 21.0 18.0
5 19.1 18.4 13 22.1 12.8
6 21.5 18.6 14 23.0 15.5
7 22.1 18.6 15 12.0 18.0
8 20.4 15.3

Perform a complete analysis of Darwin’s data that includes both a significance test and a
confidence interval. Follow the four-step statistical problem-solving process (page 19).

CALCULATOR CORNER Inference for means

You can perform the calculations for confidence intervals and sig-
nificance tests on the TI-84/TI-Nspire. Here is a brief summary of
the process using the data from the caffeine study of Example 10.15
(page 507).
1. Type the 11 difference values (placebo 2 caffeine) into a list.
TI-84
Put the data in list L1.
TI-Nspire
Open a Lists & Spreadsheet page. Enter the data in a list called “diff.”
 10. 2 Inference about a Population Mean 511

2. Plot your data. Inspect the shape, center, and spread and look for possible outliers.
TI-84
Define Plot 1 to be a boxplot of the values in list L1. Use ZoomStat to display
the graph.
TI-Nspire
Press b, then choose 3:Data, 5:Quick Graph. You should see a dotplot
of the data. To change it to a boxplot, press b, then choose 1:Plot Type,
2:Box Plot.

3. Proceed to inference. There are no outliers or strong skewness, so we can use t

procedures for inference. For a significance test, our hypotheses are
H0 : m 5 0
Ha : m . 0
TI-84
Press A, choose TESTS, and then choose 2:T-Test. Adjust your settings
as shown and choose Calculate.

TI-Nspire
You can perform significance tests in a Calculator page or a Lists & Spreadsheet
page. We’ll give instructions for a Calculator page.
• Press b, and choose 5:Statistics, 7:Stat Tests . . . , and 2:t
Test. Choose “Data” as the input method.
• Enter the values shown in the dialog box.

CALCULATOR CORNER (continued)

512 CHAPTER 10 Inference in Practice

CALCULATOR CORNER (continued)

To construct a confidence interval:

TI-84
Press A, choose TESTS, and then choose 8:TInterval. Enter 0.95 for
the confidence level (“C-Level”) and choose Calculate.
TI-Nspire
Press b, and choose 5:Statistics, 6:Confidence Intervals. . . ,
and 2:t Interval. Choose “Data” as the input method. In the dialog box,
select “diff ” as the List and enter C Level 0.95. Then choose OK.

As you prepare to conclude your study of statistics, here’s a final Data Explo-
ration that allows you to put your understanding of inference to the test (pardon
the pun).

DATA EXPLORATION
I’m Getting a Headache!
The makers of Aspro brand aspirin want to be sure When the production process is working properly,
that their tablets contain the right amount of active Aspro tablets have an average of m 5 320 milligrams
ingredient (acetylsalicylic acid). So they inspect a (mg) of active ingredient. Here are the amounts (in
sample of 36 tablets from a batch of production. mg) of active ingredient in the 36 selected tablets:

319 328 321 324 322 320 324 321 320 324 322 317
321 320 322 318 326 316 316 326 325 320 316 319
319 321 322 319 326 320 324 320 318 321 322 318

What do these data tell us about the mean what you have learned in this chapter to prepare a
acetylsalicylic acid content of the tablets in this one-page response to these two questions. Be sure
batch? Should the company distribute these tablets to include appropriate graphical and numerical
to drugstores or dispose of the entire batch? Use evidence to support your answers.
 10. 2 Inference about a Population Mean 513

ws
Ne
ws
Ne

APPLICATION 10.2 B]
RO
g
<
Se
a

Nitrogen in tires—a lot of hot air?

In Chapter 1 (page 21), we described an III. Analyze data. The amount of pressure lost
www
experiment performed by Consumers Union (in pounds per square inch) during the year for the
to test whether filling car tires with nitrogen instead air-filled and nitrogen-filled tires of each brand is
of compressed air would reduce pressure loss. Let’s shown in the table below.
put together what you have learned about inference 2. The sampling distribution. What inference
with the statistical problem-solving process. procedure is appropriate in this setting? Make
I. Ask a question of interest. Does filling auto- a graph to check that conditions for using that
mobile tires with nitrogen instead of compressed procedure are satisfied.
air reduce pressure loss? 3. The data. Calculate the test statistic assuming
1. The Hypotheses. State an appropriate pair of that H 0 is true.
hypotheses for a significance test. 4. The P-value. State the number of degrees of
II. Produce data. Consumers Union designed freedom. Then sketch and shade an appropriate
a study to test whether nitrogen-filled tires would density curve, and find the P-value.
maintain pressure better than air-filled tires. They IV. Interpret results.
obtained two tires from each of several brands and
5. Conclusion. What conclusion do you draw
then filled one tire in each pair with air and one
from your significance test? Calculate and interpret
with nitrogen. All tires were inflated to a pressure of
a confidence interval. What additional information
30 pounds per square inch and then placed outside
do you get from the confidence interval?
for a year. At the end of the year, Consumers Union
measured the pressure in each tire.
Brand Air Nitrogen Brand Air Nitrogen

BF Goodrich Traction T/A HR 7.6 7.2 Pirelli P6 Four Seasons 4.4 4.2
Bridgestone HP50 (Sears) 3.8 2.5 Sumitomo HTR H4 1.4 2.1
Bridgestone Potenza EL400 2.1 1.0 Yokohama Avid H4S 4.3 3.0
Bridgestone Potenza G009 3.7 1.6 BF Goodrich Traction T/A V 5.5 3.4
Bridgestone Potenza RE950 4.7 1.5 Bridgestone Potenza RE950 4.1 2.8
Continental Premier Contact H 4.9 3.1 Continental ContiExtreme
Cooper Lifeliner Touring SLE 5.2 3.5 Contact 5.0 3.4
Dayton Daytona HR 3.4 3.2 Continental ContiProContact 4.8 3.3
Falken Ziex ZE-512 4.1 3.3 Cooper Lifeliner Touring SLE 3.2 2.5
Fuzion Hrl 2.7 2.2 General Exclaim UHP 6.8 2.7
General Exclaim 3.1 3.4 Hankook Ventus V4 H105 3.1 1.4
Goodyear Assurance Tripletred 3.8 3.2 Michelin Energy MXV4 Plus 2.5 1.5
Hankook Optimo H418 3.0 0.9 Michelin Pilot Exalto A/S 6.6 2.2
Kumho Solus KH16 6.2 3.4 Michelin Pilot HX MXM4 2.2 2.0
Michelin Energy MXV4 Plus 2.0 1.8 Pirelli P6 Four Seasons 2.5 2.7
Michelin Pilot XGT H4 1.1 0.7 Sumitomo HTR 4.4 3.7

This data set is available at the book’s Web site,

www.whfreeman.com/sta2e.
514 CHAPTER 10 Inference in Practice

Section 10.2 Summary

We estimate a population mean m using the sample mean x of an SRS from the pop-
ulation. Confidence intervals and significance tests for m are based on the sampling
distribution of x. This distribution has mean m and standard deviation sy !n.
When the sample size n is large, the central limit theorem says that the sampling
distribution of x will be approximately Normal.
To perform inference about a population mean m, we can use z critical values
if the sample size is large. If the sample size is small, or if we want more accurate
results, we use a t distribution with n 2 1 degrees of freedom for confidence inter-
vals and tests. Our results will be quite accurate when the population is Normally
distributed and will be accurate enough when no outliers or strong skewness are
present.

Section 10.2 10.51 Explaining confidence A student reads that a 95% confidence interval
Exercises for the mean NAEP quantitative score for men aged 21 to 25 is 267.8 to 276.2.
Asked to explain the meaning of this interval, the student says, “95% of all young
men have scores between 267.8 and 276.2.” Is the student right? Justify your
answer.
10.52 What’s the average height? One hundred and fifty students attend the
lecture of a college statistics course on the first day of class. The course instruc-
tor wants to estimate the average height of the students in the room and asks the
nine students in the front row to write down their heights. The average height of
these nine students turns out to be 64 inches with a standard deviation of
3 inches.
(a) What is the parameter in this problem? What is the sample statistic?
(b) Would it be appropriate to use the Normal distribution and the formula
x 6 z* sy !n to make a confidence statement about the average height of the
students in the class, based on the data given in the problem? If yes, construct the
appropriate confidence interval. If no, explain what is wrong.
10.53 Airline passengers get heavier In response to the increasing weight of
airline passengers, the Federal Aviation Administration (FAA) told airlines to
assume that passengers average 190 pounds in the summer, including clothes
and carry-on baggage. But passengers vary, and the FAA did not specify a stan-
dard deviation. A reasonable standard deviation is s 5 35 pounds. Weights are
not Normally distributed, especially when the population includes both men
and women, but they are not very non-Normal. A commuter plane carries 30
passengers.
(a) Explain why you cannot calculate the probability that a randomly selected pas-
senger on this flight weighs more than 200 pounds.
(b) You can calculate the probability that the mean weight of the passengers on
the flight exceeds 200 pounds. Explain why. Then do it.
10.54 One tail or two? What null and alternative hypotheses should you test in
each of the following situations?
(a) Experiments on learning in animals sometimes measure how long it takes
mice to find their way through a maze. The mean time is 18 seconds for one
 10. 2 Inference about a Population Mean 515

particular maze. A researcher thinks that a loud noise will

cause the mice to complete the maze faster. She measures
how long each of several mice takes with a noise as
stimulus.
(b) Last year, your company’s service technicians took
an average of 2.6 hours to respond to trouble calls from
business customers who had purchased service contracts.
Do this year’s data show a significantly different average
response time?
10.55 Pleasant smells, I Do pleasant odors help
work go faster? Twenty-one subjects worked a
paper-and-pencil maze wearing a mask that was either
unscented or carried the smell of flowers. Each subject
worked the maze three times with each mask, in ran-
dom order. (This is a matched pairs design.) Here are
the differences in their average times (in seconds), un-
scented minus scented. If the floral smell speeds work,
the difference will be positive because the time with the
By permission of Dave Coverly and Creators Syndicate, Inc.
scent will be lower.15

27.37 23.14 4.10 24.40 19.47 210.80 20.87

8.70 2.94 217.24 14.30 224.57 16.17 27.84
8.60 210.77 24.97 24.47 11.90 26.26 6.67

(a) We hope to show that work is faster on the average with the scented mask.
State null and alternative hypotheses in terms of the mean difference in times
m for the population of all adults.
(b) Using a calculator, find the mean and standard deviation of the 21 observa-
tions. Did the subjects work faster with the scented mask? Is the mean improve-
ment big enough to be important?
(c) Make a stemplot of the data (round to the nearest whole second). Are there
outliers or other problems that might hinder inference?
(d) Test the hypotheses you stated in (a). Is the improvement statistically significant?
10.56 Pleasant smells, II Return to the data in Exercise 10.55. Calculate a 95%
confidence interval for the mean improvement in time to solve a maze when
wearing a mask with a floral scent. Are you confident that the scent does improve
mean working time?
10.57 Study more! A student group claims that first-year students at a univer-
sity study 2.5 hours per night during the school week. A skeptic suspects that they
study less than that on the average. A survey of a random sample of 269 first-year
students finds that x 5 137 minutes and s 5 65 minutes. What conclusion do you
draw? Give appropriate evidence to support your conclusion.
10.58 IQ at BCU The admissions director at Big City University has a new idea. He
wants to use the IQ scores of current students as a marketing tool. The director gives
an IQ test to an SRS of 50 of the university’s 5000 freshmen. The mean IQ score for
the sample is x 5 112 and the standard deviation is s 5 15. What can the director
say about the mean score m of the population of all 5000 freshmen? Show your work.
516 CHAPTER 10 Inference in Practice

CHAPTER 10 REVIEW
This chapter discusses a few more specific inference procedures. Section 10.1 intro-
duces the chi-square distributions and two types of chi-square tests. The goodness
of fit test is used to determine whether a specified distribution for a single cat-
egorical variable is correct. Relationships between two categorical variables can be
summarized in two-way tables. The chi-square test for two-way tables tests the null
hypothesis that there is no association between two categorical variables.
Section 10.2 deals with inference for a population mean m. Confidence inter-
vals and significance tests rely on the sampling distribution of x, which has a Nor-
mal shape if the population distribution is Normal or the sample size is large. When
no outliers or strong skewness are present, the t distributions are used to calculate
intervals and perform tests.
Here is a review list of the most important skills you should have developed
from your study of this chapter.

A. CHI-SQUARE GOODNESS OF FIT

1. Explain what null hypothesis the chi-square goodness of fit statistic tests in a
specific setting.
2. Calculate expected cell counts, the chi-square statistic, and its degrees of free-
dom from the data.
3. Use Table 10.1 for chi-square distributions to assess significance. Interpret the
test result in context.

B. TWO-WAY TABLES
1. Use percents and bar graphs to describe the relationship between any two
categorical variables starting from the counts in a two-way table.
2. Explain what null hypothesis the chi-square statistic tests for a specific two-way
table.
3. Calculate expected cell counts, the chi-square statistic, and its degrees of free-
dom from a two-way table.
4. Use Table 10.1 for chi-square distributions to assess significance. Interpret the
test result in the setting of a specific two-way table.

C. SAMPLING DISTRIBUTIONS
1. Use the Normal sampling distribution of a sample mean x to find probabilities
involving x.

D. CONFIDENCE INTERVALS
1. Use the formula x 6 z*sy !n to obtain an approximate confidence interval for
a population mean m when the sample size is large. Use the formula x 6 t*sy !n
to obtain a more accurate confidence interval for a population mean m when there
are no outliers or strong skewness in the data.

E. SIGNIFICANCE TESTS
1. Carry out a one-sided or a two-sided test about a population mean m. Use the z
distribution if the sample size is large to get an approximately correct P-value. Use
 Chapter 10 Review Exercises 517

the t distribution with n  1 degrees of freedom to obtain a more accurate P-value

if there are no outliers or strong skewness.

CHAPTER 10 REVIEW EXERCISES

10.59 Longer-lasting batteries A company that produces AA batteries makes the
following statistical announcement: “Our AA batteries last 7.5 hours plus or minus
20 minutes, and our confidence in that interval is 95%.”16 They tested a random
sample of 40 batteries using a special device designed to imitate real-world use and
recorded data on battery lifetime.
(a) Determine the sample mean and standard deviation.
(b) A reporter translates the statistical announcement into “plain English” as follows:
“If you buy one of this company’s AA batteries, there is a 95% chance that it will last
between 430 and 470 minutes.” Comment on this interpretation.
(c) Your friend, who has just started studying statistics, claims that, if you select 40 more
AA batteries at random from those manufactured by this company, there is a 95%
probability that the mean lifetime will fall between 430 and 470 minutes. Do you agree?
(d) Give a statistically correct interpretation of the confidence interval that could be
published in a newspaper report.
10.60 Representative sample? For a class project, a statistics student is required
to take an SRS of students from his large high school to take part in a survey. The
student’s sample consists of 54 freshmen, 66 sophomores, 56 juniors, and 30 seniors.
The school roster shows that 29% of the students enrolled at the school are freshmen,
27% are sophomores, 25% are juniors, and 19% are seniors. Use a goodness of fit test
to examine how well this student’s sample reflects the population at his high school.
10.61 Call the paramedics! Vehicle accidents can result in serious injuries to drivers
and passengers. Police, firefighters, and paramedics respond to these emergencies as
quickly as possible. Slow response times can have serious consequences for accident
victims. Several cities have begun to monitor paramedic response times. In one such
city, the mean response time to all accidents involving life-threatening injuries last
year was m 5 6.7 minutes. The city manager shares this information with emergency
personnel and encourages them to “do better” next year. At the end of the follow-
ing year, the city manager selects a simple random sample of 400 calls involving
life-threatening injuries and examines the response times. For this sample, the mean
response time was x 5 6.48 minutes with a standard deviation of s 5 2 minutes. Do
these data provide good evidence that response times have decreased since last year?
Carry out a significance test to find out.
10.62 Tall girls Based on information from the National Center for Health Statistics, the
heights of 10-year-old girls closely follow a Normal distribution with mean 54.5 inches
and standard deviation 2.7 inches.
(a) Suppose we select one 10-year-old girl at random. Find the probability that she is
over 56 inches tall. Show your work.
(b) Suppose we select nine 10-year-old girls at random. Find the probability that the
mean height x of this sample exceeds 56 inches. Show your work.
10.63 Popular kids, I Who were the popular kids at your elementary school? Did
they get good grades or have good looks? Were they good at sports? A study was
performed to examine the factors that determine social status for children in grades 4,
518 CHAPTER 10 Inference in Practice

5, and 6. Researchers administered a questionnaire to a sample of 478 elementary

school students. One of the questions they asked was “What would you most like to
do at school: make good grades, be good at sports, or be popular?” The two-way table
below summarizes the students’ responses.17

Goal
Gender Grades Popular Sports
Female 130 91 30
Male 117 50 60

(a) Calculate appropriate percents for comparing male and female students’ goals.
(b) Make a well-labeled bar graph to compare male and female responses. Write a few
sentences describing the relationship between gender and goals.
10.64 Popular kids, II Refer to the previous exercise. Is there convincing evidence
of an association between gender and goals for elementary school students? Carry out
a test and report your conclusion.

By permission of xkcd.com

10.65 Good news for chocolate lovers? A German study concluded that dark
chocolate might reduce blood pressure.18 The subjects in the study were patients
with untreated mild hypertension. Six subjects ate a three-ounce white-chocolate
bar each day for two weeks, while seven subjects ate a three-ounce dark-chocolate
bar. Dark chocolate contains plant substances called polyphenols, which scientists
think are responsible for the heart-healthy attributes of red wine. Polyphenols also
have been shown to lower blood pressure in animals. Prior to the study, the partici-
pants had average systolic and diastolic blood pressure readings of about 153 over
84. The systolic blood pressure readings for the dark-chocolate group dropped an
average of 5 points to 148. Assume a standard deviation of 6 points for the dark-
chocolate group.
(a) Is there sufficient evidence to conclude that eating a three-ounce dark chocolate
bar every day will lower one’s systolic blood pressure? Begin to answer this question
by writing null and alternative hypotheses for this experiment.
(b) Calculate the test statistic and the P-value. Is there sufficient evidence to reject
your null hypothesis?
(c) Write your conclusion in plain language.
(d) Do you have any concerns about the design of this experiment? If you were repli-
cating this experiment, would you do anything differently? Explain.
 Chapter 10 Review Exercises 519

10.66 Sleepy students? A sample of 28 college students responded to the survey

question “How much sleep did you get last night?” Here are the data (in hours):

9 6 8 6 8 8 6 6.5 6 7 9 4 3 4
5 6 11 6 3 6 6 10 7 8 4.5 9 7 7

(a) Calculate a 95% confidence interval for the population mean m.

(b) What do we need to know about these students in order to interpret your result
in (a)?
10.67 Acupuncture and pregnancy A study was performed to determine if the an-
cient Chinese art of acupuncture could help infertile women become pregnant.19 One
hundred and sixty healthy women undergoing artificial insemination were recruited
for the study. Half of the subjects were randomly assigned to an acupuncture treatment
group, and the other half were assigned to a control group. The subjects in the treat-
ment group received acupuncture 25 minutes before artificial insemination and again
25 minutes afterward. Subjects in the control group were instructed to lie still for
25 minutes after artificial insemination. Results are shown in the two-way table below.
Acupuncture group Control group
Pregnant 34 21
Not pregnant 46 59
Total 80 80

(a) Was this an observational study or an experiment? Justify your answer.

(b) Is there convincing evidence of a relationship between pregnancy status and
whether acupuncture was administered? Carry out a significance test and report your
conclusion.
(c) Why can’t we conclude that acupuncture caused an increase in pregnancy?
10.68 Mercury in tuna As we discussed in Chapter 2 (page 56), some of the tuna that
people eat may contain high levels of mercury. The Food and Drug Administration will
take action (such as removing the product from store shelves) if the mercury concen-
tration in a six-ounce can of tuna is 1.00 parts per million (ppm) or above. Defenders
of Wildlife collected a sample of 164 cans of tuna from stores across the United States
and sent them to a laboratory for testing. The Fathom histogram displays the mercury
concentration in the sampled cans. For these data, x 5 0.285 ppm and s 5 0.3 ppm.

