Introduction

to Probability
Mark Daniel Ward
Associate Professor of Statistics
Purdue University

Ellen Gundlach
Education Specialist and Continuing Lecturer in Statistics
Purdue University

W. H. Freeman and Company

A Macmillan Education Imprint
Publisher: Terri Ward
Senior Acquisitions Editor: Karen Carson
Marketing Manager: Cara LeClair
Developmental Editor: Katrina Mangold
Editorial Assistant: Victoria Garvey
Photo Editor: Cecilia Varas
Cover Designer: Diana Blume
Text Designer: Mark Daniel Ward
Project Editor: Liz Geller
Copyeditor: Elizabeth Marraffino
Illustrations: Mark Daniel Ward; Eli Ensor
Illustration Coordinator: Matthew McAdams
Production Supervisor: Susan Wein
Printing and Binding: Courier Printing
Cover Photo: Matthias Pahl/Shutterstock

Library of Congress Control Number: 2015935799

ISBN-13: 978-0-7167-7109-8
ISBN-10: 0-7167-7109-8
©2016 by W. H. Freeman & Company
All rights reserved

Printed in the United States of America

First printing

W. H. Freeman and Company

41 Madison Avenue
New York, NY 10010
www.macmillanhighered.com
To my wife, Laura,
and to our children:
Bruce, Audrey, Mary, Luke, and Dean.
—MDW

To my boys, Callum and Philip;

to my parents, John and Dianne;
and to Judd.
—EG
this page left intentionally blank
Contents

Preface xiii

Notation Review xix

Math Review xxi

I Randomness 1
1 Outcomes, Events, and Sample Spaces 3
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Complements and DeMorgan’s Laws . . . . . . . . . . . . . . . . 11
1.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2 Probability 17
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Equally Likely Events . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 Complements; Probabilities of Subsets . . . . . . . . . . . . . . . 24
2.4 Inclusion-Exclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5 More Examples of Probabilities of Events . . . . . . . . . . . . . 27
2.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

3 Independent Events 37
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2 Some Nice Facts about Independence . . . . . . . . . . . . . . . . 42
3.3 Probability of Good Occurring before Bad . . . . . . . . . . . . . 43
3.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4 Conditional Probability 48
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2 Distributive Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.3 Conditional Probabilities Satisfy the Probability Axioms . . . . . 54
4.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

5 Bayes’ Theorem 60
5.1 Introduction to Versions of Bayes’ Theorem . . . . . . . . . . . . 60
5.2 Multiplication with Conditional Probabilities . . . . . . . . . . . 66
5.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

v
vi Contents

6 Review of Randomness 73
6.1 Summary of Randomness . . . . . . . . . . . . . . . . . . . . . . 73
6.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

II Discrete Random Variables 77

7 Discrete Versus Continuous Random Variables 78

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
7.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

8 Probability Mass Functions and CDFs 87

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
8.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
8.3 Properties of the Mass and CDF . . . . . . . . . . . . . . . . . . 91
8.4 More Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
8.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

9 Independence and Conditioning 102

9.1 Joint Probability Mass Functions . . . . . . . . . . . . . . . . . . 102
9.2 Independent Random Variables . . . . . . . . . . . . . . . . . . . 107
9.3 Three or More Independent Random Variables . . . . . . . . . . 112
9.4 Conditional Probability Mass Functions . . . . . . . . . . . . . . 113
9.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

10 Expected Values of Discrete Random Variables 120

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
10.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
10.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

11 Expected Values of Sums of Random Variables 130

11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
11.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
11.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

12 Variance of Discrete Random Variables 142

12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
12.2 Variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
12.3 Five Friendly Facts with Independence . . . . . . . . . . . . . . . 152
12.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

13 Review of Discrete Random Variables 161

13.1 Summary of Discrete Random Variables . . . . . . . . . . . . . . 161
13.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
 Contents vii

III Named Discrete Random Variables 171

14 Bernoulli Random Variables 173

14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
14.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
14.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178

15 Binomial Random Variables 182

15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
15.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
15.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190

16 Geometric Random Variables 197

16.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
16.2 Special Features of the Geometric Distribution . . . . . . . . . . 204
16.3 The Number of Failures . . . . . . . . . . . . . . . . . . . . . . . 204
16.4 Geometric Memoryless Property . . . . . . . . . . . . . . . . . . 205
16.5 Random Variables That Are Not Geometric . . . . . . . . . . . . 207
16.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

17 Negative Binomial Random Variables 214

17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
17.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
17.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219

18 Poisson Random Variables 224

18.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
18.2 Sums of Independent Poisson Random Variables . . . . . . . . . 229
18.3 Using the Poisson as an Approximation to the Binomial . . . . . 230
18.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233

19 Hypergeometric Random Variables 241

19.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
19.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
19.3 Binomial Approximation to the Hypergeometric . . . . . . . . . . 247
19.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248

20 Discrete Uniform Random Variables 254

20.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
20.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
20.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257

21 Review of Named Discrete Random Variables 259

21.1 Summing-up: How To Tell Random Variables Apart . . . . . . . 259
21.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
21.3 Review Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
viii Contents

IV Counting 271

22 Introduction to Counting 272

22.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
22.2 Order and Replacement in Sampling . . . . . . . . . . . . . . . . 276
22.3 Counting: Seating Arrangements . . . . . . . . . . . . . . . . . . 280
22.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285

23 Two Case Studies in Counting 293

23.1 Poker Hands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
23.2 Yahtzee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295

V Continuous Random Variables 299

24 Continuous Random Variables and PDFs 300

24.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
24.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
24.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313

25 Joint Densities 319

25.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
25.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
25.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328

26 Independent Continuous Random Variables 333

26.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
26.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
26.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339

27 Conditional Distributions 345

27.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
27.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
27.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351

28 Expected Values of Continuous Random Variables 355

28.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
28.2 Some Generalizations about Expected Values . . . . . . . . . . . 358
28.3 Some Applied Problems with Expected Values . . . . . . . . . . . 359
28.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361

29 Variance of Continuous Random Variables 366

29.1 Variance of a Continuous Random Variable . . . . . . . . . . . . 366
29.2 Expected Values of Functions of One Random Variable . . . . . . 370
29.3 Expected Values of Functions of Two Random Variables . . . . . 372
29.4 More Friendly Facts about Continuous Random Variables . . . . 376
29.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
 Contents ix

30 Review of Continuous Random Variables 384

30.1 Summary of Continuous Random Variables . . . . . . . . . . . . 384
30.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386

VI Named Continuous Random Variables 389

31 Continuous Uniform Random Variables 391

31.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
31.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
31.3 Linear Scaling of a Uniform Random Variable . . . . . . . . . . . 403
31.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404

32 Exponential Random Variables 412

32.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
32.2 Average and Variance . . . . . . . . . . . . . . . . . . . . . . . . . 413
32.3 Properties of Exponential Random Variables . . . . . . . . . . . . 417
32.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423

33 Gamma Random Variables 428

33.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428
33.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
33.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434

34 Beta Random Variables 438

34.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
34.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
34.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443

35 Normal Random Variables 445

35.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
35.2 Transforming Normal Random Variables . . . . . . . . . . . . . . 449
35.3 “Backward” Normal Problems . . . . . . . . . . . . . . . . . . . . 461
35.4 Summary: “Forward” Vs. “Backward” Normal . . . . . . . . . . . 464
35.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465

36 Sums of Independent Normal Random Variables 470

36.1 Sums of Independent Normal Random Variables . . . . . . . . . . 470
36.2 Why a Sum of Independent Normals Is Normal (Optional) . . . . 476
36.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479

37 Central Limit Theorem 483

37.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
37.2 Laws of Large Numbers . . . . . . . . . . . . . . . . . . . . . . . 484
37.3 Central Limit Theorem . . . . . . . . . . . . . . . . . . . . . . . . 484
37.4 CLT for Sums of Continuous Random Variables . . . . . . . . . . 485
37.5 CLT for Sums of Discrete Random Variables . . . . . . . . . . . . 488
x Contents

37.6 Approximations of Binomial Random Variables . . . . . . . . . . 492

37.7 Approximations of Poisson Random Variables . . . . . . . . . . . 496
37.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500

38 Review of Named Continuous Random Variables 505

38.1 Summing-up: How To Tell Random Variables Apart . . . . . . . 505
38.2 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507

VII Additional Topics 515

39 Variance of Sums; Covariance; Correlation 516

39.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516
39.2 Motivation for Covariance . . . . . . . . . . . . . . . . . . . . . . 517
39.3 Properties of the Covariance . . . . . . . . . . . . . . . . . . . . . 518
39.4 Examples of Covariance . . . . . . . . . . . . . . . . . . . . . . . 521
39.5 Linearity of the Covariance . . . . . . . . . . . . . . . . . . . . . 530
39.6 Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531
39.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537

40 Conditional Expectation 541

40.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541
40.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 542
40.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 550

41 Markov and Chebyshev Inequalities 554

41.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554
41.2 Markov Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . 554
41.3 Chebyshev Inequality . . . . . . . . . . . . . . . . . . . . . . . . . 557
41.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 561

42 Order Statistics 565

42.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 565
42.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 566
42.3 Joint Density and Joint CDF of Order Statistics . . . . . . . . . 568
42.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 578

43 Moment Generating Functions 582

43.1 A Brief Introduction to Generating Functions . . . . . . . . . . . 582
43.2 Moment Generating Functions . . . . . . . . . . . . . . . . . . . . 584
43.3 Moment Generating Function: Discrete Case . . . . . . . . . . . 586
43.4 Moment Generating Function: Continuous Case . . . . . . . . . . 588
43.5 Appendix: Building a Generating Function . . . . . . . . . . . . 593
43.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 594

44 Transformations of One or Two Random Variables 596

44.1 The Distribution of a Function of One Random Variable . . . . . 596
 Contents xi

44.2 The Distributions of Functions of Two Random Variables . . . . 601

44.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 609

45 Review Questions for All Chapters 613

Answers to Exercises 621

Bibliography 647

Index 649
this page left intentionally blank
Preface

We want to briefly justify why there should be another probability book, when
so many others are available.
Motivation: Students from majors in the mathematical sciences and in other
areas will be more engaged with the material if they are studying problems
that are relevant to them. While testing drafts of the book in the classroom,
the students who used this book were asked to contribute questions. As a
result, many of the exercises in this text began with questions motivated by the
students’ own interests.
Example- and exercise-oriented approach: Our book serves as a student’s
first introduction to probability theory, so we devote significant attention to a
wealth of exercises and examples. We encourage students to practice their skills
by solving lots of questions. Our exercises are split into practice, extensions,
and advanced types of questions. We recommend assigning a small number of
questions to students on a daily basis. This promotes more interactive discus-
sion in class between the students and the instructor. It consistently empowers
the students to try their hand at some problems of their own. It reduces stress
and “cramming” at exam time, as the students consistently develop their under-
standing during the course. It also provides a firm foundation for the students’
long term understanding of probability. The exercises, theorems, definitions,
and remarks are all numbered in one list (instead of numbered separately) be-
cause we believe that they all should be used in tandem to understand the
chapter material.
Relationship between events and random variables: The jump from events
to random variables is often a “leap” in other texts. In the present book, we
devote significant attention to outcomes, events, sample spaces, and probabili-
ties, before moving on to random variables. Random variables are introduced
explicitly as real-valued functions on the sample space, i.e., as functions that
depend on the outcome. The notation X(ω) is used to introduce a random
variable at first (where ω is an outcome), so that students can more easily make
a transition from studying outcomes to studying random variables that depend
on such outcomes.
Jointly distributed random variables: Many probability texts first emphasize
properties of one discrete random variable, followed by properties of one contin-

xiii
xiv Preface

uous random variable, and finally return to jointly distributed random variables.
We believe, in contrast, that a firm understanding of jointly distributed random
variables, from the very beginning, is most helpful for the students’ comprehen-
sive understanding of the material. Using jointly distributed random variables
at an early stage of the book also allows for more intuitive definitions of some of
the concepts. For instance, many probability texts introduce Binomial random
variables by explaining the mass, which requires a good grasp of Binomial co-
efficients, i.e., of nk . We believe, however, that Binomial random variables are

introduced more intuitively by X = X1 + · · · + Xn , where the Xj ’s are indicator

random variables. This requires the students to have familiarity with more than
one random variable at a time (i.e., to understand joint distributions), but it
allows the students to be more versatile in their thinking. It promotes the un-
derstanding of “big picture” kinds of insights. Students begin to think, from the
very start, about the ways in which random variables are related, and the ways
in which collections of random variables can give rise to new random variables.

Counting: Many probability books start with counting, which means this
material is taught during the first two or three weeks of a course. Unfortu-
nately, this means that a student who is weighing her/his interest in a course
will not even begin to grasp the concepts of randomness until the registration
period is over. This leads to attrition. Moreover, some questions in counting
are best understood from a probabilistic point of view, using (for instance) the
linearity of expectation, which is not available to the students at the start of the
course. As an example, consider how many couples are expected to sit together,
when people sit uniformly at random in a circle. This question can be answered
very succinctly, using indicator functions and the linearity of expectation. (The
probability mass function, in contrast, is cumbersome to compute.) In general,
we believe that our approach to counting is significantly enhanced by the use
of sums of indicator random variables. Therefore, we focus on counting after
we finish a thorough treatment of discrete random variables, but before moving
onwards to continuous random variables. This allows the students to feel con-
fident in their understanding of the discrete world before they tackle difficult
counting questions. We emphasize to students that combinatorics is a deep and
beautiful subject (much of the Ward’s research is motivated by problems in ap-
plied discrete mathematics). We also try to emphasize the connections between
discrete random variables and counting. We firmly believe that this is best
accomplished when the students already understand discrete random variables.

Comparison/summary chapters: A first course in probability theory can feel

like a whirlwind tour. Thus, we have checkpoints throughout the book, where
material is summarized and reviewed. This helps to ground the reader and
build confidence. It also helps the students discriminate between the commonly
confused distributions and counting techniques, mass functions vs. CDFs, etc.
These summaries are useful while reviewing for examinations, e.g., the Actuarial
P/1 exam given by the Society of Actuaries and the Casualty Actuarial Society.
Passing the P/1 exam requires knowledge of all the material in this text. We
 Preface xv

also guide the students through ways to tell which kind of distribution they are
working with. We give suggestions about how to grasp nuances in a problem that
make it a Binomial or Geometric or Negative Binomial situation. These guides
help students home in on what separates these distributions. We summarize
each distribution for quick reference, but we also go into details to explain each
discrete distribution’s mass, expected value, variance, etc.
Our students have had an excellent experience using this probability book.
Several of our own students have already passed the SOA/CAS P/1 exam after
having learned probability using only the early drafts of this book. Our students
seem to enjoy the many examples and friendly tone. We hope that you and your
students also find our book approachable and thorough. We are delighted by the
kind reception that our students and colleagues have given to the book during
its pilot tests.
We have divided our book into seven main parts:
Part I: Randomness. We introduce outcomes, events, sample spaces, basic
probability rules, independence, conditional probabilities, and Bayes’ Theorem.
Part II: Discrete Random Variables. We discuss the difference between dis-
crete and continuous random variables and introduce probability mass function,
cumulative distribution function, expected value, variance, and joint distribu-
tions for discrete random variables.
Part III: Named Discrete Random Variables. We consider ways to dis-
tinguish between—and perform calculations with—the most common discrete
random variables: Bernoulli, Binomial, Geometric, Negative Binomial, Poisson,
Hypergeometric, and Discrete Uniform. We include a review chapter to help
students see the similarities and differences between all of these distributions.
Part IV: Counting. We use indicator variables and the linearity of expecta-
tion as tools to help tackle several different types of counting problems: sam-
pling with and without replacement; when order matters and doesn’t matter;
and rearrangement problems. We have case studies on poker and Yahtzee, two
popular games many students will recognize.
Part V: Continuous Random Variables. We reinforce the difference between
discrete and continuous random variables. Then we introduce the probability
density function, cumulative distribution function, expected value, variance,
and joint and conditional distributions for continuous random variables.
Part VI: Named Continuous Random Variables. We show ways to utilize
(and quickly make distinctions between) the most common continuous random
variables: Continuous Uniform, Exponential, Gamma, Beta and Normal. We
show how the Central Limit Theorems and Laws of Large Numbers work. We
include a review chapter to help students see the similarities and differences
between all of these continuous distributions and between some of the continuous
and discrete distributions.
xvi Preface

Part VII: Additional Topics. Here we cover more advanced topics that could
be optional, depending on how much time an instructor has in a semester or
quarter. We treat the distribution of a function of one continuous random vari-
able, the variance of sums of random variables, correlation, conditional expecta-
tion, Markov and Chebyshev Inequalities, order statistics, moment generating
functions, and the joint density of two random variables that are functions of
another pair of random variables.

Acknowledgements
We thank our students for their feedback, as well as their contributions to the
exercises in this text. These students are:
Probability: The Science Of Uncertainty (HONR 399, Fall 2010) J. Blair,
J. Covalt, S. Fancher, C. Fleming, B. Goosman, W. Hess, E. Hoffman, C. Hol-
comb, A. Hurlock, E. Jenkins, J. Ling, D. Mo, B. Morgan, C. Mullen, S. Muss-
mann, D. Rouleau, C. Sanor, V. Savikhin, S. Sheafer, S. Spence, R. Stevick,
J. Wu, S. Yap, M. Zachman, B. Zhou.
Probability (MA/STAT 416, Fall 2011) K. Amstutz, C. Ben, M. Boing,
K. Breneman, P. Coghlan, B. Copeland, O. El-Ghirani, M. Fronek, Z. Gao,
A. Gerardi, T. He, X. He, K. Hopkins, J. Jaagosild, N. Johnson, K. Kemmer-
ling, J. Kwong, N. Lawrence, K. Leow, B. Lewis, K. Li, S. Li, Y. Liang, E. Luo
Cao, M. Mazlan, J. Milligan, E. Myers, Y. Nazarbekov, C. Ng, R. Schwartz,
S. Scott, S. Seffrin, J. Sheng, D. Snyder, T. Sun, B. Wilson, E. Winkowska,
R. Xue, Y. Yang, G. Zhao.
We heartily thank our colleague Dr. Frederi Viens (Purdue University) for
piloting the book in his STAT/MA 416 course in Fall 2011 and again in Spring
2013. We extend our thanks also to Dr. David Galvin (University of Notre
Dame), Dr. Patricia Humphrey (Georgia Southern University), Dr. Hosam Mah-
moud (The George Washington University), and Dr. Meike Neiderhausen (Uni-
versity of Portland), for their willingness to test our book in Fall 2012 in several
different types of courses at their own universities. These colleagues and their
students provided very valuable feedback about early drafts of the book. We
benefited a great deal from their guidance, insight, and suggestions.
This book was genuinely a team effort. We are very grateful to Deborah
Sutton for preparing the answers to the odd-numbered exercises. Dr. Patri-
cia Humphrey provided numerous insights about all aspects of the content and
checked the accuracy of the entire book. Pat also produced the solutions in
the student and instructor manuals. We appreciate Jackie Miller’s insights and
ideas at the beginning of this project. We are thankful for the support of Terri
Ward, our publisher at W. H. Freeman, and for the contributions of the entire
W. H. Freeman team. In particular, Liz Geller and Elizabeth Marraffino pro-
vided hundreds of helpful suggestions for improvement. We are also especially
thankful for the marketing expertise of Karen Carson, Cara LeClair, and Kat-
 Preface xvii

rina Mangold. We are grateful to Diana Blume and Matt McAdams for their
help with the design, layout, and art, and to Susan Wein for making sure our
files were technically sound and printer-ready.
We are also thankful for the feedback of the following reviewers:
David Anderson, University of Wisconsin
Alireza Arasteh, Western New Mexico University
G. Jogesh Babu, Pennsylvania State University
Christian Beneš, Brooklyn College of the City University of New York
Kiran R. Bhutani, The Catholic University of America
Ken Bosworth, Idaho State University
Daniel Conus, Lehigh University
Michelle Cook, Stephen F. Austin State University
Dana Draghicescu, City University of New York, Hunter College
Randy Eubank, Arizona State University
Marian Frazier, Gustavus Adolphus College
Rohitha Goonatilake, Texas A&M University
Patrick Gorman, Kutztown University
Ross Gosky, Appalachian State University
Marc Goulet, University of Wisconsin–Eau Claire
Susan Herring, Sonoma State University
Christopher Hoffman, University of Washington
Patricia Humphrey, Georgia Southern University
Ahmad Kamalvand, Huston-Tillotson University
Judy Kasabian, El Camino College
Syed Kirmani, University of Northern Iowa
Charles Lindsey, Florida Gulf Coast University
Susan Martonosi, Harvey Mudd College
Ronald Mellado Miller, Brigham Young University–Hawaii
Joseph Mitchell, Stony Brook University
Sumona Mondal, Clarkson University
Cindy Moss, Skyline College
Meike Niederhausen, University of Portland
Will Perkins, Georgia Institute of Technology
Rebecca L. Pierce, Ball State University
David J. Rader, Jr., Rose-Hulman Institute of Technology
Aaron Robertson, Colgate University
Charles A. Rohde, Johns Hopkins University
Seyed Roosta, Albany State University
Juana Sanchez, University of California, Los Angeles
Yiyuan She, Florida State University
Therese Shelton, Southwestern University
A. Robert Sinn, University of North Georgia
Clifton D. Sutton, George Mason University
Mahbobeh Vezvaei, Kent State University
xviii Preface

Frederi Viens, Purdue University

P. Patrick Wang, University of Alabama
Daniel Weiner, Boston University
Alison Weir, University of Toronto, Mississauga
Reem Yassawi, Trent University
Norbert Youmbi, Saint Francis University
M. D. Ward’s work is supported by the National Science Foundation grants
#0939370, #1140489, and #1246818.

We welcome readers’ feedback. Please contact us at [email protected] .

We will maintain a list of all the corrections and suggestions sent to us, at
http://www.stat.purdue.edu/~mdw/book.html .
Notation Review

Notation for Named Sets of Numbers:

Z≥0 the set of nonnegative integers Z≥0 = {0, 1, 2, . . .}

Z the set of integers, Z = {. . . , −2, −1, 0, 1, 2, . . .}
N the natural (a.k.a. counting) numbers, N = {1, 2, . . .}
R>0 the set of positive real numbers
R≥0 the set of nonnegative real numbers
R the set of real numbers, including positive
and negative numbers, i.e., whole numbers, fractions,
decimals, roots (although not imaginary), and
transcendental numbers like π and e.

Notation for Events:

sets: {things in the set | conditions on those things};

e.g., S = {(x, y) | x + y = 2}.
∅ the empty set, i.e., event with no outcomes
e.g., A ∩ B = ∅ if A, B have no outcomes in common;
see the definition of ∩ below to clarify further
S the sample space, i.e., event with all outcomes
ω an outcome in the sample space
∈ inclusion in an event (i.e., x ∈ A if outcome x is in event A)
⊂ subset (i.e., A ⊂ B if every outcome of A is also in B)
|A| the number of outcomes (also called the size) of event A

B
A

xix
xx Notation Review

Notation for Building New Events Using Known Ones:

c complement of an event, which corresponds with the word “not”
i.e., x is in Ac exactly when x is not in A

A Ac

\ setminus (i.e., B \ A = B ∩ Ac ;
the event containing outcomes in B that are not in A)
∪ union of events, which corresponds with the word “or,”
i.e., A ∪ B is the event containing outcomes in A or B or both.
The set A ∪ B corresponds to the region containing
shading, lines, or both, in the figure below.
∩ intersection of events, which corresponds with the word “and,”
i.e., A ∩ B is the event containing outcomes in A and B.
The set A ∩ B is the region that contains lines
overlapping the shading in the figure below.

S
B
A

Notation for Random Variables:

X a random variable (always written in capital letters)

x a value that a random variable might take on
P (A) probability that event A occurs
P (X = x) is a shorthand notation for P ({ω | X(ω) = x})
E(X) = µ expected value of X
Var(X) = σ 2 variance of X
Math Review

Geometric Sums
For −1 < a < 1, recall these two finite summations of geometric terms,
r
2 3 r
X 1 − ar+1
1 + a + a + a + ··· + a = aj =
1−a
j=0

and
r
2 3 4 r
X a − ar+1
a + a + a + a + ··· + a = aj = .
1−a
j=1

For −1 < a < 1, these yield two infinite summations of geometric terms,
∞
2 3
X 1
1 + a + a + a + ··· = aj =
1−a
j=0

and
∞
2 3 4
X a
a + a + a + a + ··· = aj = .
1−a
j=1

Exponential Function
For any real-valued x, the power series definition of the exponential function
evaluated at x is
∞
X xn
ex = ,
n!
n=0

where n! := (1)(2) · · · (n).