(a) Explain why it would be reasonable to construct a confidence interval based on z

critical values here, in spite of the histogram’s shape.
(b) What is the typical mercury concentration in cans of tuna sold in stores? Construct
and interpret a 95% confidence interval.
This page intentionally left blank
 NOTES AND DATA SOURCES

PREFACE Statistics Education (GAISE) Report: A Pre-K–12 Curricu-

1. From the National Council of Teachers of Mathematics lum Framework, American Statistical Association, 2007.
(NCTM) Standards Web site, http://standards.nctm.org/. 13. Tattoo survey data from The Harris Poll®, No. 15, www.
2. The quotations are from a summary of the committee’s harrisinteractive.com, February 12, 2008.
report that was unanimously endorsed by the Board of 14. We obtained the tire pressure loss data from the Consumer
Directors of the American Statistical Association. The Reports Web site: http://blogs.consumerreports.org/cars/
full report is George Cobb, “Teaching statistics,” in L. A. 2007/10/tires-nitrogen-.html.
Steen (ed.), Heeding the Call for Change: Suggestions for 15. The article describing this study is Nikhil Swaminathan,
Curricular Action, Mathematical Association of America, “Gender jabber: do women talk more than men?” Scientific
1990, pp. 3–43. American, July 6, 2007.
16. Application 1.2 is based on P. A. Mackowiak, S. S. Wasser-
man, and M. M. Levine, “A critical appraisal of 98.6 degrees
CHAPTER 1 F, the upper limit of the normal body temperature, and other
1. “Computer games ‘burn up calories,’” BBC News, http://
legacies of Carl Reinhold August Wunderlich,” Journal of
news.bbc.co.uk, February 19, 2007.
the American Medical Association, 268, No. 12 (September
2. “Traffic safety facts 2006: young drivers,” published by the
23–30, 1992), pp. 1578–1580.
National Highway Traffic Safety Administration on their
17. We found Allen Shoemaker’s data in “What’s normal?
Web site, www.nhtsa.dot.gov.
temperature, gender, and heart rate,” Journal of Statistics Edu-
3. Nanci Hellmich, “Extra weight shaves years off lives,” USA
cation, 4, No. 2 (July 1996), www.amstat.org/publications/jse/.
Today, January 7, 2003.
18. Barbara A. Dennison, Tara A. Erb, and Paul L. Jenkins,
4. Joachim Schüz et al., “Cellular telephone use and cancer risk:
“Among low-income preschool children television viewing
update of a nationwide Danish cohort,” Journal of the Na-
and television in bedroom associated with overweight risk,”
tional Cancer Institute, 98 (2006), pp. 1707–1713.
Pediatrics 109 (2002), pp. 1028–1035.
5. Example 1.2 is suggested by Maxine Pfannkuch and Chris
19. Samantha’s paper towel experiment is described at www.
J. Wild, “Statistical thinking and statistical practice: themes
selah.k12.wa.us/soar/sciproj2001/SamanthaP.html.
gleaned from professional statisticians,” unpublished man-
20. The cereal data came from the Data and Story Library,
uscript, 1998.
http://lib.stat.cmu.edu/DASL/.
6. The study is by J. David Cassidy et al., “Effect of eliminat-
21. T. M. Amabile, “Children’s artistic creativity: detrimental
ing compensation for pain and suffering on the outcome of
effects of competition in a field setting.” Personality and
insurance claims for whiplash injury,” New England Jour-
Social Psychology Bulletin 8 (1982), pp. 573–578.
nal of Medicine, 342 (2000), pp. 1179–1186.
22. Data are from the NBA Web site, www.nba.com.
7. The estimates of the census undercount come from Howard
23. The original paper is T. M. Amabile, “Motivation and cre-
Hogan, “The 1990 post-enumeration survey: operations and
ativity: effects of motivational orientation on creative
results,” Journal of the American Statistical Association, 88
writers,” Journal of Personality and Social Psychology
(1993), pp. 1047–1060.
48, No. 2 (February 1985), pp. 393–399. The data for
8. Read “Is vitamin C really good for colds?” in Consumer
Exercise 1.42 came from Fred Ramsey and Dan Schafer,
Reports, February 1976, pp. 68–70, for a discussion of this
The Statistical Sleuth, 2nd edition, Duxbury, 2002.
conclusion.The article also reviews the Toronto study and
the need for controlled experiments.
9. J. E. Muscat et al., “Handheld cellular telephone use and CHAPTER 2
risk of brain cancer,” Journal of the American Medical 1. Data on income and education from the March 2008 CPS
Association, 284 (2000), pp. 3001–3007. supplement, downloaded from the Census Bureau Web site
10. S. G. Stolberg, “Link found between behavioral problems using the government’s FERRET system. These are raw sur-
and time in child care,” New York Times, April 19, 2001. The vey data, which include many people with no income or nega-
study is the National Institute of Child Health and Human tive income. To simplify comparisons, we report only four
Development Study of Early Child Care. education classes.
11. Activity 1.2 is based on a suggested science fair project en- 2. The Harris Poll online survey is described in “Despite un-
titled “Do the eyes have it?” at the Science Buddies Web derstanding risks many U.S. adults still use cell phones while
site, www.sciencebuddies.org. driving,” published June 6, 2006, at www.harrisinteractive.
12. Our steps in the statistical problem-solving process were in- com.
fluenced by Guidelines for Assessment and Instruction in 3. Data from the 2009 Statistical Abstract of the United States.
N-1
N-2 Notes and Data Sources

4. Data provided by the College Board. CHAPTER 3

5. Data provided by Professor Dennis Pearl, Ohio State University, 1. Will Shortz, “A few words about sudoku, which has none,”
Columbus. New York Times, August 28, 2005.
6. The 2008–2009 academic year tuition and fees are from the 2. Information on bone density in the reference populations
College Board’s Annual Survey of Colleges, 2008–2009. was found on the Web at http://courses.washington.edu/
7. Data from the government’s Survey of Earned Doctorates, bonephys/opbmd.html.
in the online supplement to Jeffery Mervis, “Wanted: a better 3. Data from Gary Community School Corporation, courtesy
way to boost numbers of minority PhDs,” Science, 281 of Celeste Foster, Purdue University School of Education.
(1998), pp. 1268–1270. 4. The IQ scores in Figure 3.15 were collected by Darlene
8. Figure 2.11 is based on an episode in the Annenberg/Corpo- Gordon, Purdue University School of Education.
ration for Public Broadcasting telecourse Against All Odds: 5. Ulric Neisser, “Rising scores on intelligence tests,” American
Inside Statistics. Scientist, September–October 1997, online edition at www.
9. Data from the report “Is our tuna family-safe?” prepared by americanscientist.org/.
Defenders of Wildlife, 2006. 6. We found the information on birth weights of Norwegian
10. Percent low-birth-weight data from the 2002 Statistical children on the National Institute of Environmental Health
Abstract of the United States. Sciences Web site: http://eb.niehs.nih.gov/bwt/subcfreq.htm.
11. Thanks to Sanderson Smith from Santa Barbara City Col-
lege for the idea for Activity 2.2A.
12. Data from Guidelines for Assessment and Instruction in CHAPTER 4
Statistics Education: Pre-K–12 Report, published by the 1. From the Electronic Encyclopedia of Statistical Examples
American Statistical Association, 2007. and Exercises (EESEE) story “Is Old Faithful Faithful?”
13. Data from the Higher Education Research Institute’s report 2. Christer G. Wiklund, “Food as a mechanism of density-
“The American freshman: national norms for fall 2007,” dependent regulation of breeding numbers in the merlin
published January 2008. Falco columbarius,” Ecology, 82 (2001), pp. 860–867.
14. Data from Jessica Utts and Robert Heckard, Mind on Statis- 3. Information about the sources used to obtain the data for
tics, 3rd edition, Cengage Learning, 2007. Figure 4.6 can be found under “Documentation” at the Gap-
15. Advanced Placement exam results from AP Central Web minder Web site, www.gapminder.org.
site, http://apcentral.collegeboard.com. 4. From the EESEE story “Is It Tough to Crawl in March?”
16. We got the idea for Activity 2.3 from David Lane’s case 5. Figure 4.9 is based on data from the Web site of Professor
study “Who is buying iMacs?” which we found online at Kenneth French of Dartmouth, mba.tuck.dartmouth.edu/
www.ruf.rice.edu/~lane/case_studies/imac/index.html. pages/faculty/ken.french/data_library.html.
17. Data from the Department of Energy Web site, www.eia. 6. Data from Daren Starnes’s Advanced Placement Statistics
doe.gov/oil_gas/petroleum/data_publications/wrgp/mogas_ class, 1998–1999.
history.html. 7. We got the 1999 fuel economy data from Professor Dennis
18. The information in Exercise 2.53 comes from the report Pearl, The Ohio State University. He downloaded the data
“Cell phones key to teens’ social lives, 47% can text with from the EPA Web site at www.epa.gov.
eyes closed,” published online at www.marketingcharts. 8. The World Almanac and Book of Facts (2009).
com. The results come from a Harris Interactive online 9. From a study of 4511 white, non-Hispanic children reported
survey conducted in July 2008. by Judith Blake, “Number of siblings and educational at-
19. R. J. Newan, “Snow job on the slopes,” US News and World tainment,” Science, 245 (1989), pp. 32–36.
Report, December 17, 1994, pp. 62–65. 10. T. J. Lorenzen, “Determining statistical characteristics
20. “Colleges inflate SATs and graduation rates in popular of a vehicle emissions audit procedure,” Technometrics, 22
guidebooks,” Wall Street Journal, April 5, 1995. (1980), pp. 483–493.
21. Letter by J. L. Hoffman, Science, 255 (1992), p. 665. The 11. Alan S. Banks et al., “Juvenile hallux abducto valgus asso-
original article appeared on December 6, 1991. ciation with metatarsus adductus,” Journal of the American
22. The graphs in Application 2.3 came from the article “SEO Podiatric Medical Association, 84 (1994), pp. 219–224.
lie factor,” posted online by Marios Alexandrou at www. 12. These figures are cited by Dr. William Dement at “The Sleep
allthingssem.com/seo-lie-factor/. Well” Web site, www.standford.edu/~dement/.
23. Science, 189 (1975), p. 373. 13. From the EESEE story “Blood Alcohol Content.”
24. Lafayette (Ind.) Journal and Courier, October 23, 1988. 14. From the EESEE story “What’s Driving Car Sales?”
25. Condé Nast Traveler magazine, June 1992. 15. Laura L. Calderon et al., “Risk factors for obesity in Mexican-
26. Purdue (West Lafayette, Ind.) Exponent, September 21, 1977. American girls: dietary factors, anthropometric factors,
27. Fine Gardening, September/October 1989, p. 76. physical activity, and hours of television viewing,” Journal of
28. New York Times, April 21, 1986. the American Dietetic Association, 96 (1996), pp. 1177–1179.
29. College Board Online, www.collegeboard.com. 16. M. S. Linet et al., “Residential exposure to magnetic fields
30. Data from W. S. Gosset, “The probable error of a mean,” and acute lymphoblastic leukemia in children,” New England
Biometrika, 6 (1908), pp. 1–25. Journal of Medicine, 337 (1997), pp. 1–7.
 Notes and Data Sources N-3

17. Quotation from a Gannett News Service article appearing in 10. New York Times, August 21, 1989.
the Lafayette (Ind.) Journal and Courier, April 23, 1994. 11. The results of the survey were announced in the article
18. We obtained the cricket chirp data for Application 4.2 from “Glance at breakdown of video-game poll results,” at www.
Rex Boggs’s Exploring Data Web site at http://exploringdata. foxnews.com, on May 9, 2006.
cqu.edu.au/datasets.htm. 12. Gregory Flemming and Kimberly Parker, “Race and reluc-
19. See Note 15. tant respondents: possible consequences of non-response for
20. The cricket chirp data come from the GLOBE Chief Scien- pre-election surveys,” Pew Research Center for the People
tist’s Blog at www.globe.gov/fsl/scientistsblog/2007/10/. and the Press, 1997, found at www.people-press.org.
21. The data plotted in Figure 4.32 come from G. A. Sacher and 13. For more detail on nonsampling errors, along with refer-
E. F. Staffelt, “Relation of gestation time to brain weight for ences, see P. E. Converse and M. W. Traugott, “Assessing
placental mammals: implications for the theory of verte- the accuracy of polls and surveys,” Science, 234 (1986),
brate growth,” American Naturalist, 108 (1974), pp. 593– pp. 1094–1098.
613. We found them in Fred L. Ramsey and Daniel 14. For more detail on the limits of memory in surveys, see
W. Schafer, The Statistical Sleuth: A Course in Methods of N. M. Bradburn, L. J. Rips, and S. K. Shevell, “Answering
Data Analysis, Duxbury, 1997, p. 228. autobiographical questions: the impact of memory and in-
22. We found the data on cherry blossoms in the paper “Linear ference on surveys,” Science, 236 (1987), pp. 157–161.
equations and data analysis,” which was posted on the North 15. The nonresponse rate for the CPS comes from “Technical
Carolina School of Science and Mathematics Web site, notes to household survey data published in Employment and
www.ncssm.edu. Earnings,” found on the Bureau of Labor Statistics Web site
at www.bls.gov/CPS/. The General Social Survey reports its
response rate on its Web site, www.norc.org/homepage.htm.
PART B OPENER The claim that “pollsters say response rates have fallen as low
1. We found these year-end sales data at the Recording Indus-
as 20 percent in some recent polls” is made in Don Van Natta,
try Association of America’s Web site, www.riaa.com.
Jr., “Polling’s ‘dirty little secret’: no response,” New York
Times, November 11, 1999.
CHAPTER 5 16. The quotation is typical of Gallup polls, from the Gallup
1. “Acadian ambulance officials, workers flood call-in poll,” Web site, www.gallup.com.
Baton Rouge Advocate, January 22, 1999. 17. P. H. Lewis, “Technology” column, New York Times, May
2. The hand-washing survey described in Exercise 5.4 was 29, 1995.
conducted by Harris Interactive on behalf of the American 18. The responses on welfare are from a New York Times/CBS
Society for Microbiology and the Soap and Detergent News Poll reported in the New York Times, July 5, 1992.
Association. Harris Interactive contacted 1001 U.S. adults Those for Scotland are from “All set for independence?”
via telephone during August 2007. We found the results of Economist, September 12, 1998.
the survey at www.washup.org. 19. The most recent account of the design of the CPS is Design
3. Harris Interactive conducted the observational study de- and Methodology, Current Population Survey Technical
scribed in Exercise 5.4 on behalf of the American Society Paper 66, October 2006. Available in print or online at
for Microbiology and the Soap and Detergent Association. www.census.gov/apsd/techdoc/cps/cps-main.html. The ac-
Harris Interactive observed 6076 adults in public rest- count in Example 5.18 omits many complications, such as
rooms at major U.S. public attractions (a train station, a the need to separately sample “group quarters” like college
baseball stadium, an aquarium, and a farmers’ market) dormitories.
during August 2007. 20. From the online “Supplementary Material” for G. Gaskell
4. “Smoking habits stable; most would like to quit,” Gallup et al., “Worlds apart? The reception of genetically modified
Poll press release by Jeffrey M. Jones, July 18, 2006. Read foods in Europe and the U.S.,” Science, 285 (1999),
online at www.gallup.com/poll/. pp. 384–387.
5. D. Horvitz in his contribution to “Pseudo-opinion polls: 21. From the New York Times, August 18, 1980.
SLOP or useful data?” Chance, 8, No. 2 (1995), pp. 16–25. 22. John Simons, “For risk takers, system is no longer sacred,”
6. From “Working women count,” a report released by the Wall Street Journal, March 11, 1999.
U.S. Department of Labor on October 15, 1997. For more 23. D. Goleman, “Pollsters enlist psychologists in quest for
information, visit their Web site, www.dol.gov. unbiased results,” New York Times, September 7, 1993.
7. Gallup Organization, “Gambling in America–1999: a com- 24. Richard Morin, “It depends on what your definition of ‘do’
parison of adults and teenagers,” Topline and Trends detailed is,” Washington Post, December 21, 1998.
report; read online at www.gallup.com. 25. See www.harrisinteractive.com.
8. Corky Siemaszko, “Fools of the road,” Daily News, November 26. Gallup Poll described on the Gallup Web site, www.gallup.
14, 2007. com.
9. Warren McIsaac and Vivek Goel, “Is access to physician 27. Jay Schreiber and Megan Thee, “Fans concerned about
services in Ontario equitable?” Institute for Clinical Evalu- steroid use and believe it’s widespread, poll shows,” New
ative Sciences in Ontario, October 18, 1993. York Times, March 31, 2008.
N-4 Notes and Data Sources

CHAPTER 6 17. The “three-quarters” estimate is cited by Martin Enserink,

1. Allan H. Schulman and Randi L. Sims, “Learning in an on- “Can the placebo be the cure?” Science, 284 (1999),
line format versus an in-class format: an experimental study,” pp. 238–240. An extended treatment is Anne Harrington
T.H.E. Journal, June 1999, pp. 54–56. (ed.), The Placebo Effect: An Interdisciplinary Exploration,
2. L. L. Miao, “Gastric freezing: an example of the evaluation Harvard University Press, 1997.
of medical therapy by randomized clinical trials,” in J. P. 18. Kristin L. Nichol et al., “Effectiveness of live, attenuated
Bunker, B. A. Barnes, and F. Mosteller (eds.), Costs, Risks intranasal influenza virus vaccine in healthy, working adults,”
and Benefits of Surgery, Oxford University Press, 1977, Journal of the American Medical Association, 282 (1999),
pp. 198–211. pp. 137–144.
3. Details of the Carolina Abecedarian Project, including ref- 19. “Cancer clinical trials: barriers to African American partici-
erences to published work, can be found online at www.fpg. pation,” Closing the Gap, newsletter of the Office of Minor-
unc.edu/~abc/. ity Health, December 1997–January 1998.
4. Linda Rosa et al., “A close look at therapeutic touch,” 20. Michael H. Davidson et al., “Weight control and risk factor
Journal of the American Medical Association, 279 (1998), reduction in obese subjects treated for 2 years with orlistat:
pp. 1005–1010. a randomized controlled trial,” Journal of the American
5. See Note 2. Medical Association, 281 (1999), pp. 235–242.
6. From the Electronic Encyclopedia of Statistical Examples 21. Edward P. Sloan et al., “Diaspirin cross-linked hemoglobin
and Exercises (EESEE) story “Anecdotes of Placebos.” (DCLHb) in the treatment of severe traumatic hemorrhagic
7. Samuel Charache et al., “Effects of hydroxyurea on the fre- shock,” Journal of the American Medical Association, 282
quency of painful crises in sickle cell anemia,” New England (1999), 1857–1864.
Journal of Medicine, 332 (1995), pp. 1317–1322. 22. See Note 6.
8. Marilyn Ellis, “Attending church found factor in longer 23. “Advertising: the cola war,” Newsweek, August 30, 1976,
life,” USA Today, August 9, 1999. p. 67.
9. Dr. Daniel B. Mark, in Associated Press, “Age, not bias, may 24. Mary O. Mundinger et al., “Primary care outcomes in patients
explain differences in treatment,” New York Times, April 26, treated by nurse practitioners or physicians,” Journal of the
1994. Dr. Mark was commenting on Daniel B. Mark et al., American Medical Association, 238 (2000), pp. 59–68.
“Absence of sex bias in the referral of patients for cardiac cath- 25. Carol A. Warfield, “Controlled-release morphine tablets
eterization,” New England Journal of Medicine, 330 (1994), in patients with chronic cancer pain,” Cancer, 82 (1998),
pp. 1101–1106. See the correspondence from D. Douglas pp. 2299–2306.
Miller and Leslee Shaw, “Sex bias in the care of patients with 26. From the EESEE case study “Is Caffeine Dependence Real?”
cardiovascular disease,” New England Journal of Medicine, 27. Thomas B. Freeman et al., “Use of placebo surgery in con-
331 (1994), p. 883, for comments on an opposed study. trolled trials of a cellular-based therapy for Parkinson’s dis-
10. “Family dinner linked to better grades for teens: survey ease,” New England Journal of Medicine, 341 (1999),
finds regular meal time yields additional benefits,” written pp. 988–992. Freeman supports the Parkinson’s disease trial.
by John Mackenzie for ABC News’s World News Tonight, The opposition is represented by Ruth Macklin, “The ethical
September 13, 2005. problems with sham surgery in clinical research,” New Eng-
11. Washington Post, March 9, 1989. land Journal of Medicine, 341 (1999), pp. 992–996.
12. C. S. Fuchs et al., “Alcohol consumption and mortality 28. The difficulties of interpreting guidelines for informed
among women,” New England Journal of Medicine, 332 consent and for the work of institutional review boards in
(1995), pp. 1245–1250. medical research are a main theme of Beverly Woodward,
13. W. E. Paulus et al., “Influence of acupuncture on the pregnancy “Challenges to human subject protections in U.S. medical
rate in patients who undergo assisted reproductive therapy,” research,” Journal of the American Medical Association,
Fertility and Sterility, 77, No. 4 (2002), pp. 721–724. 282 (1999), pp. 1947–1952. The references in this paper
14. The original study on the Mozart Effect is described in F. H. point to other discussions.
Rauscher, G. L. Shaw, and K. N. Ky, “Music and spatial task 29. Report of the Tuskegee Syphilis Study Legacy Committee,
performance,” Nature, 365 (1993), p. 611. For an opposing May 20, 1996. A detailed history is James H. Jones, Bad
perspective, see Kenneth Steele, “The ‘Mozart Effect’: an Blood: The Tuskegee Syphilis Experiment, Free Press, 1993.
example of the scientific method in operation,” Psychology 30. Dr. Hennekens’s words are from an interview in the Annenberg/
Teacher Network, November–December 2001. Corporation for Public Broadcasting video series Against All
15. The fact about rats comes from E. Street and M. B. Carroll, Odds: Inside Statistics.
“Preliminary evaluation of a new food product,” in J. M. 31. Stanley Milgram, “Group pressure and action against a per-
Tanur et al. (eds.), Statistics: A Guide to the Unknown, 3rd son,” Journal of Abnormal and Social Psychology, 69 (1964),
edition, Wadsworth, 1989, pp. 161–169. pp. 137–143.
16. The placebo effect examples are from Sandra Blakeslee, 32. Stanley Milgram, “Liberating effects of group pressure,”
“Placebos prove so powerful even experts are surprised,” Journal of Personality and Social Psychology, 1 (1965),
New York Times, October 13, 1998. pp. 127–134.
 Notes and Data Sources N-5

33. Dr. C. Warren Olanow, chief of neurology, Mt. Sinai School 9. From the EESEE story “What Makes a Pre-teen Popular?”
of Medicine, quoted in Margaret Talbot, “The placebo pre- 10. Data from the Stanford School of Medicine Blood Center
scription,” New York Times Magazine, January 8, 2000, Web site at http://bloodcenter.stanford.edu/about_blood/
pp. 34–39, 44, 58–60. blood_types.html.
34. For extensive background, see Jon Cohen, “AIDS trials eth- 11. We got these data from the Energy Information Administra-
ics questioned,” Science, 276 (1997), pp. 520–523. Search tion on their Web site at http://tonto.eia.doe.gov/dnav/pet/
the archives at www.sciencemag.org for recent episodes in pet_sum_mkt_dcu_SCT_m.htm.
the continuing controversies of Exercises 6.60 and 6.69. 12. Thanks to Michael Legacy for suggesting this exercise.
35. See Paul Meier, “The biggest public health experiment 13. From the EESEE story “Baby Hearing Screening.”
ever,” in Judith M. Tanur et al. (eds.), Statistics: A Guide to
the Unknown, Holden-Day, 1972, pp. 2–13, for a discussion
of the experiment that established the effectiveness of the
CHAPTER 8
1. Andrew Pollack, “In the gaming industry, the house can
Salk polio vaccine, a case that fits this exercise.
have bad luck, too,” New York Times, July 25, 1999, reports
36. Gina Kolata and Kurt Eichenwald, “Business thrives on
that Mirage Resorts issued a quarterly profit warning due in
unproven care leaving science behind,” New York Times,
part to its bad luck at baccarat.
October 3, 1999. Background and details about the first clini-
2. The Apgar score data come from the National Center for
cal trials appear in a National Cancer Institute press release
Health Statistics Monthly Vital Statistics Report, 30, No. 1,
“Questions and answers: high-dose chemotherapy with bone
Supplement, May 6, 1981.
marrow or stem cell transplants for breast cancer,” April 15,
3. S. Newcomb, “Note on the frequency of the use of digits in
1999. That one of the studies reported there involved falsified
natural numbers,” American Journal of Mathematics, 4
data is reported by Denise Grady, “Breast cancer researcher
(1881), p. 39.
admits falsifying data,” New York Times, February 5, 2000.
4. F. Benford, “The law of anomalous numbers,” Proceedings of
37. Exercise 6.76 is based on Christopher Anderson, “Mea-
the American Philosophical Society, 78 (1938), pp. 551–572.
suring what works in health care,” Science, 263 (1994),
5. The idea for Application 8.3 came from David Salsburg, The
pp. 1080–1082.
Lady Tasting Tea, W. H. Freeman and Co., 2001.
38. See Note 18.

CHAPTER 7 CHAPTER 9
1. More historical data can be found in the opening chapters of 1. Janice E. Williams et al., “Anger proneness predicts coronary
F. N. David, Games, Gods and Gambling, Charles Griffin heart disease risk,” Circulation, 101 (2000), pp. 2034–2039.
and Co., 1962. The historical information given here comes 2. The 2007 national school-based Youth Risk Behavioral
from this excellent and entertaining book. Survey was carried out by the Centers for Disease Control
2. For a discussion and amusing examples, see A. E. Watkins, to monitor the prevalence of youth behaviors that most in-
“The law of averages,” Chance, 8, No. 2 (1995), pp. 28–32. fluence health.
3. R. Vallone and A. Tversky, “The hot hand in basketball: on 3. Report of the American Society for Microbiology: “2007
the misperception of random sequences,” Cognitive Psy- hand washing fact sheet.” We found the report online at
chology, 17 (1985), pp. 295–314. www.asm.org.
4. Estimated probabilities from R. D’Agostino, Jr., and R. 4. Data from the Web site www.drugfreeamerica.org.
Wilson, “Asbestos: the hazard, the risk, and public policy,” 5. For the state of the art in confidence intervals for p, see Alan
in K. R. Foster, D. E. Bernstein, and P. W. Huber (eds.), Agresti and Brent Coull, “Approximate is better than ‘exact’
Phantom Risk: Scientific Inference and the Law, MIT Press, for interval estimation of binomial proportions,” American
1994, pp. 183–210. See also the similar conclusions in B. T. Statistician, 52 (1998), pp. 119–126. The authors note that
Mossman et al., “Asbestos: scientific developments and impli- the accuracy of our confidence interval for p can be greatly
cations for public policy,” Science, 247 (1990), pp. 294–301. improved by simply “adding 2 successes and 2 failures.”
5. The quotation is from R. J. Zeckhauser and W. K. Viscusi, That is, p^ 5 (count of successes 1 2) / (n 1 4). Texts on
“Risk within reason,” Science, 248 (1990), pp. 559–564. sample surveys give confidence intervals that take into ac-
6. Ivars Peterson, “Random home runs,” Science News, 159, count the fact that the population has finite size and also give
No. 26 (June 30, 2001). Available at www.sciencenews.org. intervals for sample designs more complex than an SRS.
7. “Euro coin accused of unfair flipping,” New Scientist, January 6. Diane Rado and Darnelle Little, “Schools toying with test
4, 2002. results,” Chicago Tribune, September 29, 2003, available
8. We got the data from a talk entitled “Two-Way Tables: Intro- online at www.chicagotribune.com.
ducing Probability Using Real Data,” presented by Gail Burrill 7. This and similar results of Gallup polls are from the Gallup
at the Mathematics Education into the 21st Century Project, Organization Web site, www.gallup.com.
Czech Republic, September 2003. She cited her source as 8. John Paul McKinney and Kathleen G. McKinney, “Prayer
H. Kranendonk, P. Hopfensperger, and R. Scheaffer, Exploring in the lives of late adolescents,” Journal of Adolescence, 22
Probability, Dale Seymour Publications, 1999. (1999), pp. 279–290.
N-6 Notes and Data Sources

9. Hunting data provided by the Virginia Department of Game 6. R. Shine, T. R. L. Madsen, M. J. Elphick, and P. S. Harlow,
and Inland Fisheries, Richmond, Virginia. “The influence of nest temperatures and maternal brooding
10. Julie Ray, “Few teens clash with friends,” May 3, 2005, on on hatchling phenotypes in water pythons,” Ecology, 78
the Gallup Poll Web site, www.gallup.com. (1997), pp. 1713–1721.
11. Jeffrey Selingo, “What Americans think about higher edu- 7. S. V. Zagona (ed.), Studies and Issues in Smoking Behavior,
cation,” Chronicle of Higher Education, 49, No. 34 (May 2, University of Arizona Press, 1967, pp. 157–180.
2003), pp. A10–A15. 8. Francine D. Blau and Marianne A. Ferber, “Career plans
12. Dorothy Espelage et al., “Factors associated with bullying and expectations of young women and men,” Journal of
behavior in middle school students,” Journal of Early Ado- Human Resources, 26 (1991), pp. 581–607.
lescence, 19, No. 3 (August 1999), pp. 341–362. 9. R. W. Mannan and E. C. Meslow, “Bird populations and veg-
13. J. Stutts et al., “Distractions in everyday driving,” Univer- etation characteristics in managed and old-growth forests,
sity of North Carolina Highway Safety Research Center, northeastern Oregon,” Journal of Wildlife Management, 48
June 2003. (1984), pp. 1219–1238.
14. The newspaper report summarizes findings from “A com- 10. Exercise 10.30 comes from the Electronic Encyclopedia of
parison of sustained-release bupropion and placebo for Statistical Examples and Exercises (EESEE) story “Domes-
smoking cessation,” New England Journal of Medicine, tic Violence.”
337, No. 17 (October 23, 1997), pp. 1195–1202. 11. Francisco L. Rivera-Batiz, “Quantitative literacy and the
15. Jane A. Cauley et al., “Lipid-lowering drug use and breast likelihood of employment among young adults,” Journal of
cancer in older women: a prospective study,” Journal of Human Resources, 27 (1992), pp. 313–328.
Women’s Health, 12, No. 8 (October 2003), pp. 749–756. 12. E. C. Strain et al., “Caffeine dependence syndrome: evi-
16. Linda Lyons, “Teens: sex can wait,” December 14, 2004, dence from case histories and experimental evaluation,”
from the Gallup Organization Web site, www.gallup.com. Journal of the American Medical Association, 272 (1994),
pp. 1604–1607.
13. W. S. Gosset, “The probable error of a mean,” Biometrika, 6
CHAPTER 10 (1908), pp. 1–25. We obtained the sleep data from the Data
1. M. M. Roy and N. J. S. Christenfeld, “Do dogs resemble their and Story Library (DASL) Web site, http://lib.stat.cmu.edu/
owners?” Psychological Science, 15, No. 5, (May 2004), DASL/. They cite as a reference R. A. Fisher, The Design of
pp. 361–363. Experiments, 3rd edition, Oliver and Boyd, 1942, p. 27.
2. There are many computer studies of the accuracy of chi- 14. D. J. Hand et al., A Handbook of Small Data Sets, Chapman
square critical values. For a brief discussion and some & Hall, 2006.
references, see Section 3.2.5 of David S. Moore, “Tests of 15. A. R. Hirsch and L. H. Johnston, “Odors and learning,” Jour-
chi-squared type,” in Ralph B. D’Agostino and Michael A. nal of Neurological and Orthopedic Medicine and Surgery,
Stephens (eds.), Goodness-of-Fit Techniques, Marcel Dekker, 17 (1996), pp. 119–126. We found the data in the EESEE
1986, pp. 63–95. case study “Floral Scents and Learning.”
3. Janice E. Williams et al., “Anger proneness predicts 16. From program 19, “Confidence Intervals,” in the Against
coronary heart disease risk,” Circulation, 101 (2000), All Odds video series.
pp. 2034–2039. 17. Exercise 10.63 is based on the EESEE story “What Makes
4. Richard M. Felder et al., “Who gets it and who doesn’t: a Pre-teens Popular?”
study of student performance in an introductory chemical 18. Dirk Taubert, Reinhard Berkels, Renate Roesen, and Wolfgang
engineering course,” 1992 ASEE Annual Conference Pro- Klaus, “Chocolate and blood pressure in elderly individuals
ceedings, American Society for Engineering Education, with isolated systolic hypertension,” Journal of the American
1992, pp. 1516–1519. Medical Association, 290 (2003), pp. 1029–1030.
5. We found these data at Patrick Meirmans’s Web site, where he 19. W. E. Paulus et al., “Influence of acupuncture on the pregnancy
attributes the study to a paper by H. R. Schiffman in a 1982 rate in patients who undergo assisted reproductive therapy,”
issue of the Journal of Counseling and Psychology. Fertility and Sterility, 77, No. 4 (2002), pp. 721–724.
 SOLUTIONS TO ODD-NUMBERED EXERCISES

Chapter 1 evidence that the color red causes bees to sting more
1.1 (a) Students in a statistics class. (b) Categorical: gen- often. (d) There are ethical concerns associated with
der, homeroom teacher, grade level, calculator num- conducting this experiment.
ber. Quantitative: score on Test 1. 1.25 The y scale does not start at 0, and the values 13, 14,
1.3 Answers will vary; for example, whether a household and 15 occur twice, distorting the difference due to
uses its recycling bin. gender.
1.5 (a) Likely a survey of pizza shops was taken in which 1.27 Variability due to fill type is present since each brand
data showing average daily pizza sales were collected listed actually refers to two tires, one for each fill
and then an estimate of the area involved in that type; variability due to brand is present since many
calculation was made. (b) Assuming 300,000,000 different brands are being compared; variability due
people in the United States, this is about 1 square foot to tire is included since several of the brands are re-
of pizza per person per year (43,560 sq. ft./acre  peated and there are at least two tires of each listed
18 acres  365 days/300,000,000 < 1), which seems brand.
plausible. 1.29 Take a random sample of teens across the area of in-
1.7 (a) Population: all wood being considered for pur- terest. Make sure their responses are anonymous.
chase. Sample: the 5 pieces of wood from each batch. 1.31 (a) Answers will vary. One possibility: “Do a majority
(b) Population: all claims filed with the insurance of Americans eat meat on a daily basis?” (b) Obser-
company in a given month. Sample: the sample of vational study, since no treatments were applied.
claims selected. (c) Individuals: surveyed Americans. Variable mea-
1.9 Population: all apples in the truckload. Sample: three sured: how often the individual eats meat per week;
large buckets of apples selected. Individuals: apples. categorical.
Variable(s): most likely weight, color, number of 1.33 (a) Paper towel brand (four brands were compared);
blemishes. towels of the same brand (four towels of each brand
1.11 (a) Television shows. (b) Viewers (in millions), are tested). (b) I. Question of interest: Which paper
household rating (in points; 1 point 5 1,145,000 towel brand absorbs best? II. Produce data: Four
households), network, rank. (c) 25,000 “Peoplemeter” towels of each brand are placed in a container with
homes. 200 ml of water, and the amount absorbed is squeezed
1.13 An experiment, because a treatment (tasting muffins) out and recorded for each towel. III. Analyze data:
is imposed on the subjects. Scott had the highest average absorbency but also the
1.15 Answers will vary, examples are given: (a) What greatest variability.
make of car is most popular at your school? (b) Does
listening to relaxing music for ten minutes immedi- Bounty
Brand

ately prior to a test improve students’ scores? Scott

1.17 (a) Answers will vary. (b) An observational study Sparkle
using police records.
1.19 An observational study; no treatment was imposed. Western family
1.21 (a) An observational study; two groups were mea- 25 30 35 40 45 50 55 60
sured on the same variable; no treatment was imposed. Absorbency
(b) Population: people in leadership positions. Vari-
ables: fitness level and leadership ability. (c) Volun- IV. Interpret results: Absorbency varies among the
teers are likely to have a higher fitness level. four brands; Scott had the greatest absorbency.
1.23 (a) An experiment, since each student is assigned to 1.35 (a) An experiment, since the children were ran-
wear either red or white clothes. (b) Clothing color domly assigned to the two treatments (judged by
and number of bee stings. (c) Since a chance process experts and prizes awarded or shared at an art par-
was used to determine the students’ clothing color, ty). (b) To avoid judgment based on perceived
the two groups should be pretty balanced in terms of effort due to the assigned group. (c) The two groups
all other variables that might affect the number of bee and the judges. (d) Not necessarily. We do not
stings (size, smell, hair color, etc.). If the average know how the students were assigned to groups. If
number of bee stings is much larger for the students a chance process was not used, it is possible that
wearing red clothes, then the experiment gives strong the two groups differed systematically from each
S-1
S-2 Solutions to Odd-Numbered Exercises

other before the treatments were administered, in

which case the large difference in ratings could be Divorced
due to this initial difference rather than to the treat- 11.5% Never
ments themselves. married
1.37 (a) Population: passengers who take a trip that in- Widowed
22.0%
volves more than one airline. Sample: 12% of the 9.8%
tickets sold. (b) They will probably not result in ex-
actly the correct amount of money for each airline.
With a good sample, however, the sample results
should be close to the truth about the population. Married
1.39 (a) Answers will vary. One possibility: “Does the pill 56.7%
reduce the amount of methane gas produced by
cows?” (b) Randomly select cows on a farm; ran-
domly assign half to receive the pill; the other half
receive nothing. Provide the cows with the same feed
and record the number of burps from each cow dur- 2.3 (a) 8.7%.
ing a fixed time period. Record the amount of meth- (b)
ane gas released in the burps if possible.
1.41 (a) Answers will vary. One possibility: “How much 30%

Percent choosing
25.20%
allowance does a child aged 5–12 receive in a week?” 25%
(b) Observational study, since no treatment was 19.30%
20% 16.60%
applied. (c) Individuals: children aged 5–12 who 15.30%
15%
participated in the study. Variable: child’s allowance. 7.80%
10% 7.10%
(d) Most likely due to rounding.
1.43 I. Question of interest: Are poems about laughter more 5%
creative if written for internal, rather than external, 0%
Arts and
humanities

Business

Education

Engineering and
science

Professional

Social science
reasons? II. Produce data: Students were randomly
divided into an external-reasons group and an internal-
reasons group and were asked to write a poem about
laughter. Their poems were rated by 12 poets using a
creativity scale. III. Analyze data: The creativity scores
were categorized by group and plotted on a dotplot.
IV. Interpret results: Based on the dotplot, the average
ratings were higher for the internal-reasons group. 2.5 (a) No, the data compare four separate quantities, not
four parts of a whole. (b) The percentage of drivers
who use cell phones while driving decreases for the
older generations.
Chapter 2
Cell phone use while driving

2.1 (a) 49,680,000. 90% 86%

79% 76%
(b) 80%
70,000 65,128 70%
60%
60,000
Count, in thousands

50% 48%
50,000 40%
40,000 30%
30,000 20%
25,262
10%
20,000 0%
11,208 13,210
Gen. Y Gen. X Baby Matures
10,000
boomers
0 Age group
Never Married Widowed Divorced
married

(c) Pie chart would be correct since categories repre-

sent distinct parts making up a whole.
 Solutions to Odd-Numbered Exercises S-3

2.7 (a) 2.23 (a) Yes, since the categories represent distinct parts
making up a whole.
2.0% 2.0%
2.0%

0 1 2 3 4 5 6 7 8.0%
Goals

(b) Skewed to the right with a center of 1 goal and a

34.0%
spread of 7. 7 may be an outlier.
2.9 Centered around 19 mpg. Without the outlying values
of 10 and 11, the data are somewhat skewed right.
2.11 (a) 16%, possibly high due to the Mormon church. 52.0%
(b) Roughly symmetric around 13% with a spread of
about 3.5%. (c) Less, since the range is only 4.6%,
whereas it is 10% for the older adults.
2.13 Skewed to the right with a center between 5 and 10
and a spread of about 55.
2.15 In a histogram the class widths should be the same.
In Joe’s histogram the class widths vary. Category
2.17 (a)
Cell phone
20 In person
Internet chat or MSN
15 MySpace, Facebook, blogs
Frequency

Telephone (landline)
Text messaging
10
(b) Majority of the students prefer communicating
5 with their friends in person.
2.25 (a)
0 50
0 8 16 24 32 44
Salary (100,000s) 40
Frequency

(b) Skewed to the right since there are a few very 30

high salaries. (c) $27,610,000.
2.19 (a) 20

10
3 3
0
R L A
3 4 5 6 7 8 9 10 11
Hours of sleep (b) 6% is our best estimate based on the sample per-
cent who are left-handed.
(b) Roughly symmetric with a center of 6 hours and 2.27 $50,233 is the income such that half of the incomes
a spread of 8 hours. No outliers. of U.S. households in 2007 were less and half were
2.21 (a) Skewed to the right with a center between 7 and 8 more than this value.
and a spread of about 5% (b) When making summary 2.29 (a) 7. About half of the students made/received fewer
statements about the data, it is usually more informa- than 7 phone calls, and about half made/received
tive to use percents rather than counts. (c) Possibly more than 7. (b) Q1 5 3.5, Q3 5 11.5, IQR 5 8. The
the poverty and teen pregnancy rates. middle 50% of the data have a range of 8. (c) 35 is an
outlier.
S-4 Solutions to Odd-Numbered Exercises

2.31 (a) Answers may vary slightly: Minimum 5 28, (c)

Q1 5 0, Median 5 2.5, Q3 5 22, Maximum 5 102.
(b) About 25% of the students sent/received more
calls than texts. The difference, texts 2 calls, was
less than 2.5 for half of the students and more than
2.5 for half of the students. About 25% of the students
sent/received 22 or more texts than calls.
2.33 Mean 5 1600. The average daily metabolic rate of
the 7 men taking part in the study of dieting was 1600
calories.
2.35 The mean is higher because there is a relatively small
percent of households that make much higher in-
comes. The mean is affected by these values, but not 0 10 20 30 40 50
the median. Electoral votes
2.37 (a) 2. A left-skewed distribution implies mean , 2.47 (a)
median; a small spread implies a relatively small SD. Calculus Statistics
(b) 3. A symmetric distribution implies that the mean 25%
is close to the median; a large spread implies a rela-

Percentage
tively large SD. (c) 1. A right-skewed distribution 20%
implies mean . median; a large spread implies a
relatively large SD. 15%
2.39 Since the distribution is skewed to the right, the me- 10%
dian and IQR will give better summaries of the center
and spread of the distribution. 5%
2.41 (a) Mean 5 87.188, Median 5 87.5. (b) Standard
deviation 5 3.2, IQR 5 3.25. 0%
5 4 3 2 1
2.43 Distributions are similar except silver was the most
popular for full-size/intermediate cars and not as Grade
popular for SUVs/trucks. White was the most popu- (b) The distributions are different. For AP Calculus
lar option for SUVs/trucks and not as popular for the most common exam grade was a 1, followed by
full-size/intermediate cars. Beige/brown was the 5, 4, 3, and 2. For AP Statistics, the most common
least popular for both car types. exam grade was a 3, followed by 4, 1, 2, and 5. We
Full-size/ cannot tell which exam was easier, since the groups
25% of students taking the exams were not the same.
intermediate cars
2.49 The distribution differs for the two regions. The
SUVs/trucks
Percent favoring

20% median poverty rate for the northern states was ap-
proximately 3% less than for the southern states.
15% Q3 for the North was less than Q1 for the South. The
spread of poverty rates was also less for the north-
10% ern states.
2.51 (a) Both will increase by $1000. (b) The extremes
5% and quartiles will increase by $1000; the standard de-
viation will not change.
0% 2.53 The areas of the phones should be in proportion, not
Silver Blue White Black Red Gray Beige/ just their heights.
Car color Brown
2.55 (a) It appears that Democrats were more than 5 times
2.45 (a) Minimum 5 3, Q1 5 4, Median 5 8, Q3 5 12, as likely to support the court’s decision than Repub-
Maximum 5 55. (b) 27, 31, 34, and 55 are outliers. licans or Independents.
 Solutions to Odd-Numbered Exercises S-5

(b) 2.65 A 14.8% increase. Poverty did not necessarily be-

come more common during these years, because the
60 population also increased during this time.
2.67 (a) In either case, the stemplot shows a symmetric
Percent agree

50
distribution centered at the upper 60s.
40 4 69 4 69
30 5 36678 5 3
6 003344567778 5 6678
20 7 0112347889 6 003344
10 8 01358 6 567778
9 00 7 011234
0 7 7889
Democrats Republicans Independents
8 013
Results by party 8 58
9 00
2.57 We now see that the trend has been much steadier
over time. (b) The shape is symmetric without outliers. Mean 5
60,000 70.39 and SD 5 12.23. The spread is about 50 years.
Total lottery ticket sales

50,000
10
40,000

Frequency
30,000

20,000 5

10,000

0
0
1980 1985 1990 1995 2000 2005 50 60 70 80 90
Year Age at death
2.59 The size of the graphics is not in proportion to the (c) Answers will vary.
percents. The last picture is below the dotted line, 2.69 The incorrect value is for coronary heart disease. It
which gives the incorrect impression of a decrease should be 27.5%, not 2.75%.
for this point. 2.71 No. The cones make the differences appear larger
2.61 (a) The vertical scale does not begin at 0, overempha- than they really are. The area of the pictures should
sizing the difference in cholesterol. be proportional, not their heights.
(b) 2.73 90%
Parent and teen Teen only
80% Parent only Neither
200 70%
60%
Percentage
Cholesterol

150 50%
40%
100
30%
50 20%
10%
0 0%
Week 1 Week 2 Week 3 Week 4 Desktop Cell iPod/ Laptop PDA
phone MP3
The drop in cholesterol is much smaller.
2.63 If every single adult had failed the test, 100% would 2.75 I. Question of interest: Does the drug increase sleep
have failed. It is not possible to have more than 100% time? II. Produce data: An experiment was carried
failing the test. out with 10 patients. The number of additional hours
of sleep gained by each subject after taking the drug
was recorded. III. Analyze data: Minimum 5 20.1,
S-6 Solutions to Odd-Numbered Exercises

Q1 5 0.8, Median 5 1.75, Q3 5 4.4, Maximum 5 5.5, 3.11 (a) The overall shape of the distribution is symmetric
Mean 5 2.33, Standard Deviation 5 2.00. IV. Inter- and mound-shaped, so mean 5 median 5 Point A.
pret results: The data show that the drug was effec- (b) The overall shape of the distribution is skewed to
tive. On average, the participants gained at least some the left, so mean 5 Point A and median 5 Point B.
additional sleep. However, there should have been a 3.13 (a) The distribution is skewed to the right, so median 5
control group taking a placebo to ensure that the Point B, mean 5 Point C. (b) The distribution is
results can be attributed to the treatment. symmetric, so mean 5 median 5 Point A. (c) The
distribution is skewed to the left, so mean 5 Point A,
median 5 Point B.
3.15 (a)
10
10

8
7

Frequency
6
6
5
4 4 4
4
3 3

2
1 1 1 1
0 1 2 3 4 5 6
Additional sleep time, in hours 0
3 4 5 6 7 8 9
Chapter 3 Unemployment rate
3.1 (a) 87% of the girls her daughter’s age weigh the same or (b) Mean 5 5.598%; median 5 5.45%; standard
less than her daughter does and 67% of girls her daugh- deviation 5 1.391%. The distribution of unemploy-
ter’s age are her daughter’s height or shorter. (b) Accord- ment rates is fairly symmetric, with a center of about
ing to the Los Angeles Times, the speed limits on Califor- 5.5% and a spread of 5.6%. (c) 80th percentile. 80%
nia highways are such that 85% of the vehicle speeds on of the unemployment rates are less than or equal to
those stretches of road are no greater than the speed limit. Illinois’s unemployment rate. Illinois has a pretty
3.3 (a) Francine’s bone density is 1.45 standard devia- high unemployment rate compared with the remain-
tions below the mean hip bone density of 956 g/cm2 ing 49 states. (d) Texas; z 5 20.36.
for 25-year-old women. (b) 5.517 g/cm2. 3.17 (a) Mean 5 170 cm; standard deviation 5 7.5 cm.
3.5 (a) 79.3 percentile. Roughly 79% of the team had sala- (b) 188.75 cm.
ries less than or equal to Lidge’s salary. (b) z 5 0.79. 3.19 (a) z 5 0.67. Paul is slightly taller than average for his
Lidge’s salary was 0.79 standard deviations above age. His height is 0.67 standard deviations above the
the mean salary of $3,388,617. average male height for his age. (b) 75% of boys
3.7 (a) The total area under the curve must equal 1. Since Paul’s age are shorter than or equal to Paul in height.
the base of the rectangle is 1, the height must also be 1. 3.21 89 and 133.
(b) 0.5. (c) 0.5. They are the same since the density 3.23 144 has a z-score of 3. We would expect 0.15% to be
curve is symmetric. (d) (0.4)(1) 5 40%. above 144. It is not surprising that none of the stu-
3.9 One possibility: dents in the sample had an IQ score this high, be-
Percent of Shakespeare’s words

cause we would only expect a value this large about

25 1.5 times out of every thousand.
Median  4 3.25 (a) 327 to 345 days. (b) z 5 1 1 16% of horse preg-
20 Mean ⬇ 4.2 nancies last longer than 339 days.
3.27 (a) 0.3557.
15

10

0
1 2 3 4 5 6 7 8 9 101112
Number of letters in word z  0.37
 Solutions to Odd-Numbered Exercises S-7

(b) 0.6443. (b) z 5 0.13.