Sum of Integers, Sum of Squares
It is also helpful to know that, for positive integers n,

(n)(n + 1)
1 + 2 + ··· + n =
2
and
(n)(n + 1)(2n + 1)
12 + 22 + · · · + n2 =
6
xxi
xxii Math Review

Floor and Ceiling Functions

bxc round x down to the closest, lower integer, e.g., b14.37c = 14
dxe round x up to the closest, higher integer, e.g., b14.37c = 15
Binomial

Coefficient
n n!
k = k!(n−k)! number of ways to choose k out of n objects, when the
order of selection does not matter, e.g., 53 = 10 since there are 10 ways to

choose 3 out of 5 objects:

1 2 3 4 5
1 2 3 4 5 1 2 3 4 5
1 2 3 4 5
1 2 3 4 5 1 2 3 4 5
1 2 3 4 5
1 2 3 4 5 1 2 3 4 5
1 2 3 4 5
Gamma Function
Γ(n) = (n − 1)! The gamma function extends the notion of factorials
beyond the set of nonnegative integers. We will only use this one fact about the
gamma function in this text.
Double
R b R Integration
d
a c f (x, y) dy dx This denotes the double integral of f (x, y) over the
range where a ≤ x ≤ b and c ≤ y ≤ d. Under most conditions used in this book,
the order of integration can often be switched. (In more advanced courses, this
must be done with caution, and suitable convergence theorems must hold—
but we do not cover such topics in this text.) If the order of integration is
RdRb
switched, we have instead c a f (x, y) dx dy. We urge students to make sure
that the outer integral corresponds to the outer variable, and the inner integral
corresponds to the inner variable.
Dice
We assume that all dice in this text are six-sided, numbered 1, . . . , 6, unless
stated otherwise.
Cards
We assume that all decks of playing cards have 52 cards, consisting of four suits
(spades ♠, hearts ♥, diamonds ♦, clubs ♣), 13 values each: A, 2, 3, . . . , 10, J,
Q, K.
Part I

Randomness

At the beginning of this course in probability, we consider the basic aspects

of randomness. We first discuss the outcomes that are possible when some-
thing random occurs, with an emphasis on how these outcomes can be collected
into events. Then we give the basic, fundamental notions of probability theory.
These are very simple to state, but they constitute the groundwork on which
the rest of our study of probability theory is based. We also consider indepen-
dence of events, as well as the way that the occurrence of one event will affect
the probability of occurrence of another event. The first part of the book con-
cludes with the introduction of Bayes’ Theorem, which allows us to manipulate
probabilities and conditional probabilities.
Even small children learn basic probability ideas simply from observing the
world around them. We will formalize these probability ideas and introduce
mathematical calculations to go with them. Drawing pictures to help visualize
the information in the stories is strongly recommended.
By the end of this part of the book, you should be able to:

1. Define basic terms related to probability and events.

2. Use proper set notation for events.

3. Characterize the possible outcomes, when something random occurs.

4. Describe the events into which outcomes can be grouped.

5. Assign probabilities to events, and perform calculations using probability

rules.

6. Calculate whether two or more events are independent.

7. Calculate the probability of an event occurring, given that another event

occurred.

8. Calculate the conditional probability of an event using Bayes’ Theorem.

1
2 Part I. Randomness

Math skills you will need:Pbasic understanding of set notation, unions, in-
tersections, and summation notation.
Additional resources: Calculators may be used to assist in the calculations.
Colored pencils may be helpful for drawing Venn diagrams clearly.
Chapter 1

Outcomes, Events, and Sample

Spaces

On Monday in math class, Mrs. Fibonacci says, “You know, you can think of
almost everything as a math problem.” On Tuesday I start having problems.
—Math Curse by Jon Scieszka and Lane Smith (Viking, 1995)

In a National Public Radio story from November 30, 2012, “That’s So Random:
The Evolution of an Odd Word,” Neda Ulaby writes about the many misuses of
the word “random” in our modern culture, including snippets from the comedian
Spencer Thompson’s routine, “I Hate When People Misuse the Word Random.”
For example, Thompson explains that if your friends talk about a “random
party” they went to, it probably wasn’t as random is they think since it was likely
to be held within a reasonably small community and planned with some people
that your friends already knew. What do mathematicians and statisticians
mean by the word “random”?

1.1 Introduction
Probability theory is the study of randomness and all things associated with
randomness. Examples abound everywhere. From the time that we are children,
we play guessing games, roll dice, and flip coins. We frequently encounter the
unknown and the uncertain. We turn on an mp3 player in a “shuffle” mode, or
listen to the radio, eagerly waiting to see what song will come on next. The
time until an something happens is often random, e.g., until a traffic light turns
green, an email arrives, the telephone rings, or a text message buzzes. The
sex of a baby remains unknown until birth (or an ultrasound). An athlete
runs a race, but the exact finishing time is unknown beforehand. Millions of

3
 4 Chapter 1. Outcomes, Events, and Sample Spaces

people play lotteries and other games of chance, often wagering large amounts
of money. Throughout this book, we study probability using examples that will
be familiar to the reader.
Definition 1.1. When something happens at random there are several po-
tential outcomes. Exactly one of these outcomes occurs. An event is defined
to be a collection of some outcomes.
Even though the Two extreme events have names: The empty set ∅ consists of no outcomes
empty set never (so the empty set never happens). The sample space S consists of all outcomes
happens, we will (so the sample space always happens).
need it to
understand disjoint
events, i.e., events
that have no
outcome in common. Example 1.2. You roll a 6-sided die.
The sample space is S = {1, 2, 3, 4, 5, 6}. Only one of these six outcomes ac-
tually occurs; for instance, 2 is a possible outcome, or 5 is a possible outcome,
etc. We cannot “solve” for which outcome occurs because, as we know from
practical experience, we do not know (in advance) which outcome will occur.
The outcome is random.
(Q: How many One event is {1, 3, 5}, i.e., the event that the outcome is odd. The event
events are there that the roll is 2 or higher is {2, 3, 4, 5, 6}. The event that 4 does not appear is
altogether? Hint: {1, 2, 3, 5, 6}. The event {3} consists of only one outcome. Event {1, 6} has the
It’s a power of 2.) smallest and largest possible outcomes.
One event is a subset of another if every outcome from the first event is
contained in the second event too. Subsets are denoted with the “⊂” symbol.
For instance, an event with one outcome (such as 5) is a subset of a larger event
(such as {1, 2, 5}), which is a subset of the sample space:

{5} ⊂ {1, 2, 5} ⊂ {1, 2, 3, 4, 5, 6}.

Definition 1.3. Event A is a subset of event B, written A ⊂ B, if every

outcome in A is also an outcome in B.

B
A

Example 1.4. A student buys a book and opens it to a random page. He notes
the number of typographical errors on the page.
 1.1. Introduction 5

The sample space is S = Z≥0 , i.e., the set of nonnegative integers.

The event that the page contains at most 2 errors is {0, 1, 2}.

Example 1.5. A new mother delivers one baby.

The sample space is S = {boy, girl}. Although there is just one baby, we can
describe four events:

∅, {boy}, {girl}, {boy, girl} = S.

Example 1.6. A new mother delivers at least one baby.

One possible outcome is that the the mother has triplets, which are all girls;
we denote this outcome as (g, g, g). If she delivers a boy and then a girl, the
outcome is (b, g). So the sample space is

S = {(b), (g),
(b, b), (b, g), (g, b), (g, g),
(b, b, b), (b, b, g), (b, g, b), (b, g, g), (g, b, b), (g, b, g), (g, g, b), (g, g, g), . . .}.

Note: A new mother may have a single baby, twins, triplets, octuplets or A set of octuplets (8
any other (relatively small) number of babies at one time. We only listed the babies) was born in
possibilities up to triplets explicitly, but the other possibilities are included in 1998 and also in
S too; hence, the “. . .” at the end of S. 2009 in the United
States.
Let A be the event that the mother has at least one boy and at least one
girl. So A does not contain the outcomes (b) or (b, b) or (b, b, b) etc., and does
not contain the outcomes (g) or (g, g) or (g, g, g) etc. Thus

A = {(b, g), (g, b), (b, b, g), (b, g, b), (b, g, g), (g, b, b), (g, b, g), (g, g, b), . . .}.

Example 1.7. You wait at a red traffic light and record the time (in seconds)
until the light turns green.

The sample space is the set of all positive real numbers, R>0 . One event is
[5, 10], the event consisting of all outcomes between 5 and 10 seconds (inclusive).
Another event is (12.7, ∞), i.e., the waiting time is strictly more than 12.7
seconds. Another event is {32.7} seconds, the event consisting of only the
outcome 32.7 seconds. Events can be built using unions and intersections, e.g.,
(0, 60) ∪ (120, 180) is the event consisting of all outcomes less than 1 minute and
also consisting of all outcomes of 2 to 3 minutes.
6 Chapter 1. Outcomes, Events, and Sample Spaces

Example 1.8. You notice the color of the next car to pass on the street.

The sample space is the set of all possible colors in the scheme used to classify
this car’s color, for instance, perhaps it is classified according to the sample
space

S = {red, yellow, green, blue, orange, silver, brown, black, white, other}.

As we see in the examples with the baby’s sex or car’s color, outcomes do not
have to be numbers.
At the most fundamental level, it is essential to consider how we classify the
outcomes. There are often several valid viewpoints. As an example:

Example 1.9. We hit or miss the bullseye with a dart (two possible outcomes).

5 20 1
12 18
9 4 r
14 13 x
11 6 (x, y)
bullseye 8 10
16 15
7 2
19 3 17 miss

Figure 1.1: Different sample spaces for a dart throw. Left: Two outcomes in
the sample space. Middle: Twenty-one outcomes in the sample space (the 21st
outcome denotes missing the board altogether). Right: Sample space consists
of the outcomes, according to location, given as coordinates.

The sample space is S = {hit, miss}. This is depicted on the left side of Fig-
ure 1.1. There are four events:

∅, {hit}, {miss}, {hit, miss} = S.

(The empty set never happens because it has no outcomes. Sometimes event
{hit} happens; sometimes event {miss} happens. Event {hit, miss} = S always
happens.)
 1.1. Introduction 7

Example 1.9 (continued) When throwing a dart, we hit one of twenty re-
gions, or we miss the entire board (twenty-one possible outcomes). Notice: this
classification of the outcomes is very different than the “hit” or “miss” setup.
The sample space is S = {miss, 1, 2, 3, . . . , 20}, consisting of the twenty-one
possible outcomes: either we “miss” the board altogether or we hit one of the
20 specified regions. This is depicted in the middle of Figure 1.1. (The board
has metal ridges between the regions, so that a dart cannot land exactly on the
boundary of two regions.)
Example 1.9 (continued) When throwing a dart, we note the exact location
where the dart lands.
The sample space is

S = {(x, y) | x, y ∈ R},

consisting of the outcome listed according to the x (horizontal) and y (vertical)

locations where it landed (using the origin at the center of the dartboard as a
reference point). This is depicted on the right side of Figure 1.1.
The set notation we used for S is nice, because we are unable to list all of the
possible points, in this last version of Example 1.9; in fact, there are infinitely
many points in the sample space S = {(x, y) | x, y ∈ R}. Set notation is also
nice because we can add other conditions to the allowable points. For instance,
if we want to consider only the situation in which the dart hits the dartboard,
then the sample space could be narrowed to

{(x, y) | x2 + y 2 ≤ r2 },

where r is the radius of the dartboard. (In this case, we have not handled darts
that miss the board entirely.) For instance, if r = 9 inches, then the sample
space includes outcomes such as (x, y) = (3.6, −1.35), etc.

Set notation is a way to describe a collection of things, sometimes using an

annotation about conditions on these things (maybe it should be called “set
annotation”). The things that are inside the set are written on the left, and any
conditions about these things go on the right.

Notation 1.10. The notation for a set uses braces, with the contents of the
set, often followed by a line and then any conditions on the contents of the
set.
things conditions on
 

in the set these things

The dartboard example should illustrate that it is really important to un-

derstand what outcomes are possible when something random happens. This
takes some practice at the outset. Sometimes it is helpful to write a list—even
8 Chapter 1. Outcomes, Events, and Sample Spaces

an incomplete list—of the different outcomes that are possible from a random
phenomenon. (As a rule of thumb, we often encourage students to write five
different possible outcomes, if the problem is complicated, just to develop some
intuition.) With the darts in Example 1.9, we can certainly write down both
outcomes in the first scenario, i.e., “bullseye” or “not bullseye.” In the second
scenario, the list of all possible outcomes would be “miss,” 1, 2, 3, . . . , 20. In
the third scenario, as soon as we begin to try to write down all of the possible
locations on the board by their (x, y) coordinates, we quickly realize that this
is a hopeless task. It will not be possible for us to write down every potential
outcome, so the concise set notation is crucial to use.

Definition 1.11. We use the union notation “∪” when a new set is formed
that contains each outcome found in any of the component events. E.g., A ∪ B
contains each outcome that is found in A, or in B, or in both.

Definition 1.12. We use the intersection notation “∩” to construct a new

event that contains only the outcomes found in all of the components. E.g.,
A ∩ B contains each outcome that is found in both A and B; it is insufficient
to be in just one of these sets.

Example 1.13. A student shuffles a deck of cards thoroughly (one time) and
then selects cards from the deck without replacement until the ace of spades
appears.
“Without replacement” means that the cards are not put back into the deck
after they are drawn. So on the first draw there are 52 cards available, but on
the second draw there are only 51 cards available, and 50 cards available on
the third draw, etc. So the ace of spades is certain to appear sometime during
the 52 draws. Also, because they are selected without replacement, the chosen
cards will be distinct.
The event that exactly three draws are needed to see the ace of spades is
{(x1 , x2 , x3 ) | x3 = A♠, and the xj ’s are distinct}.
The sample space S consists of all possible draws of distinct cards that end with
the ace of spades:
S = {(A♠)} ∪ {(x1 , x2 ) | x2 = A♠, and the xj ’s are distinct }
∪ {(x1 , x2 , x3 ) | x3 = A♠, and the xj ’s are distinct }
∪ {(x1 , x2 , x3 , x4 ) | x4 = A♠, and the xj ’s are distinct }
..
.
∪ {(x1 , x2 , . . . , x52 ) | x52 = A♠, and the xj ’s are distinct } .
Equivalently, if Bk =S{(x1 , x2 , . . . , xk ) | xk = A♠, for distinct xj ’s }, then the
sample space is S = 52k=1 Bk .
 1.1. Introduction 9

Example 1.14. A student draws cards from a standard deck of playing cards
until the ace of spades appears. After every unsuccessful draw, the student
replaces the card and shuffles the deck thoroughly before selecting a new card.
The set of outcomes in which the ace of spades first appears on the kth draw is
Bk = {(x1 , . . . , xk ) | only xk is A♠}
Notice that we dropped the condition about the cards being distinct.
The set
S∞of all possibilities in which the student actually finds the ace of Since the cards are
spades is j=1 Bk . The astute reader will notice that we did not yet mention replaced after each
the possibility that the aces of spades never appears. We write this event as draw, this scenario
is quite different
C = {(x1 , x2 , x3 , . . .) | none of the xk ’s is A♠}. from Example 1.13.
So the entire sample space is
[ 
S= Bk ∪ C.
k≥1

Example 1.15. A traffic engineer records times (in seconds) between the next
six cars that pass.
For example, consider when the next six cars arrive:
car 1 car 2 car 3 car 4 car 5 car 6

0.62 1.31 0.58 1.77 1.97 1.10

x
0 0.62 1.93 2.51 4.28 6.25 7.35

The sample space is

S = {(x1 , . . . , x6 ) | xj ∈ R>0 for each j}.
The engineer uses x1 for the time until the first car passes, and x2 is the time
between the first and second cars, and in general, xj is the time between (j −1)st
and jth cars.
The event where there are at least 3 seconds between all pairs of consecutive
cars is
{(x1 , . . . , x6 ) | xj ≥ 3 for each j}.
The event in which the cars have consecutively longer and longer inter-arrival
times (i.e., the distance between cars 1 and 2 is shorter than the distance be-
tween cars 2 and 3, which is shorter than the distance between cars 3 and 4,
etc.), is
{(x1 , . . . , x6 ) | xj < xj+1 for each j}.
Many other possible events can be written. The possibilities are endless.
10 Chapter 1. Outcomes, Events, and Sample Spaces

Example 1.16. A meteorologist records the quantity of rain (in centimeters)

in a city on a certain day.
The sample space is S = R≥0 . The event in which it rains between 3 to 5
centimeters (inclusive) is [3, 5] = {x | 3 ≤ x ≤ 5}. The event in which it rains
more than 2 centimeters is (2, ∞) = {x | x > 2}.

Example 1.17. A student hears ten songs (in a random shuffle mode) on her
music player, noting how many of these songs belong to her favorite type of
music.
If she uses F to denote when a song belongs to her favorite type of music, and
N for not-favorite, then the sample space consists of all ten-tuples of F ’s and
N ’s. In other words, the sample space is
S = {(x1 , . . . , x10 ) | xj ∈ {F, N }}.
The event that none of the first three songs is her favorite type of music is
A = {(N, N, N, x4 , . . . , x10 ) | xj ∈ {F, N }}.
Songs 1, 2, 3 must be of type “N ,” but each of songs 4, 5, 6, 7, 8, 9,
10 have two possible assignments of types, either N or “F .” So A contains
1 × 1 × 1 × 2 × 2 × 2 × 2 × 2 × 2 × 2 = 27 outcomes. (In Part IV of the
book, and in the chapters on discrete random variables, we will investigate
more thoroughly the ways that counting is used in probability theory.)
The event that the even-numbered songs are from her favorite type of music
is
B = {(x1 , F, x3 , F, x5 , F, x7 , F, x9 , F ) | xj ∈ {F, N }}.
Event B contains 25 = 32 outcomes.
No selection of songs would be an outcome in both A and B, so A ∩ B = ∅;
in particular, the second song is of type N if the outcome is in event A, but the
second song must be of type F if the outcome is in event B.
The event that the last five songs are from type F is
C = {(x1 , x2 , x3 , x4 , x5 , F, F, F, F, F ) | xj ∈ {F, N }};
there are 25 = 32 outcomes in event C.
An outcome is in B ∩ C if and only if the 2nd, 4th, 6th, 7th, 8th, 9th, and
10th songs are of type “F .” So the event B ∩ C can be written as
B ∩ C = {(x1 , F, x3 , F, x5 , F, F, F, F, F ) | xj ∈ {F, N }}.
So event B ∩ C has 23 = 8 outcomes.
 1.2. Complements and DeMorgan’s Laws 11

1.2 Complements and DeMorgan’s Laws

In Chapter 2, we introduce probabilities of events. If events are disjoint (have
no overlap), we will see that we can calculate the probability of each, and then
sum the probabilities. If the events are overlapping, however, we need to use
unions and intersections to handle the overlaps. DeMorgan’s Laws give us the
dexterity of moving our view between unions and intersections.
Before introducing DeMorgan’s Laws, however, we need to introduce the
idea of the complement of an event.

Example 1.18. Consider two events: A is the event that the amount of rainfall
on a given day is strictly less than 2.8 inches, and B is the event that the amount
of rainfall is 2.8 inches or more. Thus

A = [0, 2.8) B = [2.8, ∞).

Then any possible outcome (amount of rainfall) is in either A or B but not

both. So A and B are complements, i.e., B contains exactly the outcomes not
found in A, and vice versa.

Whenever the entire sample space is split into two events without overlap, the Some students call
two events are complements of each other. Thus B is the complement of A; “\” the “throwaway”
this is written as B = Ac . Similarly, A = B c . operator, i.e., S \ A
consists of all of S,
Definition 1.19. For an event A, the complement is the set of all outcomes “throwing away” any
in the sample space S that are not in A. The complement of A is written as outcomes in A.
Ac or as S \ A, using the “setminus” notation given below.

A Ac

Notation 1.20. The setminus “ \” notation can be used for any pair of
events. The event B \ A contains all outcomes found in B but not found in A.

B∩A B\A
12 Chapter 1. Outcomes, Events, and Sample Spaces

Example 1.21. If S = {0, 1, 2, 3, . . .} is the set of all nonnegative integers,

and A = {1, 2, 3, . . .} is the event containing all strictly positive integers, then
Ac = {0}, i.e., Ac contains only 1 outcome, namely, the integer 0.

Now we are prepared to discuss DeMorgan’s Laws, as seen in the following

example.

Example 1.22. A student is randomly selected, and she is asked about her
movie preferences. Let A1 , A2 , A3 be the event that she enjoys adventure, com-
edy, or romance movies, respectively. She might enjoy more than one genre.

c
Event 3j=1 Aj occurs if she likes at least one of these genres. Thus,
S S3
j=1 Aj
occurs if she dislikes all three genres; this is the same event as 3j=1 Acj , i.e.,
T
S3
c T3
the event that she dislikes each of the three genres. So j=1 Aj = j=1 Acj .
A picture of this scenario is given on the left side of Figure 1.2.

A B A B

C C

Figure 1.2: Left: VisualizationS3 of DeMorgan’s

c T3 firstc law, with three events.
The blue shaded area shows j=1 A j = j=1 Aj . Right: Visualization
ofTDeMorgan’s second

c S3 law, with three events. The blue shaded area shows
3
j=1 Aj = j=1 Aj .
c

c
The event 3j=1 Aj happens if she likes all three genres. Thus,
T T3
j=1 Aj
occurs if she dislikes at least one genre; this is the same event as 3j=1 Acj ,
S
T3
c
i.e., the event that she dislikes at least one genre. So = 3j=1 Acj .
S
j=1 Aj
A picture of this scenario is given on the right side of Figure 1.2.

These ideas about complements of unions and intersections of events hold much
more generally, for both finite and infinite collections of events. Consider a finite
collection of events A1 , A2 , . . . , An or an infinite collection of events A1 , A
S2 , . . ..
The event that contains the outcomes found in at least one of the Aj ’s is j Aj .

c
The complement of this event is j Aj ; it contains the outcomes missing
S
from all of the Aj ’s, i.e., the outcomes that are in every Acj and thus in j Acj .
T
 1.3. Exercises 13

Theorem 1.23. DeMorgan’s first law (complement of the union

equals the intersection of the complements)
For a finite or infinite collection of events A1 , A2 , . . . ,
[
c \
Aj = Acj .
j j

Similarly, the event that contains the outcomes found in all of the Aj ’s is j Aj .
T

c
The complement of this event is j Aj ; it contains the outcomes missing from
T
at S
least one of the Aj ’s, i.e., the outcomes that are in at least one Acj and thus
in j Acj .

Theorem 1.24. DeMorgan’s second law (complement of the intersec-

tion equals the union of the complements)
For a finite or infinite collection of events A1 , A2 , . . . ,
 \ c [
Aj = Acj ,
j j

1.3 Exercises
In some of these scenarios, several different interpretations are possible. These
early exercises are intended to inspire discussion. A key goal is to effectively
communicate your understanding of the sample space and of the various at-
tributes of the scenario that the outcomes exhibit.

1.3.1 Practice

In each of Exercises 1.1 to 1.6, find:

(1) one specific outcome; (2) one specific event; and (3) the sample space.

Exercise 1.1. Skydiver. A skydiver jumps out of a plane and lands somewhere
at random inside a circle with radius one mile. What is his landing location?

Exercise 1.2. Q library books. A library worker named Jim is going through
the returned books. Books are constantly arriving, and Jim’s quirky boss,
Quinten Quirrell, forces his workers to sort books until they sort a book with a
title beginning with the letter Q. How many books will Jim have to sort until
he gets a break?