0.45

z  0.37 z  ??
(c) 0.9842. 3.33 Step 1: State the problem. Let x 5 the distance that
Tiger’s ball travels. x has a Normal distribution with
m 5 304 and s 5 8. Find the probability that x # 317.
Step 2: Standardize and draw a picture. For x 5 317,
z 5 1.63.

z  2.15
(d) 0.0158.

3 2 1 0 1 2 3
Step 3: Use the table. From Table A, the probability
that z # 1.63 is 0.9484. The probability that Tiger is
safe hitting his driver is about 0.95. Step 4: Conclu-
z  2.15 sion. Since the probability that Tiger’s drive misses
the creek is about 0.95, we can conclude that Tiger is
3.29 (a) 0.8587.
safe hitting his driver.
3.35 Step 1: State the problem. Let x 5 a randomly se-
lected female’s score on the SAT Math test. x has a
Normal distribution with m 5 500 and s 5 111. Find
the probability that x . 533. Step 2: Standardize and
draw a picture. For x 5 533, z 5 0.30.

z  1.33 z  1.65
(b) 0.2718.

3 2 1 0 1 2 3

Step 3: Use the table. From Table A, the probability

that z , 0.30 is 0.6179. The proportion of observations
z  0.5 z  1.79
greater than 0.30 is then 0.3821. Step 4: Conclusion.
3.31 (a) z 5 20.84. Approximately 38% of females scored higher than
the male mean.
3.37 (a) Step 1: State the problem. Let x 5 the weight of
0.20 a randomly selected bag of potatoes from the ship-
ment. x has a Normal distribution with m 5 10 and
s 5 0.5. Find the percent of bags weighing less than
10.25 pounds. Step 2: Standardize and draw a pic-
ture. For x 5 10.25, z 5 0.5.
z  ??
S-8 Solutions to Odd-Numbered Exercises

3.41 In a Normal distribution, 68% of the values are within

one standard deviation of the mean. Since the distribu-
tion is symmetric, 34% of these values must be below
the mean. That leaves 16% of values that fall more
than one standard deviation below the mean. Since the
percentile rank of an observation is the percent of
terms below it, the percentile rank of an observation
one standard deviation below the mean is 16. Similarly,
3 2 1 0 1 2 3 95% of the observations are within two standard de-
viations of the mean, leaving 97.5% below a value that
Step 3: Use the table. From Table A, the percent of
is two standard deviations above the mean.
observations less than 0.5 is 0.6915. Step 4: Conclu-
3.43 (a) z 5 20.47.
sion. Approximately 69% of the bags in the shipment
weighed less than 10.25 pounds.
(b) Step 1: State the problem. Let x 5 the weight of a 32%
randomly selected bag of potatoes from the shipment.
x has a Normal distribution with m 5 10 and s 5 0.5.
Find the percent of bags weighing between 9.5 and
10.25 pounds. Step 2: Standardize and draw a picture.
From part (a), P(x , 10.25) 5 P(z , 0.5). For x 5 9.5,
z 5 21. The area we want to find is depicted below. z  ??
(b) z 5 0.25.

0.4

3 2 1 0 1 2 3
z  ??
Step 3: Use the table. From part (a), P(z , 0.5) 5 0.6915.
From Table A, the percent of observations less than 21 is (c) z 5 21.28.
0.1587. Thus, the percent of bags weighing between
9.5 and 10.25 pounds is 0.5328. Step 4: Conclusion.
Approximately 53% of the bags in the shipment
weighed between 9.5 and 10.25 pounds.
3.39 68% of women are between 62.5 and 67.5 inches tall, 0.10
95% are between 60 and 70 inches tall, and 99.7%
are between 57.5 and 72.5 inches tall.
z  ??

(d) z 5 0.95.

99.7% 0.83

95% z  ??

3.45 (a) 2.28%. (b) 25.14%.

68% 3.47 (a) About 68%. $215 and $395 represent one stan-
dard deviation above and below the mean of $305.
57.5 60 62.5 65 67.5 70 72.5 (b) z 5 20.2 1 about 42%.
 Solutions to Odd-Numbered Exercises S-9

3.49 (a) 0.9878.

z  0.73
z  2.25
(c) z 5 0.18 and z 5 0.79. The area between z 5 0.18
(b) 0.9494. and z 5 0.79 is 0.2138.

2.25 1.77 0.18 0.79

(c) 0.9981.
Chapter 4
4.1 (a) Response: grade. Explanatory: time of study.
(b) Relationship. (c) Response: yield. Explanatory:
rain. (d) Relationship.
4.3 (a) There is a moderate, positive, linear relationship
between IQ scores and GPA. (b) About 103 and 0.5.
(c) A, B: moderate IQ but low GPA. C: low IQ but
z  2.89 high GPA.
4.5 (a) The number of males returning each year may
(d) 0.0730.
vary dramatically, so it would be more appropriate to
use percents for comparing various years.
(b)

80
Percent

1.44 2.89
60
3.51 (a) Since z 5 22.29, about 1% of babies will be
identified as low birth weight. (b) Q1: 20.67 5
(x 2 3668) / 511 1 x 5 (20.67 3 511) 1 3668 5
3325.63; median 5 mean 5 3668. Q3: 0.67 5
25 27.5 30 32.5 35 37.5 40
(x 2 3668) / 511 1 x 5 (0.67 3 511) 1 3668 5
Pairs
4010.37.
3.53 (a) Bimodal. Mean 5 median 5 B. (b) Skewed to the (c) There is a definite downward trend, but it is more
right. Median 5 A and mean 5 B. curved than linear. Yes.
3.55 (a) Within 2 standard deviations of the mean, that is, 4.7 (a) States with higher (lower) median household
between 221% and 45%. (b) z 5 20.73. The area is income will also have higher (lower) mean personal
0.2327. The market is down for about 23% of the income. (b) Except for the District of Columbia, the
years. relationship is moderately strong, positive, and linear.
(c) UT: given the mean income per person, the expected
median household income is higher than expected.
D.C. and CT: the median household income is lower
S-10 Solutions to Odd-Numbered Exercises

than expected given the high mean income per 4.25 (a) MA is explanatory.
person. 60
4.9 (a) Stocks; highest, 50%; lowest, 230%. Treasury
bills: highest, 15%; lowest, 1%. (b) The relationship
is weak at best. No evidence that high interest rates
are bad for stocks. 40
4.11 (a)

HAV
50 Represents multiple
45 states 20
% Proficient

40
35 0
0 7 14 21 28 35 42
30
MA
25
(b) Weak, positive, linear relationship; (12, 50) is an
20 outlier. (c) There is a positive association, but it is not
230 235 240 245 250 255 very strong. (d) Not really.
4.27 (a) 219.87. We predict the amount of gas consumed to
Mean NAEP score
decrease by 19.87 cubic feet for every degree increase
(b) Linear, positive, fairly strong. (c) Answers will in the average monthly temperature. (b) 1425. When the
vary. average monthly temperature is 08F, the predicted gas
4.13 (a) Positive but not close to 1. consumption is 1425 cubic feet. (c) 828.9 cubic feet.
4.29 (a) 4.7 laps. (b) No. The number of slices eaten is
72 outside the observed data range.
4.31 y 5 1 2 x. The sum of the squares of the vertical
distances is 3, but it is 18 for the other line.
4.33 Residuals: 21.1, 23.13, 20.11, 21.57, 0.91, 3.85,
0.48, 1.26; sum 5 0.59.
Men

68 4.35 False. No. 70% of the variation in y is explained by

the linear regression model.
4.37 (a) x 5 number of beers consumed. y 5 BAC.
0.20
64

64.5 66 67.5 69 70.5

Women
BAC

(b) r 5 0.57. 0.10

4.15 The strength of the linear relationship between GPA
and first-exam score.
4.17 We would expect a strong association between a
woman’s height and her height as a girl, a moderate
association between a man’s height and his adult 0
son’s height, and a weak association between a man’s 0 2 4 6 8 10
height and the height of his wife.
Beers
4.19 (a) The direction is positive, but the form appears
curved, not linear. The strength is moderate. (b) Hip- (b) r 5 0.894. Yes, since the relationship appears
popotamus, Asian elephant: life spans are longer than very linear. (c) Predicted BAC 5 20.013 1 0.018
expected given their gestation periods. Giraffe: tends to (number of beers). Slope: for every additional beer
follow the curvilinear shape, possibly a little shorter consumed, a student’s BAC increases by 0.018, ac-
life span than expected. (c) The point would be at about cording to our model. Since 0 is outside the range of
(280, 70) and would not be part of the general pattern. the data, it should not be interpreted.
4.21 (a) 21 to 1. (b) Any nonnegative number. 4.39 Temperature (z) is likely to be a lurking variable: in
4.23 (a) Negative; slightly curved; yes, at about (4, 2.9). warmer weather, more people go swimming and thus
(b) No. there are more drowning deaths (y). Ice cream sales
(x) are likely to be higher during this time as well.
 Solutions to Odd-Numbered Exercises S-11

4.53 Time spent standing is a confounding variable.

x y x y (x and z are confounded)

?

?
z z

4.41 For example, motivation level, intelligence, socio-

4.55 (a) Longer crickets should weigh more than shorter
economic status.
crickets. (b) It would not change.
4.43 Stronger students are more likely to choose such
4.57 (a)
math courses; weaker students may avoid them.
4.45 (a) The line shows a general decrease but not a strong 15
relationship. (b) The general pattern is concave up-
ward (bowl-shaped). This pattern strengthens the
conclusion to avoid hospitals that treat few heart at- 10
tacks, but the hospitals that treat the most patients are
not necessarily better than those that treat a moderate y
number of patients.
4.47 Inactive girls are more likely to be obese. 3.2%. 5
4.49 Close to 21.
140
0
0 2 4 6 8 10 12
x
Weight

(b) r 5 0.48. (c) The outlier at (10, 1).

70
4.59 (a) Dolphins: body weight, 190 kg; brain weight,
1600 g. Hippos: body weight, 1400 kg; brain weight,
600 g. (b) Dolphins and hippos are outliers in the
plot. Dolphins have unusually high brain weights for
their body weights. Hippos are heavy but have un-
0 usually low brain weights.
0 5 10 15 20 25 4.61 74% of the variation in brain weight is explained by
Day the least-squares regression of brain weight on body
4.51 (a) 256.1 grams; prediction outside the range of the weight.
available data is risky. (b) 99.6% of the variation in 4.63 (a) Fairly strong, negative, linear association.
the soap weights is explained by the least-squares Predicted number of days 
regression of weight on day. (c) Yes. 33.12–4.69 temperature
30
Days in April to 1st bloom

4 25

20
2
Residual

15
0
10
2 5

4 0
2 3 4 5 6 7
Temperature in Celsius
5 10 15 20
Day
S-12 Solutions to Odd-Numbered Exercises

(b) Predicted number of days 5 33.12 2 4.69 (tem- 5.15 (a) All black adult residents of Miami; SRS of 300
perature). For every 1 degree increase in average adults from predominately black neighborhoods.
March temperature, in degrees Celsius, we predict (b) Out of fear, people may not respond honestly;
the number of days in April until first bloom to de- also a black police officer is asking the questions.
crease by 4.69. The y intercept is outside the range of 5.17 Answers will vary.
data and therefore has no meaningful interpretation. 5.19 (a) Label from 0001 to 3478. (b) 2940, 0769, 1481,
(c) 16.7. (d) 22.015. (e) There is no discernible pat- 2975, 1315.
tern in the residuals. They are clustered about 0 in a 5.21 It’s a convenience sample.
random fashion. (f) r 2 5 0.72. 72% of the variation 5.23 Supporters of an individual’s right to bear arms have
in the number of days in April until the first cherry tremendous political influence. Opinion polls on such
blossom appears is explained by the least-squares re- subjects do not always result in legislative action.
gression of the number of days in April until first 5.25 Yes, since the library surveyed card holders.
bloom on the average temperature in March. 5.27 Population: all smokers who signed a card saying
they intended to quit smoking. Parameter: proportion
6 signing the card who had not smoked in the past six
months. Sample: random sample of 1000 people who
4 signed the card. Statistic: 21%.
5.29 Statistic, parameter.
2 5.31 This is a comparative experiment because the re-
Residual

0 searchers assigned the ducks to the groups being

compared (inside boxes versus outside boxes).
2 5.33 (a) High bias, high variability. (b) Low bias, low
variability. (c) Low bias, high variability. (d) High
4 bias, low variability.
6
5.35 Possible answers: (a) Hunter College survey was an
observational study, not a national survey; intersec-
tions in New York are not likely to be representative
2 3 4 5 6 7 of the nation as a whole. (b) No. The discrepancy is
Temperature in Celsius due to sampling differences, not sample size.
5.37 Answers will vary.
5.39 The margin of error is half as big with the larger sam-
Chapter 5
ple (0.1 instead of 0.2).
5.1 (a) 29,777. (b) Voluntary response sample.
5.41 Larger.
5.3 (a) Convenience sample. (b) It’s unlikely that the
5.43 Smaller sample size means larger margin of error.
first 100 students to arrive are representative of the
5.45 Chicago; in general, margin of error decreases as the
student population in general. Probably higher, since
sample size increases.
students who arrive first are likely to be those who
5.47 Parameter, statistic.
got a good night’s sleep the night before.
5.49 0.004.
5.5 Possible answers: (a) A call-in poll. (b) Interviewing
5.51 (a) 0.69. The proportion of teens aged 13 to 17 na-
students as they enter the student center.
tionwide who have received personal messages on-
5.7 Apartment complexes 16, 32, 18: Fairington, Water-
line from people they don’t know. (b) We are 95%
ford Court, and Fowler.
confident that the proportion of teens aged 13 to 17
5.9 (a) Use labels 001 to 440. (b) 395, 020, 118, 167,
nationwide who have received personal messages
300, 360, 241, 065, 188, 365.
online from people they don’t know is between 66%
5.11 Using line 101: 19, 22, 05 (Petrucelli, Shen, and
and 72%.
Brockman).
5.53 (a) Quick method: 0.02, 0.03. (b) We are 95% confi-
5.13 (a) Assign 1 to heads and 2 to tails. Enter “2” next to
dent that the proportion of American adults who play
“Population 51 to” and click Reset. Enter a sample
computer or video games is between 38% and 42%.
size of 1 and click Sample. Number returned identifies
Of those American adults who play electronic games,
heads or tails. (b) Enter “52” next to “Population 5 1
we are 95% confident that between 7% and 13% play
to” and click Reset. Enter a sample size of 52 and
for 10 or more hours per week. (c) Possible answers:
click Sample. The ordering of the 52 numbers cor-
sampling method (how were the adults selected, and
responds to the ordering of the cards. (c) Assign
are they representative of the population?) and inac-
each adult to a number between 1 and 500. Enter
curate responses. (d) Answers will vary.
“500” next to “Population 5 1 to” and click Reset.
5.55 For example, undercoverage or nonresponse.
Enter a sample size of 12 and click Sample. Num-
bers returned correspond to jurors.
 Solutions to Odd-Numbered Exercises S-13

5.57 (a) 11.3%. (b) 2.6%; yes. (c) No, since a voluntary is the price the student would expect to pay. The sub-
response sample was used. jects are the students in the course. The treatments
5.59 Answers will vary. are the two price histories.
5.61 The margin of error depends on the sample size, not 6.3 A mother’s prior attitude toward her baby, which
the population size. probably influenced her choice of feeding method, is
5.63 (a) The sample was chosen in stages; it is not an SRS the lurking variable; the prior attitude is confounded
of all Europeans. (b) The 17 countries. (c) There is with the effect of the nursing.
random probability sampling in at least one stage of
the sampling process.
5.65 Closed questions allow for easier tabulation of results Bottle-feed Mother’s
but may omit response choices that individuals would vs. breast-feed Causes? attitude toward
prefer to select. Open questions give individuals (explanatory child (response
more freedom to respond, but the tabulation of re- variable) variable)
sults could then be more difficult.
5.67 (a) 69, 169, 269, 369, and 469. (b) 1/100. (c) Not
every sample of five addresses is equally likely to be
chosen.
5.69 (a) 40. (b) Stratified random samples, to allow for Mother’s
separate conclusions about males and females. choice of feeding
5.71 (a) We are told that each region was proportionately method (lurking
represented in the sample. (b) An SRS could result in variable)
heavy representation from some areas and light (or
no) representation from other areas. This could lead
to a biased estimate. 6.5 (a) Therapeutic touch practitioners; hand position
5.73 (a) 0.004. (b) Some people probably said that they (left or right); choice of hand (correct or incorrect).
voted when they actually didn’t. (b) The setting lacked realism; conditions were not
5.75 (a) Sampling error because of undercoverage. (b) similar to those typically faced by therapeutic touch
Nonsampling error due to nonresponse. (c) Sampling practitioners.
error due to convenience sample. 6.7 (a) The subjects were 22,071 male physicians; the ex-
5.77 (a) You are sampling only from the lower-priced ticket planatory variable is medication (aspirin or placebo);
holders. (b) Undercoverage is a sampling error. the response variable is whether or not the subject had
5.79 (a) Population of West Lafayette, Indiana. (b) Larger, a heart attack.
since this voluntary response poll probably overrep- (b)
resents those who are opposed to the one-way street.
5.81 (a) Assign labels 0001 through 3500 and use random Group 1 Treatment 1
Compare
digits. (b) Randomly select 1 of the first 14 students, Random (11,037 doctors) (Aspirin)
number of
then take every 14th student from that point on. allocation
heart attacks
(c) Choose SRSs of size 200 and 50 from each group. Group 2 Treatment 2
(d) Possible answer: choose a stratified sample to en- (11,034 doctors) (Placebo)
sure that attitudes of bussed and local students are
represented fairly. 6.9 (a) Randomly allocate the 200 rooms to two groups
5.83 The first question pictures a pugnacious president of 100 rooms. First group gets the flat rate; second
“fighting” to save his job; the second portrays a more group gets the varied rates. Compare Internet tele-
reasonable alternative: allow a process to occur that phone usage.
will determine the outcome.
5.85 (a) Adults who used the Internet in the week prior to Group 1 Treatment 1
the survey. (b) 0.041. We are 95% confident that the Random 100 rooms Fixed rate Compare
assignment amount and
true proportion of all adults who used the Internet in
time of use
the week prior to the survey and who believe that the Group 2 Treatment 2
Internet has made their life “much better” is in the 100 rooms Variable rate
interval 0.22 to 0.30.
(b) Rooms 119, 033, 199, 192, and 148 are the first
Chapter 6 five selected.
6.1 (a) An experiment because the researcher imposes a 6.11 Possible answer: students registering for a course
treatment: one of two price histories. (b) The explan- should be randomly assigned to a classroom or online
atory variable is whether the student sees a steady version of the course. Scores on a standardized test
price or temporary price cuts. The response variable can then be compared.
S-14 Solutions to Odd-Numbered Exercises

6.13 The difference in blood pressures was so great that it during the experiment knew who received hydroxy-
was unlikely to have occurred by chance (if calcium urea and who received the placebo.
is not effective). 6.29 (a) The placebo effect. (b) Use a three-treatment,
6.15 The effects of factors other than the nonphysical completely randomized design. (c) No. (d) Since the
treatment have been eliminated or accounted for, patient assesses the effectiveness of the treatment,
so the differences in improvement observed between the experiment does not need to be double-blind.
the subjects can be attributed to the differences in 6.31 (a) No. It’s possible that women who see an ultra-
treatments. sound picture are more likely to take better care of
6.17 This is an observational study, not an experiment, so themselves during pregnancy. (b) No. Without a pla-
we are unable to measure the effects of possible lurk- cebo, it’s unknown whether the results are due to the
ing and/or confounding variables. A possible lurking treatment or the idea of the treatment. (c) A blinded
variable is the amount of parental involvement the experiment using a fake ultrasound as the placebo.
teens receive. 6.33 (a) Difference in score (after minus before) on the
6.19 If the economy worsened due to a recession during Beck Depression Inventory. (b) Give the Beck De-
the five-year period, then unemployment could rise pression Inventory to all subjects before and after.
even if the training program was effective. Consumer Randomly assign subjects into three groups of size
spending is a possible lurking variable that would be 110. Group 1 gets Saint-John’s-wort, Group 2 gets
confounded with the effectiveness of the training placebo, and Group 3 gets Zoloft. (c) Use a double-
program. blind design.
6.35 For example, do rat tumors arising from high expo-
sures for a relatively short period indicate that hu-
Participates in mans would get tumors from low doses for a longer
job training Causes? Unemployment period?
or not rate (response 6.37 (a) Subjects are the pieces of dirty laundry; the ex-
(explanatory variable) planatory variables are the brands of laundry deter-
variable) gent and the water temperature; the response is a
cleanliness score. (b) Four treatments. Thirty pieces.
Detergent Brand
Temp. A B
Changes in Hot Treatment 1 Treatment 2
economy, consumer Cold Treatment 3 Treatment 4
spending, demand
(lurking variables)
(c) Randomly assign 30 pieces of dirty laundry to
each of the four treatment combinations.
6.39 (a) Answers will vary on the random assignment.
(b) Label the five varieties A, B, C, D, and E. Let the
6.21 Time of day would be a lurking variable. Students in
digits 1 to 5 correspond to the letters. Go to any line
the 8:30 a.m. class may be more motivated that those
of Table B, select the first four unique digits between
in the 2:30 p.m. class.
1 and 5 and assign the corresponding letters to the
6.23 “Significant” means that a difference as large as the
rows, west to east. The remaining variety will be as-
one we observed is unlikely to occur by chance.
signed to the easternmost plot of the row. Use a new
6.25 Randomly assign 25 students to Group 1 (Brochure
line in Table B for each row.
A) and 25 students to Group 2 (Brochure B). Have
6.41 (a) A block, because the diagnosis is an existing dif-
every student complete the questionnaire and com-
ference between subjects. (b) A treatment that is be-
pare responses.
ing randomly assigned to the subjects.
6.43 (a) In order to account for an effect due to whether a
Group 1 Treatment 1
particular step height was used first or second.
(25 students) (intellectual
excitement) Compare (b) Answers will vary.
Random 6.45 Use a matched pairs (block) design with each dummy
assignment questionnaires as a block. Randomize the order of use of the two air
Group 2 Treatment 2 bags. Repeat several times if possible with each dummy.
(25 students) (more money) 6.47 (a) The explanatory variables are storage and cooking.
Storage levels are fresh, a month at room tempera-
6.27 “Placebo-controlled” means that the control group ture, or a month refrigerated. Cooking levels are
received a placebo. “Double-blind” means that nei- cooked immediately or cooked after an hour at room
ther the patients nor those who interacted with them temperature. The response variables are the ratings of
 Solutions to Odd-Numbered Exercises S-15

the color and flavor. (b) Prepare a random selection (c) The effect of the surgery is confounded with the
of potatoes for each of the ways described in (a). initial health of the patient. (d) Randomly assign half
Rate and compare the treatment groups in terms of the patients to surgery and the other half to nonsurgical
flavor and color. (c) The different combinations of treatment.
treatments will be presented in a random order. 6.77 Matched pairs design. Randomize the order of left-
6.49 (a) Nine treatments: 5008-front, 5008-middle, . . . , 10008- and right-hand thread for each subject.
back. (b) Randomly assign five converters to each of 6.79 4561 healthy, working adults; flu vaccine adminis-
the nine groups. (c) Converters 19, 22, 39, 34, and 05 tered; number of days of lost work, number of days
get 5008-front. with health care provider visits.
6.51 (a) This is a matched pairs design. Blocking accounts 6.81 Institutional review board, informed consent, and
for any possible effects of the treatment order on the confidentiality.
response. (b) To avoid bias that may result from giving
all subjects the same treatment first. (c) Yes. Chapter 7
6.53 (a) The subjects should be told what kinds of ques- 7.1 Answers will vary. One example: in 25 spins we ob-
tions the survey will ask and about how much time it tained 16 tails and 9 heads. Thus: (a) P(heads) 5 0.36.
will take. (b) For example, respondents may wish to (b) P(tails) 5 0.64.
contact the organization if they feel they have been 7.3 21/200 5 0.105.
treated unfairly by the interviewer. (c) Respondents 7.5 1/6.
should not know the poll’s sponsor (responses might 7.7 Answers will vary.
be affected), but sponsors may be revealed when 7.9 No. It’s simply one of several thousand “surprising”
results are published without affecting the results. random phenomena that happen to occur.
6.55 Answers will vary. 7.11 (a) Answers will vary. For example, almost all pas-
6.57 This offers anonymity, since names are never sengers in an airplane crash are killed, but many peo-
revealed. ple survive automobile accidents. (b) Answers will
6.59 Answers will vary. vary. For example, news coverage gets public atten-
6.61 Answers will vary. tion. The news fixation with airplane crashes can lead
6.63 Answers will vary. to the belief that airplane flying is more dangerous.
6.65 Answers will vary. 7.13 (a) Assign Democrats to 0 to 4 and Republicans to 5
6.67 Answers will vary. to 9. (b) Assign Democrats to 0 to 5 and Republicans
6.69 Answers will vary. to 6 to 9. (c) Assign Democrats to 0 to 3, Republicans
6.71 Answers will vary. to 4 to 7, and undecideds to 8 and 9. (d) Assign 00 to
6.73 (a) Use a completely randomized design with equal 52 to Democrats and 53 to 99 to Republicans.
numbers of patients assigned to each of the six pos- 7.15 (a) Four Democrats and six Republicans. (b) Three
sible treatments. Democrats and seven Republicans. (c) Two Demo-
Dosage
crats, four Republicans, and four undecided. (d) There
are six Democrats and four Republicans.
Method of administration 5 10
Injection A B
7.17 (a) The face of the card obtained on each draw is inde-
Skin patch C D pendent of the other draws. Each card in the deck has
Intravenous drip E F the same probability of being drawn. (b) Assign the
digits 00 to 51 to represent the 52 cards in a deck,
Treatment A ignore the digits 52 to 99. Let 00, 01, 02, and 03 repre-
sent the four aces; the assignment of the remaining 48
Treatment B
cards is immaterial. (c) Answers will vary. Beginning
Treatment C on line 101 and moving to the next new line for each
Random Compare
repetition, we find that an ace is obtained in 10 draws
allocation results
Treatment D in 6 of the 10 repetitions, giving an estimate of 0.6.
Treatment E
7.19 (a) The roulette wheel has no memory. The probability
of getting a red on the next trial is the same as on
Treatment F each of the prior trials. (b) The gambler is wrong.
The probability of a red or black card coming up next
(b) With more subjects there is less chance variation depends on what cards have already been drawn.
in the results and a greater likelihood of detecting That is, the deck does not have a memory.
actual differences between treatments. (c) Block by 7.21 When the weather conditions are like those seen today,
gender, then randomly assign about equal numbers to it has rained the following day about 30% of the time.
the six treatments within each block. 7.23 Answers will vary. However, 140 heads in 250 spins
6.75 (a) Type of treatment; survival time. (b) No, this is an is quite unusual for a fair coin.
observational study; we are examining available data.
S-16 Solutions to Odd-Numbered Exercises

7.25 31/365 < 0.085, assuming that it is equally likely (b) MS c h FB. (c) P(MS c h FB) 5 (1 2 0.54) 1
for a person to be born on any day of the year. 0.85 2 0.43 5 0.88.
7.27 (a) HHHH HHHT HHTH HTHH THHH HHTT (d) MySpace
HTHT HTTH THHT THTH TTHH HTTT THTT
Facebook Yes No
TTHT TTTH TTTT. (b) 1/16. (c) P(A) 5 4/16 5 0.25.
7.29 P(D) 5 6/36 5 1/6; P(M) 5 1 2 P(not M) 5 1 2 3/36 5 Yes 0.42 0.43
33/36; P(R) 5 15/36. No 0.12 0.03
7.31 (a) Since the events are deemed to be mutually exclusive,
the probability is 0.45 1 0.23 5 0.68. (b) 1 2 0.68 5 7.45 (a) The surveyed students in fourth, fifth, and sixth
0.32. grade. Which grade the student is in and which is most
7.33 (a) P(C) 5 40/52. P(D) 5 39/52. (b) Drawing a important to the student: good grades, athletic ability, or
spade, club, or diamond that is not a face card. P(C being popular. (b) P(sixth grader)  135/335 < 0.40.
and D) 5 30/52. (c) Drawing a spade, club, or dia- P(good grades most important)  168/335 < 0.50.
mond OR drawing a heart that is not a face card. P(C P(sixth-grader and good grades most important) 
or D) 5 (39 1 10)/52 5 49/52 or P(C or D) 5 40/52 1 69/335 < 0.21. P(sixth-grader or good grades most
39/52 2 30/52 5 49/52. important)  135/335  168/335  69/335 < 0.70.
7.35 (a) The 100 U.S. senators. The senators’ gender and po- (c) There are more than two responses for each vari-
litical affiliation. (b) P(Democrat) 5 (40 1 11)/100 5 able measured.
51/100. P(female) 5 (11 1 5)/100 5 16/100. P(female 7.47
and Democrat) 5 11/100. P(female or Democrat) 5 M R
51/100 1 16/100 2 11/100 5 56/100.
7.37 (a) Black Not black Total
1 6 18
Even 10 10 20
Not even 8 10 18
Total 18 20 38
3
(b) P (B) 5 18/38. P(E) 5 20/38. (c) The event that 7.49 (a) AA AB BA BB.
the ball lands in a black, even spot. P(B and E) 5 (b) Blood type Probability
10/38. (d) The event that the ball lands in either a
black spot or an even spot. P(B or E) 5 18/38 1 A 1/4 5 0.25
20/38 2 10/38 5 28/38. AB 2/4 5 0.5
B 1/4 5 0.25
7.39
D F
7.51 (a) P(B or O) 5 P(B) 1 P(O) 5 0.54.
(b)
B P
40 11 5

0.015 0.085 0.755

44
7.41
B E
0.145

7.53 (a) P(D 0 F) 5 11/16. P(D 0 F) is the probability that a

randomly selected senator is a Democrat, given that
the senator is female. (b) P(F 0 D) 5 11/51. P(F 0 D) is
8 10 10

the probability that a randomly selected senator is

female, given that the senator is a Democrat. (c) No,
P(D d F) 5 11/100 50.11 2 0. (d) No, P (D 0 F) 5
10
7.43 (a) 11/16 2 P(D) 551/100.
FB MS
7.55 Answers will vary. For example, define A to be the
event that the red die is even and B to be the event
that the green die is even. Then P(A d B) 5 9/36 5
43% 42% 12% 1/4 is equal to P(A) # P(B)  1/2 # 1/2 5 1/4.
7.57 Answers will vary.

3%
 Solutions to Odd-Numbered Exercises S-17

7.59 (a) 7.61 (a) 2nd card

4th toss 1st card ♥
H 13
52
1
2 ♥
3rd toss 13 not
H 52 39 ♥
1 T 52
1
2 2 Deal a card
2nd toss 13 ♥
H 1 H 39 not 52
2 52 ♥
1 1 T
2 2
1 39 not
T
1st toss 52 ♥
2
H 1 H
2 (b) P (A d B) 5 (13/52)(13/52) 5 0.0625. (c) Yes,
1 H since we are drawing with replacement. P(B 0 A) 5
1
2 T 13/52 5 P(B).
P(MS 0 FB) 5 P(MS d FB)/P(FB) 5 0.42/0.85 5
2 T 1
1 2 7.63
2 1 H 0.49.
1 T 2
7.65 (a) Method of
2
1 T payment
Toss a 2 CC
fair coin
1 H Type of gasoline
2 0.28
H
1
2 1 T 0.72 other
H 2 Regular
1 0.88
1 1 H CC
2
2 T 2
Midgrade 0.34
T 1 0.02
T
2 0.66 other
Premium
1 H CC
1 2 0.10
1 H
2
2 0.42
T 1
T
2
0.58 other
1 1 H
2 2 (b) P(CC) 5 (0.88)(0.28) 1 (0.02)(0.34) 1 (0.10)(0.42)
T 5 0.2952. (c) P(Premimum 0 CC) 5 P(Premium and
CC)/P(CC) 5 [(0.10)(0.42)]/0.2952 < 0.1423.
1
2
T

(b) 15/16. Only 1 out of 16 outcomes has no heads.

S-18 Solutions to Odd-Numbered Exercises

7.67 Survived transplant? 7.73 (a)

No Yes 4th roll
“1”
Successful Return to
transplant dialysis Total 1
Survive 0 378 180 558 6
5 years 3rd roll
Die within 100 162 180 442 not
“1”
5 years “1”
1 5
Total 100 540 360 1000
6 6
2nd roll “1”
7.69 (a) Test result 1
“1”
 6
AIDS?
1 5 not
99.9% 6 “1”
6 5
Y 6
0.1% 1st roll “1”
1
5%  “1”
6
Randomly not
selected person 1 “1”
5 not 6 “1”
from study 6 “1” 5
population  1 6
0.2% 6 not 1 “1”
95%
N 5 “1” 6
6 not
99.8% Roll a 5 “1”
 6-sided 6
die
(b) P(1) 5 (0.05)(0.999) 1 (0.95)(0.002) 5 0.05185. 1 “1”
(c) P(HIV 0 1)  P(HIV and )/P()  [(0.05) 6
1 “1” not
(0.999)]/0.05185 < 0.963.
P(B 0 T) , P(B) , P(T) , P(T 0 B). 5 6
7.71 “1” 5 “1”
6 6
1
6 “1”
not 5 not
“1” 6 “1” 1
6
5 not
6 “1”
5
6 1 “1”
not “1” 6
“1”
1 not
6 5
“1”
6
5
6 not “1”
“1”
1
6
5
6 not
“1”

(b) P(at least one “1”) 5 1 2 P(no “1”s) 5 1 2 (5/6)

(5/6) (5/6)(5/6) < 0.52.
 Solutions to Odd-Numbered Exercises S-19

7.75 Cat? (b)

Dog? Yes No Total Value di 0 1 2 3 4 5 6 7 8 9
Probability pi 1/10 1/10 1/10 1/10 1/10 1/10 1/10 1/10 1/10 1/10
Yes 15 15 30
No 5 5 10
(c) 0.10
Total 20 20 40
7.77 (a) P(different birthdays) 5 P(2nd individual was not

Probability pi
born on the 1st individual’s birthday) 5 364/365 <
0.9973. P(same birthdays) 5 P(2nd individual was born
0.05
on the 1st individual’s birthday) 5 1/365 < 0.0027.
(b) P(3 different birthdays) 5 (364/365) (363/365) <
0.9918. P(at least two have the same birthday) 5 1 2
P(3 different birthdays) < 0.0082. (c) P(30 different
364 # 363 # c # 336 0
birthdays) 5 < 0.2937. P(at 0 1 2 3 4 5 6 7 8 9
36529
least 2 of 30 have the same birthday) 5 1 2 P(30 dif- Value di
ferent birthdays) < 0.7063. 8.3 P(Y # 3) 5 0.022.
7.79 (a) The winning die is boldfaced for each outcome. 8.5 The expected value of D is 4.5. In the long run, the
Teacher’s 8-sided die average value of the randomly chosen digits will be 4.5.
Your
6-sided 8.7 (a) If boys and girls are equally likely to be born, a
die 1 2 3 4 5 6 7 8 string of 16 boys is quite unusual (1/65,536), so such a
run at any small hospital should be newsworthy. How-
1 (1, 1) (1, 2) (1, 3) (1, 4) (1, 5) (1, 6) (1, 7) (1, 8) ever, given the large number of births across the nation,
2 (2, 1) (2, 2) (2, 3) (2, 4) (2, 5) (2, 6) (2, 7) (2, 8) it is likely that some hospital is going to have a run of
3 (3, 1) (3, 2) (3, 3) (3, 4) (3, 5) (3, 6) (3, 7) (3, 8) 16 boys in a row, so the fact that it happened some-
4 (4, 1) (4, 2) (4, 3) (4, 4) (4, 5) (4, 6) (4, 7) (4, 8) where isn’t all that surprising. (b) The law of averages
5 (5, 1) (5, 2) (5, 3) (5, 4) (5, 5) (5, 6) (5, 7) (5, 8) says nothing about short-run streaks. The overall pro-
6 (6, 1) (6, 2) (6, 3) (6, 4) (6, 5) (6, 6) (6, 7) (6, 8) portion of boys and of girls over time will be about
50%, but streaks of a particular gender, even of length
(b) P(A) 5 27/48 5 0.5625. (c) P(A h B) 5 27/48 16 and certainly of length 5, are to be expected.
1 8/48 2 5/48 5 30/48 5 0.625. (d) No. P(A 0 B) 5 8.9 It is possible that exactly 1000 will be women but it
5/8 5 0.625 2 P(A) 5 0.5625. won’t always be true, so statement (a) is false. With a
7.81 (a) 1 2 0.3 2 0.3 2 0.3 5 0.1. (b) Top 10%: 0, 1, 2. sample of size 2000, we would expect the count of
Top quarter but not top 10%: 3, 4, 5. Top half but not women to be quite close to the theoretical value of
top quarter: 6, 7, 8. Bottom half: 9. 1000, so (b) is true.
7.83 (a)P(smokes)5 31/10050.31.(b)P(smokes 0 male) 5 8.11 (a) Table B: Assign each day of the year a three-digit
19 / 60 < 0.32. (c) P(smokes) 2 P(smokes 0 male) number, 001 to 365. Ignoring 000 and 366 to 999, pro-
but they are very close. It’s possible that these events ceed through the table noting three-digit numbers in the
are independent but that the sample does not give an range until you have obtained 30 birth dates. Check for
exact equality for these probabilities. duplicates. Calculator: randInt (1,365,30) S
7.85 P(safe launch)  P(all six o-rings function properly)  L1. Check for duplications. (b) Answers will vary.
(0.977)6 < 0.87. For example, in one set of 20 repetitions of the calcu-
7.87 Answers will vary. One example: Define A to be the lator simulation, we obtained 15 samples that contained
event that an even digit less than 8 is selected (0, 2, 4, 6) a duplication. You probably shouldn’t take the bet.
and B to be the event that an odd digit greater than 1 is 8.13 (a) 50% of samples will have a sample proportion of
selected (3, 5, 7, 9). These events cannot occur simulta- 0.15 or less since half of the area under the curve is to
neously and are therefore mutually exclusive, but to- the left of 0.15. (b) About 68%. (c) 0.32. (d) Yes. The
gether they do not contain the entire sample space and probability of getting at least 17% is 0.0132.
thus cannot be complements. 8.15 (a) 4.
(b)
Chapter 8 Sample {1,4} {1,9} {2,4} {2,9} {4,9}
8.1 (a) D is a random variable because it takes numeri- x 2.5 5 3 5.5 6.5
cal values (0 to 9) that describe the outcomes of
randomly selecting digits from Table B or using
your calculator.
S-20 Solutions to Odd-Numbered Exercises

(c) 8.47 • Binary? Success 5 seed germinates. Failure 5

seed does not germinate.
• Independent? Possibly, although it’s possible that
if one seed does not germinate, it’s more likely
1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5
that others in the packet will not grow either.
(d) 4. They are the same. • Number? 20 seeds
8.17 P(X . 112) 5 P(z . (112 2 100) / 15 5 0.8) 5 • Success? The probability that each seed germinates
0.2119. is 85%, assuming the advertised percent is true.
8.19 (a) 20/132. The probability of getting an ice cream bar Assuming that independence does hold, this is a bi-
for less than $0.10 is approximately 0.15. (b) P(X , nomial setting.
$0.40) 5 45/132. (c) If all of the 132 balloons are
8.49 • Binary? Success 5 person is left-handed. Failure 5
popped, the store makes $43.96. Without the promo-
person is right-handed.
tion, the store makes $52.80. The store would make
• Independent? Since students are selected randomly,
more money without the promotion, assuming they
would still sell 132 ice cream bars. their handedness is independent.
8.21 The expected price of an ice cream bar is $0.33. It is • Number? There is not a fixed number of trials for
to her benefit to play the raffle since her expected this chance process since you continue to sample
price is less than the usual price of $0.40. until you find a left-handed student.
8.23 (a) Since the number of trials is not fixed, this is not a
y P(Y  y)
binomial setting.
$300 0.9998 8.51 P(X 5 6) 5 6C 6(1/6)6(5/6)6 2 6 < 0.00002.
2$199,700 0.0002 8.53 (a) P(X 5 7) 5 7C 7(0.44)7 (0.56)7 2 7 < 0.0032.
(b) P(X 5 4) 5 7C 4(0.44)4 (0.56)7 2 4 < 0.2304.
(b) The expected value of Y 5 $260. On average, the 8.55 (a) n ? p 5 (20)(0.85) 5 17 seeds. (b) P(X 5 17) 5
insurance company expects to earn $260 for each 20 2 17
17
20C 17(0.85) (0.15) < 0.2428. (c) P(X , 17) 5
home insurance policy it sells.
1 2 P(X $ 17) 5 1 2 0.6477 5 0.3523.
8.25 z 5 (0.51 2 0.47) / 0.016 5 2.5. P(Z . 2.5) 5 0.0062.
8.57 Let X represent the number of heads observed in the
8.27 (a) Letters only: 26 ? 26 ? 26 5 17,576 possibilities.
six tosses. (a) P(X 5 3) 5 6C3(0.5)3(0.5)623 5 0.3125.
(b) Numbers or letters: 36 ? 36 ? 36 5 46,656
(b) P(X 5 4) 5 6C4(0.5)4(0.5)624 5 0.2344.
possibilities.
8.59 (a)
8.29 1959: 26 ? 26 ? 26 ? 10 ? 10 ? 10 5 17,576,000 differ-
• Binary? Success 5 owner greets the dog first.
ent license plates. 1999: 26 ? 26 ? 26 ? 10 ? 10 ? 26 5
Failure 5 owner does not greet the dog first.
45,697,600 different license plates.
8.31 (a) 6! 5 720. (b) 5! 5 120. • Independent? One owner’s greeting habits do not
8.33 (a) 28 ? 28 5 784. (b) 28P2 5 756. (c) 28C2 5 378. affect the greeting habits of other owners.
(d) 28C2 1 28 5 378 1 28 5 406. • Number? 12 dog owners are randomly selected.
8.35 (a) 10P3 5 720. (b) 1 2 0.72 5 0.28 (this is 1 minus • Success? The probability of greeting the dog first
the probability that the winning number has 3 differ- remains constant from one owner to the next, 0.66.
ent digits). This is a binomial setting.
8.37 (a) 6P0 5 1. There is one way to arrange none of the (b) P(X 5 7) 5 12C7(0.66)7(0.34)1227 5 0.1963. (c)
6 items. (b) 6P3 5 120 and 6C3 5 20. Answers will P(X $ 5) 5 1 2 P(X # 4) 5 1 2 0.0213 5 0.9787.
vary. Note that once the 3 friends are chosen (20 pos- 8.61 Let X 5 number of left-handed students in the SRS
sibilities) there are 6 (3 ? 2 ? 1) ways to order the three of 15 students. (a) P(X 5 3) 5 15C3(0.1)3(0.9)1523 5
positions so that there are 20 ? 6 5 120 permutations 0.1285. (b) P(X # 3) 5 0.1285 1 0.2669 1 0.3432 1
of 3 out of the 6. (c) 6C2 5 15 and 6C4 5 15. These 0.2059 5 0.9445. (c) P(X $ 4) 5 1 2 P(X # 3) 5 1 2
values are the same because once you pick 2 friends 0.9445 5 0.0555. Since the probability of this event
out of the 6, the remaining 4 are automatically is quite small, it would be surprising for the SRS to
grouped. contain 4 or more left-handed students. Note that the
8.39 (a) 26 ? 26 5 676. (b) 2 ? 26 ? 26 5 1352. probability of obtaining exactly 4 left-handed stu-
8.41 (a) 28 ? 28 5 784. (b) 28P2 5 756. (c) 756/784 < 0.96. dents is 0.0428.
8.43 1/49C6 < 0.00000007. 8.63 Let X 5 number of orange M&M’s in a sample of
8.45 (a) 6/(6 ? 6 ? 6 ? 6 ? 6) 5 6/7776 < 0.00077. (b) There size 8. (a) P(X 5 0) 5 8C0(0.2)0(0.8)820 5 0.1678,
are 6 ? 5 ? 4 ? 3 ? 2 5 720 rolls such that all five dice not surprising. (b) P(X 5 2) 5 8C2(0.2)2(0.8)822 5
show a different number of spots on the up-faces. 0.2936, not surprising. (c)P(X5 4) 5 8C4(0.2)4(0.8)824
The probability is therefore (6 ? 5 ? 4 ? 3 ? 2)/(6 ? 6 ? 5 0.0459, surprising.
6 ? 6 ? 6) 5 720/7776 < 0.09. 8.65 Note that X 5 the number of times that the ball falls
to the right is a binomial random variable with n 5 6
 Solutions to Odd-Numbered Exercises S-21