Exercise 1.3. Gatorade. Chris has an 18-pack of Gatorade sports drink: 6

orange, 6 lemon-lime, and 6 fruit punch. He blindly grabs one out of the pack
over and over if necessary—without replacement—until he finds an orange one.
How many bottles will he have to pull out of the pack?
14 Chapter 1. Outcomes, Events, and Sample Spaces

Exercise 1.4. A random hand. You are dealt a hand of five cards (without
replacement) from a standard deck of fifty-two playing cards. You note the suits
and values of the cards (the order does not matter). Which cards are you dealt?

Exercise 1.5. Cell phone minutes. Your parents restricted your cell phone
minutes to 400 minutes this month. You call your boyfriend 75 times during
the month. What are the lengths of your calls, if you don’t exceed your allotted
400 minutes?

Exercise 1.6. Crayons. A little girl picks out crayons (without replacement)
from her 24-pack of Crayolas until she gets to the pink crayon. How many
crayons are needed?

1.3.2 Extensions
Exercise 1.7. Moving chairs. Four chairs are placed in a row; two of them
are red and look identical; the other two are blue and look identical.

a. How many outcomes are in the sample space? What are they?
b. How many different events are there?

Exercise 1.8. Abstract art. A painter has four different jars of paint colors
available, exactly one of which is purple. She wants to paint something abstract,
so she blindfolds herself, randomly dips her brush, and paints on the canvas.
She continues trying paint jars until she finally gets some purple onto the canvas
(her assistant will tell her when this happens). Assume that she does not repeat
any of the jars because her assistant removes a jar once it has been used.

a. How many outcomes are in the sample space? What are they?
b. How many different events are there?
c. Another painter borrows the four jars of paint and performs the same
experiment; i.e., selects paint at random; but she allows the jars to be reused,
perhaps over and over many times (assume each contains an unlimited amount
of paint). List a few of the outcomes in the sample space, when repetitions are
allowed.
d. In the scenario from part c, write an expression for the sample space.

Exercise 1.9. Waiting for a text message. On Wednesday evening at 5

PM, a student waits for a text message to arrive (hoping to arrange plans for
the weekend). Each time a text arrives, the student quickly answers and then
begins to wait for the next message. The student quits waiting after receiving
three messages.

a. What is the sample space that describes the set of waiting times between
the messages that the student receives?
 1.3. Exercises 15

b. Assume the time required to type a text response also takes a random
amount of time. What is the sample space that describes the set of all waiting
times and also the lengths of typing the responses as well?
c. Write an expression for the event that the waiting times get longer and
longer, but the times used to type responses get shorter and shorter.

Exercise 1.10. Double die rolling. Two friends are playing a board game
that requires each of them to roll a die. Each player uses her/his own die.

a. What is the sample space for a single roll if their dice are painted two
different colors?
b. What if both dice are white—does this change anything?
c. What if one person rolls both dice—does this change anything?

1.3.3 Advanced
Exercise 1.11. Choose a point in a triangle. A point is chosen at random
inside the triangle in Figure 1.3.

What is the sample space? (Use set notation for the constraints on x and y.)

y
4
y
3
2
2
1
1
x x
1 2 1 2

Figure 1.3: Left: A triangle. Right: A quadrilateral.

Exercise 1.12. Choose a point in a quadrilateral. A point is chosen at

random inside the quadrilateral in Figure 1.3.

What is the sample space? (Hint: Give bounds on x and then on y.)

Exercise 1.13. Building a loft. You are assembling a loft. One piece of wood
has 8 screw holes in a straight row, but you can only find 6 screws (which look
identical). In a hurry, you put the 6 screws in the 8 holes. How many outcomes
are in the sample space, if exactly 6 holes are selected (and the order of selection
does not matter)?
16 Chapter 1. Outcomes, Events, and Sample Spaces

Exercise 1.14. Seating arrangements. Alice, Bob, Catherine, Doug, and

Edna are randomly assigned seats at a circular table in a perfectly circular
room. Assume that rotations of the table do not matter, so there are exactly
24 possible outcomes in the sample space.
Bob and Catherine are married. Doug and Edna are married. In how many
of these 24 outcomes are both married couples sitting together and therefore
happy?

Exercise 1.15. Sum of three dice. Roll three distinguishable dice (e.g.,
assume that there is a way to tell them apart, for instance, that the dice are
three different colors). There are 6 × 6 × 6 = 216 possible outcomes.
For 3 ≤ j ≤ 18, define Aj as the event that the sum of the dice equals j.
Find |Aj |, i.e., the number of outcomes in Aj ? (For instance, |A3 | = 1 since A3
contains only the one outcome (1, 1, 1). Similarly, |A18 | = 1 since A18 contains
only one outcome, (6, 6, 6).)
Chapter 2

Probability

In the fields of observation chance favors only the prepared mind.

—Louis Pasteur (Lecture, University of Lille, December 7, 1854)

You have 22 songs on your mp3 player’s playlist, and you set the player on
shuffle mode, where songs are allowed to repeat, while you study. Country
music is your favorite, but only 12 of the songs on this playlist are country with
the rest being rock. What is the probability the first song will be country music?
What is the probability the first song will not be country music? If 3 songs play,
what is the probability that all 3 are country music? Or the probability that
exactly 2 of them are country music? Or the probability that none of them are
country music? If 3 songs play, what is the probability that only the first song
is country, and is this different from the probability that exactly one song will
be country? Why or why not?

2.1 Introduction

The probability of an event is a number between 0 and 1 (inclusive).1 In Chap-

ter 37, we further explore the idea of a probability of an event as the percentage
of time that an event occurs, in the long run. For now, however, we begin with
only three basic assumptions. We build the framework for our understanding
from these 3 intuitive ideas:
1
In more advanced courses, some events are so complicated that there is not a reasonable
way to assign them a probability. We will not study such complications here. See, e.g.,
Billingsley [1] or Durrett [2].

17
18 Chapter 2. Probability

Remark 2.1. Three Probability Axioms (Intuitive Statements)

1. Any event occurs a certain percentage of the time, so probabilities are

always between 0 and 1.

2. With probability 1, some outcome in the sample space occurs.

3. If events have no outcomes in common, then the probability of their

union is the sum of the probabilities of the individual events.

Definition 2.2. A pair of events A, B is disjoint (also called mutually

exclusive) if they have no outcome in common, i.e., if their intersection is
empty, A ∩ B = ∅. A collection of events is disjoint if every pair of events is
disjoint.

Definition 2.3. We always use “P ” to denote a probability that is defined on

the events associated with some random phenomenon.

Now we can state the fundamental ideas (mentioned earlier) in a precise way.

Axioms 2.4. Three Probability Axioms (Mathematical Statements)

1. For each event A,

0 ≤ P (A) ≤ 1.

2. For the sample space S,

P (S) = 1.

3. If A1 , A2 , . . . is a collection of disjoint events, then

∞
[  ∞
X
P Aj = P (Aj ).
j=1 j=1

Theorem 2.5. The probability of the empty set ∅ is always 0.

This theorem makes sense intuitively, but does it fit with our basic assumptions?
Yes! Here is the reasoning:

1 = P (S) by Axiom 2.4.2

= P (S ∪ ∅ ∪ ∅ ∪ · · · ) since S = S ∪ ∅ ∪ ∅ · · ·
= P (S) + P (∅) + P (∅) + · · · by Axiom 2.4.3 (A1 = S; Aj = ∅ for j ≥ 2)

By Axiom 2.4.2, P (S) = 1. The rest of the right hand side consists of nonneg-
ative terms P (∅), which must therefore each be 0. So P (∅) = 0.
 2.2. Equally Likely Events 19

Theorem 2.6. If A1 , A2 , . . . , An is a collection of finitely many disjoint events,

then the probability of the union of the events equals the sum of the proba-
bilities of the events: [ n  X n
P Aj = P (Aj ).
j=1 j=1

Again, this makes intuitive sense. To prove it, using our basic assumptions,
define Aj = ∅ for all j > n. Then we use the probability theory axioms, as
follows:
[n  [∞ 
P Aj = P Aj since Aj = ∅ for j > n
j=1 j=1
∞
X
= P (Aj ) by Axiom 2.4.3
j=1
n
X
= P (Aj ) since P (Aj ) = 0 for j > n
j=1

2.2 Equally Likely Events

We begin by considering the probabilities assigned to some of the events that
were discussed at the start of Chapter 1.

Example 2.7. When rolling a die, each of the six outcomes should be equally
likely. This means that each single-outcome event should have the same prob-
ability.

The probability of each seems (intuitively) to be 1/6. Using our simple assump-
tions,

1 = P (S) by Axiom 2.4.2

= P ({1, 2, 3, 4, 5, 6})
= P ({1} ∪ {2} ∪ {3} ∪ {4} ∪ {5} ∪ {6})
= P ({1}) + P ({2}) + P ({3}) + P ({4}) + P ({5}) + P ({6}) by Theorem 2.6

If all P ({j})’s are the same, then 1 = 6P ({j}), so P ({j}) = 1/6 for each j. So
our intuition is correct. Now we can compute any kind of probability associated
with one roll of a die. For instance, the probability a die roll is odd is:
1 1 1 1
P ({1, 3, 5}) = P ({1}) + P ({3}) + P ({5}) = + + = .
6 6 6 2
 20 Chapter 2. Probability

Example 2.8. A pregnancy that yields exactly one baby would yield an out-
come of either a boy or a girl, which are equally likely (as in the die example
above).
www.cdc.gov/ The four relevant probabilities are
nchs/data/nvsr/
nvsr61/ 1. P (∅) = 0,
nvsr61_01.pdf
suggests that the 2. P ({boy}) = 1/2,
odds are really
3. P ({girl}) = 1/2,
closer to 51.17% for
boys vs 48.83% for 4. P ({boy, girl}) = P (S) = 1.
girls, but we assume
a 50/50 ratio. In the last case, S = {boy, girl}, so we are really just emphasizing the fact that
P (S) = 1.

These observations about equally likely outcomes are handy and very general:

Theorem 2.9. If a sample space S has n equally likely outcomes, then each
outcome has probability 1/n of occurring.

This is true for just the same reasons as in the die example. Let x1 , x2 , . . . , xn
be the n outcomes. Then

1 = P (S) by Axiom 2.4.2

= P ({x1 , x2 , . . . , xn })
= P ({x1 } ∪ {x2 } ∪ · · · ∪ {xn })
= P ({x1 }) + P ({x2 }) + · · · + P ({xn }) by Axiom 2.4.3

If all P ({xj })’s are the same, then 1 = nP ({xj }), so P ({xj }) = 1/n for each j.

Corollary 2.10. If sample space S has n equally likely outcomes, and A is

an event with j outcomes, then event A has probability j/n of occurring, i.e.,
P (A) = j/n.

To prove this corollary, write y1 , . . . , yj as the j outcomes in A. Then

P (A) = P ({y1 , y2 , . . . , yj })
= P ({y1 } ∪ {y2 } ∪ · · · ∪ {yj })
= P ({y1 }) + P ({y2 }) + · · · + P ({yj }) by Axiom 2.4.3
1 1 1
= + + ··· + by Theorem 2.9
n n n
= j/n

Definition 2.11. The number of outcomes in an event A, also called the size
of A, is denoted as |A|.
 2.2. Equally Likely Events 21

Using the notation of |S| and in |A| as the number of items in S and A, respec-
tively, Corollary 2.10 can be rewritten as follows:
Corollary 2.12. If sample space S has a finite number of equally likely out-
comes, then event A has probability

P (A) = |A|/|S|,

where |S| and |A| denote the number of items in S and A, respectively.
We will study equally likely outcomes to a much greater extent, in Chapters 20
and 22. For now, however, we give a few examples.

Example 2.13. As in Example 1.9, consider a dartboard split into 20 regions.

We treat each of the 20 regions as an outcome, and the possibility of a “miss” as
a 21st potential outcome. For simplicity, suppose that the probability of a miss
is 0. (Just allow the player to try again, if initially missing the board.) Also
suppose the player is equally likely to hit any of the 20 regions on the board.
The sample space is S = {1, 2, 3, . . . , 20}. We have 20 disjoint events, each with
one outcome: A1 = {1}, A2 = {2}, and in general,
Aj = {j}, for 1 ≤ j ≤ 20.
(This is a gentle introduction to enumerating events. We will not always use
Aj = {j}.) We assumed P ({miss}) = 0. The other 20 outcomes are each
equally likely, and the jth event has just 1 outcome, so
P (Aj ) = 1/20 for each j.

Example 2.13 (continued) We can now split the dartboard into four regions.
If a new event R1 is constructed as the union of several of the Aj ’s, then
the probability of R1 is just the sum of the probabilities. E.g., if
R1 = {1, 18, 4, 13, 6} = A1 ∪ A18 ∪ A4 ∪ A13 ∪ A6 ,
(i.e., R1 is the event that the dart lands in the northeast portion), then
P (R1 ) = P (A1 ) + P (A18 ) + P (A4 ) + P (A13 ) + P (A6 ) = 5/20 = 1/4.
This could also seen by using Corollary 2.10, since R1 contains 5 of the 20
equally likely outcomes, so P (R1 ) = 5/20.
As in Figure 2.1, define northeast, southeast, southwest, and northwest re-
gions as
R1 = {1, 18, 4, 13, 6}, R2 = {10, 15, 2, 17, 3},
R3 = {19, 7, 16, 8, 11}, R4 = {14, 9, 12, 5, 20}.
22 Chapter 2. Probability

5 20 1
12 18
9 4
14 13
11 6
8 10
16 15
7 2
19 3 17 (or could miss the
dartboard altogether)

Figure 2.1: Twenty-one possible outcomes. Four colors for the northeast,
southeast, southwest, and northwest regions.

A classification of the sample space, as we made in Example 2.13, when splitting

the sample space into four regions, is called a partition of the outcomes in the
sample space. More generally, in a partition, every outcome belongs in exactly
one of the events, and the events in the partition are disjoint. Partitions will
be very helpful in Chapter 5, on Bayes’ Theorem.

Definition 2.14. If a collection of nonempty events is disjoint, and their

union is the entire sample space, then the collection is called a partition. If
j Bj = S and the Bj ’s are disjoint events, then the collection of Bj ’s is called
S
a partition.

In a partition, not all of the regions need to have the same size.

Example 2.15. For instance, a different partition of the dartboard could con-
sist of three disjoint events:

the upper portion, T1 = {12, 5, 20, 1, 18}

the lower left portion, T2 = {3, 19, 7, 16, 8, 11, 14, 9}
the lower right portion, T3 = {4, 13, 6, 10, 15, 2, 17}

See Figure 2.2. In this partition, P (T1 ) = 5/20, P (T2 ) = 8/20, and P (T3 ) =
7/20. Notice P (T1 ) + P (T2 ) + P (T3 ) = 1.

Remark 2.16. The probabilities of events in a partition always sum to 1.

 2.2. Equally Likely Events 23

5 20 1
12 18
9 4
14 13
11 6
8 10
16 15
7 2
19 3 17 (or could miss the
dartboard altogether)

Figure 2.2: Twenty-one possible outcomes. Three colors for Example 2.15
regions.

To see this,
S in a partition consisting Bj ’s, every outcome is in one of the events,
so S = j Bj . Also, each outcome is in exactly one of these events in the
partition, so the Bj ’s are disjoint. Thus P j P (Bj ). Putting these
S
P
j Bj =
together, we get [
X
1 = P (S) = P Bj = P (Bj ).
j j

So the sum of the probabilities of the events in a partition is always 1.

Example 2.17. As in Example 1.13, a student shuffles a deck of cards thor-

oughly (one time) and then selects cards from the deck without replacement
until the ace of spades appears.

Let Bk denote the event that the ace of spades is drawn on exactly the kth
draw:

Bk = {(x1 , x2 , . . . , xk ) | xk = A♠, and the xj ’s are distinct } .

We emphasize that P (Bk ) = 1/52 for each k, since the initial placement of the
ace of spades (i.e., during the initial shuffle) completely determines when the
ace of spades will appear. Since the ace of spades is equally likely to be in any
of the 52 places in the deck, then the ace of spades is equally likely to appear
on any of the 52 draws.
The Bk ’s are disjoint events, since it is impossible for an outcome to simul-
taneously be in more than one of the Bk ’s. Also, every outcome is in exactly
one of the events. So B1 , B2 , . . . , B52 form a partition of the sample space.
24 Chapter 2. Probability

Not every set of outcomes is equally likely, so we must be careful when applying
Corollary 2.10. For instance, when bowling, it is possible to knock down between
0 and 10 pins, so there are 11 outcomes (if we only keep track of the score, not
the specific pins that fall down). We have no reason to believe that all of these
11 outcomes are equally likely.

2.3 Complements; Probabilities of Subsets

Example 2.18. As in Example 1.18, consider two events: A is the event that
the amount of rainfall on a given day is strictly less than 2.8 inches, and B is
the event that the amount of rainfall is 2.8 inches or more. Thus
A = [0, 2.8) B = [2.8, ∞).
The sample space consisting of all possible amounts of rain is S = [0, ∞) so
S = A ∪ B. Also, A and B are disjoint. So
1 = P (S) = P (A ∪ B) = P (A) + P (B),
so P (B) = 1 − P (A). E.g., if the probability of “rainfall less than 2.8 inches” is
83%, then the probability of “rainfall 2.8 inches or more” must be 1−0.83 = 0.17,
i.e., 17%.

Theorem 2.19. The complement Ac of event A has probability P (Ac ) =

1 − P (A).

The argument is straightforward: An event and its complement are disjoint,

and their union is the whole sample space:
A ∪ Ac = S
So
1 = P (S) = P (A ∪ Ac ) = P (A) + P (Ac ),
and the theorem follows, P (Ac ) = 1 − P (A).
Now we consider a bound for the probability of an event that is the subset
of another event. Recall that A is a subset of B if every outcome of event A
is contained in B too (denoted by A ⊂ B). Intuitively, B seems more likely to
occur than A in this case. This intuition is correct:
Theorem 2.20. If A ⊂ B then P (A) ≤ P (B).

The event B can be expressed as a disjoint union B = A ∪ (B \ A). So

P (B) = P (A) + P (B \ A),
but probability is always positive, so P (B \ A) ≥ 0, and thus P (B) ≥ P (A).
 2.4. Inclusion-Exclusion 25

2.4 Inclusion-Exclusion
The method of inclusion-exclusion allows us to relate overlaps among subsets
to unions and intersections. This decomposition enables us to calculate proba-
bilities for events that are overlapping.

Example 2.21. Consider an observer who randomly chooses a car without

knowing its color. There are exactly 10 cars available, one from each of these
colors: red, blue, yellow, green, lime, teal, orange, silver, brown, or black. Let
A denote the event that the car is red, blue, yellow, orange, or silver; let B
denote the event that the car is orange, silver, brown, or black. So A ∩ B is the
event that the car is orange or silver. Thus

P (A) = 5/10, P (B) = 4/10, P (A ∩ B) = 2/10, P (A ∪ B) = 7/10.

These probabilities correspond to the scenario in Figure 2.3.

S
A B
red
orange brown
blue
green yellow silver black
A∩B
lime A\B B\A
teal
S \ (A ∪ B)

Figure 2.3: Observing ten colors of cars.

Notice
P (A ∪ B) = P (A) + P (B) − P (A ∩ B)
This might seem intuitively clear because A ∪ B accounts for each of the colors
red, blue, yellow, orange, silver one time, while
1. A accounts for red, blue, yellow, orange, silver

2. B accounts for orange, silver, brown, black

3. A ∩ B accounts for orange, silver

So we are just removing the duplication, i.e., correcting for the fact that the
orange and silver were accounted for twice.
26 Chapter 2. Probability

The very same argument works much more generally:

Theorem 2.22. For any two events A and B,

P (A ∪ B) = P (A) + P (B) − P (A ∩ B).

To prove this, write

P (A ∪ B) = P (A \ B) + P (A ∩ B) + P (B \ A)
= P (A \ B) + P (A ∩ B) + P (B \ A) + P (A ∩ B) − P (A ∩ B)
= P (A) + P (B) − P (A ∩ B)

Similar nice things happen if we try to characterize P (A ∪ B ∪ C) by taking

each set into account the proper amount of times. Intuitively, we might first
guess that P (A ∪ B ∪ C) and P (A) + P (B) + P (C) are close in value, but
we have double-counted all of the contributions from the outcomes in A ∩ B
and A ∩ C and B ∩ C, so we remove those. Then P (A ∪ B ∪ C) is close to
P (A) + P (B) + P (C) − P (A ∩ B) − P (A ∩ C) − P (B ∩ C), but the contribution
from A ∩ B ∩ C was originally accounted for three times and then removed three
times, so we must add it back on. (We ask for a proof in the exercises.)

Theorem 2.23. For any three events A, B, C,

P (A ∪ B ∪ C) = P (A) + P (B) + P (C)

− P (A ∩ B) − P (A ∩ C) − P (B ∩ C)
+ P (A ∩ B ∩ C).

A
A B
D

C B

Figure 2.4: Left: Overlaps among three events A, B, C. Right: Overlaps

among four events A, B, C, D.

The overlaps among A, B, C can be visualized on the left side of Figure 2.4.
The overlaps among four events, A, B, C, D can be visualized on the right side
of the same figure. It is true, furthermore, by similar reasoning, that
 2.5. More Examples of Probabilities of Events 27

Theorem 2.24. For any four events A, B, C, D,

P (A ∪ B ∪ C ∪ D) = P (A) + P (B) + P (C) + P (D)

− P (A ∩ B) − P (A ∩ C) − P (A ∩ D)
− P (B ∩ C) − P (B ∩ D) − P (C ∩ D)
+ P (A ∩ B ∩ C) + P (A ∩ B ∩ D)
+ P (A ∩ C ∩ D) + P (B ∩ C ∩ D)
− P (A ∩ B ∩ C ∩ D).

This same kind of reasoning can continue, and we get a general inclusion-
exclusion formula.
Theorem 2.25. Inclusion-Exclusion Rule
For any finite sequence of events A1 , A2 , . . . , An ,
 
[n Xn X X
P  Aj =
 P (Aj ) − P (Ai ∩ Aj ) + P (Ai ∩ Aj ∩ Ak )
j=1 j=1 i<j i<j<k
X
− P (Ai ∩ Aj ∩ Ak ∩ Al )
i<j<k<l

± · · · + (−1)n+1 P (A1 ∩ A2 ∩ · · · ∩ An )

2.5 More Examples of Probabilities of Events

Example 2.26. As in Example 1.14, consider a student who draws cards from
a deck but this time he always replaces the card after each selection and then
reshuffles the deck. He only stops if he reaches the ace of spades.
Let Bk be the set of outcomes in which the ace of spades is discovered for the
first time on the kth draw:
Bk = {(x1 , . . . , xk ) | only xk is A♠}
Then the Sset of all possibilities in which the student actually finds the ace of
spades is ∞k=1 Bk . Also let

C = {(x1 , x2 , x3 , . . .) | none of the xk ’s is A♠}

So the entire sample space can be partitioned using the Bk ’s and C, i.e.,
[ 
S= Bk ∪ C.
k≥1
 28 Chapter 2. Probability

The probability that a single draw does not contain the ace of spades is always
51/52. Since the draws do not affect each other (a phenomenon that we will
explore much further in Chapter 3 on independence), it follows that
k−1
z
   }|  {    k−1  
51 51 51 1 51 1
P (Bk ) = ··· = .
52 52 52 52 52 52
51 j
We know that the Bk ’s and C are all disjoint, and also
P∞
= 1−151


j=0 52 52
is a geometric sum;
∞
[ 
see the Math
Review.
S= Bk ∪ C.
k=1

So
∞
X
1 = P (S) = P (C ∪ B1 ∪ B2 ∪ B3 ∪ · · · ) = P (C) + P (Bk ).
k=1

We already computed P (Bk ) = (51/52)k−1 (1/52), so

∞ ∞ 
!
51 k−1 1
   
X X 1 1
P (Bk ) = = = 1.
k=1 k=1
52 52 1 − 51
52
52

In summary, 1 = P (C) + 1, so P (C) must be 0. In other words, the proba-

bility that the ace of spades never appears is 0.

Example 2.27. (continued from Example 1.17) A student hears ten songs
(in a random shuffle mode) on her music player, paying special attention to how
many of these songs belong to her favorite type of music. We assume that the
songs are picked independently of each other and that each song has probability
p of being a song of the student’s favorite type.

As in Chapter 1, use F and N to denote when a song belongs to her favorite,

or not to her favorite, type of music. So, for each song, the probability that the
song is one of her favorite type can be called p, and the probability that the
song is not one of her favorite type is 1 − p.
Now consider the event A that none of the first three songs is her favorite
type of music. This event is

A = {(N, N, N, x4 , . . . , x10 ) | xj ∈ {F, N }},

and the probability of the event is P (A) = (1 − p)3 , because all that the event
requires is that none of the first three songs is from her favorite type of music.
We do not impose any restrictions on songs 4, 5, . . . , 10, so these do not affect
the probability of event A occurring.
 2.6. Exercises 29

Now consider the event that the even-numbered songs are from her favorite
type of music. We write this event as

B = {(x1 , F, x3 , F, x5 , F, x7 , F, x9 , F ) | xj ∈ {F, N }},

and the probability of B is P (B) = p5 , because we only require that five specific
songs are from her favorite type of music.
Similarly, if

C = {(x1 , x2 , x3 , x4 , x5 , F, F, F, F, F ) | xj ∈ {F, N }},

then P (C) = p5 too.

An outcome is in B ∩ C if and only if the 2nd, 4th, 6th, 7th, 8th, 9th, and
10th songs are favorites, i.e., of type “F .” So the event B ∩ C is

B ∩ C = {(x1 , F, x3 , F, x5 , F, F, F, F, F ) | xj ∈ {F, N }},

which has probability p7 .

As one final event, let Aj denote the event that exactly j of the 10 songs
are from her favorite type of music. The probability of Aj is
 
10 j
P (Aj ) = p (1 − p)10−j ,
j

where 10j = j!(10−j)! is the number of ways to pick exactly j out of 10 songs.
10!

We will explore this idea in greater depth in Chapter 15, on Binomial random
variables.

2.6 Exercises
2.6.1 Practice
Exercise 2.1. Songs by genre. A song is chosen at random from a person’s
mp3 player. The student makes a partition of the sample space, according to
genre of music. The table below gives the number of outcomes in each part of
the partition. There are 27,333 songs altogether.
1032 Alternative 83 Electronic 56 Metal
330 Blues 508 Folk 2718 Other
275 Books & Spoken 183 Gospel 1786 Pop
1468 Children’s Music 82 Hip-Hop 403 R&B
921 Classical 564 Holiday 8286 Rock
6169 Country 537 Jazz 1432 Soundtrack
178 Easy Listening 106 Latin 216 World
Let A be the event that a song is either blues, jazz, or rock, i.e.,

A = {x | x is a blues, jazz, or rock song},

30 Chapter 2. Probability

when one song is chosen at random. Assume that all songs are equally likely to
appear.

a. What is the probability of randomly selecting a blues song?

b. What is P (A)?
c. Did you need to use inclusion-exclusion for part b? (Explain very briefly;
one sentence will do.)
d. What is P (Ac )?

Exercise 2.2. Rock climbing. I am out rock climbing, and the rock face has
4 easy, 7 challenging, and 3 extreme routes to get to the top. The routes are
poorly marked, so I just choose one at random, with all routes equally likely.
What is the probability that I do not choose an extreme route?

Exercise 2.3. Student interests. A student is chosen at random. Let A, B, C

be the events that the student is an Aeronautics major, a Basketball player, or
a Co-op student. The events are not disjoint; we are told

P (A) = P (B) = P (C) = 0.38,

and
P (A ∩ B) = P (A ∩ C) = P (B ∩ C) = 0.12,

and
P (A ∩ B ∩ C) = 0.05.

Find the probability that the student participates in at least one of these
three programs, i.e., find P (A ∪ B ∪ C).

Exercise 2.4. Dining with Dad. At a random meal during a parent weekend
in the dining hall, a student notices the food chosen by her father. Let A, B, C
be the events that his meal include Artichokes, Broccoli, or Cauliflower. These
events have the property that

P (B) = 0.39
P (C) = 0.44
P (A ∩ B) = 0.13
P (A ∩ C) = 0.12
P (B ∩ C) = 0.13
P (A ∩ B ∩ C) = 0.09
P (A ∪ B ∪ C) = 0.89

What is the probability that the father includes Artichokes in his meal, i.e.,
what is P (A)?
 2.6. Exercises 31

Exercise 2.5. Selecting a pair of shoes. A woman has 3 pairs of sneakers,

8 pairs of flip flops, 6 pairs of flats, 4 pairs of wedges, and 9 pairs of high heels.
(Hint for those of you unfamiliar with women’s footwear: The wedges and high
heels will make her look taller.)

a. What is the probability that she selects a pair of shoes that makes her
taller if she pulls a pair from her closet without looking?
b. What is the probability that she selects a pair of shoes that does not
make her look taller?
c. Create another type of partition for this woman’s collection of shoes.

Exercise 2.6. Lollipops and licorice. In a kindergarten class, there are 30

children. Altogether, 19 of them like lollipops, and 10 of them like licorice
(some like both). There are 8 students who don’t like either of these. A child
is chosen at random. What is the probability that the child likes both lollipops
and licorice?

Exercise 2.7. Application to weather. Measure the amount of rainfall that

occurs in your city for a year.

a. List 5 possible outcomes.

b. What is the sample space?
c. Devise a useful way to partition the sample space so that the partition
sheds some insight for the general public about the annual rainfall.
d. Explain how the answer to part c meets the definition of a partition; see
Definition 2.14.

Exercise 2.8. Coin flips. You flip a coin 5 times. What is the probability the
first 4 are heads and the last one is a tail?

Exercise 2.9. Pizza meat. The guys on one floor of a college dorm all decide
to get pizzas to share. They get 3 pepperoni pizzas, 2 bacon pizzas, 1 cheese
pizza, 3 sausage pepperoni pizzas, and 3 meat lovers pizzas with sausage, pep-
peroni, and bacon. What is the probability of a randomly selected slice of pizza
containing:

a. Bacon?
b. Pepperoni?
c. Sausage?

Exercise 2.10. Math and physics. In a class of 100 people, 60 of them

are math majors, and 75 of them are physics majors. There are no students
majoring in anything else in this class. This means that some of the students
are double-majoring in both math and physics. A student is picked at random.
What is the probability the student is double-majoring?
32 Chapter 2. Probability

2.6.2 Extensions
Exercise 2.11. Monkey keystrokes. A monkey is let loose in a computer lab
and starts playing with a keyboard. What is the probability that the monkey,
without any comprehension or intention, types out the word “bananas” if he
types exactly 7 keys? The typical keyboard has 101 keys, and the monkey only
presses one key at a time.

Exercise 2.12. Apples. There are 6 apples in a basket. Two of them are red,
and four are green.

a. What is the probability of selecting a red apple when choosing at random?

b. What is the probability that, if one apple is randomly chosen per day
(and then eaten, not replaced), red apples are chosen on the first two days and
green apples are chosen on the last four days?

Exercise 2.13. Shuffling and star ratings. I have 20 five-star songs and
200 four-star songs on my iPod, which has 2000 songs total.

a. What is the probability of shuffling to a five-star song?

b. What is the probability of shuffling to a four-star song?
c. What is the probability of shuffling to either a five- or four-star song?
d. What is the probability of shuffling to a song with fewer than four stars?

Exercise 2.14. Pizza toppings. Consider the following preferences: 35% of

people like olive pizza, 54% like sausage pizza, and 12% like both olive and
sausage pizza. What is the probability that a randomly chosen person likes
neither olive pizza nor sausage pizza?

Exercise 2.15. Roll a die. If you roll a die, event A contains outcomes 1, 3,
and 6; event B contains outcomes 1 and 6, and event C contains outcomes 4
and 6.

a. What is the probability of having A ∪ B ∪ C occur?

b. What is the probability of having A ∩ B ∩ C occur?

Exercise 2.16. DeMorgan’s first law. For three events, A, B, C, consider

the following probabilities: A, B, C have probability 20% of occurring, the prob-
ability of any two of these events occurring is 3% for each pair, and the probabil-
ity of all three occurring is 1%. Using DeMorgan’s first law, find the probability
of (A ∪ B ∪ C)c .

Exercise 2.17. Abstract art. A painter has three different jars of paint colors
available, in colors green, yellow, and purple. She wants to paint something
abstract, so she blindfolds herself, randomly dips her brush, and paints on the
canvas. She continues trying paint jars until she finally gets some purple onto
the canvas (her assistant will tell her when this happens) and then she stops.
 2.6. Exercises 33

Assume that she does not repeat any of the jars because her assistant removes
a jar once it has been used. So the sample space is

S = {(P ), (G, P ), (Y, P ), (Y, G, P ), (G, Y, P )}.

Find the probabilities of each of the following events:

{(P )}, {(G, P ), (Y, P )}, {(G, P ), (G, Y, P )},

{(Y, G, P ), (G, Y, P )}, {(P ), (Y, P )}

Exercise 2.18. Yahtzee. In the game Yahtzee, there are 5 dice with 6 possible
numbers on each. What is the probability for a Yahtzee on a player’s first roll?
(In other words, what is the probability that all 5 dice show the same number
the first time that they are rolled)?

Exercise 2.19. Locker combinations. You just forgot your locker combi-
nation and are too embarrassed to ask for it. You know for sure that the first
number is 22, or was it 32? It’s one of those. You’re certain that the middle
number is a one-digit number (0–9), and the last number could be anything
between 0 and 45. If the lock is a 3-number lock with numbers 0 through 45,
what is the maximum number of tries needed to open it, assuming you don’t
repeat any combinations?

Exercise 2.20. Mathematical Science majors. There are 89 students in

a volunteer group for majors in math, statistics, or actuarial science. Thirteen
students have a double-major in math and statistics. Twelve students have a
double major in math and actuarial science. Thirteen students have a double
major in statistics and actuarial science. There are 9 students doing a triple
major. Thirty-nine students are majoring in at least statistics, and forty-four
are majoring in at least actuarial science. How many students are majoring in
at least mathematics?

Exercise 2.21. Postal customers. A sequence of seven people walk into a

post office and only their sexes are noted (in sequence) as they enter.

a. How many outcomes are in the sample space?

b. Let A2 be the event that exactly two of the people are females. How
many outcomes are in A2 ?
c. Let Aj be the event that exactly j of the people are females. How many
outcomes are in Aj ?
d. If each of the seven customers is equally likely to be male or female, what
probability should be associated with event Aj ?
34 Chapter 2. Probability

Exercise 2.22. Deducing a probability. Events A, B, C are to be considered

with the following properties:

P (A) = 0.17
P (B) = 0.37
P (C) = 0.19
P (A ∩ B) = 0.07
P (B ∩ C) = 0.11
P (A ∩ B ∩ C) = 0.03
P (A ∪ B ∪ C) = 0.48

Find the probability of A ∩ C.

Exercise 2.23. Mystery probability. Suppose there are 3 events such that

P (A) = 0.20
P (B) = 0.10
P (C) = 0.40
P (A ∩ B) = 0.05
P (A ∩ C) = 0.10
P (B ∩ C) = 0.03
P (A ∩ B ∩ C) = 0.01

What is the probability that none of the events happens?

Exercise 2.24. Prove Theorem 2.23.

Exercise 2.25. Prove Theorem 2.24.

2.6.3 Advanced
Exercise 2.26. Prove Theorem 2.25.

Exercise 2.27. Is the whole smaller than the sum of the parts?
a. It is always true, for any events A, B, that P (A ∪ B) ≤ P (A) + P (B).
Why? Explain briefly with words or a very clear picture.
b. Is it always true that P (A ∪ B ∪ C) ≤ P (A) + P (B) + P (C)? If so,
explain why, either using words or a very clear picture. If not, please give a
counterexample.

Exercise 2.28. Grabbing a pen. You find a container of 27 old pens in your
school supplies and continue to test them (without replacement), until you find
one that works. If each individual pen works 25% of the time (regardless of the
other pens), what is the probability that you find one that works within the
first four tries?
 2.6. Exercises 35

Exercise 2.29. Die rolls. You roll a die three times. What is the probability
the sum of the first two rolls is equal to the third roll?
Exercise 2.30. Cookies. Consider a jar of 9 chocolate chip and 11 peanut
butter cookies. You randomly select 2 cookies to eat. All possible choices are
equally likely.
a. What is the probability that the 2 you select will both will be chocolate
chip?
b. What is the probability that at least one of your cookies will be peanut
butter?
c. What is the probability that last 2 cookies left in the jar (after 18 have
been eaten) will be chocolate chip? (Is this answer the same or different than
part a? Why or why not?)
Exercise 2.31. Seating arrangements. Alice, Bob, Catherine, Doug, and
Edna are randomly assigned seats at a circular table in a perfectly circular
room. Assume that rotations of the table do not matter, so there are exactly
24 possible outcomes in the sample space.
Bob and Catherine are married. Doug and Edna are married.
Let Aj denote the event that exactly j of the married couples are happy
because they are sitting together. Find P (A0 ) and P (A1 ) and P (A2 ).
Exercise 2.32. Socks. In your drawer you have 10 white socks, 6 black socks,
4 red socks, and 2 purple socks. Your roommate is still asleep, and you can’t
turn the light on while you’re getting dressed. You reach in blindly and grab
two socks. What is the probability of pulling out a matching pair of purple
socks?
Exercise 2.33. Maximum of three dice. Roll three distinguishable dice
(e.g., assume that there is a way to tell them apart, for instance, that the dice
are three different colors). There are 6 × 6 × 6 = 216 possible outcomes.
Let Bk be the event that the maximum value that appears on all three dice
when they are rolled is less than or equal to k. Find P (B1 ), P (B2 ), P (B3 ),
P (B4 ), P (B5 ), and P (B6 ). If you prefer, you are welcome to just give a general
formula that covers all six of these cases, i.e., you are welcome to just give a
formula for P (Bk ) itself.
Exercise 2.34. If A1 , A2 , . . . , An is a collection of events, is it always true that
n
[  Xn
P Ak ≤ P (Ak )?
k=1 k=1

Prove this inequality, or give a counterexample.

Exercise 2.35. Tickets. Consider n people who drop a ticket into a box. The
box of tickets is thoroughly shaken and randomized. Each person then draws
one ticket, without replacement. A person is a winner if she selects her own
ticket.
36 Chapter 2. Probability

a. What is the probability that nobody is a winner?

b. Find the limit of the probability in part a, as n → ∞.
Chapter 3

Independent Events

The word probability, in its mathematical acceptation, has reference to the state
of our knowledge of the circumstances under which an event may happen or fail.
—Collected Logical Works, Volume 2: The Laws Of Thought by George
Boole (Walton and Maberly, 1854)

A mother has 2 different pregnancies, each producing a single baby. What is

the chance both babies are girls? What is the chance the older child is a girl?
What is the chance exactly one of the babies is a girl? What is the chance
neither baby is a girl? Are these probabilities the same? Why or why not?

3.1 Introduction
We have an intuitive understanding of the word “independence”: Two events A
and B are independent if the occurrence of one of the events does not affect
the probability of occurrence of the other event. This is exactly right, but
we need the concept of conditional probabilities (to be covered in Chapter 4),
to use this viewpoint. In the present chapter, we define events A and B as
independent if the probability that A and B both occur equals the probability
that A occurs times the probability that B occurs. We will also discuss the
notion of independence among more than two events. Afterwards, we will give
examples of dependent events, as well as a very general fact about sequences
of independent attempts, in which we are waiting for the first “good” result to
occur.

Definition 3.1. Independence

Events A and B are called independent if

P (A ∩ B) = P (A)P (B).

37
38 Chapter 3. Independent Events

Definition 3.2. Dependence

Events A and B are called dependent if they are not independent. In other
words, A and B are dependent if

P (A ∩ B) 6= P (A)P (B).

We give a multitude of examples to clarify the concept of independence.

Example 3.3. Consider the birth of two children from two separate pregnancies
(in particular, we are not considering the birth of twins, in which one baby’s
sex might affect the other).

If A is the event that the first baby is a girl, and B is the event that the
second baby is a girl, then P (A) = 1/2, and P (B) = 1/2, and P (A ∩ B) = 1/4,
so P (A ∩ B) = P (A)P (B). Thus, events A and B are independent. This
matches our traditional understanding of the word “independent,” because the
sex of the first baby does not affect the sex of the second baby.
Let C denote the event that both children are girls. Then P (A ∩ C) = 1/4
but P (A)P (C) = 1/8, so A and C are dependent (intuitively, if C happens,
then A must happen).

A question immediately arises:

Remark 3.4. Is “independent” the same thing as “disjoint”? Answer: “No”!

E.g., consider the outcome “girl, girl” in Example 3.3, which is found in both
events A and B, and thus in A ∩ B too. So A and B are independent but are
not disjoint.
More generally, consider the picture in Figure 3.1, for two different situa-
tions.

S S

A B A B

Figure 3.1: Left: Independent events A and B. Right: Disjoint events A

and B.
 3.1. Introduction 39

Remark 3.5. Can A and B be both independent and disjoint? Only in a

very special case: if P (A) = 0 or P (B) = 0.

Why? If A and B are disjoint, then they have no overlap, i.e., A ∩ B = ∅. So

P (A ∩ B) = P (∅) = 0. On the other hand, since A and B are independent,
then P (A)P (B) = P (A ∩ B). So

P (A)P (B) = 0.

So either P (A) = 0 or P (B) = 0 (or both). In summary:

Remark 3.6. Independent vs Disjoint

If A and B both have positive probabilities, they cannot be both independent
and disjoint.

Remark 3.7. When P (A) = 0, then A is independent from any event B,

because, in such a case,

P (A ∩ B) = 0 = P (A)P (B).

Example 3.8. When rolling a die, let A denote the event consisting of outcomes
{1, 2, 3}, and let B denote the event consisting of outcomes {3, 4}, so P (A) =
1/2 and P (B) = 1/3. Also A ∩ B = {3}, so P (A ∩ B) = 1/6. So P (A ∩ B) =
P (A)P (B), and this means that A and B are independent.

We will return to this scenario in Example 4.14.

Continuing the scenario, let C = {1, 2, 3, 5, 6}. So P (C) = 5/6. Also P (B ∩
C) = P ({3}) = 1/6, but P (B)P (C) = (1/3)(5/6) = 5/18 6= 1/6. So B, C are
dependent.

Theorem 3.9. Subsets are dependent. If A ⊂ B and neither P (A) = 0

nor P (B) = 1, then A, B are dependent.

To see this, consider such events A and B. We have P (A ∩ B) = P (A). Also

P (B) < 1, so multiplying both sides by P (A) (which is strictly positive) pre-
serves the strict inequality. Thus P (A)P (B) < P (A) = P (A ∩ B), so A, B are
dependent.

Theorem 3.10. Complements are dependent. If neither P (A) = 0 nor

P (A) = 1, then A, Ac are dependent.

Note P (A ∩ Ac ) = P (∅) = 0, but P (A) 6= 0 and P (Ac ) 6= 0, so P (A)P (Ac ) 6=

0 = P (A ∩ Ac ), so A, Ac are dependent.
40 Chapter 3. Independent Events

Example 3.11. Consider the songs from Exercise 2.1. Suppose that songs are
chosen in such a way that each song is chosen at random, and repetitions are
allowed, and every outcome is equally likely (an “outcome” is a particular song,
not a genre).
1032 Alternative 83 Electronic 56 Metal
330 Blues 508 Folk 2718 Other
275 Books & Spoken 183 Gospel 1786 Pop
1468 Children’s Music 82 Hip-Hop 403 R&B
921 Classical 564 Holiday 8286 Rock
6169 Country 537 Jazz 1432 Soundtrack
178 Easy Listening 106 Latin 216 World
Let A be the event that the first song is either blues or jazz. Let B be the
event that the second song is jazz. Let C be the event that the third song is
blues or rock.
Notice A and B are independent. Also, A and C are independent. Also, B
and C are independent. In the scenario when song repetitions are allowed, the
type of one song does not affect the types of other songs.

So far, we have only characterized what it means for a pair of events to be

“independent.” Now, we see the need to discuss the conditions that are necessary
for a collection of three or more events to be collectively called “independent.”
Definition 3.12. Independence of three events
A collection three events A, B, C is called (mutually) independent if all four
of the following are satisfied:

P (A ∩ B) = P (A)P (B)
P (A ∩ C) = P (A)P (C)
P (B ∩ C) = P (B)P (C)
P (A ∩ B ∩ C) = P (A)P (B)P (C)

Example 3.11 (continued)

For instance, in the example above, about various combinations of blues and
jazz and rock songs, the collection of events A, B, C is independent.

  
330 + 537 537
P (A ∩ B) = = P (A)P (B)
27,333 27,333
  
330 + 537 330 + 8286
P (A ∩ C) = = P (A)P (C)
27,333 27,333
 3.1. Introduction 41
  
537 330 + 8286
P (B ∩ C) = = P (B)P (C)
27,333 27,333
   
330 + 537 537 330 + 8286
P (A ∩ B ∩ C) = = P (A)P (B)P (C)
27,333 27,333 27,333

Definition 3.13. Independence for a finite collection of events

A finite collection of events A1 , A2 , . . . , An are called (mutually) independent
if, for every subcollection of the events, the probability of the intersection
is equal to the product of the probabilities of the individual events in the
collection.
Independence for an infinite collection of events
An infinite collection of events A1 , A2 , . . . is called (mutually) independent if
every finite collection of the events is independent.

Example 3.14. Consider a student who flips twenty coins in a row. Let Aj
denote the event that the jth coin shows a head. Then the events A1 , . . . , A20
are independent.

Example 3.15. Consider a student who flips coins for an arbitrarily long
amount of time. As before, let Aj denote the event that the jth coin shows
a head. Again, the individual coin flips do not impact each other, so the collec-
tion of all of the Aj ’s is independent.

Example 3.16. If Aj represents the event that there are two or more errors
on the jth page of a book, then the collection of Aj ’s is perhaps independent,
because the errors on the individual pages of a book should not affect the errors
that occur on other pages of the book.

Example 3.17. The lifetimes of 100 randomly selected light bulbs are mea-
sured. Let Aj denote the event that the jth bulb lasts for at least 60 days.
Then the collection of 100 events, A1 , . . . , A100 , is independent.
42 Chapter 3. Independent Events

Example 3.18. Two hundred customers’ purchases are randomly inspected

at the grocery store. The event Aj denotes the event that the jth customer
purchased at least 3 dairy items. Then the collection of two hundred events,
A1 , . . . , A200 , are again (perhaps) independent.

When a finite or infinite sequence of random phenomenon are observed, and

the results are recorded, we often use the word trials to denote such a collection
of events. Trials are usually independent from each other, although they do not
need to be independent. For instance, if we repeatedly draw cards from a deck,
looking for the ace of spades, we might call each draw a “trial,” regardless of
whether the cards are replaced before the next trial. If the cards are replaced
(and reshuffled!), then the trials are independent. If the cards are not replaced,
then the trials are dependent.
In Examples 3.14 and 3.15, each coin flip is a trial. In Example 3.16, each
examination of a page of a book is a trial. In Example 3.17, each test of the
lifetime of a bulb is a trial. In Example 3.18, the examination of a customer’s
purchase is a trial.

Example 3.19. A student flips a coin until the tenth head appears. See Fig-
ure 3.2. Let A denote the event that at least 3 flips are needed between the 7th
and 8th heads; let B denote the event that at least 3 flips are needed between
the 8th and 9th heads. Then A and B are independent. The coin flips are trials.

H T T T ··· H T T T ··· H
7th head 8th head 9th head

Figure 3.2: The number of flips between the 7th and 8th heads do not affect
the number of flips between the 8th and 9th heads.

3.2 Some Nice Facts about Independence

Theorem 3.20. Independence among complements
When events are independent, their complements are too, i.e., if A, B are
independent, then Ac , B are independent, A, B c are independent, and Ac , B c
are independent too.
 3.3. Probability of Good Occurring before Bad 43

To see this, if A, B are independent, then

P (A)P (B) = P (A ∩ B).

Subtracting both sides from P (B) yields

P (B) − P (A)P (B) = P (B) − P (A ∩ B),

i.e., P (B)(1 − P (A)) = P (B \ A). So

P (B)P (Ac ) = P (B ∩ Ac ),

i.e., Ac , B are independent.