and p 5 0.6. (a) P(X 5 0) 5 6C0(0.6)0 (0.4)620 5 482

0.0041. (b) P(X 5 2) 5 6C2(0.6)2 (0.4)622 5 0.1382. (b) p^ 5 5 0.47
1025
(c) Slot E.
(0.47)(0.53)
Å
Slot Value x Probability pi 0.47 6 2 5 (0.439, 0.501).
1025
A 0 0.0041 (c) We are 95% confident that the true proportion of
B 1 0.0369 American women who feel that they are not getting
C 2 0.1382 enough time for themselves is in the interval (0.439,
D 3 0.2765 0.501).
E 4 0.3110 9.7 We are 95% confident that the true proportion of
F 5 0.1866 women in favor of new laws restricting guns is in the
G 6 0.0467 interval (0.62, 0.70). The 95% confidence level
means that the process used to generate this interval
8.67 It’s likely that they mean that there are n types of will successfully capture the true population propor-
pasta and m types of sauce where nm 5 42. tion 95% of the time.
8.69 (a) 107 5 10,000,000. (b) 8 ? 106 5 8,000,000. (c) 9.9 Since (1 2 0.98)/2 5 0.01, z* for a 98% confidence
There are 7 ? 96 5 3,720,087 phone numbers with no interval can be found by looking for an area of 1 2
9s. It follows that there are 8,000,000 2 3,720,087 5 0.01 5 0.99. The closest area is 0.9901, corresponding
4,279,913 phone numbers with at least one 9. to a critical value of 2.33.
8.71 (a) 9.11 (a) p^ 5 36/50 5 0.72. A 95% confidence interval is
• Binary? Success 5 makes the free-throw basket. given by
Failure 5 does not make the free-throw basket.
1 0.72 2 1 0.28 2
5 1 0.596, 0.844 2
Å
• Independent? Since the probability of making a 0.72 6 1.96
basket is the same for each free throw, the events 50
are independent. (b) A 90% confidence interval will be more precise
• Number? 7 free-throw attempts are made. (narrower) but also have lower confidence than either
• Success? The probability of making a free-throw a 95% or a 99% confidence interval. (c) z* 5 1.645.
basket remains the same for each attempt, 0.8. The 90% confidence interval is given by
This is a binomial setting. 1 0.72 2 1 0.28 2
5 1 0.616, 0.824 2
Å
(b) P(X # 4) 5 1 2 P(X . 4) 5 1 2 (0.2753 1 0.72 6 1.645
50
0.3670 1 0.2097) 5 1 2 0.852 5 0.148. (c) Assuming
We are 90% confident that the true proportion of se-
that she is an 80% free-throw shooter, the probability
niors who will attend the prom lies in the interval
that she makes 4 or fewer baskets is 0.148 (part (b)),
(0.616, 0.824).
which is not small enough to raise suspicion.
9.13 z* 5 2.58. Using p^ 5 0.5 as our estimate of the pop-
8.73 (a) By the 68–95–99.7 rule, the probability is 0.95.
(b) For 36%, z 5 (0.36 2 0.4)/0.015 5 22.67; ulation proportion, n 5 (2.58 / 0.03)2 # p^ (1 2 p^ ) 5
and for 44%, z 5 (0.44 2 0.4)/0.015 5 2.67. From (2.58 / 0.03)2 # 0.5(0.5) 5 1849.
Table A, the probability is 2?(0.0038) 5 0.0076. 9.15 p^ 5 180 / 2004 5 0.0898. Thus,
8.75 (a) 20/80 5 0.25. 1 0.0898 2 1 0.9102 2
Å
(b) the margin of error 5 1.96
Amount gained xi $3 2$1 2004
Probability pi 0.25 0.75 5 0.0125.
(c) The expected value of X is $3(0.25) 2 $1(0.75) 5 $0. 9.17 (a) The sampling distribution of p^ is approximately
Normal with mean 0.14 and
Chapter 9 1 0.14 2 1 0.86 2
Å
9.1 (a) The seniors in her school. (b) It is the true propor- standard deviation 5 5 0.0142
tion of seniors in Tanya’s school who actually plan to 600
attend the senior prom. (c) p^ 5 36/50 5 0.72. (b) z 5 (0.182 2 0.14)/0.0142 5 2.96. P(Z $ 2.96) 5
9.3 It means that the process we used to generate this 0.0015. It would be very surprising. (c) z 5 (0.12 2
interval will successfully capture the true population 0.14)/0.0142 5 21.41. The probability of obtaining a
proportion 95% of the time. sample with a sample proportion of 0.12 or less is
9.5 (a) It is the true proportion of women living in the 0.0793. It would be somewhat unlikely, but not too
United States (excluding Alaska and Hawaii) who surprising.
feel that they don’t get enough time to themselves. 9.19 (a) Undergraduate students taking courses in psychol-
ogy and communications since this is the population
S-22 Solutions to Odd-Numbered Exercises

from which the sample was drawn. (b) z* 5 2.58. A 99% (d) If the proportion of presses of the button that re-
confidence interval is then sult in a red light is truly 0.3, the probability of ob-
p^ 1 1 2 p^ 2 serving 20 or fewer red lights in 100 presses is about
p^ 6 z*
Å
1.5%. This is unlikely enough to cause one to suspect
n that the custom officers’ claim is false.
107 1 107 / 127 2 3 1 2 1 107 / 127 2 4 9.29 (a) Let p 5 the proportion of middle school students
Å
5 6 2.58 who engage in bullying behavior. Then H0 : p 5 0.75
127 127
5 (0.7591, 0.9259) and Ha : p . 0.75.
We are 99% confident that the proportion of under- 445
graduate students from courses in psychology and p^ 5 5 0.7975
558
1 0.75 2 1 0.25 2
communications who pray at least a few times a year
z 5 1 0.7975 2 0.75 2 /
0.0475
Å
is between 0.76 and 0.93. 5
9.21 (a) Mean 5 p 5 1/2 and 558 0.0183
p11 2 p2 0.5 1 0.5 2
5 2.59 1 P-value 5 0.0048
Å Å 25
standard deviation 5 5 Since the P-value is so small, the alternative hypoth-
n esis that more than 75% of middle school students
5 0.1 engage in bullying behavior is strongly supported.
(b) p^ 5 16 / 25 5 0.64. (b) Not necessarily. The questionnaire did not define
(c) bullying; the students simply responded to questions
that supposedly describe bullying. There is also some
question as to how well middle school students would
understand questions about their behavior over the
last 30 days. Additionally, the data come from only
one school in the Midwest and may not be represen-
tative of all middle schoolers in the United States.
9.31 (a) P-value 5 P(z , 21.77) 5 0.0384. Since 0.01 ,
0.5 p  0.64 P-value , 0.05, we reject the null hypothesis and
conclude that the population proportion is less than
0.64 2 0.5
(d) P 1 p^ . 0.64 2 5 P az . b 5 P 1 z .1.4 2
0.3 if our level of significance is 5%, but not if it is
0.1 1%. (b) P-value 5 2 ? P(z , 21.77) 5 0.0768. There
5 1 2 0.9192 5 0.0808. is insufficient evidence to reject the null hypothesis
It would be somewhat unlikely, but not so small to in favor of the alternative hypothesis at either the
conclude that the deck isn’t fair. 0.01 or a 0.05 significance level.
9.33 The parameter is the proportion p of students at the
0.56 2 0.5
9.23 P 1 p^ . 0.56 2 5 P az . b 5 P 1 z . 0.85 2 school who completed their math homework last
0.0707 night. H0 : p 5 0.75 and Ha : p , 0.75.
5 1 2 0.8023 < 0.20 and 9.35 (a) Significant at the 5% level means that our ob-
0.72 2 0.5
P 1 p^ . 0.72 2 5P az . b 5 P 1 z . 3.112
served value would occur by chance less than 5% of
0.0707 the time if the null hypothesis were true. This does
5 1 2 0.9991 < 0.001 not mean it will occur by chance less than 1% of the
9.25 The parameter is the proportion p of all teenagers time, which is what it means to be significant at the
who say they rarely or never argue with their friends. 1% level. (b) This means that the probability of ob-
taining a sample result as extreme as that observed,
H0 : p 5 0.72. Ha : p 2 0.72.
assuming the null hypothesis is true, is less than 5%.
9.27 (a) The parameter is the proportion p of presses of the but-
The null hypothesis is either true or it’s not true—
ton that result in a red light. H0 : p 5 0.30. Ha : p 2 0.30.
there’s no probability associated with the actual sta-
(b) Assuming the null hypothesis is true, the sampling
tus of the null hypothesis.
distribution of p^ has mean 5 p 5 0.30 and 9.37 (a) The P-value is one-sided on the high side, P-value 5
p11 2 p2 0.3 1 0.7 2 P(z . 2.36) 5 1 2 0.9909 5 0.0091. (b) Since the
Å Å 100
standard deviation 5 5 P-value is less than both 5% and 1%, we reject the
n
5 0.0458 null hypothesis and conclude p . 0.81 for either a
(c) p^ 5 20 /100 5 0.2. Assuming p 5 0.3, the prob- 5% or a 1% level of significance.
ability of obtaining a sample proportion this extreme is 9.39 (a) The hypotheses: Let p be the true proportion
of drivers who use cell phones while driving. Then
0.2 2 0.3
2 # P 1 p^ , 0.2 2 5 2 # P az , b H 0 : p 5 0.5 and H a : p , 0.5. The sampling
0.0458 distribution: If the null hypothesis is true, p^ follows
5 2 # P 1 z , 22.182 5 0.0292
 Solutions to Odd-Numbered Exercises S-23

a Normal distribution with mean 5 0.5 and stan- 9.61 (a) It stands for the true proportion of people who
dard deviation 5 "3 1 0.5 2 1 0.5 2 4 /70 5 0.0598. have body temperatures lower than 98.68F. (b) H0 : p 5
The data: p^ 5 21 / 70 5 0.30. The P-value: z 5 0.5. H a : p . 0.5.
(0.3 2 0.5) / 0.0598 5 23.34 1 P-value 5 0.0004. 9.63 (a) 37% is a sample proportion and is therefore sub-
Conclusion: This is a very small P-value and pro- ject to sampling error.
vides strong evidence against the null hypothesis. p^ 1 1 2 p^ 2 0.37 1 0.63 2
(b) p^ 6 z*
Å Å
Based on these results, we conclude that fewer than 5 0.37 61.96
n 1000
half of all drivers use cell phones while driving.
(b) The drivers were aware that they were being 5 0.37 6 0.030
taped and might well have allowed themselves fewer We are 95% confident that the proportion of adults aged
distractions than if they were not being taped. Dis- 18 and older who claim that football is their favorite
carding the first three hours probably helped, but it is sport to watch on television is between 0.34 and 0.40.
still likely that driver behavior was influenced by the (c) Yes. (d) Yes. People may be swept into the excite-
presence of the cameras. ment during the playoff season and claim football is their
9.41 The idea of this section that the company is address- favorite sport to watch on television yet have a different
ing is “the design of the data production matters.” opinion when a different sport is “in season.”
This is because the small bags are not truly SRSs 9.65 You do not need a confidence interval when you
from the larger population of M&M’s. know the population proportion. You construct a con-
9.43 You use sample data to construct confidence intervals fidence interval only when you are using sample data
for unknown population values. In this case, we know to make inferences about the population.
the exact number of presidents who were only chil- p^ 1 1 2 p^ 2
Å
dren; inference is not necessary. 9.67 (a) p^ 6 z*
n
9.45 The 59 dropouts could have created bias in the results
a1 2 b
if there was a systematic reason for their dropping 775 775
out. If the unmotivated participants dropped out, 775 1100 1100
ã
those remaining could have more extreme results 5 6 2.58
1100 1100
than are typical.
5 0.705 6 0.035
9.47 The difference in the results was 11% (23% 2 12%),
which would be practically significant to a smoker We are 99% confident that the population proportion
who is trying to quit. of teenagers who play games online is between
9.49 This study gives some evidence that Zyban is effec- 66.5% and 73.5%. (b) They would be larger due to
tive at helping people quit smoking. Before drawing the smaller sample sizes for each of the proportions.
a cause-and-effect conclusion, however, we would (c) If asked the question directly, teenagers may not
need to check that the difference in the percents who respond truthfully. They may be worried about get-
quit smoking is too large to be explained by chance ting in trouble with their parents, being embarrassed
variation. in front of their peers, etc.
9.51 We would expect to incorrectly reject the null hypoth- 9.69 (a) The hypotheses: H 0 : p 5 0.5 versus H a : p . 0.5,
esis in about 5%, or 1000(0.05) 5 50, of the tests. where p 5 the proportion of all U.S. teens who
9.53 No. This is an observational study, and we have no would say that young people should wait to have
way of knowing whether taking statin drugs or some sex until marriage. The sampling distribution:
other factor reduced the likelihood of developing If the null hypothesis is true, p^ follows a Normal
distribution with mean 5 0.5 and standard deviation 5
!(0.5)(0.5) / 439 5 0.0239. The data: p^ 5
breast cancer.
246 / 439 5 0.5604. The P-value: z 5 1 0.5604 2
9.55 (a) No. We would expect about 1% of the sample,
0.5 2 / 0.0239 5 2.53 1 P-value 5 P(z . 2.53) 5
five people, to have results that deviate enough by
chance to have a P-value of less than 0.01. (b) The
four should be tested again. The chances of their be- 0.0057. Conclusion: This is a very small P-value
ing able to get significant results twice in a row if and provides strong evidence against the null hypoth-
they do not have ESP are much smaller than 0.01. esis. The data give convincing evidence that a major-
9.57 (a) The margin of error must get smaller as n gets ity of U.S. teens would say that young people should
larger because n is in the denominator. (b) As n in- wait to have sex until they’re married.
creases, z will be farther out in the tail of the Normal p^ (1 2 p^ )
(b) p^ 6 z*
distribution, and the P-value will therefore decrease. Å n
9.59 Yes. Significant results have a low probability of
occurring if the null hypothesis is true. While they 0.5604(1 2 0.5604)
Å
could have occurred by chance, they are unlikely to 5 0.5604 6 1.96
439
have done so. 5 0.5604 60.0464
S-24 Solutions to Odd-Numbered Exercises

We are 95% confident that the population proportion 80%

of teenagers who would say that young people should

Percent with C or better

70%
wait to have sex until they’re married is between
51.4% and 60.7%. (c) No. The inference methods 60%
assume that the responses are truthful; this may or 50%
may not be the case. Assuring the teens that the 40%
survey is anonymous is one way to achieve more
30%
truthful responses.
20%
10%
Chapter 10
10.1 H0 : The company’s stated color distribution for their 0%
peanut M&M’s is correct; that is, pblue 5 0.23, 2 2 to 12 12
porange 5 0.23, pgreen 5 0.15, pyellow 5 0.15, pred 5 Extracurricular activities (hours per week)
0.12, pbrown 5 0.12. Ha : The company’s stated color
distribution for their peanut M&M’s is not correct. 10.15 (a) Alcohol Did not
10.3 X2 5 11.3724. consumption Smoked smoke Total
10.5 Using a significance level of 0.05, there is enough
Yes 20 13 33
evidence to reject the null hypothesis and conclude
No 48 96 144
that the distribution of colors does not follow the
company’s stated color distribution. Total 68 109 177
10.7 The hypotheses: H0 : Mendel’s hypothesis is correct;
(b) Among the drinkers, 60.6% smoked. Among the
that is, pwrinkled 5 0.25, psmooth 5 0.75. H a : The pro-
teetotalers, 33.3% smoked. Based on these data, people
portions of smooth and wrinkled peas differ from
were more likely to resume smoking if they con-
Mendel’s hypothesized values. The sampling distri-
sumed alcohol.
bution: Degrees of freedom 5 1. The data: The
expected counts are smooth 5 417 and wrinkled 5 (c)
139. The P-value: X2 5 0.3453. The P-value is 60.60%
greater than 0.25. Conclusion: There is not enough 60%
Percent who smoke

evidence to reject Mendel’s hypothesized distribu-

50%
tion of pea type.
10.9 (a) H0 : Skittles candies are equally distributed among 40%
the five flavors: lemon, lime, orange, strawberry, and 33.30%
grape; that is, plemon 5 plime 5 porange 5 pstrawberry 5 30%
pgrape 5 0.20. H a : Skittles candy flavors are not 20%
equally distributed. (b) 12 candies of each of the five
flavors. (c) Greater than 9.49. (d) Greater than 10%
13.28. 0%
10.11 (a) H0 : The digits are equally likely to occur; that is, Drinker Nondrinker
p0 5 p1 5. . . 5 p9 5 0.10. H a : The digits are not equal- Drinking status
ly likely to occur. (b) Answers will vary.
10.17 (a) Temperature
10.13 (a)
Extracurricular activities Cold Neutral Hot
(hours per week)
Hatched 16 38 75
2 2 to 12 12 Did not hatch 11 18 29
C or better 11/20 5 0.55 68/91 < 0.747 3/8 5 0.375
D or F 9/20 5 0.45 23/91 < 0.253 5/8 5 0.625 (b) 59.3% of the cold eggs hatched, 67.9% of the
neutral eggs hatched, and 72.1% of the hot eggs
(b) Among the students who spend very little time in hatched. The researchers were mistaken in believing
extracurricular activities, slightly more than half got that the eggs would not hatch in cold water, though
a C or better. The group of students who spent some they don’t hatch as well in cold water.
time in extracurricular activities had a high percent 10.19 (a) X2 5 6.91. D or F . 12. (b) df 5 2; 0.01 , P ,
who received a grade of C or better (75%), while 0.05. The chi-square statistic is significant at the 5%
only 38% of students who spent a lot of time in extra- level but not at the 1% level. (c) We conclude that
curricular activities received a grade of C or better. at the 5% level there is a statistically significant
 Solutions to Odd-Numbered Exercises S-25

relationship between grade earned and number of 40% Gender

hours spent in extracurricular activities. Female
10.21 The hypotheses: H 0: Water temperature and hatch- Male
ing rates are not related. H a: Water temperature and 30%

Percent
hatching rates are related. The sampling distribu-
tion: The observed and expected values (O/E) are 20%
given in the following table:
Temperature
10%
Cold Neutral Hot
Hatched 16/18.63 38/38.63 75/71.74 0%

Finance
Economics
Administration
Accounting
Did not hatch 11/8.37 18/17.37 29/32.26 Major

df 5 2. The data: The observed cell counts are close

to the expected cell counts, so it does not appear that
water temperature and hatching rates are associated.
The P-value: X2 5 1.703. P-value . 0.25. Conclusion:
These results are not significant. These data do not Administration is the most popular major for women.
support the hypothesis that water temperature and Finance, followed closely by accounting, is the most
hatching rates are related. popular for men. The least popular major for men
10.23 (a) The most effective treatment was desipramine, and women is economics.
followed by lithium. 10.29 Hypotheses: H0 : Nuthatches do not prefer particular
types of trees when they’re searching for seeds and
60%
Percent staying off cocaine

insects; that is pfirs 5 0.54, ppines 5 0.40, pother 5 0.06.

50% H a : Nuthatches prefer particular types of trees when
they’re searching for seeds and insects. The sam-
40% pling distribution: df 5 2. The data: The expected
counts are firs 5 84.24, pines 5 62.4, other 5 9.36.
30%
The P-value: X2 5 7.4182. 0.01 , P , 0.05. Con-
20% clusion: We reject the null hypothesis at the 5% level
but not at the 1% level. At the 5% level, we conclude
10% that nuthatches exhibit a preference when searching
0% for seeds and insects.
Desipramine Lithium Placebo 10.31 (a) Hypotheses: H0 : The coin is fair; that is, pheads 5
Treatment 0.5, ptails 5 0.5. H a : The coin is not fair. The sam-
pling distribution: df 5 1. The data: The expected
(b) Stayed off Did not stay counts are heads 5 tails 5 100. The P-value: X2 5
Treatment type cocaine off cocaine 9.68. 0.001 , P , 0.01. Conclusion: We reject the
null hypothesis at the 1% level and conclude that the
Desipramine 14 10 coin is not fair. (b) Hypotheses: The parameter is the
Lithium 6 18 proportion of coin spins that resulted in “tails.”
Placebo 4 20 H 0 : p 5 0.5. H a : p 2 0.5. The sampling distribu-
tion: If the null hypothesis is true, p^ follows a Normal
10.25 (a) 0.01 , P-value , 0.05. (b) 0.25 , P-value. distribution with mean 5 p 5 0.5 and standard devia-
(c) 0.20 , P-value , 0.25.
tion 5 0.0354. The data: p^ 5 0.61. The P-value:
10.27 The following table gives the counts and percents, by
major, for males and females: P-value 5 2 # P(p^ . 0.61) 5 0.0018. Conclusion:
Female Male There is enough evidence, at the 1% level, to con-
clude that the coin is unfair. (c) The results from the
Count Percent Count Percent two tests are the same.
Accounting 68 30.2% 56 34.8%
Administration 91 40.4% 40 24.8%
Economics 5 2.2% 6 3.7%
Finance 61 27.1% 59 36.6%
Total 225 161
S-26 Solutions to Odd-Numbered Exercises

10.33 Answers will vary. Here’s one possibility:

SRS of size n x
Population SRS of size
mean μ and n
x
standard SRS o
deviation σ. f size
n
x

97 103 109 115 121 127 133

(b) and (c) If the null hypothesis were true, it would
Std. dev. σ not be surprising to get a value as far removed from
n
Mean μ 115 as 118.6, but it would be fairly unusual to get one
as far removed as 125.7.