Switching A and B, it follows too that if A, B are independent then A, B c
are independent too. Replacing B by B c in the previous paragraph, we see that
if A, B c are independent then Ac , B c are independent too.

Remark 3.21. These nice rules about independence between complements of

events also extend to collections of three or more events.

Example 3.22. Two randomly chosen people are selected from a large college
campus, and their heights are measured.

Let A denote the event that the height of the first person is 70 inches or
greater; let B denote the event that the height of the second person is less than
68.5 inches. Then A and B are independent.
Let C denote the event that the first student’s height is less than 68.5 inches.
Then A and C are disjoint, so by Remark 3.5, A and C are dependent too.

3.3 Probability of Good Occurring before Bad

Example 3.23. Consider a very large container of carbonated beverages, of

which 37% are apple flavor, 20% are orange flavor, and the other 43% are other
flavors (cherry, lemon, etc.). If we repeatedly reach for a beverage until we get
an apple or orange flavor, what is the probability that an apple flavored drink
appears first?

Let An denote the event that the nth trial is appleSand none

of the earlier
∞
trials are apple or orange. Then we are looking for P n=1 n .
A
44 Chapter 3. Independent Events

Notice that the An ’s are disjoint. If A3 occurs, then apple first appears on
the 3rd trial, so apple cannot appear for the first time on the 1st trial, or 2nd
trial, so neither A1 nor A2 can occur. Since the An ’s are disjoint, then
[ ∞  X ∞
P An = P (An )
n=1 n=1

Also P (An ) = (1 − 0.37 − since we need n − 1 choices that are

0.20)n−1 (0.37),
not apple or orange, followed by a choice that is apple.
So the desired probability is
∞
X ∞
X
P (An ) = (1 − 0.37 − 0.20)n−1 (0.37)
n=1 n=1
0.37
=
1 − (1 − 0.37 − 0.20)
0.37
=
0.37 + 0.20
= 0.65.
This idea works much more generally and can be helpful in many situations in
probability theory.

Theorem 3.24. Consider a sequence of independent trials, each of which can

be classified as good, bad, or neutral, which happen (on any given trial) with
probabilities p, q, and 1 − p − q, respectively. (We do not necessarily have
q = 1 − p here, although that case is allowed.) Then the probability that
something good happens before something bad happens is p/(p + q).
To see this, consider a sequence of independent events A1 , A2 , A3 , . . ., where the
nth event indicates that something good happens on the nth trial, and neither
good nor bad things happen on the previous trials. The same kind of reasoning
from the apples and S oranges argument works here.
∞
We compute P n=1 An . The An ’s are disjoint, since something good

cannot happen for the first time on two different trials! Since the An ’s are
disjoint, this gives
[∞  X ∞
P An = P (An ).
n=1 n=1
Also, An occurs if n − 1 neutral trials are followed by a good one, so
P (An ) = (1 − p − q)n−1 p.
So the desired probability is
∞ ∞
[  X 1 p
P An = (1 − p − q)n−1 p = p= .
1 − (1 − p − q) p+q
n=1 n=1
So the probability something good happens before something bad happens is
p/(p + q).
 3.4. Exercises 45

3.4 Exercises
3.4.1 Practice
Exercise 3.1. Graduation. Jack and Jill are independently struggling to
pass their last (one) class required for graduation. Jack needs to pass Calcu-
lus III, but he only has probability 0.30 of passing. Jill needs to pass Advanced
Pharmaceuticals, but she only has probability 0.46 of passing. They work in-
dependently. What is the probability that at least one of them gets a diploma?
Exercise 3.2. Japanese pan noodles. Ten students order noodles at a cer-
tain local restaurant. Their orders are placed independently. Each student is
known to prefer Japanese pan noodles 40% of the time (it is a very popular and
tasty dish!).
a. What is the probability that all ten of the students order Japanese pan
noodles?
b. What is the probability that none of the students order Japanese pan
noodles?
c. What is the probability that at least one of the students orders Japanese
pan noodles?
Exercise 3.3. Off to the races. Suppose Mike places three separate bets on
three separate horse races at three separate tracks. Each bet is for a specific
horse to win. His horse in race 1 wins with probability 1/5. His horse in race 2
wins with probability 2/5. His horse in race 3 wins with probability 3/5. What
is the probability that he made the correct bet in exactly one of these three
races?
Exercise 3.4. Early class. Consider these 3 independent trials: On Monday
you wake up 45 minutes before class, and the probability that you get to class
on time is 0.98. On Tuesday you wake up 32 minutes before class, and your
chance of being on time is 0.71. On Wednesday you wake up very, very late,
and your probability of being on time is only 0.16.
a. What is the probability that you were on time to class all 3 days?
b. What is the probability that you were never on time?
c. What is the probability that you were on time at least 1 day?
Exercise 3.5. Home for the holidays. A holiday flight from New York to
Indianapolis has a probability of 0.75 each time it flies (independently) of taking
less than 4 hours.
a. What is the probability that at least one of 3 flights arrives in less than
4 hours?
b. What is the probability that exactly 2 of the 3 flights arrive in less than
4 hours?
46 Chapter 3. Independent Events

3.4.2 Extensions
Exercise 3.6. Hoops. Your sister is playing basketball. She makes 4 tosses to
a lowered basketball hoop, and whether the ball goes in each time is independent
of the other trials. Her chance of making the basket on a trial is 60%.
For each j with 0 ≤ j ≤ 4, what is the probability that she makes exactly j
baskets?
Exercise 3.7. Abstract art. A painter has three different jars of paint colors
available, in colors green, yellow, and purple. She wants to paint something
abstract, so she blindfolds herself, randomly dips her brush, and paints on the
canvas. She continues trying paint jars, without replacement, until all three
have been used. (Her assistant helps with this blindfolded process!) So sample
space S is
S = {(G, P, Y ), (G, Y, P ), (P, G, Y ), (P, Y, G), (Y, G, P ), (Y, P, G)}.
Let A be the event that purple is found in the second jar tested by the painter.
Let B be the event that green is found before yellow. Are events A and B
independent?
Exercise 3.8. Even versus four or less. Roll a die. Let A be the event that
the outcome on the die is an even number. Let B be the event that the outcome
on the die is 4 or smaller. Let C be the event that the outcome on the die is 3
or larger.
a. Are A and B independent?
b. Are B and C independent?
Exercise 3.9. Vegetarian dilemma. In a very large collection of sandwiches,
40% are cheese, 45% have steak, and 15% have tofu. A person is vegetarian
and therefore samples the sandwiches randomly until finding a cheese or tofu
sandwich. What is the probability that they find a cheese sandwich before
finding a tofu sandwich?
Exercise 3.10. Guessing on an exam. While taking a probability exam,
you come to three questions that you have no clue how to answer. You would
have known the answers if you had taken the time to study the night before
instead of going to a party, but you did not make a good life choice, and you
vow to never party on a school night again if you fail this exam. Each question
on the exam is multiple choice with the correct answer being either a, b, c, d,
or e. (Your guesses are independent.)
What is the probability that:
a. you randomly guess the right answer to all three questions?
b. you randomly guess the right answer to none of the three questions?
c. you randomly guess the right answer to exactly one of the three questions?
d. you randomly guess the right answer to exactly two of the three questions?
e. Do the probabilities in parts a–d sum to 1?
 3.4. Exercises 47

3.4.3 Advanced
Exercise 3.11. Can the sum be greater than 1? Is it possible to have two
independent events A and B with the property that

P (A) + P (B) > 1 ?

If your answer is “no,” give a brief justification of why this is impossible.

If your answer is “yes,” please give a brief example, in which you list the
numbers P (A) and P (B) and also P (A ∪ B) and P (A ∩ B).

Exercise 3.12. Seating arrangements. Alice, Bob, Catherine, Doug, and

Edna are randomly assigned seats at a circular table in a perfectly circular
room. Assume that rotations of the table do not matter, so there are exactly
24 possible outcomes in the sample space.
Bob and Catherine are married. Doug and Edna are married.
Let T denote the event that Bob and Catherine are sitting next to each
other. Let U be the event that Alice and Bob are sitting next to each other.
Are events T and U independent?

Exercise 3.13. Political survey. On a large campus, 53% of the students are
Democrats, and 47% are Republicans. A political survey is conducted. Assume
that the students respond independently. How many students are needed, so
that the probability of at least 1 Democratic participant exceeds 99%?
Chapter 4

Conditional Probability

Probability is the very guide of life.

—The Analogy of Religion by Joseph Butler (Knapton, 1736)

Your dad is visiting you at college, and you have taken him to lunch in your
dorm’s dining hall to give him the full college student experience. Your dad
is a bit of a health nut and loves vegetables, but he thinks some vegetables
pair better than others. For example, he’s more likely to put broccoli and
cauliflower together than broccoli and green beans. Your cafeteria has a fairly
extensive selection of vegetables available in the lunch buffet. If you know
your dad already picked up cauliflower, what’s the chance he will also pick up
broccoli? How does knowing that your dad already picked up cauliflower change
the probability that he will pick up broccoli compared to when you walked in
the door before he had selected any vegetables?

4.1 Introduction
When we have some additional information about a random phenomenon, we
can take advantage of the concept of conditional probability. When we know
(or assume) something about a random phenomenon in advance, it allows us to
essentially shrink the sample space to a smaller set of possible outcomes. This
fundamentally alters the probabilities. Consider the following example:

Example 4.1. A student on a large college campus is chosen at random. We

want to know the probability that the student gets a job next summer, given
that the student is pursuing a major in French.

48
 4.1. Introduction 49

An advantage of conditional probability is that, given some information, we can

effectively narrow the sample space. This often makes it easier to compute a
probability. E.g., we are not interested in the entire sample space of students
at the university. We only care about the probability that the student gets a
job, given that they are studying French. There could be (say) 40,000 possible
outcomes when choosing a general student from the whole campus, i.e., the
original sample space is large. The (much smaller) set of students who study
French is more feasible to analyze. So we change our view from the original
sample space (all of the college’s students) to just the French majors (a much
smaller sample space), and the problem gets easier.

If A is the event that the randomly selected student gets a job, and B is the
event that the randomly selected student is a French major, the conditional
probability of A given B is written as P (A | B). In P (A|B), the bar
is read as “given.”
The reason for conditional probability is that we sometimes have additional
information that we want (or need) to incorporate into the problem. In general,
we denote conditional probability as follows:

Definition 4.2. The conditional probability of event A, given event B

(i.e., B has occurred, or will occur, or is assumed to occur, etc.) is written as
P (A | B).
In other words, conditional probability is written as

P (event under consideration | a given event).

In general, if event B has nonzero probability (i.e., P (B) > 0), then the
conditional probability P (A | B) of A given B is defined as

P (A ∩ B)
P (A | B) = .
P (B)

Equivalently,
P (A ∩ B) = P (B)P (A | B).

The statement P (A ∩ B) = P (B)P (A | B) provides another very natural

interpretation of conditional probability. What is the probability that A and B
simultaneously occur? Event B needs to occur (hence, the presence of P (B))
and—once B is given—the likelihood that A occurs too is P (A | B). Thus
P (A ∩ B) is the product of these two percentages, P (B)P (A | B).
Conditional probabilities reduce our sample space to just the part where B
has occurred. This shrinks our world and often makes calculations easier to
perform, i.e., having the knowledge that some event B occurred can often be
helpful for computing the probability that another event occurred too.
50 Chapter 4. Conditional Probability

Remark 4.3. Consider event B with P (B) > 0. Recall A and B are inde-
pendent exactly when P (A)P (B) = P (A ∩ B), i.e., exactly when

P (A ∩ B)
P (A) = ,
P (B)

but the right-hand side is always equal to the conditional probability P (A | B).
So A and B are independent exactly when

P (A) = P (A | B),

i.e., when B’s occurrence does not affect the probability of A occurring.

Theorem 4.4. If P (B) > 0, then A and B are independent if and only if
P (A) = P (A | B), i.e., when B’s occurrence does not affect the probability of
A occurring.

Example 4.5. When a die is rolled, let B be the event that the outcome is
“odd.” Then, for example,

P ({3, 5} | B) = 2/3, P ({1, 3, 5} | B) = 1, P ({4, 6} | B) = 0.

Since we know that B occurs, the sample space has essentially been reduced
from the original sample space, with 6 outcomes,

1 2 3 4 5 6

to a smaller, conditional sample space, with only 3 outcomes,

1 2 3 4 5 6

All of the probabilities from the original model are scaled by a factor of 1
P (B)
to get the conditional probabilities.

Example 4.6. Suppose that a friend will call one time during the next 60
minutes. We measure (in minutes) the waiting time until she calls. Let A =
{x | x ≤ 30}, and let B = {x | x ≤ 10}.
Given that A occurs, the probability of B occurring is 1/3; in other words,

P (B | A) = 1/3.

B A

0 5 10 15 20 25 30 35 40 45 50 55 60
 4.1. Introduction 51

Once we know that A occurs, we can ignore any of the outcomes that are
bigger than 30. Also, conditional probabilities satisfy all of the requirements of
probabilities, as we will see at the end of this chapter. So, given that A occurs,
if we know B occurs 1/3 of the time, it must be the case that B c occurs the
other 2/3 of the time. Thus, the conditional probability of B not occurring
(given A occurred) must be 2/3, i.e.,

P (B c | A) = 2/3.

Example 4.7. On a dartboard let C = {9, 12, 5, 20, 1, 18, 4} be the event that
the dart lands in the upper portion of the dartboard. If A = {9, 12, 5}, we have
P (A | C) = 3/7, as in Figure 4.1.

5 20 1
12 18
9 4
14 13
11 6
8 10
16 15
7 2
19 3 17 (or could miss the
dartboard altogether)

Figure 4.1: Event C is shaded in blue. More specifically, the event A ∩ C is

shaded in dark blue, and the event Ac ∩ C is shaded in light blue.

We can also consider some events that are not completely within C. For
instance, say B = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10}. Some of B overlaps with C and
some does not; this is depicted in Figure 4.2.

5 20 1
12 18
9 4
14 13
11 6
8 10
16 15
7 2
19 3 17 (or could miss the
dartboard altogether)

Figure 4.2: Event C is shaded in blue; event B is shaded with lines; so event
B ∩ C is the blue part that is shaded with lines.
52 Chapter 4. Conditional Probability

If we compute P (B | C), we will end up basically ignoring that part of B

outside of C, because these will not be relevant. We compute

P (B ∩ C) P ({1, 4, 5, 9}) 4/20

P (B | C) = = = = 4/7.
P (C) P (C) 7/20

So the outcomes 8, 7, 3, 2, 10, 6 from B were essentially not used in this solution.
These are not possible outcomes, because they are outside C. Event C can be
viewed as the (new) sample space, when calculating probabilities conditioned
on C.

Example 4.8. In Exercise 2.4, at a random meal during a parent weekend in

the dining hall, a student notices the food chosen by her father. Let A, B, C
be the events that his meal include Artichokes, Broccoli, or Cauliflower. These
events have the property that

P (B) = 0.39, P (C) = 0.44, P (B ∩ C) = 0.13.

If he orders cauliflower, the conditional probability he also orders broccoli is

0.295:
P (B ∩ C) 0.13
P (B | C) = = = 0.295.
P (C) 0.44
If he orders broccoli, the conditional probability he also orders cauliflower is
0.333:
P (B ∩ C) 0.13
P (C | B) = = = 0.333.
P (B) 0.39
Notice P (B | C) and P (C | B) are not usually the same.

Example 4.8 demonstrates that we cannot swap the events A and B in the
conditional probability P (A | B).

Example 4.9. Consider events A and B (pictured in the Venn diagram in

Figure 4.3) such that

P (A) = 0.10 and P (B) = 0.42 and P (A ∩ B) = 0.05

Now we calculate
P (A ∩ B) 0.05
P (A | B) = = = 5/42 = 0.1190.
P (B) 0.42
 4.2. Distributive Laws 53

P (B \A) = 0.37
B
P (A ∩ B = 0.05)
P (A\B = 0.05) A S

Figure 4.3: Venn diagram for events A and B.

This is a very small probability. Intuitively: The event B is very large, compared
to A ∩ B. So if we know that B has occurred, the probability of A ∩ B is very
small. On the other hand,

P (A ∩ B) 0.05
P (B | A) = = = 1/2.
P (A) 0.10

Intuitively: The event A is twice as large as A ∩ B. The overlapping part of A

with B is half of the size of A itself. Once we know that A has occurred, the
outcome will be in A ∩ B half the time (and of course in A \ B the other half
of the time).

4.2 Distributive Laws

The distributive laws for events are helpful to mention at this point because
they will be used when showing that conditional probabilities satisfy all of the
properties of probability.

Theorem 4.10. Distributive Laws For any events A1 , A2 , . . .,

[  [
Aj ∩ B = (Aj ∩ B);
j j

and \  \
Aj ∪B = (Aj ∪ B);
j j
Sn
The unions Sand intersections over j can be finite, written as j=1 Aj , or can
be infinite, ∞
j=1 Aj .

The first of theSdistributive laws is perhaps easier to understand: An out-

come is found in exactly when the outcome is found in B and in


A
j j ∩ B
at least one of the
S Aj ’s, or equivalently, found in Aj ∩ B for at least one j, or
equivalently, in j (Aj ∩ B).
54 Chapter 4. Conditional Probability

For the second distributive law, an outcome is in Aj ∪ B if and only

T

j
if:

1. the outcome is in all of the Aj ’s, or

2. the outcome is in B, or

3. both are true, i.e., the outcome is in all of the Aj ’s and also in B.

This is the same as the requirement for an outcome to be in j (Aj ∪ B); the
T
outcome must be in B, or if not in B, it must be in all of the Aj ’s.

Example 4.11. A student is chosen at random. Let A1 , A2 , A3 be the events

that the student is majoring in biology, chemistry, or mathematics, respectively.
(Students can pursue more than one major.) Let B be the event that the student
is on the volleyball team.
S3
Event j=1 Aj ∩ B occurs if and only if the student pursues at least one of

these
S3 three majors and is T3also on
the volleyball team; this is the same event as
j=1 (Aj ∩ B). Event j=1 Aj ∪ B occurs if and only if the student pursues
all three T
of these majors or is on the volleyball team (or both); this is the same
event as 3j=1 (Aj ∪ B).

4.3 Conditional Probabilities Satisfy the Probability

Axioms
In this section, we show that, when conditioning on any event B, the con-
ditional probabilities satisfy all of the three axioms of probabilities,
discussed in Section 2.1.

Remark 4.12. The probabilities of the form P (A) that we have studied in
Chapters 2 and 3 were unconditional. They did not assume that any event
occurred. Such probability could be treated as “conditional” if we just make
S the condition, i.e., P (A) is the same as P (A | S).

The intuitive way to think about this is that, when B is a given event
throughout a problem or scenario, it is just the same as if B is written into the
fabric of what is known. So everything is conditioned on B. The event B can
just be viewed as replacing the original sample space S.
 4.3. Conditional Probabilities Satisfy the Probability Axioms 55

Theorem 4.13. Consider an event B with P (B) > 0.

1. For any event A,

0 ≤ P (A | B) ≤ 1;

2. For the sample space S,

P (S | B) = 1;

3. For any disjoint events A1 , A2 , . . .,

∞
[  ∞
X
P Aj B = P (Aj | B).
j=1 j=1

To see (1), notice 0 ≤ P (A ∩ B) ≤ P (B) since A ∩ B is a subset of B.

Dividing throughout by P (B) gives 0 ≤ P P(A∩B)
(B) ≤ 1.
P (S∩B)
For (2), we observe P (S | B) = P (B) , but S ∩ B = B, so P (S | B) =
P (B)
P (B)= 1.
For (3), consider disjoint events A1 , A2 , . . .. By the definition of conditional
events,
∞ S∞


P j=1 Aj ∩ B
[ 
P Aj B = . (4.1)
P (B)
j=1

Now we observe that

∞
[  ∞
[
Aj ∩B = (Aj ∩ B)
j=1 j=1

by the first distributive law, explained above. So Equation (4.1) becomes

∞ S∞

P j=1 (Aj ∩ B)
[ 
P Aj B = .
P (B)
j=1

S∞
The
P∞ Aj ∩ B are disjoint since the Aj ’s are disjoint. So P


j=1 (Aj ∩ B) =
j=1 P (Aj ∩ B). So we conclude that

∞ P∞ ∞ ∞
j=1 P (Aj ∩ B)
[  X P (Aj ∩ B) X
P Aj B = = = P (Aj | B).
P (B) P (B)
j=1 j=1 j=1

Example 4.14. When rolling a die, let A denote the event consisting of out-
comes {3, 5}, and let B denote the event consisting of outcomes {1, 2, 3}. Find
P (A | B) and P (Ac | B).
56 Chapter 4. Conditional Probability

We have P (A | B) = P P(A∩B) 1/6

(B) = 1/2 = 1/3. Since conditional probabilities satisfy
the rules of probability, then the conditional probability of the complement
must be 2/3, i.e., P (Ac | B) = 2/3. (We can check this directly: P (Ac | B) =
P (Ac ∩B)
P (B) = 2/6
1/2 = 2/3.)

Example 4.15. When rolling a die, let A denote the event consisting of out-
comes {1, 2, 3}, and let B denote the event consisting of outcomes {3, 4}.

Then P (A | B) = P P(A∩B) 1/6

(B) = 1/3 = 1/2 = P (A). Since P (A | B) = P (A), then
A and B are independent (as we also verified in Example 3.8).

Example 4.16. Let A denote the event that the amount of rain on July 1,
2011, in Lafayette, Indiana, is 0.10 inches or more. Let B denote the event
that the amount of rain on July 1, 2012 (i.e., one year later), in Lafayette, is
0.10 inches or more. The occurrence of A should not affect the likelihood of
occurrence of B, so P (B | A) = P (B), and thus A and B are independent.

4.4 Exercises
4.4.1 Practice
Exercise 4.1. Pick ten songs. As in Example 2.27, a student hears ten songs
(in a random shuffle mode) and uses F to denote when a song belongs to her
favorite type of music, and N for not-favorite, so the sample space consists of
all ten-tuples of F ’s and N ’s. For each song, the probability that the song is
one of her favorite type can be called p, and the probability that the song is not
one of her favorite type is 1 − p. Define

A = {(N, N, N, x4 , . . . , x10 ) | xj ∈ {F, N }},

B = {(x1 , F, x3 , F, x5 , F, x7 , F, x9 , F ) | xj ∈ {F, N }},
C = {(x1 , x2 , x3 , x4 , x5 , F, F, F, F, F ) | xj ∈ {F, N }},

so that, for instance, P (A) = (1 − p)3 , and P (B) = P (C) = p5 . Find the
following conditional probabilities:

P (B | C), P (C | B), P (A | B), P (B | A), P (A | C), P (C | A).

 4.4. Exercises 57

Exercise 4.2. Dining with Dad. Consider the events from Exercise 2.4, i.e.,
at a random meal during a parent weekend in the dining hall, a student notices
the food chosen by her father. Let A, B, C be the events that his meal include
Artichokes, Broccoli, or Cauliflower. These events have the property that:
P (A) = 0.35; P (B) = 0.39; P (C) = 0.44; P (A ∩ B) = 0.13; P (A ∩ C) =
0.12; P (B ∩ C) = 0.13.
Find the following conditional probabilities:
P (B | C), P (C | B), P (A | B), P (B | A), P (A | C), P (C | A).

Exercise 4.3. Songs by genre. As in Exercise 2.1 and Example 3.11, a

song is chosen at random from a person’s mp3 player. The student makes a
partition of the sample space, according to genre of music. The table below
gives the number of outcomes in each part of the partition. There are 27,333
songs altogether.
1032 Alternative 83 Electronic 56 Metal
330 Blues 508 Folk 2718 Other
275 Books & Spoken 183 Gospel 1786 Pop
1468 Children’s Music 82 Hip-Hop 403 R&B
921 Classical 564 Holiday 8286 Rock
6169 Country 537 Jazz 1432 Soundtrack
178 Easy Listening 106 Latin 216 World
Let A be the event that a song is either blues, jazz, or rock, i.e.,

A = {x | x is a blues, jazz, or rock song},

when one song is chosen at random. Let B, J, R denote the events that the song
is a blues, jazz, or rock song, respectively.

Find P (B | A) and P (J | A) and P (R | A). Hint: Since B, J, R are disjoint

and A = B ∪ J ∪ R, these three answers should sum to 1.

Exercise 4.4. Golf. In the PGA, on par 3 holes, golfers hit the green in one
shot 80% of the time. In fact, 20% of the time, they hit the green in one shot
and then need only one putt to complete the hole; so 60% of the time, they hit
the green in one shot but are unsuccessful on their putt. What is the probability
that a PGA golfer only needs one putt, given that he hits the green in one shot?

Exercise 4.5. Parity of spinning. A spinner has the left side (numbers 1, 2,
3, 4, and 5) colored red and the right side colored white (numbers 6, 7, 8, and
9), with all numbers equally likely.

a. What is the probability the spinner lands on an odd number?

b. Given that the spinner landed on an odd number, what is the chance the
spinner landed on a white number?
c. Given that the spinner landed on a white number, what is the chance it
landed on an odd number?
58 Chapter 4. Conditional Probability

Exercise 4.6. Conditioning on cards. Draw one card from a shuffled deck
of 52 cards.

a. What is the probability that the card is a spade if you know the card is
a 7?
b. What is the probability that the card is a spade if you know the card is
black?
c. What is the probability that the card is a 7 if you know the card is a
spade?
d. What is the probability that the card is a 7 if you know the card is black?
e. What is the probability that the card is black if you know the card is a
spade?
f. What is the probability the card is black if you know the card is a 7?

Exercise 4.7. Puppets. You are previewing movies for your young nephew.
You have 1,284 movies available to view, 272 of which are G-rated. Your nephew
enjoys movies with puppets, which make up 94 of your G-rated movies. There
are only 30 movies with puppets that are not G-rated. If you happen to pick a
movie that has puppets, what is the probability that it is G-rated?

4.4.2 Extensions

Exercise 4.8. Dice. You roll two dice. Let A be the event that the sum of the
dice is an even number. Let B be the event that the two results are different.
If B has occurred, what is the probability A has also occurred?

Exercise 4.9. Pair of dice. Roll a pair of dice. Given that the two dice have
different values, find the probability that the sum of the dice is an even number.

Exercise 4.10. Pair of dice. Roll a pair of dice. Given that the sum of the
pair of dice is 9 or larger, find the probability that the sum of the pair of dice
is exactly 10.

Exercise 4.11. Random sexes. A couple has two children. At least one is a
boy. What is the probability that the couple has one child of each sex?

Exercise 4.12. More random sexes. A couple has three children. They are
not all girls.

a. What is the probability of exactly one boy?

b. What is the probability of exactly two boys?
c. What is the probability of exactly three boys?
 4.4. Exercises 59

4.4.3 Advanced
Exercise 4.13. Seating arrangements. Alice, Bob, Catherine, Doug, and
Edna are randomly assigned seats at a circular table in a perfectly circular
room. Assume that rotations of the table do not matter, so there are exactly 24
possible outcomes in the sample space. Bob and Catherine are married. Doug
and Edna are married. Given that Bob and Catherine are sitting next to each
other, find the conditional probability that Doug and Edna are sitting next to
each other.

Exercise 4.14. Randomly choose a page. Randomly open a 300-page book,

and mark a page.

a. Given that at least one of the digits on the chosen page is a 5, find the
probability that the page is 255.
b. Given that at least two of the digits on the chosen page are 5’s, find the
probability that the page is 255.

Exercise 4.15. Even more random sexes. A couple has n children, where
n ≥ 1 is fixed. They are not all girls. What is the probability of exactly j boys
(where j ≥ 1)?
 Chapter 5

Bayes’ Theorem

Are no probabilities to be accepted, merely because they are not certainties?

—Sense and Sensibility by Jane Austen (Egerton, 1811)

We all know that using cell phones while driving is dangerous, but many people
do it anyway. If an accident occurs, the first question an insurance company asks
the driver is, “Was anybody using a cell phone when the accident occurred?”
What is the probability of a randomly selected driver having an accident in the
month of September? How does that probability change if we know that driver
is a “regular” cell phone user? If a driver has an accident, what is the chance
the driver is a “regular” cell phone user? Why are these last two probabilities
not the same?

5.1 Introduction to Versions of Bayes’ Theorem

Consider two events that we are interested in, say A and B. Suppose that we
know how the occurrence of event A affects the occurrence of event B. How
can we use this information to evaluate the way that the occurrence of event B
affects the occurrence of event A? Bayes’ Theorem tells us exactly how to do
this.
Bayes’ Theorem lets us flip conditional probabilities. If the conditional
For “backward,” probability we know is backward from the conditional probability we want, use
think “Bayes.” Bayes’ Theorem.
The explanation for Bayes’ Theorem is straightforward; no memorizing is
necessary. We first write P (A ∩ B) in two ways, as P (B)P (A | B) and also
P (A)P (B | A):
P (B)P (A | B) = P (A)P (B | A).
Then we divide by P (B) throughout, and we get Bayes’ Theorem:

60
 5.1. Introduction to Versions of Bayes’ Theorem 61

Theorem 5.1. Bayes’ Theorem

For any two events A and B with nonzero probabilities,

P (A)P (B | A)
P (A | B) = .
P (B)

Example 5.2. In a certain household, 20% of the milk has two-percent milkfat,
and the other 80% of the milk is whole milk. The whole milk is spoiled 5% of
the time; overall, the milk is spoiled 4.7% of the time. Find the conditional
probability that, if you just poured a spoiled cup of milk from a random jug, it
is whole milk.

Let A denote the event that the milk came from a whole milk carton, and let
B denote the event that the milk was spoiled. Then we want to find P (A | B).
We are given P (B | A) = 0.05, i.e., the probability that a whole milk carton will
be spoiled is 5%. We are also told that P (A) = 0.80 and that P (B) = 0.047.
So we get

P (A)P (B | A) (0.80)(0.05)
P (A | B) = = = 0.85.
P (B) 0.047

So if the milk in a cup is spoiled, it is whole milk with probability 0.85.

Often we are not given P (B) directly, and we have to decompose P (B) by
writing:

P (B) = P (A ∩ B) + P (Ac ∩ B)
= P (A)P (B | A) + P (Ac )P (B | Ac )

We consider several examples like this.

Example 5.3. In a recent study about driving safety, drivers were classified
as either “regularly” or “rarely” using a cell phone while driving. Each person
who regularly talks on a cell phone while driving has a probability of 1/250 of
causing an accident in September. Each person who rarely talks on a cell phone
while driving has a probability of 1/2000 of causing an accident in September.
Also, 35% of people were classified as regularly using a cell phone while driving.
Suppose that a randomly selected person causes an accident in September.
What is the probability that the person is a “regular” cell phone user while
driving?
62 Chapter 5. Bayes’ Theorem

We write A for the event that the person who caused the accident was a
“regular” user, and B for the event that the person selected causes an accident
in September. We want to know P (A | B). We are given:

A is the event that B is the event that

the person is a the person selected causes an
“regular” cell phone user accident in September

P (A) = 0.35 P (B | A) = 1/250

so P (A ) = 0.65
c
P (B | Ac ) = 1/2000

We want P (A | B), which we get from

P (A)P (B | A)
P (A | B) = ,
P (B)

and we now have all of the pieces except P (B). There are 2 ways in which an
accident can happen: with a driver who regularly uses a cell phone or with a
driver who rarely uses a cell phone. We need to combine these possibilities for
the overall probability that there will be an accident.
We can compute P (B) by writing

P (B) = P (A ∩ B) + P (Ac ∩ B)
= P (A)P (B | A) + P (Ac )P (B | Ac )
= (0.35)(1/250) + (0.65)(1/2000)
= 0.001725

Thus,
P (A)P (B | A) (0.35)(1/250)
P (A | B) = = = 0.81.
P (B) 0.001725
So, given that a randomly selected person causes an accident in September, that
person is a “regular” user with probability 0.81.

There are several alternative formulations which can be derived directly from
Bayes’ Theorem. The first one is a generalization of the example about driver
safety.

Remark 5.4. We can decompose B into two parts: A∩B and B \A = Ac ∩B.
Since B = (A ∩ B) ∪ (Ac ∩ B) is a disjoint union, then if P (A) is not 0 or 1,

P (B) = P (A ∩ B) + P (Ac ∩ B) = P (A)P (B | A) + P (Ac )P (B | Ac ).

Substituting this into the denominator of the original statement of Bayes’ The-
orem, we get the following alternative formulation:
 5.1. Introduction to Versions of Bayes’ Theorem 63

Theorem 5.5. Bayes’ Theorem (Decomposition of Sample Space into

Two Parts)
If P (B) 6= 0 and 0 < P (A) < 1, then,

P (A)P (B | A)
P (A | B) = .
P (A)P (B | A) + P (Ac )P (B | Ac )

This is a really nice version because we can now compute P (A | B) using only
three numbers: P (A), P (B | A), and P (B | Ac ). The last piece of information,
P (Ac ), we can get for free, because we always have P (Ac ) = 1 − P (A).

Example 5.6. If a student chooses a playlist created by his roommate on his

mp3 player, he finds a song from his favorite type of music 62% of the time. If
he chooses a playlist created by his girlfriend, he finds a song from his favorite
type of music 88% of the time. These are his only two playlists. He chooses
his roommate’s playlist 40% of the time. If the student is listening his favorite
type of music, what is the probability he is using his roommate’s playlist?

Let A and Ac be the events that the song was chosen from his roommate’s or
girlfriend’s playlist, respectively. Let B be the event a song is from the student’s
favorite type of music. We want to find P (A | B). We are given:

A is the event that the song B is the event that the song
is from the roommate’s playlist is one of the favorite type

P (A) = 0.4 P (B | A) = 0.62

thus P (A ) = 0.6
c
P (B | Ac ) = 0.88

We can compute P (A | B) directly. Here are the steps:,

P (A ∩ B)
P (A | B) =
P (B)
P (A ∩ B)
=
P (A ∩ B) + P (Ac ∩ B)
P (A)P (B | A)
=
P (A)P (B | A) + P (Ac )P (B | Ac )
(0.40)(0.62)
=
(0.40)(0.62) + (0.60)(0.88)
= 0.32.

Thus, given that his favorite type of music is playing, he chose his roommate’s
playlist 32% of the time.
64 Chapter 5. Bayes’ Theorem

In Example 5.6, we split the sample space into two pieces, e.g., whether a song
was chosen from the playlist of a person’s roommate or girlfriend, or whether
the randomly chosen person regularly or rarely uses the cell phone while driving.
Sometimes it is helpful to be able to split the sample space into more than two
groups. Here is an example:

Example 5.7. Consider three types of computer monitors. The probabilities

that each type lasts more than five years are, respectively, 0.8, 0.7, 0.3. It is
known that 10% of monitors are type 1; 30% of monitors are type 2; and 60% of
monitors are type 3. What is the probability a randomly chosen monitor lasts
over 5 years? Given that a monitor lasts over 5 years, what is the probability
that it was a monitor of type 1? Type 2? Type 3?

Let B be the probability a randomly chosen monitor lasts over 5 years. Let Aj
be the event that a monitor is of type j. We want P (Aj | B). We are given:

Aj is the event that B is the event that

a monitor type is j a monitor lasts 5 years

P (A1 ) = 0.1 P (B | A1 ) = 0.8

P (A2 ) = 0.3 P (B | A2 ) = 0.7
P (A3 ) = 0.6 P (B | A3 ) = 0.3

The probability that a randomly chosen person’s monitor lasts over 5 years is

P (B) = P (A1 ∩ B) + P (A2 ∩ B) + P (A3 ∩ B)

= P (A1 )P (B | A1 ) + P (A2 )P (B | A2 ) + P (A3 )P (B | A3 )
= (0.1)(0.8) + (0.3)(0.7) + (0.6)(0.3)
= 0.47

Given that a randomly chosen person’s monitor lasts over 5 years, the proba-
bilities that it is of type 1, or type 2, or type 3 (respectively) are:

P (A1 )P (B | A1 ) (0.1)(0.8)
P (A1 | B) = = = 0.17
P (B) 0.47
P (A2 )P (B | A2 ) (0.3)(0.7)
P (A2 | B) = = = 0.45
P (B) 0.47
P (A3 )P (B | A3 ) (0.6)(0.3)
P (A3 | B) = = = 0.38
P (B) 0.47
The probabilities 0.17, 0.45, and 0.38 must add to 1 because the three possible
events A1 ∩ B, A2 ∩ B, A3 ∩ B are disjoint and their union is all of B.
 5.1. Introduction to Versions of Bayes’ Theorem 65

The following version of Bayes’ Theorem allows us to split the sample space
into a finite number of different pieces:

Remark S 5.8. Let A1 , A2 , . . . , An be any finite partition of the sample space

(i.e., nk=1 Ak = S and the Ak ’s are disjoint). Then a natural partition of B
into n parts is A1 ∩ B, A2 ∩ B, . . . , An ∩ B, i.e., B = (A1 ∩ B) ∪ (A2 ∩ B) ∪
· · · ∪ (An ∩ B) is a union of disjoint events, so

P (B) = P (A1 ∩B)+· · ·+P (An ∩B) = P (A1 )P (B | A1 )+· · ·+P (An )P (B | An ).

Substituting into the denominator of the first version of Bayes’ Theorem, we get:
Theorem 5.9. Bayes’ Theorem (Decomposition of Sample Space into
Finitely Many Parts)
If P (B) > 0, and if A1 , A2 , . . . , An form a partition of S, with all P (Aj ) > 0,
then
P (Ak )P (B | Ak )
P (Ak | B) = .
P (A1 )P (B | A1 ) + P (A2 )P (B | A2 ) + · · · + P (An )P (B | An )
Thus, to compute P (Ak | B) we only need the P (Ak )’s and P (B | Ak )’s.
Finally, sometimes we need to split the sample space into infinitely many
pieces. A version of Bayes’ Theorem is possible for this situation too:
Remark 5.10. Let A1 , A2 , . . . be a partition of the sample space (i.e.,
∞
j=1 Aj = S and the Aj ’s are disjoint). Then we have the decomposition of B
S
into infinitely many parts, A1 ∩B, A2 ∩B, etc. So B = (A1 ∩B)∪(A2 ∩B)∪· · ·
is a union of disjoint events. Thus
∞
X ∞
X
P (B) = P (A1 ∩ B) + P (A2 ∩ B) + · · · = P (Aj ∩ B) = P (Aj )P (B | Aj ).
j=1 j=1

Substituting this into the denominator of the original statement of Bayes’ The-
orem, we get the following alternative formulation:
Theorem 5.11. Bayes’ Theorem (Decomposition of Sample Space
into Infinitely Many Parts)
If P (B) > 0, and if A1 , A2 , . . . form a partition of S, with all P (Aj ) > 0, then

P (Ak )P (B | Ak )
P (Ak | B) = P∞ .
j=1 P (Aj )P (B | Aj )

Example 5.12. Consider a game with two stages. In the first stage, a player
flips a fair coin until a head appears (usually this only requires a small number
of flips). Say that it takes j flips to get this first head. Then, in the second
stage, he random draws an integer between 1 and 2j to get his winnings. E.g.,
if the player takes 3 flips to get a head, he wins an amount between 1 and 8.
66 Chapter 5. Bayes’ Theorem

Given that the player wins 1 dollar, what is the probability that the coin was
flipped exactly 6 times, i.e., that j = 6?
Let Aj be the event that exactly j coin flips are needed to get a head on the
coin. Let Bk be the event that the player wins k dollars. We want P (A6 | B1 ).
We have
P (B1 | A6 )P (A6 )
P (A6 | B1 ) = P∞ .
j=1 P (B1 | Aj )P (Aj )

Also, P (Aj ) = 1/2j . Given that Aj occurred, the player will win one of the
amounts between 1 and 2j inclusive, and these 2j outcomes are equally likely,
so the player wins 1 with probability 1/2j . So we have
1 1 1
6 26 6
P (A6 | B1 ) = P∞2 1 1 = P∞ 4
1 j


j=1 2j 2j j=1 4

As a side computation for the series above, we compute

∞  j ∞  
!
1X 1 j
 
X 1 1 1 1 1 1
= = 1 = = .
4 4 4 4 1− 4 4 3/4 3
j=1 j=0

So we conclude that
1/46 3
P (A6 | B1 ) = = 6 = 0.000732.
1/3 4

5.2 Multiplication with Conditional Probabilities

We close this chapter with one more idea that is useful in working with condi-
tional problems. We assume throughout this section that all of the intersections
of events here have positive probability, so that we are not dividing by zero.

Example 5.13. We already know that

P (A1 ∩ A2 )
P (A2 | A1 ) = ,
P (A1 )

or equivalently
P (A1 ∩ A2 ) = P (A1 )P (A2 | A1 ).

Informally, if we want A1 and A2 to occur:

we just need A1 to occur, and then,

given that A1 occurs, we need A2 to occur too.
 5.2. Multiplication with Conditional Probabilities 67

Example 5.14. Next we note that

P (A1 ∩ A2 ∩ A3 )
= P (A3 | A1 ∩ A2 ).
P (A1 ∩ A2 )
Multiplying on the left- and right-hand sides of the equation from Example 5.13,
yields
P (A1 ∩ A2 ∩ A3 ) = P (A1 )P (A2 | A1 )P (A3 | A1 ∩ A2 ).
Informally, if we want A1 and A2 and A3 to occur:

we just need A1 to occur, and then,

given that A1 occurs, we need A2 to occur too, and then,
given that A1 and A2 occur, we need A3 to occur too.

1 ∩A2 ∩A3 ∩A4 )

Example 5.15. Similarly, we note that P (A P (A1 ∩A2 ∩A3 ) = P (A4 | A1 ∩A2 ∩A3 ).
Multiplying on the left- and right-hand sides of the equation from Example 5.14,
yields

P (A1 ∩ A2 ∩ A3 ∩ A4 ) = P (A1 )P (A2 | A1 )P (A3 | A1 ∩ A2 )P (A4 | A1 ∩ A2 ∩ A3 ).

Informally, if we want A1 and A2 and A3 and A4 to occur:

we just need A1 to occur, and then,

given that A1 occurs, we need A2 to occur too, and then,
given that A1 and A2 occur, we need A3 to occur too, and then,
given that A1 and A2 and A3 occur, we need A4 to occur too.

Similar reasoning continues, and we make the following general assumption:

Remark 5.16. Multiplication with Conditional Probabilities
For any events A1 , A2 , . . . , An with A1 ∩ A2 ∩ · · · ∩ An 6= ∅,

P (A1 ∩ A2 ∩ · · · ∩ An ) = P (A1 )P (A2 | A1 )P (A3 | A1 ∩ A2 )

× P (A4 | A1 ∩ A2 ∩ A3 )
· · · × P (An | A1 ∩ A2 ∩ · · · ∩ An−1 ).

Example 5.17. Consider a student’s playlist that has 10 rock songs and 12
country songs. The student chooses at random—with all selections equally
likely, and no repeats—three songs from the playlist. What is the probability
that all three are country songs?
68 Chapter 5. Bayes’ Theorem

Let A1 , A2 , A3 be the event that the first, second and third songs, respectively,
are country songs. Then the desired probability is

P (A1 ∩ A2 ∩ A3 ) = P (A1 )P (A2 | A1 )P (A3 | A1 ∩ A2 )

The probability of A1 is 12/22 since there are initially 22 songs, 12 of which are
country, and all of the selections are equally likely.

Next, given that A1 occurs, there are 21 songs remaining—10 rock songs
and 11 country songs—so P (A2 | A1 ) = 11/21.

Finally, given that A1 ∩ A2 occurs, there are 20 songs remaining—10 rock

songs and 10 country songs—so P (A3 | A1 ∩ A2 ) = 10/20.

So the desired probability is

P (A1 ∩ A2 ∩ A3 ) = P (A1 )P (A2 | A1 )P (A3 | A1 ∩ A2 )

= (12/22)(11/21)(10/20)
= 1/7.

This kind of example works much more generally:

Example 5.18. Consider any scenario in which there are N items of one type
and M items of another type, and we need to choose n items altogether. Suppose
all choices are equally likely. If we want to calculate the probability that all n
of the chosen items are of the second type, we can do this systematically.
We write A1 , A2 , . . . , An for the events that the first, second, third, ..., nth items
are of the second type. Then we use the fact that

P (A1 ∩ A2 ∩ A3 ∩ · · · ∩ An ) = P (A1 )P (A2 | A1 )P (A3 | A1 ∩ A2 )

· · · × P (An | A1 ∩ A2 ∩ · · · ∩ An−1 ).
 5.3. Exercises 69

As above, we have

M
P (A1 ) =
N +M
M −1
P (A2 | A1 ) =
N +M −1
M −2
P (A3 | A1 ∩ A2 ) =
N +M −2
.. ..
. .
M − (n − 1)
P (An | A1 ∩ A2 ∩ · · · ∩ An−1 ) =
N + M − (n − 1)

Multiplying this system of equations yields:

(M )(M − 1) · · · (M − n + 1)
P (A1 ∩ A2 ∩ · · · ∩ An ) = .
(N + M )(N + M − 1) · · · (N + M − n + 1)

5.3 Exercises
5.3.1 Practice

Exercise 5.1. Car safety. A car safety institute observes that 12% of cars on
the road are manufactured by Honda. They also observe that 98% of Honda
vehicles are classified as “safe” at the time of inspection. The percentage of all
cars (regardless of brand) classified as “safe” is 72%. A car is randomly chosen
on the road and inspected. It is classified as “safe.” What is the probability
that it is a Honda?

Exercise 5.2. Happy courses. Twenty percent of students will participate

in an Honors course this semester. Students who have an Honors course on
their schedule are known to have a 99% chance of fully enjoying their semester.
Students who are not in an Honors course during the semester have only a 30%
chance of fully enjoying their semester. With these assumptions, if a randomly
chosen student is fully enjoying his semester, what is the probability that he
has an Honors course on his schedule?

Exercise 5.3. Stubborn dogs. I am at a dog training class. Twenty percent of

the dogs are hounds. The other eighty percent of the dogs are of the toy variety.
Ninety percent of the hounds are stubborn and difficult to train. Thirty percent
of the dogs in general are stubborn. What is the probability that a stubborn
dog is a hound?
70 Chapter 5. Bayes’ Theorem

Exercise 5.4. mp3 players. Eighty percent of all mp3 players are iPods. Five
percent of iPods are defective. Seven percent of all mp3 players are defective.
A randomly chosen mp3 player is taken to a repair shop because it is defective.
What is the probability it is an iPod?
Exercise 5.5. Chalkboards and dry-erase boards. In the Department
of Mathematical and Computational Science, 83% of lecture halls have chalk-
boards, and the other 17% have dry-erase boards. Of the classes taught in
rooms with chalkboards, 75% are mathematics courses, 15% are computer sci-
ence courses, and 10% are statistics courses. Of the classes taught in rooms
with dry-erase boards, 65% are computer science courses, 8% are mathemat-
ics courses, and 27% are statistics courses. What percentage of courses at the
college are mathematics or statistics?
Exercise 5.6. Math and art. In the theory of general intelligence, it is stated
that being good in one intelligence, like math, increases the chance of one being
good in another intelligence. Suppose 10% of the people are good at art, and
that 40% of the people who are good at art are also good at math. If a person
is not good at art, they have only a 30% chance of being good at math. What
is the probability that a person who is good at math will also be good at art?
Exercise 5.7. Students with music players. A student is chosen at random.
You want to know the probability that the student has an iPod. This could be
hard to determine since there are over 40,000 students on the given campus.
Fortunately, a current survey recorded that 47% of first-year students have
iPods, and you know that 32% of the students are first-year students. The
survey indicates that 56.2% of upperclass students (i.e., non first-years) have
iPods.
a. What is the probability that the student you randomly select has an
iPod?
b. Given that the selected student has an iPod, what is the probability that
he/she is an upperclass student?
c. Given that the selected student does not have an iPod, what is the
probability that he/she is an upperclass student?
Exercise 5.8. Smoke detectors. It is estimated that 82% of homes have
working smoke detectors. On average, 22% of fires result in fatalities, but the
presence of a working smoke detector cuts the risk to just 7%.
a. If a random fire resulted in a fatality, what is the probability that the
house had a working smoke detector?
b. In homes without a working smoke detector, what is the risk of fatalities?