118.6 125.7

Values of x
10.35 (a) The histogram appears to be more or less Normal
(the bar centered at 0.54 makes it appear somewhat
less so) and centered at the expected mean of 0.50.
15 97 103 109 115 121 127 133
10.39 (a) 95% CI: 275 6 1.96(58 / !1077) 5 (271.54,
278.46).
(b) 90% CI: 275 6 1.645(58 / !1077) 5 (272.09,
277.91). 99% CI: 275 6 2.575 (58 / !1077) 5
10
Frequency

(270.45, 279.55).
(c) The margin of error is 2.91 for a 90% CI, 3.46 for
5 a 95% CI, and 4.55 for a 99% CI. Increasing the con-
fidence level has the effect of increasing the margin
of error of the confidence interval.
10.41 The hypotheses: Let m 5 the mean increase, in dol-
0 lars, of credit card charges this year compared with
0.42 0.44 0.46 0.48 0.50 0.52 0.54 0.58 last year for customers who charge at least $2400.
Numbers H 0 : m 5 0. H a : m . 0. The sampling distribution:
(b) The population mean has a theoretical value of If the null hypothesis is true, the sampling distribution
0.50. of x for samples of size 200 would be approximately
1 Normal with mean 0 and standard deviation 7.637.
The data: The sample mean is 332. The correspond-
ing standardized score is 43.47. The P-value:
P-value < 0. Conclusion: We reject the null hypoth-
esis and conclude that it is likely that the mean
amount charged has increased this year compared
with last year.
10.43 (a) The sample size is 200 and the central limit theo-
rem tells us that, for large samples, the sampling distri-
0 bution of x will be approximately Normal regardless
0.5 1 of the shape of the population from which the indi-
vidual observations are drawn. (b) Answers will vary.
10.37 (a) x is approximately Normally distributed with One possibility is that the economy improved, so
mean 115 and standard deviation 6. people spent more. Randomly divide the customers
into two groups. One of these groups has their fees
waived and the other group does not. Compare the
two groups in terms of the increase in the amount
charged from the prior year.
 Solutions to Odd-Numbered Exercises S-27

10.45 (a) t* 5 3.106. (b) t* 5 1.699. (c) t* 5 2.110. (d) If the null hypothesis is true, the sampling distri-
10.47 95% CI: 7.364 6 (2.228)(2.086) 5 (2.72, 12.01). bution of x has mean 0.957 and standard deviation
The confidence interval gives us a range of plausible 12.55. The standardized score is 0.35, so P-value .
values for the difference in depression scores while 0.05. This is not significant and does not provide evi-
taking a placebo versus taking a caffeine capsule. dence that wearing a scented mask reduces the time
10.49 The hypotheses: The parameter of interest is m, the needed to complete a task.
mean number of additional hours of sleep gained af- 10.57 The hypotheses: Let m 5 average nightly study time
ter taking the drug. H 0 : m 5 0. H a : m . 0. The for first-year students at a university. H 0 : m 5 2.5
sampling distribution: We will use a t distribution hours, or 150 minutes. H a : m , 150 minutes. The
with df 5 10 2 1 5 9. The data: x 5 2.33 and s 5 sampling distribution: If the null hypothesis is true,
2.0022. The corresponding standard error is 0.6332, the sampling distribution of x for samples of size 269
t 5 3.68. The P-value: 0.0025 , P-value , 0.005. would be approximately Normal with mean 150 min-
Conclusion. We reject the null hypothesis. The data utes and standard deviation 3.963. The data: The
support the hypothesis that the drug is effective in sample mean is 137 minutes. The standardized score
increasing sleep time. is 23.28. The P-value: P-value 5 0.0005. Conclu-
10.51 No, the student is not correct. The interval gives a sion: We reject the null hypothesis and conclude that
range of plausible values for the mean NAEP quanti- first-year students at the university study less than
tative score for men aged 21 to 25. It says nothing 2.5 hours per night on average.
about the individual scores. 10.59 (a) x 5 7.5 hours. s 5 64.536 minutes. (b) His inter-
10.53 (a) Because the distribution of individual weights is pretation is incorrect. The confidence interval is a
not Normally distributed. (b) Since the sample size is statement regarding the population mean lifetime of
large, and the data are not very non-Normal, the cen- the batteries, not the individual lifetimes. (c) No. The
tral limit theorem tells us that the distribution of the confidence interval gives information about the value
mean weight is approximately Normal. The standard of the population mean. It tells us nothing about the
deviation 5 6.39. P(x . 200) 5 0.0594. value of another sample mean. (d) We are 95% con-
10.55 (a) Let m be the average difference in speed of com- fident that the mean lifetime of the company’s AA
pleting a task wearing an unscented mask minus batteries is between 430 and 470 minutes.
wearing a scented mask. H 0 : m 5 0. H a : m . 0. 10.61 The hypotheses: Let m 5 mean response time to all
(b) x 5 0.957, s 5 12.55. On average, there is about accidents involving life-threatening injuries this year
a one-second improvement, which is not enough to in a particular city. H 0 : m 5 6.7 minutes. H a : m ,
be considered practically important. (c) There aren’t 6.7 minutes. The sampling distribution: If the null
any outliers. hypothesis is true, the sampling distribution of x for
Stem-and-leaf of diff N  21 samples of size 400 would be approximately Normal
Leaf unit  1.0 with mean 6.7 minutes and standard deviation 0.1.
2 5 The data: x 5 6.48 minutes. z 5 22.2. The P-value:
P-value 5 0.0139. Conclusion: Using a 5% signifi-
2 cance level, we reject the null hypothesis and conclude
1 7 that the mean response time this year was less than
6.7 minutes, the previous year’s average response
1 1 0
time.
0 8 7 6
10.63 (a) Goal
0 4 4 3 0
Gender Grades Popular Sports
0 24
Female 130/51.8% 91/36.2% 30/12.0%
0 789
Male 117/51.5% 50/22% 60/26.4%
1 24
1 69 (b) The most popular response for both males and
females was “make good grades.” The second most
2
popular response was “to be popular” for females
2 5 and “to be good at sports” for males.
S-28 Solutions to Odd-Numbered Exercises

Female 10.67 (a) This was an experiment since a treatment was

50% “applied” to the subjects. (b) The hypotheses: H0 :
Male
Treatment and pregnancy are unrelated. Ha: Treat-
40% ment and pregnancy are related. The sampling
Percent

distribution: The data (O/E) are

30%
Acupuncture Control
20% Pregnant 34/27.5 21/27.5
Not pregnant 46/52.5 59/52.5
10%
df 5 1. The data: There are slightly more pregnan-
0% cies in the acupuncture group and fewer in the con-
Goal Grades Popular Sports
trol group than we would expect if the null hypothesis
10.65 (a) Let m 5 the mean decrease in systolic blood pres- were true. The P-value: X2 5 4.68. 0.01 , P , 0.05.
sure for patients who eat dark chocolate. H 0 : m 5 0. Conclusion: This result is significant at the 5% level
H a : m . 0. (b) The standard score for x is 2.20, but not at the 1% level. Using a 5% significance level,
which gives a P-value of 0.035. (c) The P-value of we reject the null hypothesis and conclude that preg-
0.035 is significant at the 5% level and provides good nancy and treatment are related. (c) Since the subjects
evidence that there has been a reduction in systolic were aware of which treatment they were receiving,
blood pressure for patients who eat dark chocolate. there was no placebo group. It’s possible that the ob-
(d) The sample size was very small. We would have served results are due in part to the placebo effect.
more confidence in the results if the sample size were
increased to control for variability.
 INDEX

Note: A page number in boldface indicates a Buffon, Count (George-Louis Leclerc), 314, Charts. See Graphs; Pie charts
definition; in italics, a figure; and followed by t, 445, 458 – 459 Chi-square (x2) distributions, 474 – 475t, 475
a table. Chi-square (x2) statistics, 473
C and goodness of fit, 471– 476
A C. See Levels of confidence (C) Chi-square (x2) tests, 470 – 493, 472
a. See y intercepts (a) nCk. See Combinations ( nCk) with calculators, 487– 488
a. See Significance levels (a) x2. See Chi-square (x2) distributions; Chi-square for two-way tables, 485– 488
Adjustment, statistical, 268 –269 (x2) statistics; Chi-square (x2) tests Clinical trials, 260 –263. See also Experiments
Alternative hypotheses (Ha), 442 – 444 Calculators. See also Applets; Computers ethics in, 292–298
one- vs. two-sided, 447 binomial distributions, 409–410 problems encountered in, 277–279
Anonymity, 295 boxplots, 71–72 Coin tossing, 313–315, 350
Apgar, Virginia, 372 chi-square (x2) tests for two-way tables, histograms for, 371
Applets. See also Calculators; Computers 487– 488 tree diagrams for, 352, 352–353
confidence intervals, 427– 428 confidence intervals, 433– 434 Combinations ( nCk), 392–393
correlation and regression, expected values, 378 with calculators, 397
175–176, 178 goodness of fit tests, 478– 479 Common responses, 185 –186, 186
data transformation, 74 –75 histograms, 52–53 Comparative observational studies, 6 –7, 268
expected values, 376 Normal curve areas, 132–134 Complements, 341
Galton’s board, 406 permutations and combinations, 397 Completely randomized experimental designs,
histograms, 49, 56 prediction, 174 –175 281–282
means and medians, 68 probability simulations, 327–328 Computers. See also Applets; Calculators
Normal curves, 118–119, 127–128 regression, 171, 174 –175 and interviewing, 239
probability, 312–313 residual plots, 181 and least-squares regression lines, 171
sampling distributions, 493–494 scatterplots, 152–154 and random numbers, 207
significance tests, 442 – 443 significance tests, 450– 451 and simple random samples, 211–214
simple random samples, 211, 216 simple random samples, 212–213 Conditional probability, 349–350
Approximately Normal distributions, 422 t tests, 510 –512 applications of, 357–361
areas under curves, 107–109, 124 –132 Cards, 334–336 and independence, 350 –352
with calculators, 132–134 expected values for, 377–378 and tree diagrams, 352–356
and sampling distributions, 381, 499 and probabilities, 394 – 395 Conditional probability formula, 355
Associations tree diagrams for, 353, 353–354, 354 Confidence intervals, 420 – 438, 427. See also
and correlation, 160–164 Venn diagrams for, 339 –340, 340 Intervals; Levels of confidence (C)
positive vs. negative, 151 Casinos, 369, 375 advantages of, 460
Astragali, 316, 316 Categorical variables, 4 with calculators, 433– 434
bars graphs and pie charts for, 36 – 41 and critical values, 431– 432, 431t
B and two-way tables, 480 – 482 and estimation, 421– 426
b. See Slopes (b) Causation, 183–186, 185, 186 interpreting, 427– 430
Bar charts. See Bar graphs evidence for, 186 –188 for population means, 498–500, 499
Bar graphs, 12, 12, 20, 36 – 41, 37, 38. and experiments, 262–264 for population proportions, 431– 435
See also Graphs and prediction, 186 and t intervals, 505–507
Behavioral experiments, 299–301 Census Bureau, U.S., 15–16 Confidence levels. See Levels of confidence (C)
Bell-shaped curves. See Normal curves Censuses, 10 –13 Confidence statements, 230 –232
Benford, Frank, 384 Centers of distributions, 46, 50 Confidentiality, 294, 295–296
Benford’s Law, 384 –385 under data transformations, 74 –75 Confounding, 185–186, 186, 260
Bias, 203, 224–226, 227 of density curves, 109 –110, 110 and experiments, 260, 268
managing, 225–226 and means, 66 – 68 Control groups, 13, 263 –264, 286
Binomial distributions, 404 – 408, 405 and medians, 60 – 61 Convenience samples, 203
with calculators, 409 – 410 Central limit theorem, 494, 495 – 497 Correlation (r), 158, 161
Binomial settings, 400 – 404, 401 Chance and causation, 183 –186
“BINS” rules for, 401– 402 history of, 315–317 defined, 157–160
Binomial theorem, 400 – 404 myths about, 317–320 nonsense, 184
“BINS” rules, 401– 402 and Normal distributions, 117 and outliers, 161
Block designs, 285–287, 286 Chance, games of. See Cards; Dice; Game and regression, 175–178
Boxplots, 63– 66, 64. See also Graphs shows; Lotteries; Roulette square of (r 2), 180 –183
with calculators, 71–72 Change, percent (%), 90 strategies for interpreting, 160 –164

I-1
I-2 Index

Counting, 387–399 Drug testing, 233–234 General addition rule, 336

combinations, 392–393 and conditional probability, 359–361 General multiplication rule, 354–355
permutations, 391–392 Generalizing, 279–281, 280
Counts, 36 Goodall, Jane, 6, 13
observed vs. expected, 472 E Goodness of fit, 471– 476
CPS. See Current Population Survey (CPS) Empirical rule. See 68–95–99.7 rule with calculators, 478– 479
Critical values (t*) [t distributions], 504–505 Equal treatments, 275–276 chi-square (x2) tests for, 472, 477– 480
Critical values (z*) [Normal distributions], Error curves. See Normal curves Gosset, William S., 503
431– 432, 431t Errors. See also Sampling errors Graphs, 36 –59. See also Data
for population means, 499 margins of, 228–231 bar, 36 – 41
Current Population Survey (CPS), 8, 11, 244 nonsampling, 237, 239–243 boxplots, 63– 66
roundoff, 37 and comparing distributions, 75 –77
Estimation, 220 –228 density curves, 107–109
D
and confidence intervals, 421– 426 dotplots, 41– 43
da Vinci, Leonardo, 168
Ethics, in data gathering, 292–305 good practices for, 36, 86, 89–92
Data, 2–18. See also Graphs
Events, 330–331. See also Independent events histograms, 49–57
drawing conclusions from, 18–28, 89–92
mutually exclusive or disjoint, 344 –345 line, 83 – 84
ethics in collecting, 292–305
and Venn diagrams, 339, 339 misleading, 22, 82, 82 – 86, 84, 85
and individuals and variables, 3– 6
Expected counts, 472, 484 pictograms, 85 – 86
and Normal distributions, 116
for chi-square (x2) tests, 476 pie charts, 36 – 41
and observational studies, 6 –7
for two-way tables, 482– 485 regression lines, 169 –175
plausibility of, 89–92
Expected values, 373–374 residual plots, 178 –180, 179, 181
and sampling, 7–10
for binomial distributions, 406 scatterplots, 142–168
standardizing, 103–104
and simulation, 377–380 stemplots, 44 – 49
transformations of, 74 –78
Experiments, 13–14, 257–273, 259. strategies for interpreting, 43,
vs. personal experiences, 3
See also Observational studies 45– 46, 150 –157
Decision making
behavioral and social science,
and conditional probability, 358–359
299–301
Degrees of freedom (df), 474 – 475t, 475 H
and causation, 186 –188
for t distributions, 504, 504 H0. See Null hypotheses (H0)
clinical trials, 260 –263
for two-way tables, 485 Ha. See Alternative hypotheses (Ha)
design of, 265–267, 281–288
Density curves, 107–109, 108, 108. Histograms, 49–57, 50, 50. See also Graphs
double-blind, 275–277
See also Graphs with calculators, 52–53
ethics in, 292–305
areas under, 381 and density curves, 107–109
generalizing from, 279–281
center and spread of, 109–110, 110 Hypotheses
pitfalls in design of, 259–262
Normal distributions as, 116 –118 null vs. alternative, 441– 444
problems encountered in, 277–279
t distributions as, 504 for two-way tables, 482– 485
randomized comparative, 262–265, 275
Dependent variables. See Response variables (y)
real-world, 273–292
Deviations, 43, 150
Explanatory variables (x), 145, 259 I
in scatterplots, 150 –157
Independent events, 323 –324, 350–351
df. See Degrees of freedom (df)
and binomial settings, 401
Dice, 316, 348, 350 F and conditional probability, 350–352
probability models for, 331–332, 332 Factorial (n!), 389 multiplication rule for, 356
Direct cause-and-effect, 185, 186 False negatives, 359 – 361 Independent variables. See Explanatory
Directions of relationships, 150, 151 False positives, 359 – 361 variables (x)
in scatterplots, 150–157 Fermat, Pierre de, 316 Individuals, 3 – 6
Disjoint events, 344 –345 First quartile (Q1), 62 – 63 Inference. See Statistical inference
Distributions, 37. See also Binomial distribu- Fisher, Ronald A., 411 Informed consent, 294 –295
tions; Chi-square (x2) distributions; Five-number summaries, 63 – 66 Institutional review boards, 294
Normal distributions; Probability Forms of relationships, 150, 151 Interactions, 282
distributions; Sampling distributions; in scatterplots, 150 –157 Interquartile range (IQR), 62 – 63
t distributions
Intersections, 341
comparing, 75–77
Intervals, 427. See also Confidence intervals
describing mathematically, 59–81 G
IQ tests. See Stanford-Binet (IQ) tests
measuring location in, 102–116 Gallup polls, 7–8, 201, 435
IQR. See Interquartile range (IQR)
overall patterns of, 43, 45–46, 50 Galton, Francis, 117, 180, 406
skewness of, 50 –51, 474 Galton’s board, 406, 406 – 408
symmetry of, 50 –51, 109–110 Game shows, 329, 387 K
uniform, 110 –111 Games of chance. See Cards; Dice; Lotteries; Kerrich, John, 315
Dotplots, 12, 13, 41– 43, 42. See also Graphs Roulette
Double-blind experiments, 275–277, 276 Gauss, Carl Friedrich, 117 L
Draft lottery, 214 –216 Gaussian distributions. See Normal Lack of realism, in experiments, 280 –281
Dropouts, 278 distributions Law of averages, 319–320
 Index I-3

Law of large numbers, 375–376 Nielsen ratings, 8 Pie charts, 20, 36 – 41, 38, 38. See also Graphs
Least-squares method, 170 95% confidence, 228–229, 425 Pierce, George Washington, 189–190
Least-squares regression lines, 171–175 95% confidence interval, 422, 423–425, 424, Placebo effect, 260 – 261
Leaves, 45, 45 430. See also Confidence intervals Placebos, 13, 260
Leclerc, George-Louis, Count Buffon, 314, Nonadherers, 277 ethics of, 293
445, 458 – 459 Nonresponse, 201, 240 –241 Polls, 7– 8, 201
Levels of confidence (C), 230. See also Nonsampling errors, 237, 239–243 quality of, 237, 249–250
Confidence intervals nonresponse, 240 –241 sample sizes of, 435–436
for parameters, 427 processing, 239 Population distributions, 226
for population means, 505 response, 239–240 Population means (m), 495
Life insurance, 315, 375 Nonsense correlations, 184 tests for, 500–502
Line graphs, 83, 83 – 84. See also Graphs Normal curves, 108, 109, 116–118 Population proportions
Lines. See Regression lines with calculators, 132–134 confidence intervals for, 431– 435
Lotteries, 311–312 and sampling distributions, 380–382 significance tests for, 450 – 451
and chance, 319 Normal distributions, 116 –136. Populations, 7, 221
draft, 214 –216 See also t distributions inferences about, from samples, 220 –222
and probabilities, 393 –394 approximate, 422 sampling from large, 232–234
Lurking variables, 184 –186, 260 calculations for, 128–135 Positive associations, 151
and experiments, 260 critical values for, 431t Prediction
as density curves, 116–118 with calculators, 174 –175
properties of, 74 and causation, 186
M and sampling distributions of sample and regression lines, 169 –170
M. See Medians (M) means, 495–497 and residuals, 178 –180
m. See Population means (m) standard, 123–128 and square of the correlation, 180 –183
McKellar, Danica, 326
nPk. See Permutations (nPk) Probability, 313–315
Margins of error, 228 –231, 230. See also Null hypotheses (H0), 442 – 444 with calculators, 327–328
Confidence intervals; Levels of conditional, 349–361
confidence (C) and risk, 320 – 322
and sampling errors, 241 O rules of, 331–334
Matched-pairs designs, 284 –285 Observational studies, 6–7. See also
simulations of, 322–324
Matching, 268 Experiments
and two-way tables, 334 –337
Means (x– ), 66 – 68, 110 design of, 267–269
unconditional, 350
of density curves (m), 109 –110, 110, 121 Observed counts, 472
using counting methods, 393–395
of Normal curves, 117, 117–118 Old Faithful, 141–142
and Venn diagrams, 339 –343
and outliers, 67 One-sided alternatives, 447
Probability distributions, 370 –386, 371
population vs. sample, 495 Outliers, 43, 150
and expected values, 373 –374
of sampling distributions, 422 – 425 and correlation, 161
sampling, 380 –382
sampling distributions of, 495 – 497 and interquartile ranges, 64
Probability models, 329–331, 330. See also
and skewness, 67– 68 resistance to, 67, 72–74
Binomial distributions; Counting;
vs. medians, 72–74, 151 in scatterplots, 147, 150–157
Probability distributions
Medians (M), 60 – 61, 110 sensitivity to, 72–74, 161, 177, 500
Probability samples, 245–246
of density curves, 109–110, 110 Overall patterns, 43, 45 – 46, 150
Probability theory, 316
and outliers, 67 in scatterplots, 150–151
Processing errors, 239
and skewness, 68 Pseudo-random numbers, 207
vs. means, 72–74, 151 Punnett squares, 477
Mendel, Gregor, 479
P
p. See Parameters ( p)
Meta-analysis, 281
p^. See Statistics ( p^ ) [measurements] Q
M&M candies
nPk. See Permutations (nPk) Q1. See First quartile (Q1)
chi-square (x2) tests for, 470 – 471
P-values, 444 – 447 Q3. See Third quartile (Q3)
significance tests for, 453– 454, 475– 476
calculating, 447– 448 Quantitative variables, 4
Monty Hall problem, 329
for chi-square (x2) distributions, 474 and correlation, 162
Multiplication counting principle, 389
and sample sizes, 458 – 459 and dotplots, 41– 43
Multiplication rule for independent events, 356
Parameters ( p), 221, 421 and histograms, 49–57
Mutually exclusive events, 344 –345
Pascal, Blaise, 316 and stemplots, 44 – 49
dependence of, 351
Patterns, overall, 43, 45 – 46 Quartiles, 61– 63, 62. See also Percentiles
Pearson, Karl, 314 Questions, wording of, 242 –243
N Percent change (%), 90 Quincunx, 406, 406
n!. See Factorial (n!) Percentiles, 103
nCk. See Combinations (nCk) z-scores from, 127 R
Negative associations, 151 Permutations (nPk), 391–392 r. See Correlation (r)
Newcomb, Simon, 384 with calculators, 397 r2. See Square of the correlation (r2)
Newman, Chris, 3 Pictograms, 85, 85–86. See also Graphs Random digits, 207t
I-4 Index

Random samples, 11. See also Simple random Sampling errors, 237–239, 238 Standard Normal table, 124
samples (SRS) random, 237–239 using, 124 –127
stratified, 244 –246 undercoverage, 238 Standardized values, 104. See also
Random sampling errors, 237–239, 238 Sampling frames, 238 z-scores
Random variables, 371 Sampling variability, 20, 201, 222–228 Standardizing data, 103–104
Randomized comparative experiments, Scales, importance of, 84–85 Stanford-Binet (IQ) tests, 136
262–265, 263, 275 Scatterplots, 143 –144, 144. See also Graphs Statistical adjustment, 268–269
Randomness, 315 with calculators, 152–154 Statistical inference, 421. See also
Rates, 36 –37 defined, 143–149 Chi-square (x2) tests; Confidence
Regression, 168–194 strategies for interpreting, 150 –157 intervals; Significance tests
with calculators, 171, 174 –175 Seasonal variations, 84 and Normal distributions, 117
and correlation, 175–178 Sham surgeries, 293 for two-way tables, 482– 485
and outliers, 177 Shapes of distributions, 46, 50 use and misuse of, 453 – 464
Regression equations, 170 –175 under data transformations, 74 –75 Statistical problem-solving
Regression lines, 169, 169 –170 Shoemaker, Allen, 503 process, 19–23
least-squares, 171–175 Significance, statistical, 267 Statistical significance, 267
Relationships levels of, 447 levels of, 447
and causation, 183–186 Significance levels (a), 447, 460 – 461 Statistics [discipline], 3
and correlation, 157–164 and chi-square (x2) distributions, 474 – 475t use and misuse of, 82–94, 453 – 464
and prediction, 178–183 Significance tests, 438 – 453, 439 Statistics ( p^ ) [measurements], 221, 421
and regression, 168–178 with calculators, 450 – 451 sampling distributions for, 380–382
and scatterplots, 142–157 chi-square (x2) tests, 470 – 493 Stemplots, 44 – 49, 45. See also Graphs
between variables, 143 misuse of, 457– 459 Stems, 45, 45
Residual plots, 178 –180, 179. See also Graphs and P-values, 444 – 447 Strata, 244 –246
with calculators, 181 reasoning behind, 439 – 441 Stratified random samples, 244 –246,
Residuals, 178 –180 stating hypotheses for, 441– 444 286–287
Resistance to outliers, 67, 72–74 t tests, 507–512 Strengths of relationships, 150, 151
Response errors, 239 –240 Simple random samples (SRS), 206 –210 in scatterplots, 150 –157
Response variables (y), 145, 259 with calculators, 212–213 Studies. See Clinical trials; Experiments;
Risk, 320 –322 and probability distributions, 380 – 382 Observational studies
and conditional probability, 358 –359 and sampling distributions, 422 – 426 Subjects, 13, 259
Roulette, 338, 338 –339, 373 and t intervals, 505 Sudoku, 101–102
Roundoff errors, 37 with technology, 211–214 Surveys. See Censuses; Polls
Runs, 317–318 Simulations, probability, 322 –324 Symmetry of distributions, 50–51
with calculators, 327–328 for density curves, 109–110, 110
S and expected values, 377–380
s. See Standard deviation (s) 68–95–99.7 rule, 118–123, 119, 120
T
s. See under Standard deviation (s) and sampling distributions, 423
t*. See Critical values (t*) [t distributions]
Sample means (x– ), 495 Skewness, 50–51
t distributions, 503 –505
Sample proportions p^ , 421– 422 of chi-square (x2) distributions, 474
t intervals, 505–507
Sample sizes of density curves, 109 –110, 110
t tests, 507–510
and confidence levels, 435– 437 and medians vs. means, 67– 68, 110
with calculators, 510 –512
and P-values, 458 – 459 Slopes (b), 171–172
two-sample, 509
Sample spaces, 329–330 Slot machines, 377
Table of random digits, 207t
and Venn diagrams, 339, 339 Social science experiments, 299–301
Technology. See Applets; Calculators;
Sample surveys, 7–10, 201 Spreads of distributions, 46, 50
Computers
pitfalls of, 237–238 under data transformations, 74 –75
Third quartile (Q3), 62– 63
real-world, 236 –252 of density curves, 109 –110, 110
Treatments, 23, 259
Samples, 7, 202–219, 221. See also Random and quartiles, 61– 63
equal, 275–276
samples; Simple random samples (SRS) and standard deviations, 66, 68 –71
Tree diagrams, 352, 352–356
biased, 202–206, 214–217 Square of the correlation (r 2), 180 –183.
Trends, 83
convenience, 203 See also Regression
Two-sided alternatives, 447
and inferences about populations, 220 –222 SRS. See Simple random samples (SRS)
Two-way tables, 334–337, 335, 480
information from, 219 –236 Standard deviations (s), 66, 68 –71, 69
chi-square (x2) tests for, 485–488
from large populations, 232–234 of density curves (s), 121
and relationships between categorical
probability, 245–246 of Normal curves (s), 117, 117–118
variables, 480 – 482
real-world, 243 –246 and outliers, 72–74
stratified random, 244 – 246 of sample means, 495
voluntary response, 203 of sampling distributions, 422 – 425 U
Sampling distributions, 380 –382, 381, 422 and skewness, 72–74 Unconditional probability, 350
means and standard deviations of, 422– 425 Standard Normal distributions, Undercoverage, 238 –239
of sample means (x– ), 495– 498 123 –128, 124 Uniform distributions, 110 –111, 111
 Index I-5

Unions, 341 Variances, 69 X

Units Variations, seasonal, 84 x. See Explanatory variables (x)
importance of, 36 Venn diagrams, 339, 339–343 x– . See Means (x– ); Sample means (x– )
for mutually exclusive or disjoint events, x2. See Chi-square (x2) distributions
V 344 –345, 345
Variability, 224 –226, 227 terminology for, 341
Vitruvius, 168 Y
managing, 225–226 y. See Response variables (y)
in sampling, 20, 201, 222–228 Voluntary response samples, 203
Vos Savant, Marilyn, 329, 348 y intercepts (a), 171–172
Variables, 3 –6
categorical, 4
explanatory (x), 145 Z
lurking, 184 –186 W z*. See Critical values (z*) [Normal
quantitative, 4 Wald, Abraham, 162 distributions]
random, 371 Woods, Tiger, 128 z-scores, 103–107, 104, 121
relationships between, 143 Wording of questions, 242–243 from percentiles, 127
response (y), 145 Wunderlich, Carl, 25 and sampling distributions, 382
 Table entry

z Table entry for z is the area under the standard Normal curve to the left of z.

Table A Standard Normal probabilities (continued)

z .00 .01 .02 .03 .04 .05 .06 .07 .08 .09

0.0 .5000 .5040 .5080 .5120 .5160 .5199 .5239 .5279 .5319 .5359
0.1 .5398 .5438 .5478 .5517 .5557 .5596 .5636 .5675 .5714 .5753
0.2 .5793 .5832 .5871 .5910 .5948 .5987 .6026 .6064 .6103 .6141
0.3 .6179 .6217 .6255 .6293 .6331 .6368 .6406 .6443 .6480 .6517
0.4 .6554 .6591 .6628 .6664 .6700 .6736 .6772 .6808 .6844 .6879
0.5 .6915 .6950 .6985 .7019 .7054 .7088 .7123 .7157 .7190 .7224
0.6 .7257 .7291 .7324 .7357 .7389 .7422 .7454 .7486 .7517 .7549
0.7 .7580 .7611 .7642 .7673 .7704 .7734 .7764 .7794 .7823 .7852
0.8 .7881 .7910 .7939 .7967 .7995 .8023 .8051 .8078 .8106 .8133
0.9 .8159 .8186 .8212 .8238 .8264 .8289 .8315 .8340 .8365 .8389
1.0 .8413 .8438 .8461 .8485 .8508 .8531 .8554 .8577 .8599 .8621
1.1 .8643 .8665 .8686 .8708 .8729 .8749 .8770 .8790 .8810 .8830
1.2 .8849 .8869 .8888 .8907 .8925 .8944 .8962 .8980 .8997 .9015
1.3 .9032 .9049 .9066 .9082 .9099 .9115 .9131 .9147 .9162 .9177
1.4 .9192 .9207 .9222 .9236 .9251 .9265 .9279 .9292 .9306 .9319
1.5 .9332 .9345 .9357 .9370 .9382 .9394 .9406 .9418 .9429 .9441
1.6 .9452 .9463 .9474 .9484 .9495 .9505 .9515 .9525 .9535 .9545
1.7 .9554 .9564 .9573 .9582 .9591 .9599 .9608 .9616 .9625 .9633
1.8 .9641 .9649 .9656 .9664 .9671 .9678 .9686 .9693 .9699 .9706
1.9 .9713 .9719 .9726 .9732 .9738 .9744 .9750 .9756 .9761 .9767
2.0 .9772 .9778 .9783 .9788 .9793 .9798 .9803 .9808 .9812 .9817
2.1 .9821 .9826 .9830 .9834 .9838 .9842 .9846 .9850 .9854 .9857
2.2 .9861 .9864 .9868 .9871 .9875 .9878 .9881 .9884 .9887 .9890
2.3 .9893 .9896 .9898 .9901 .9904 .9906 .9909 .9911 .9913 .9916
2.4 .9918 .9920 .9922 .9925 .9927 .9929 .9931 .9932 .9934 .9936
2.5 .9938 .9940 .9941 .9943 .9945 .9946 .9948 .9949 .9951 .9952
2.6 .9953 .9955 .9956 .9957 .9959 .9960 .9961 .9962 .9963 .9964
2.7 .9965 .9966 .9967 .9968 .9969 .9970 .9971 .9972 .9973 .9974
2.8 .9974 .9975 .9976 .9977 .9977 .9978 .9979 .9979 .9980 .9981
2.9 .9981 .9982 .9982 .9983 .9984 .9984 .9985 .9985 .9986 .9986
3.0 .9987 .9987 .9987 .9988 .9988 .9989 .9989 .9989 .9990 .9990
3.1 .9990 .9991 .9991 .9991 .9992 .9992 .9992 .9992 .9993 .9993
3.2 .9993 .9993 .9994 .9994 .9994 .9994 .9994 .9995 .9995 .9995
3.3 .9995 .9995 .9995 .9996 .9996 .9996 .9996 .9996 .9996 .9997
3.4 .9997 .9997 .9997 .9997 .9997 .9997 .9997 .9997 .9997 .9998

T-1
Table B Random digits

Line

101 19223 95034 05756 28713 96409 12531 42544 82853

102 73676 47150 99400 01927 27754 42648 82425 36290
103 45467 71709 77558 00095 32863 29485 82226 90056
104 52711 38889 93074 60227 40011 85848 48767 52573
105 95592 94007 69971 91481 60779 53791 17297 59335
106 68417 35013 15529 72765 85089 57067 50211 47487
107 82739 57890 20807 47511 81676 55300 94383 14893
108 60940 72024 17868 24943 61790 90656 87964 18883
109 36009 19365 15412 39638 85453 46816 83485 41979
110 38448 48789 18338 24697 39364 42006 76688 08708
111 81486 69487 60513 09297 00412 71238 27649 39950
112 59636 88804 04634 71197 19352 73089 84898 45785
113 62568 70206 40325 03699 71080 22553 11486 11776
114 45149 32992 75730 66280 03819 56202 02938 70915
115 61041 77684 94322 24709 73698 14526 31893 32592
116 14459 26056 31424 80371 65103 62253 50490 61181
117 38167 98532 62183 70632 23417 26185 41448 75532
118 73190 32533 04470 29669 84407 90785 65956 86382
119 95857 07118 87664 92099 58806 66979 98624 84826
120 35476 55972 39421 65850 04266 35435 43742 11937
121 71487 09984 29077 14863 61683 47052 62224 51025
122 13873 81598 95052 90908 73592 75186 87136 95761
123 54580 81507 27102 56027 55892 33063 41842 81868
124 71035 09001 43367 49497 72719 96758 27611 91596
125 96746 12149 37823 71868 18442 35119 62103 39244
126 96927 19931 36809 74192 77567 88741 48409 41903
127 43909 99477 25330 64359 40085 16925 85117 36071
128 15689 14227 06565 14374 13352 49367 81982 87209
129 36759 58984 68288 22913 18638 54303 00795 08727
130 69051 64817 87174 09517 84534 06489 87201 97245
131 05007 16632 81194 14873 04197 85576 45195 96565
132 68732 55259 84292 08796 43165 93739 31685 97150
133 45740 41807 65561 33302 07051 93623 18132 09547
134 27816 78416 18329 21337 35213 37741 04312 68508
135 66925 55658 39100 78458 11206 19876 87151 31260
136 08421 44753 77377 28744 75592 08563 79140 92454
137 53645 66812 61421 47836 12609 15373 98481 14592
138 66831 68908 40772 21558 47781 33586 79177 06928
139 55588 99404 70708 41098 43563 56934 48394 51719
140 12975 13258 13048 45144 72321 81940 00360 02428
141 96767 35964 23822 96012 94591 65194 50842 53372
142 72829 50232 97892 63408 77919 44575 24870 04178
143 88565 42628 17797 49376 61762 16953 88604 12724
144 62964 88145 83083 69453 46109 59505 69680 00900
145 19687 12633 57857 95806 09931 02150 43163 58636
146 37609 59057 66967 83401 60705 02384 90597 93600
147 54973 86278 88737 74351 47500 84552 19909 67181
148 00694 05977 19664 65441 20903 62371 22725 53340
149 71546 05233 53946 68743 72460 27601 45403 88692
150 07511 88915 41267 16853 84569 79367 32337 03316

T-2
 Probability
p

Table entry for p and C is the critical value t* with probability p lying to its right and
t* probability C lying between 2t* and t*.

Table C t distribution critical values

Upper tail probability p

df .25 .20 .15 .10 .05 .025 .02 .01 .005 .0025 .001 .0005

1 1.000 1.376 1.963 3.078 6.314 12.71 15.89 31.82 63.66 127.3 318.3 636.6
2 0.816 1.061 1.386 1.886 2.920 4.303 4.849 6.965 9.925 14.09 22.33 31.60
3 0.765 0.978 1.250 1.638 2.353 3.182 3.482 4.541 5.841 7.453 10.21 12.92
4 0.741 0.941 1.190 1.533 2.132 2.776 2.999 3.747 4.604 5.598 7.173 8.610
5 0.727 0.920 1.156 1.476 2.015 2.571 2.757 3.365 4.032 4.773 5.893 6.869
6 0.718 0.906 1.134 1.440 1.943 2.447 2.612 3.143 3.707 4.317 5.208 5.959
7 0.711 0.896 1.119 1.415 1.895 2.365 2.517 2.998 3.499 4.029 4.785 5.408
8 0.706 0.889 1.108 1.397 1.860 2.306 2.449 2.896 3.355 3.833 4.501 5.041
9 0.703 0.883 1.100 1.383 1.833 2.262 2.398 2.821 3.250 3.690 4.297 4.781
10 0.700 0.879 1.093 1.372 1.812 2.228 2.359 2.764 3.169 3.581 4.144 4.587
11 0.697 0.876 1.088 1.363 1.796 2.201 2.328 2.718 3.106 3.497 4.025 4.437
12 0.695 0.873 1.083 1.356 1.782 2.179 2.303 2.681 3.055 3.428 3.930 4.318
13 0.694 0.870 1.079 1.350 1.771 2.160 2.282 2.650 3.012 3.372 3.852 4.221
14 0.692 0.868 1.076 1.345 1.761 2.145 2.264 2.624 2.977 3.326 3.787 4.140
15 0.691 0.866 1.074 1.341 1.753 2.131 2.249 2.602 2.947 3.286 3.733 4.073
16 0.690 0.865 1.071 1.337 1.746 2.120 2.235 2.583 2.921 3.252 3.686 4.015
17 0.689 0.863 1.069 1.333 1.740 2.110 2.224 2.567 2.898 3.222 3.646 3.965
18 0.688 0.862 1.067 1.330 1.734 2.101 2.214 2.552 2.878 3.197 3.611 3.922
19 0.688 0.861 1.066 1.328 1.729 2.093 2.205 2.539 2.861 3.174 3.579 3.883
20 0.687 0.860 1.064 1.325 1.725 2.086 2.197 2.528 2.845 3.153 3.552 3.850
21 0.686 0.859 1.063 1.323 1.721 2.080 2.189 2.518 2.831 3.135 3.527 3.819
22 0.686 0.858 1.061 1.321 1.717 2.074 2.183 2.508 2.819 3.119 3.505 3.792
23 0.685 0.858 1.060 1.319 1.714 2.069 2.177 2.500 2.807 3.104 3.485 3.768
24 0.685 0.857 1.059 1.318 1.711 2.064 2.172 2.492 2.797 3.091 3.467 3.745
25 0.684 0.856 1.058 1.316 1.708 2.060 2.167 2.485 2.787 3.078 3.450 3.725
26 0.684 0.856 1.058 1.315 1.706 2.056 2.162 2.479 2.779 3.067 3.435 3.707
27 0.684 0.855 1.057 1.314 1.703 2.052 2.158 2.473 2.771 3.057 3.421 3.690
28 0.683 0.855 1.056 1.313 1.701 2.048 2.154 2.467 2.763 3.047 3.408 3.674
29 0.683 0.854 1.055 1.311 1.699 2.045 2.150 2.462 2.756 3.038 3.396 3.659
30 0.683 0.854 1.055 1.310 1.697 2.042 2.147 2.457 2.750 3.030 3.385 3.646
40 0.681 0.851 1.050 1.303 1.684 2.021 2.123 2.423 2.704 2.971 3.307 3.551
50 0.679 0.849 1.047 1.299 1.676 2.009 2.109 2.403 2.678 2.937 3.261 3.496
60 0.679 0.848 1.045 1.296 1.671 2.000 2.099 2.390 2.660 2.915 3.232 3.460
80 0.678 0.846 1.043 1.292 1.664 1.990 2.088 2.374 2.639 2.887 3.195 3.416
100 0.677 0.845 1.042 1.290 1.660 1.984 2.081 2.364 2.626 2.871 3.174 3.390
1000 0.675 0.842 1.037 1.282 1.646 1.962 2.056 2.330 2.581 2.813 3.098 3.300
z* 0.674 0.841 1.036 1.282 1.645 1.960 2.054 2.326 2.576 2.807 3.091 3.291

50% 60% 70% 80% 90% 95% 96% 98% 99% 99.5% 99.8% 99.9%

Confidence level C
T-3
 Table D Critical values of z*

Confidence level C Critical value z* Confidence level C Critical value z*

50% 0.67 90% 1.64

60% 0.84 95% 1.96
70% 1.04 99% 2.58
80% 1.28 99.9% 3.29

Table E Chi-square critical values

Significance level a

df 0.25 0.20 0.15 0.10 0.05 0.01 0.001

1 1.32 1.64 2.07 2.71 3.84 6.63 10.83
2 2.77 3.22 3.79 4.61 5.99 9.21 13.82
3 4.11 4.64 5.32 6.25 7.81 11.34 16.27
4 5.39 5.99 6.74 7.78 9.49 13.28 18.47
5 6.63 7.29 8.12 9.24 11.07 15.09 20.51
6 7.84 8.56 9.45 10.64 12.59 16.81 22.46
7 9.04 9.80 10.75 12.02 14.07 18.48 24.32
8 10.22 11.03 12.03 13.36 15.51 20.09 26.12
9 11.39 12.24 13.29 14.68 16.92 21.67 27.88

T-4
Table entry

z Table entry for z is the area under the standard Normal curve to the left of z.

Table A Standard Normal probabilities

z .00 .01 .02 .03 .04 .05 .06 .07 .08 .09

3.4 .0003 .0003 .0003 .0003 .0003 .0003 .0003 .0003 .0003 .0002
3.3 .0005 .0005 .0005 .0004 .0004 .0004 .0004 .0004 .0004 .0003
3.2 .0007 .0007 .0006 .0006 .0006 .0006 .0006 .0005 .0005 .0005
3.1 .0010 .0009 .0009 .0009 .0008 .0008 .0008 .0008 .0007 .0007
3.0 .0013 .0013 .0013 .0012 .0012 .0011 .0011 .0011 .0010 .0010
2.9 .0019 .0018 .0018 .0017 .0016 .0016 .0015 .0015 .0014 .0014
2.8 .0026 .0025 .0024 .0023 .0023 .0022 .0021 .0021 .0020 .0019
2.7 .0035 .0034 .0033 .0032 .0031 .0030 .0029 .0028 .0027 .0026
2.6 .0047 .0045 .0044 .0043 .0041 .0040 .0039 .0038 .0037 .0036
2.5 .0062 .0060 .0059 .0057 .0055 .0054 .0052 .0051 .0049 .0048
2.4 .0082 .0080 .0078 .0075 .0073 .0071 .0069 .0068 .0066 .0064
2.3 .0107 .0104 .0102 .0099 .0096 .0094 .0091 .0089 .0087 .0084
2.2 .0139 .0136 .0132 .0129 .0125 .0122 .0119 .0116 .0113 .0110
2.1 .0179 .0174 .0170 .0166 .0162 .0158 .0154 .0150 .0146 .0143
2.0 .0228 .0222 .0217 .0212 .0207 .0202 .0197 .0192 .0188 .0183
1.9 .0287 .0281 .0274 .0268 .0262 .0256 .0250 .0244 .0239 .0233
1.8 .0359 .0351 .0344 .0336 .0329 .0322 .0314 .0307 .0301 .0294
1.7 .0446 .0436 .0427 .0418 .0409 .0401 .0392 .0384 .0375 .0367
1.6 .0548 .0537 .0526 .0516 .0505 .0495 .0485 .0475 .0465 .0455
1.5 .0668 .0655 .0643 .0630 .0618 .0606 .0594 .0582 .0571 .0559
1.4 .0808 .0793 .0778 .0764 .0749 .0735 .0721 .0708 .0694 .0681
1.3 .0968 .0951 .0934 .0918 .0901 .0885 .0869 .0853 .0838 .0823
1.2 .1151 .1131 .1112 .1093 .1075 .1056 .1038 .1020 .1003 .0985
1.1 .1357 .1335 .1314 .1292 .1271 .1251 .1230 .1210 .1190 .1170
1.0 .1587 .1562 .1539 .1515 .1492 .1469 .1446 .1423 .1401 .1379
0.9 .1841 .1814 .1788 .1762 .1736 .1711 .1685 .1660 .1635 .1611
0.8 .2119 .2090 .2061 .2033 .2005 .1977 .1949 .1922 .1894 .1867
0.7 .2420 .2389 .2358 .2327 .2296 .2266 .2236 .2206 .2177 .2148
0.6 .2743 .2709 .2676 .2643 .2611 .2578 .2546 .2514 .2483 .2451
0.5 .3085 .3050 .3015 .2981 .2946 .2912 .2877 .2843 .2810 .2776
0.4 .3446 .3409 .3372 .3336 .3300 .3264 .3228 .3192 .3156 .3121
0.3 .3821 .3783 .3745 .3707 .3669 .3632 .3594 .3557 .3520 .3483
0.2 .4207 .4168 .4129 .4090 .4052 .4013 .3974 .3936 .3897 .3859
0.1 .4602 .4562 .4522 .4483 .4443 .4404 .4364 .4325 .4286 .4247
0.0 .5000 .4960 .4920 .4880 .4840 .4801 .4761 .4721 .4681 .4641
(continued)