5.3.2 Extensions
Exercise 5.9. Coins and dice. Consider the following game: The player flips
a fair coin. If it shows a head, he gets to roll a 4-sided die. If it shows a tail,
 5.3. Exercises 71

he gets to roll a 6-sided die. In either case, let A denote the event that he gets
1 on the die roll. Let H denote the event that his coin flip shows a head, i.e.,
that he uses the 4-sided die. Let T denote the event that his coin flip shows a
tail, i.e., that he uses the 6-sided die.
a. Find P (H | A).
b. Find P (T | A).
Hint: To verify your answers, note P (H | A) + P (T | A) = 1.
Exercise 5.10. French classes. In a certain school, 4 levels of French are
taught with 40% of the students being enrolled in level 1, 30% enrolled in
level 2, 20% enrolled in level 3, and 10% enrolled in level 4. The percentage of
people who enjoy their French class is 70% in level 1, 80% in level 2, 85% in
level 3, and 90% in level 4.
Given that a person enjoys his/her French class, let pj be the probability
that the student is enrolled in level j. Find pj for j = 1, 2, 3, 4. (Check: Your
four answers should sum to 1 altogether.)
Exercise 5.11. Sex and switching majors. At a certain university, 60% of
undergraduate students are male, and 40% are female. Ten percent of females
change their majors at least once. Overall, 30% of students change their majors
at least once.
a. Find the probability that, given a randomly selected student who changes
majors, the student is a female.
b. What percentage of males change their majors at least once?
Exercise 5.12. Engineering majors. At First Street Towers, an equal num-
ber of students live on each floor. On floors 1, 2, 3, 4, 5, the percentage of
students who study engineering are, respectively, 80%, 52%, 74%, 67%, and
29%. Upon meeting an Engineering student who lives in First Street Towers,
what is the probability that the student lives on the 4th floor?
Exercise 5.13. Babies. Allison delivers one baby, and a year later she delivers
a second baby. Let C be the event that at least one of the babies is a girl (either
the first, or the second, or both). Let D be the event that both of the babies
are girls. Find P (D | C).

5.3.3 Advanced
Exercise 5.14. Pair of dice. Roll a blue die and a red die. Given that the
blue die has an odd value, what is the probability that the sum of the two dice
is exactly 4?
Exercise 5.15. Pair of dice. Roll a blue die and a red die. Given that the
blue die has a value of 4 or smaller, what is the probability that the sum of the
two dice is 7 or larger?
72 Chapter 5. Bayes’ Theorem

Exercise 5.16. Fuses. Two fuses in series are built to shut down if an overload
occurs. If the first fuse shuts down properly 90% of the time, there is no need
for the second fuse to do anything. If the first fuse fails to shut down properly,
the second fuse shuts down properly 95% of the time. What is the probability
the whole system operates correctly during an overload (with one fuse or the
other shutting down properly)?

Exercise 5.17. Weather. The weather on any given day can either be sunny,
cloudy, or partially cloudy. Each day is also classified as dry or rainy. The
probability of a sunny day is 0.48, and the probability of a cloudy day is 0.39.
The probability of having a sunny and dry day is 0.48. The probability of a
cloudy and dry day is 0.14. The probability of a partially cloudy and dry day
is 0.09.

a. Find the probability of a dry day.

b. Find the probability of a rainy day.
c. Find the probability of a sunny day given that it is a dry day.
d. Find the probability of a dry day given that it is a sunny day.
e. Find the probability of a cloudy day given that it is a rainy day.
f. Find the probability that it is a rainy day given that it is a cloudy day.

Exercise 5.18. Coin flips and then dice. Claire flips a coin until she gets
a head for the first time. Say it takes her n times. Then (afterwards) she rolls
exactly n dice. What is the probability that none of the dice show the value 1?
(For instance, if it takes her 7 flips to get a head for the first time, then she
rolls 7 dice. Hint: It is enough to use Remark 5.8. You do not need Bayes’
Theorem itself for this problem.)

Exercise 5.19. Randomly chosen beverage. Lakisha has two identical-

looking beverage containers. She randomly picks one (each is equally likely to
be chosen). Then she reaches inside the chosen container and picks a drink for
herself and a drink for her boyfriend. Suppose that one cooler has 3 sodas and
2 fruit juices; the other cooler has 2 sodas and 2 fruit juices. If her drink is a
fruit juice, find the probability that her boyfriend’s drink is also a fruit juice.
Chapter 6

Review of Randomness

6.1 Summary of Randomness

How do we classify random things?

• When something random occurs, exactly one outcome happens.

• An event is a collection of outcomes.

• Empty set ∅ contains no outcomes. Sample space S contains all outcomes.

• An event A and its complement Ac are disjoint (have no overlap) and

contain all of the events from the sample space S, i.e., A ∪ Ac = S.

c
c
• DeMorgan’s laws = j Acj and = j Acj allow us to
S T T S
j Aj j Aj
work with the complement of a union or an intersection.

What are the basic probability axioms?

• Any event occurs a certain percentage of the time, so probabilities are
always between 0 and 1.

• With probability 1, some outcome in the sample space occurs.

• If events have no outcomes in common, then the probability of their union

is the sum of the probabilities of the individual events.

Can we add the probabilities of events?

• If events are disjoint, we can sum their probabilities.

• If A1 , A2 , . . . is a collection of disjoint events, then

∞
[  X∞
P Aj = P (Aj ).
j=1 j=1

73
74 Chapter 6. Review of Randomness

Can we add the probabilities of events if the events are overlapping?

If A, B, C are overlapping events, then P (A ∪ B ∪ C) = P (A) + P (B) +
P (C) − P (A ∩ B) − P (A ∩ C) − P (B ∩ C) + P (A ∩ B ∩ C). More general versions
of the inclusion-exclusion rule are available as well.
How do we know if two events are independent?
Events A and B are independent if P (A ∩ B) = P (A)P (B).
Can two events be independent and disjoint?
No, if A and B both have positive probabilities, they cannot be both inde-
pendent and disjoint.
Can subsets be independent?
If A ⊂ B and neither P (A) = 0 nor P (B) = 1, then A, B are dependent.
Can complements be independent?
If neither P (A) = 0 nor P (A) = 1, then A, Ac are dependent.
How do we calculate conditional probabilities?
If event B has nonzero probability (i.e., P (B) > 0), then the conditional
probability P (A | B) of A given B is defined as

P (A ∩ B)
P (A | B) = .
P (B)

Equivalently,
P (A ∩ B) = P (B)P (A | B).

How can we decompose an event in parts?

If A is an event, then we can decompose B into two parts: A ∩ B and
B \ A = Ac ∩ B. Since B = (A ∩ B) ∪ (Ac ∩ B) is a disjoint union, then if P (A)
is not 0 or 1,

P (B) = P (A ∩ B) + P (Ac ∩ B) = P (A)P (B | A) + P (Ac )P (B | Ac ).

Or we can decompose B into finitely many parts: Let A1 , A2 , . . . , An be any

finite partition of the sample space. Then we can partition B into n parts as
A1 ∩ B, A2 ∩ B, . . . , An ∩ B, i.e., B = (A1 ∩ B) ∪ (A2 ∩ B) ∪ · · · ∪ (An ∩ B) is
a union of disjoint events, so

P (B) = P (A1 ∩B)+· · ·+P (An ∩B) = P (A1 )P (B | A1 )+· · ·+P (An )P (B | An ).

When do we want to use Bayes’ Theorem? How do we use it?

If you want P (A | B) but you instead know (or are told) P (B | A), Bayes’
Theorem lets you turn this around and compute P (A | B).
If P (B) 6= 0 and 0 < P (A) < 1, then,

P (A)P (B | A)
P (A | B) = .
P (A)P (B | A) + P (Ac )P (B | Ac )
 6.2. Exercises 75

If P (B) > 0, and if A1 , A2 , . . . , An form a partition of S, with all P (Aj ) > 0,

then
P (Ak )P (B | Ak )
P (Ak | B) = .
P (A1 )P (B | A1 ) + P (A2 )P (B | A2 ) + · · · + P (An )P (B | An )

6.2 Exercises
Exercise 6.1. Probabilities in a chain of subsets. Four events A, B, C, D
could be compared in the following way:

A ⊂ B ⊂ C ⊂ D.

The probability of A is 0.03, and the probability of C is 0.27. With the given
information, what are potential probabilities of B and D?
Exercise 6.2. Waiting for the bus. A customer waits for a bus to appear.
a. List 5 possible outcomes.
b. What is the sample space?
c. Write a partition for the customer’s waiting time, using five-minute in-
tervals for the partition. (Assume that the customer can wait as long as needed
for the bus.)
d. Explain how the answer to part c meets the definition of a partition; see
Definition 2.14.
e. Modify your answer to part c, with the assumption that the customer
has a maximum time that they can afford to wait before giving up.
Exercise 6.3. Randomly choose a page. Randomly open a 300-page book,
and mark a page. What is the probability that the number of the page contains
the lucky number 5?
Exercise 6.4. Cell phone. A student loses her first cell phone. Her new
phone number is randomly chosen by the store, but the area code is fixed, so
there are exactly 7 randomly selected digits. (Assume all 107 possibilities are
equally likely.) Give the probabilities of the following events:
a. Her phone number ends in a 2.
b. Her phone number ends in an odd number.
c. Her phone number ends in a 5 or a 7.
Exercise 6.5. Egg-citing! There are 45 egg boxes in a store. Twenty are
Brand A, fifteen are brand B, and ten are Brand C. Brand A and C each have
half green boxes and half yellow boxes. Brand B is all yellow.
a. If you have Brand B or C, what is the chance that you have a green box?
b. If you have a yellow box, what is the chance the brand is B?
76 Chapter 6. Review of Randomness

Exercise 6.6. Skittles. Chris has 32 Skittles candies. Nine are red, three
are blue, seven are yellow, five are orange, and eight are purple. Exactly four
are sour, and all of these sour Skittles are purple. Chris picks one Skittle at
random.

a. What is the probability he selected a red or blue one?

b. What is the probability he selected a sour one?
c. Given that he selected one that is not sour, what is the probability it was
purple?

Exercise 6.7. Homework. Joe works on homework 25% of the time. He is

on the computer 40% of the time. He likes to do homework on the computer, so
30% of the time that he is on the computer is spent doing homework. What is
the probability that he is on his computer, given that he is doing his homework?

Exercise 6.8. Depression and anxiety. According to the National Insti-

tute of Mental Health (from www.nimh.nih.gov/statistics/1ANYANX_ADULT.
shtml), in any year, 18.1% of all U.S. adults suffer from anxiety disorders.
Among people who have anxiety disorders, 30.2% are between 18 and 29 years
old. According to the 2010 U.S. Census, approximately 10% of the U.S. adults
are between 18 and 29 years old.
a. Calculate the chance that a randomly selected 18–29 year old American
has an anxiety disorder.
b. Calculate the chance that a randomly selected American is 18–29 years
old AND has an anxiety disorder.
Part II

Discrete Random Variables

In this part of the book, we study discrete random variables and their distribu-
tions. We will discuss the differences between discrete and continuous random
variables and introduce some useful formulas for calculating other information
about discrete random variables. (We will return to continuous random vari-
ables starting in Chapter 24.) The rules and formulas in the next few chapters
will work for all discrete random variables. This will build a base of under-
standing about common themes of discrete random variables that will be useful
in the subsequent part of the book, on named discrete random variables.
By the end of this part of the book, you should be able to:

1. Distinguish between discrete and continuous random variables when read-

ing a story.

2. Calculate probabilities for discrete random variables.

3. Calculate and graph a probability mass function (PMF).

4. Calculate and graph a cumulative distribution function (CDF).

5. Calculate the mean, variance, and standard deviation of discrete random

variables.

6. Calculate the mean and variance of sums of discrete random variables.

7. Calculate the mean and variance of functions of discrete random variables.

will need: Binomial coefficients (“choose” parentheses), fac-

Math skills you P
torials, summation notation, step functions, and the floor function bxc.
Additional resources: We will be calculating probabilities by hand, but
there are mathematical and statistical software programs (such as Excel, Maple,
Mathematica, Matlab, Minitab, R, SPSS, etc.), which could help you to perform
these calculations.

77
Chapter 7

Discrete Versus Continuous

Random Variables

While writing my book I had an argument with Feller. He asserted that everyone
said “random variable” and I asserted that everyone said “chance variable.” We
obviously had to use the same name in our books, so we decided the issue by a
stochastic procedure. That is, we tossed for it and he won.
—Joe Doob, from “A Conversation with Joe Doob,” by J. Laurie Snell, from
Statistical Science, volume 12, number 4, November 1997

We do not measure everything with the same types of variables. Can what we
are measuring be counted? Or does it fall within a certain range without specific
levels? For instance, consider the difference in how we measure the outcome of
a die roll versus the distance a dart lands from the bullseye. If our types of
variables are different, we will need different ways of calculating probabilities.

7.1 Introduction
We have already seen that, in a random phenomenon, there are many possible
outcomes; exactly one of these outcomes occurs. The set of all possible outcomes
is the sample space. Now we introduce random variables, which are the main
topic of study throughout the rest of this book:

Definition 7.1. A random variable assigns a real number to each outcome

in the sample space. The random variable’s value is completely determined by
the outcome. A random variable is a function from the sample space to the
set of real numbers.

78
 7.1. Introduction 79

Often an outcome contains much more information than a random variable

with which it is associated. A random variable is very simple; it is a one-number
way to summarize an outcome.

Example 7.2.a If we roll three dice, there are 63 = 216 possible outcomes. One
possible random variable, which we call X, is the sum of the three dice. 1 die has 6 possible
outcomes; 2 dice
If the outcome is ω = (6, 4, 3), then sum of the dice is X(ω) = 13. If the have 6 × 6 = 36
outcome is ω = (5, 2, 2), the sum is X(ω) = 9. In general, if the outcome is possible outcomes; 3
ω = (a, b, c), then the sum is X(ω) = a + b + c. In this example, X is always dice have
6 × 6 × 6 = 216
an integer from 3 and 18, depending on the outcome. The outcome completely
possible outcomes.
and uniquely determines the value of the random variable.

Example 7.2.b As before, roll three dice. Let the random variable Y be the
maximum of the three dice.
If the outcome is ω = (a, b, c), then Y (ω) = max(a, b, c). For instance, if the
outcome is ω = (4, 5, 2), then Y (ω) = 5.

Example 7.2.c Let the random variable Z denote the value of the first die
rolled.
If the outcome is ω = (a, b, c), then Z(ω) = a. E.g., if the outcome is
ω = (6, 1, 2), then Z(ω) = 6.

When working with random variables, we must resist the urge to “solve” for the
value of the random variable. Unlike in algebra, we will not spend time studying
equations such as x2 −x−6 = 0 and trying to solve for x. As a constant reminder
that we are not solving such equations, we always use a capital letters (such
as X) to denote a random variable.
The outcome of a random phenomenon is, well, random! Different outcomes
happen, so random variables can have different values. Every random variable
depends on the underlying outcome.
Also, random variables must assign a real number to each outcome. E.g., if
a randomly chosen person ω (the outcome) in a classroom is selected, we could
define a random variable X(ω) as the person’s age. So X(ω), the person’s age,
is a random variable that completely depends on ω, the person selected.
 80 Chapter 7. Discrete Versus Continuous Random Variables

Remark 7.3. Suppressing Outcomes in Random Variable Notation

We have been writing ω as the outcome and X(ω) as the random variable,
but often the outcome is well-understood and the random variable is the thing
we want to study. Thus, we just write X (instead of X(ω)), to simplify our
notation. We understand, however, that the there is actually an underlying
outcome ω, and that X actually means X(ω).

Since a random variable must assign a real number to each outcome, then
a random variable cannot be a person’s name, the color of a car, a country, a
suit in a deck of cards, heads or tails, etc. These types of things might be the
underlying outcomes, but they cannot be the random variables themselves.
Random variables can be either discrete or continuous. Parts II and III of
the book are all about discrete random variables. Parts V and VI are dedicated
to continuous random variables.
The list of all values taken on by a discrete random variable is either finite:

e.g., {1, 2, 3, 4, 5, 6} or {−1, 0, 1} or {2, 3, 4, . . . , 12} or {−5.62, 2.33, 7.7}, etc.

or countably infinite:

e.g., {1, 1/2, 1/4, 1/8, . . .} or {1, 2, 3, 4, . . .} or {. . . , −2, −1, 0, 1, 2, . . .}, etc.

Discrete random variables can assume any kind of real numbers; e.g., negative
numbers are allowed; decimals, fractions, transcendental numbers, etc., are all
allowed too. All that matters for a random variable to be discrete is that the
set of possible values is either finite or can be put into a countably infinite list.
Continuous random variables, on the other hand, to be covered in Part V,
take values on continuous intervals (or on the union of continuous intervals):

e.g., [0, ∞) or (−∞, ∞) or [0, 1] or [−3.7, 10.2), etc.

7.2 Examples

Some examples of random variables that are discrete are:

• let X be the number of male puppies in a litter of Golden Retrievers;

• let X be “1” if a newborn baby is a girl, or “0” if it is a boy;

• let X be “1” if the next car to pass is blue, or “2” if red, or “3” if silver, or
Remember, random “−1” otherwise;
variables can be
negative too! • let X be the number of the region, 1 through 20, on which a dart lands.
Examples of random variables that are continuous are:
• let X be the length of your left foot (in inches);
 7.2. Examples 81

• let X be the time, in seconds, for a traffic light to turn green.

Here are some more examples of random variables.

Example 7.4. Flip a coin three times, and let X denote the total number of
heads that appear.
The set of values X can assume is {0, 1, 2, 3}, so X is a discrete random variable.
We can make a chart of all of the possible outcomes and the associated values
of X:
outcome prob. of outcome value of X prob. of each X value
(H, H, H) 1/8 3  1/8 = P (X = 3)
(H, H, T ) 1/8 2 
(H, T, H) 1/8 2 3/8 = P (X = 2)
(T, H, H) 1/8 2

1

(T, T, H) 1/8 
(T, H, T ) 1/8 1 3/8 = P (X = 1)
(H, T, T ) 1/8 1

(T, T, T ) 1/8 0 1/8 = P (X = 0)

Each of the outcomes (H, H, T ) and (H, T, H) and (T, H, H) will cause X to
be 2. Each of these outcomes has probability 1/8, so the event containing all
three of these outcomes has probability 3/8, i.e.,

P ({(H, H, T ), (H, T, H), (T, H, H)}) = 3/8.

We abbreviate this by writing

P (X = 2) = 3/8.

So the probability of each possible value of X can be written succinctly as

x 0 1 2 3
P (X = x) 1/8 3/8 3/8 1/8

Example 7.5. a Let ω = (x, y) be the Cartesian coordinates where a dart lands
on a dart board. We could let X(ω) = x be a random variable that denotes the
first coordinate and let Y (ω) = y be a random variable that denotes the second
coordinate.
Both X and Y are continuous random variables that each assume values (for
instance) on the interval (−9, 9) if the dartboard has radius 9 inches. If ω =
(3.6, −1.35) is the location where the dart lands, then X(ω) = 3.6 and Y (ω) =
−1.35. Again, we often write (more simply) just X = 3.6 and Y = −1.35, in
such a case.
82 Chapter 7. Discrete Versus Continuous Random Variables

Example 7.5.b We could also consider other random variables, p for instance,
if ω = (x, y) is the location of the dart’s landing, then Z(ω) = x2 + y 2 is the
distance of the dart from the center of the board. More simply, we can drop the
notation for ω and just write Z as the distance to the center of the dartboard.
Notice Z is a continuous random variable.

Example 7.6. A traffic engineer observes the next three cars that pass. He
uses X as the time until the arrival of the third car, Y as the time between the
arrivals of the second and third cars, and Z as the speed of the third car. Since
X, Y, Z each take values on the interval (0, ∞), then X, Y, Z are each continuous
random variables.

Example 7.7. A student flips a coin until the 10th head appears. Each outcome
is a string of heads and tails. For instance, an outcome ω might be

ω = (H, H, H, T, H, T, T, T, H, T, T, T, H, H, T, H, T, H, T, H),

He writes:

X1 as the number of flips until the 1st head,

X2 as the number of flips after the 1st head until the 2nd head,
.. ..
. .
Xj as the number of flips after the (j − 1)st head until the jth head,
.. ..
. .

In this case, outcome ω causes these random variables to have the following
values:
X1 = 1, X2 = 1, X3 = 1, X4 = 2, X5 = 4,

X6 = 4, X7 = 1, X8 = 2, X9 = 2, X10 = 2.
Each of the Xj ’s takes on a positive integer value and is therefore a discrete
random variable. Finally, X1 + · · · + X10 is the total number of flips needed
until the 10th head appears.
 7.2. Examples 83

Example 7.8. A student is selected at random and her mp3 player is examined.
Let X denote the number of songs on her music player and let Y denote the
number of songs on the first playlist on the music player, or Y = 0 if there are
no playlists on the mp3 player. (Let X = 0 and Y = 0 if she doesn’t have an
mp3 player at all.) Notice that X and Y are random variables. Also, we know
Y ≤ X, since the number of songs on the playlist is limited by the number of
songs on the mp3 player altogether.
We cannot assign an artist’s name, or a genre of song, as a random variable,
because these are not real numbers. Nonetheless, we could (say) assign a nu-
meric scheme, such as Z = 1 if the majority of the songs are blues songs, Z = 2
if the majority of the songs are rock songs, Z = 3 if the majority of the songs
are jazz songs, or Z = 4 otherwise (including if there is a “tie” for the majority).
Both X and Y assume values that are nonnegative integers and thus are
discrete random variables. Since Z takes values in the range {1, 2, 3, 4}, then Z
is a discrete random variable too.

At this point, it should be clear that there are ample, real-world possibilities
for assigning random variables according to all kinds of random phenomenon.

Example 7.9. A student is selected at random. Let W be her height (in

inches); let X denote her body temperature; let Y be her age; and let Z denote
the zip code of her mailing address.
Her height is a continuous random variable that is in the range [22.5, 107.1]
inches. Her body temperature is a continuous random variable number that
is close to 98.6◦ F a large percentage of the time and is almost always in the
range (say) 92–104◦ F , even in the most extreme situations. Her age is a discrete
random variable in the set {0, 1, 2, 3, . . . , 122}, if rounded (down) to the nearest
integer, or (on the other hand) her age is a continuous random variable in the Guinness World
range [0, 122.3] if the age is not rounded to the nearest integer. Her zip code is Records (2005)
a discrete random variable with values in the set {00000, 00001, . . . , 99999}. claim adult humans
range in height from
22.5 inches to 107.1
inches.
Sometimes it is helpful to make a list of all of the possible outcomes for a random
phenomenon, and then to list the associated random variable that goes along
with each outcome.

Example 7.10. Roll a pair of dice, and let X denote the sum of the two values
that appear. So if the outcome is (i, j), then X = i + j.
84 Chapter 7. Discrete Versus Continuous Random Variables

The sum X is a discrete random variable that has values in the set {2, 3, . . . , 12}.
There are several outcomes that cause X to be equal to 5. These outcomes are
those in the event
{(1, 4), (2, 3), (3, 2), (4, 1)}.
Each event has a probability associated with it. In this case,

P ({(1, 4), (2, 3), (3, 2), (4, 1)}) = 4/36,

but we usually use a more succinct notation and write

P (X = 5) = 4/36

as a shorthand for the equation above.

Next we compute all of the probabilities associated with the possible values
of X:

1 2 3 4 5 6

1 2 3 4 5 6 7

2 3 4 5 6 7 8

3 4 5 6 7 8 9

4 5 6 7 8 9 10

5 6 7 8 9 10 11

6 7 8 9 10 11 12

x 2 3 4 5 6 7 8 9 10 11 12
1 2 3 4 5 6 5 4 3 2 1
P (X = x) 36 36 36 36 36 36 36 36 36 36 36

Example 7.11. Select a random song on a student’s music player. Let X be

equal to 1 if the song is a blues song, or otherwise let X be 0; let Y be equal
to 1 if the song is jazz, or otherwise Y is 0; let Z = 1 if the song is rock, or
otherwise Z = 0.

Random variables that are 1 when an event occurs or 0 when the event does
not occur are called indicator random variables, sometimes abbreviated
as indicators; they are also called Bernoulli random variables. They will be
studied in Chapter 14. Indicator random variables are used extensively when
working with sums of random variables, to be covered in Chapter 11.
 7.3. Exercises 85

7.3 Exercises
7.3.1 Practice
In Exercises 7.1 to 7.15, identify whether the random variables are discrete or
continuous (an exercise might contain one or more of both types). Also identify
the possible values that each random variable could assume. In some cases,
answers will differ according to interpretation of the problems by the students.

Exercise 7.1. Pair of dice. Roll two dice. Let X be the absolute value of
the difference of the two values that appear. (If D1 , D2 are the two die values,
then X = |D1 − D2 |.)

Exercise 7.2. Onions. Pick a basket of onions. Let X be the weight of the
onions (in pounds). Let Y be the number of onions in the basket.

Exercise 7.3. Grades. Let X be the cumulative grade point average of a

randomly chosen student. Let Y be the score on the student’s most recent
exam. Let Z be the number of exams that the student takes during the current
semester.

Exercise 7.4. Harmonicas. When ordering a new box of harmonicas, let X

denote the time (in hours) until the box arrives, and let Y denote the number
of harmonicas that work properly.

Exercise 7.5. Music genres. Select a random student’s music player. Let X
denote the number of blues songs on the music player; let Y denote the number
of jazz songs; let Z denote the number of rock songs.

Exercise 7.6. Average distance. Let X be the average distance (in miles)
that a student drives in a week.

Exercise 7.7. Sexes of babies. Consider the births of ten consecutive babies.
Let X be the number of babies that are girls. Let Y be the number of the first
baby that is a boy (if none of the babies are boys, then just let Y = 0, or suggest
your own, alternative value for such a case).

Exercise 7.8. Calling boyfriend/girlfriend. Consider the time X (in min-

utes) from now until your boyfriend/girlfriend calls on the telephone. Let Y be
the length of the call. Let Z be the number of times that he/she calls in one
evening.

Exercise 7.9. Lightbulbs. Consider a box of two lightbulbs. Let X be the

lifetime of the first bulb removed from the box, and let Y denote the lifetime of
the second bulb removed from the box.

Exercise 7.10. More lightbulbs. Consider a collection of 100 lightbulbs. Let

Xj denote the lifetime of the jth bulb. Let Y denote the sum of the lifetimes
of all 100 bulbs.
86 Chapter 7. Discrete Versus Continuous Random Variables

Exercise 7.11. Social networking. Place “friend” requests on a social net-

working site until the first person accepts your request. Let X be the number
of people needed until this first acceptance. Let Y be the time until the first
request is accepted.

Exercise 7.12. Puzzle pieces. A pile of puzzle pieces falls onto the floor.
Let X denote the number of pieces that are edge pieces, and let Y denote the
number of pieces that are interior (i.e., not edge) pieces.

Exercise 7.13. Milk. A half-gallon container of milk is tested by a quality

control company. Let X be a rating of the milk quality, given as an integer
from 1 to 10; let Y denote the exact volume of milk in the selected half-gallon
container; let Z be “1” if the half-gallon of milk is 1% milkfat, or “2” if it is 2%
milkfat, or “3” if it is fat-free, or “4” otherwise.

Exercise 7.14. Snake eyes. Roll a pair of dice until “snake eyes” (i.e., a pair
of 1’s) appear. Let X denote the total number of rolls required. Let Y denote
the sum of all of the dice rolled during this process.

Exercise 7.15. Student age and name. Select a random student from your
course; let X denote the age, in days, of the selected student; let Y denote the
length of the student’s name.

7.3.2 Extensions
Exercise 7.16. Toy ballet dancers. When opening a container of 32 toy bal-
let dancers performing pirouettes, let X denote the number of broken dancers,
and let Y denote the number of whole (unbroken) dancers.
Write Z = X/Y , i.e., Z is the ratio of broken dancers to whole dancers.
What complication arises in this definition of Z as a random variable?

Exercise 7.17. Pick two cards. Pick two cards at random from a well-shuffled
deck of 52 cards (pick them simultaneously, i.e., grab two cards at once—so they
are not the same card!). There are 12 cards which are considered face cards (4
Jacks, 4 Queens, 4 Kings). Let X be the number of face cards that you get.

a. Find P (X = 0).
b. Find P (X = 1).
c. Find P (X = 2).

7.3.3 Advanced
Exercise 7.18. Mixed random variables. Do you think that there are
random variables which are neither discrete nor continuous? If yes, try to
construct a simple example. If no, then discuss why not.
Chapter 8

Probability Mass Functions and

CDFs

The 50-50-90 rule: Anytime you have a 50-50 chance of getting something right,
there’s a 90% probability you’ll get it wrong.
—Andy Rooney

If we toss a single coin, we have a 1/2 chance of getting a head. How do the
probabilities change if we toss a coin 3 times and total up the number of heads?
Is the chance of getting 0, 1, 2, or 3 heads exactly the same? Why or why
not? Is the probability of getting exactly 1 head the same as the probability of
getting head, tail, tail?

8.1 Introduction
In many situations, it is very helpful to consider the probability that a random
variable assumes a specific value or is found in an entire range of numbers. For
instance, it might be helpful to know whether a pregnant mother will have 2 or
more babies during her delivery; if X denotes the number of babies to be born,
the probability of this event is written as P (X ≥ 2). If X is the number of
guitar strings in a package to be shipped, and it is supposed to have 6 strings,
the manufacturing company will be very interested in P (X = 6), i.e., if the
package contains the right number of strings. When baking cookies, X could
be the number of cookies that your roommates would like to eat. If there are 8
cookies in a container, then P (X ≤ 8) is the probability that you have enough
cookies to feed them.
We use the concepts P (X = x) and P (X ≤ x) so often that we give them
each a special name:

87
88 Chapter 8. Probability Mass Functions and CDFs

Definition 8.1. Probability Mass Function (a.k.a. PMF or mass)

If X is a random variable, the probability that X is exactly equal to x is

pX (x) = P (X = x).

This is called the probability mass function (PMF), or just the mass,
of X.

Definition 8.2. Cumulative Distribution Function (a.k.a. CDF)

If X is a random variable, the probability that X does not exceed x is written
as
FX (x) = P (X ≤ x).
This is called the cumulative distribution function (CDF) of X.

8.2 Examples

Example 8.3. Roll a die, and let X denote the value that appears.

pX (x) pX (x)

1 1
5/6 5/6
4/6 4/6
3/6 3/6
2/6 2/6
1/6 1/6
x x
1 2 3 4 5 6 1 2 3 4 5 6

Figure 8.1: Left: Mass pX (x) = P (X = x) of the value on a die roll. Right:
Same plot but with the values of 0 not shown in the plot.

In the mass of a die roll X, the left hand side of Figure 8.1 illustrates the
fact that
pX (x) = P (X = x) = 1/6
for all x in the set {1, 2, 3, 4, 5, 6}, and pX (x) = P (X = x) = 0 otherwise. Since
the “otherwise” encompasses most values of x, we usually suppress the values
 8.2. Examples 89

for which the mass is 0. Thus, the right hand side of Figure 8.1 is the way that
we will usually show such a mass (with the 0 values omitted from the plot).
To compute the cumulative distribution function of X, we compute
FX (1) = P (X ≤ 1) = P (X = 1) = 1/6;
FX (2) = P (X ≤ 2) = P (X = 1) + P (X = 2) = 2/6;
FX (3) = P (X ≤ 3) = P (X = 1) + P (X = 2) + P (X = 3) = 3/6;
FX (4) = P (X ≤ 4) = P (X = 1) + · · · + P (X = 4) = 4/6;
FX (5) = P (X ≤ 5) = P (X = 1) + · · · + P (X = 5) = 5/6;
FX (6) = P (X ≤ 6) = P (X = 1) + · · · + P (X = 6) = 1.
So if we begin to draw the cumulative distribution function in this case, we
know 6 of the values of FX (x) = P (X ≤ x); see the left side of Figure 8.2.

FX (x) FX (x) FX (x)

1 1 1
5/6 5/6 5/6
4/6 4/6 4/6
3/6 3/6 3/6
2/6 2/6 2/6
1/6 1/6 1/6
x x x
1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6

Figure 8.2: Left: Starting to construct the CDF FX (x) = P (X ≤ x) for the
value of a die roll, for x in {1, 2, 3, 4, 5, 6}. Middle: The CDF FX (x) including
x < 0 and x > 6. Right: The CDF FX (x) for all values of x.

Of course, we know that the value on a die is 6 or smaller. So P (X ≤ 100) =

1 and P (X ≤ 12) = 1 and P (X ≤ 7) = 1 and P (X ≤ 6.2) = 1, etc., etc. Thus,
P (X ≤ x) = 1 for any x ≥ 6.
Similarly, we know that the value on a die is 1 or larger. So P (X ≤ −10) = 0
and P (X ≤ −3) = 0 and P (X ≤ −1) = 0 and P (X ≤ 0) = 0 and P (X ≤ 0.7) =
0, etc., etc. In fact, P (X ≤ x) = 0 for any x < 1.
All that remains is to fill in the picture in between the integers from 1 to 6.
Here is an example. We determine P (X ≤ 3.1). If X ≤ 3.1, then X = 1 or
X = 2 or X = 3, so
P (X ≤ 3.1) = P (X = 1) + P (X = 2) + P (X = 3) = 3/6.
This is the same value as P (X ≤ 3). The reason is that X cannot be between
3 and 3.1, so the CDF does not increase in this interval. The CDF FX (x) =
P (X ≤ x) is constant for 3 ≤ x < 3.1. This reasoning similarly shows that
FX (x) = P (X ≤ x) = 3 for any x with 3 ≤ x < 4. We use similar reasoning to
finish the CDF. See the right side of Figure 8.2.
90 Chapter 8. Probability Mass Functions and CDFs

Example 8.4. Flip a coin three times; let X denote the total number of heads.

The mass of X is

pX (0) = P (X = 0) = 1/8,
pX (1) = P (X = 1) = 3/8,
pX (2) = P (X = 2) = 3/8,
pX (3) = P (X = 3) = 1/8.

Thus, the CDF of X is

FX (0) = P (X ≤ 0) = 1/8,
FX (1) = P (X ≤ 1) = 4/8,
FX (2) = P (X ≤ 2) = 7/8,
FX (3) = P (X ≤ 3) = 1.

With similar reasoning to the last example, the value of the CDF FX (x) does
not increase in between the integers x. The plots of the mass and CDF of the
total number of heads X are given in Figure 8.3. Notice that the size of the
“jumps” in the CDF function are equal to the probabilities that X is found at
a specific value:

1. The jump in FX (x) at x = 0 of size 1/8 corresponds to P (X = 0) = 1/8;

2. The jump in FX (x) at x = 1 of size 3/8 corresponds to P (X = 1) = 3/8;

3. The jump in FX (x) at x = 2 of size 3/8 corresponds to P (X = 2) = 3/8;

4. The jump in FX (x) at x = 3 of size 1/8 corresponds to P (X = 3) = 1/8.

pX (x) FX (x)
1 1
7/8 7/8
6/8 6/8
5/8 5/8
4/8 4/8
3/8 3/8
2/8 2/8
1/8 1/8
x x
1 2 3 1 2 3

Figure 8.3: Left: The mass pX (x) of X, the number of heads in three tosses
of a fair coin. Right: The CDF FX (x) of X.
 8.3. Properties of the Mass and CDF 91

The reasoning from Example 8.4 works in general. As we move from left to right
across the mass, we start with probability 0 on the extreme left-hand side of the
CDF. After sweeping all the way across the mass, we eventually accumulate all
of the probability from the mass, so that we get probability 1 on the extreme
right-hand side of the CDF.

8.3 Properties of the Mass and CDF

Remark 8.5. The mass pX (x) = P (X = x) and the CDF FX (x) = P (X ≤ x)

are both probabilities, so they both take on values between 0 and 1, i.e.,

0 ≤ pX (x) ≤ 1, and 0 ≤ FX (x) ≤ 1, for all x.

Remark 8.6. Since discrete random variables assume only a finite or count-
able number of values, we can sum over all the nonzero masses, and we must
get sum 1: X
pX (x) = 1.
x : pX (x)6=0

We usually drop the notation about restricting to x’s for which pX (x) = 0:
X
pX (x) = 1.
x

Alternatively, if X takes on only the finite values x1 , x2 , . . . , xn , then we

have
Xn n
X
P (X = xj ) = pX (xj ) = 1,
j=1 j=1

or if X takes on infinitely many values, say x1 , x2 , . . ., then we have

∞
X ∞
X
P (X = xj ) = pX (xj ) = 1.
j=1 j=1

For example,
P in Example 8.4, the random variable X can only take on values
0, 1, 2, 3, so 3j=0 pX (j) = 1.

Remark 8.7. The CDF is a non-decreasing function. From a visual perspec-

tive, this means that as we look left to right across the plot, the CDF is always
increasing or flat. So if a ≤ b, then FX (a) ≤ FX (b).
92 Chapter 8. Probability Mass Functions and CDFs

To see this, consider any two real numbers a ≤ b. If an outcome ω causes X

to be less than or equal to a, then X is automatically smaller than b too (since
a ≤ b). So the event X ≤ a is contained in the event X ≤ b, i.e.,

{ω | X(ω) ≤ a} ⊂ {ω | X(ω) ≤ b}.

(Subsets were discussed in Section 1.1.) Our shorthand for this is

{X ≤ a} ⊂ {X ≤ b}.

Whenever one event is contained in another, then the probability of the first
event is smaller than the probability of the second event. Thus

P ({X ≤ a}) ≤ P ({X ≤ b}),

i.e., FX (a) ≤ FX (b), as claimed.

Remark 8.8. Recall from Example 8.4 that, as we move from left to right
across the mass, we pick up the probabilities that build the CDF.
On the extreme left-hand side of the CDF, we would have probabilities
near (or at) 0, and on the extreme right-hand side of the CDF, we would have
probabilities near (or at) 1. Unfortunately, for instance, the mass might be
nonzero for arbitrarily large x, e.g., consider the example of the number of coin
tosses to see the first head, given below in Example 8.10. So the best thing we
can say with precision about the extreme behavior of the CDF is:
Remark 8.9. Concerning the right-hand behavior of the CDF,

lim FX (x) = 1,
x→∞

and similarly for the left-hand behavior of the CDF,

lim FX (x) = 0.
x→−∞

Intuitively, for instance, FX (100) = P (X ≤ 100) is less than FX (10,000) =

P (X ≤ 10,000), which is less than FX (1,000,000) = P (X ≤ 1,000,000), etc.,
etc., with the limit always equal to 1. Consider what would go wrong if
limx→∞ FX (x) < 1. E.g., if limx→∞ FX (x) = 0.95, then X would be equal
to +∞ five percent of the time, but that is not allowed, because random vari-
ables only take on real (finite) values.

8.4 More Examples

Example 8.10. Flip a coin until the first head appears; let X denote the total
number of flips until the first head appears.
 8.4. More Examples 93

The mass of X is depicted on the left of Figure 8.4; the values are:
pX (1) = P (X = 1) = P ({H}) = 1/2,
pX (2) = P (X = 2) = P ({T, H}) = 1/4,
pX (3) = P (X = 3) = P ({T, T, H}) = 1/8,
pX (4) = P (X = 4) = P ({T, T, T, H}) = 1/16,
and in general
j−1
z }| {
pX (j) = P (X = j) = P ({T, T, . . . , T , H}) = 1/2j .
Thus, the CDF of X (shown on the right of Figure 8.4) has the following values
at the integers:
FX (1) = P (X ≤ 1) = 1/2,
FX (2) = P (X ≤ 2) = 3/4,
FX (3) = P (X ≤ 3) = 7/8,
FX (4) = P (X ≤ 4) = 15/16,
and in general, for each positive integer x,
FX (x) = P (X ≤ x) = P ({first head within x tosses}) = 1 − 1/2x .
Another viewpoint is that, for positive integers x, we have FX (x) = P (X ≤
x) = 1 − P (X > x), but X > x only if the first x tosses are T , which has
probability 1/2x . Thus FX (x) = 1 − 1/2x .

pX (x) FX (x)
1 1
7/8 7/8
6/8 6/8
5/8 5/8
4/8 4/8
3/8 3/8
2/8 2/8
1/8 1/8
x x
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Figure 8.4: Left: The mass pX (x) of X, the number of tosses until the first
head. Right: The CDF FX (x) of X.

Remark 8.11. In Example 8.10, the observant reader will notice that, for
the outcome ω = (T, T, T, T, T, T, . . .), we did not define the value of X. This
outcome requires an infinite number of tosses until the first head is reached.
This outcome has probability 0, so it does not affect any of our calculations.
94 Chapter 8. Probability Mass Functions and CDFs

Example 8.12. As in Example 7.10, roll a pair of dice, and let X denote the
sum of the two values that appear. In other words, if the outcome is (i, j), then
we let X = i + j.

The sum X is a discrete random variable with values in {2, 3, 4, . . . , 12}. We

computed the mass back in Example 7.10; we reproduce it here:

x 2 3 4 5 6 7 8 9 10 11 12
1 2 3 4 5 6 5 4 3 2 1
P (X = x) 36 36 36 36 36 36 36 36 36 36 36

We plot the mass and CDF of the sum of two dice in Figure 8.5.

pX (x) FX (x)
1 1
5/6 5/6
4/6 4/6
3/6 3/6
2/6 2/6
1/6 1/6
x x
1 2 3 4 5 6 7 8 9 101112 1 2 3 4 5 6 7 8 9 101112

Figure 8.5: Left: The mass pX (x) of X, the sum of values on two dice.
Right: The CDF FX (x) of X.

Example 8.13. Throw a dart at a dartboard with 20 equally likely regions,

numbered 1 through 20. Let X denote the number of the region where the dart
lands. Assume that the person throwing the dart can always land the dart on
the board (otherwise, they are allowed to try again until they hit the board; the
probability of missing over and over—forever—will be 0, so it will not matter).

The mass of X is pX (x) = 1/20 for x = 1, 2, 3, . . . , 20, and pX (x) = 0 otherwise.

The CDF of X is FX (x) = x/20 for each integer x with 1 ≤ x ≤ 20. Thus

for x < 1,

0

FX (x) = bxc/20 for 1 ≤ x < 20,
for x ≥ 20.

1


Here bxc is obtained by rounding x down to an integer, e.g., b14.37c = 14.

 8.4. More Examples 95

Example 8.14. Suppose that 20.2% of cars are blue, 32.7% of cars are red,
3.06% of cars are silver, and the rest of the cars on the road are other colors.
Let X be “1” if the next car to pass is blue, or “2” if the next car is red, or “3”
if the next car is silver, or “0” otherwise.

The mass of X at the values 1, 2, 3 is given to be (respectively)

pX (1) = 0.202; pX (2) = 0.327; pX (3) = 0.0306.

Since the rest of the mass is at x = 0, and the mass adds to 1, the mass at 0
must be
pX (0) = 1 − 0.202 − 0.327 − 0.0306 = 0.4404.
For any values of x not in {0, 1, 2, 3}, we must have pX (x) = 0. The CDF is


0 for x < 0,
for

0.4404 0 ≤ x < 1,



FX (x) = 0.6424 for 1 ≤ x < 2,
for

0.9694 2 ≤ x < 3,




for

1 x ≥ 3.


Example 8.15. Draw a card from a well-shuffled deck until the ace of spades
appears. If a draw is unsuccessful, then replace and reshuffle the deck before
making the next selection. Let X be the number of draws needed until the ace
of spades appears for the first time.

(The event that the ace of spades never appears has probability 0. In such a
case, we could, for instance, make X = −1, or any other suitable solution. It
will not matter, because such an event has probability 0 anyway.)
The mass of X is pX (j) = (51/52)j−1 (1/52) for positive integers j, and
pX (x) = 0 otherwise.

Example 8.16. Suppose that a basketball player makes 80% of her free throws
successfully. She shoots as many times as necessary, until scoring the first
basket. Let X denote the number of necessary attempts.

(The probability that she never scores is 0, so we can safely ignore this outcome;
the other probabilities will not be affected.)
The mass of X is pX (j) = (0.20)j−1 (0.80) for positive integers j, and
pX (x) = 0 otherwise.
96 Chapter 8. Probability Mass Functions and CDFs

Example 8.17. Draw a card from a well-shuffled deck until the ace of spades
appears. If a draw is unsuccessful, do not replace the card—just continue to
draw. Let X be the number of draws needed until the ace of spades appears for
the first time.

Here, X will be one of the integers between 1 and 52, inclusive. As discussed
in Example 2.17, the placement of the ace of spades is equally likely to be
anywhere in the deck, so any of the first 52 draws are equally likely to be the
ace of spades. So the mass of X is pX (j) = 1/52 for j = 1, 2, 3, . . . , 52, and
pX (x) = 0 otherwise. The CDF of X is FX (x) = x/52 for each integer x with
1 ≤ x ≤ 52. Thus

for x < 1,

0

FX (x) = bxc/52 for 1 ≤ x < 52,
for x ≥ 52.

1


Again bxc is the integer obtained by rounding x down to an integer.

Example 8.18. A con artist has a trick die which is weighted so that a 1 will
come up half of the time. The other numbers are equally likely to appear. Let
X denote the number that appears when you roll the die once.

The possible values X assumes are still {1, 2, 3, 4, 5, 6}. We are given

pX (1) = P (X = 1) = 1/2.

Since the other values are equally likely to appear, we must have

pX (2) + · · · + pX (6) = P (X = 2) + · · · + P (X = 6) = 1/2,

so pX (j) = P (X = j) = 1/10 for each j in {2, 3, 4, 5, 6}.

8.5 Exercises
8.5.1 Practice
Exercise 8.1. Snacks. A student makes a trip once per day to the store, and
he always buys a snack. The student eats the snack 70% of the time, but the
other 30% of the time, his roommate eats it first.
 8.5. Exercises 97

a. During a period of four days, the student keeps track of whether he gets
to eat his snack. What is the sample space of possible outcomes?
b. Let X be the number of times, within the four day period, that he gets
to eat his snack. What is the mass of X?
c. What is the CDF of X?

Exercise 8.2. Post office. As in Exercise 2.21, a sequence of seven people

walk into a post office and only their sexes are noted. Assume that each of the
seven customers is equally likely to be a man or a woman. Let X denote the
number of customers that are female.

a. Find the mass of X.

b. Find the CDF of X.

Exercise 8.3. Songs by genre. As in Exercises 2.1 and 4.3, and in Ex-
ample 3.11, a randomly chosen song is from the blues genre with probabil-
ity 330/27333; from the jazz genre with probability 537/27333; from the rock
genre with probability 8286/27333; or from some other genre with probability
18180/27333. Let X be “1” if a randomly selected song is from the “blues” genre,
or “2” if “jazz,” or “3” if “rock,” or “−1” otherwise. Find the mass and CDF of
X.

Exercise 8.4. Milkfat. Suppose that 13% of all milk cartons have 1% milkfat,
and 28% of all milk cartons have 2% milkfat, and 18% of all milk cartons are
fat-free, and 41% of all milk cartons are of some other type. Randomly choose
a carton of milk, and let X be “1” if the carton of milk is 1% milkfat, or “2” if
the carton of milk is 2% milkfat, or “3” if the carton of milk is fat-free, or “4”
otherwise.

a. What is the mass of X?

b. Make a plot of the probability mass function.
c. What is the CDF of X?
d. Make a plot of the CDF.

Exercise 8.5. Super Breakfast Challenge. The Super Breakfast Challenge

(SBC) consists of bacon, eggs, oatmeal, orange juice, milk, and several other
foods, and it costs $12.99 per person to order at a local restaurant. It is known
to be very difficult to consume the entire SBC. Only 10% of people are able
to eat all of the SBC. The other 90% of people will be unable to eat the whole
SBC (it is too much food!).

A probability student hears about the SBC and goes to the local restaurant.
He observes the number of customers, X, that attempt to eat the SBC, until
the first success. So if there are 4 failures and then 1 success (i.e., the outcome
is (F, F, F, F, T )), then X = 5.
98 Chapter 8. Probability Mass Functions and CDFs

a. What is the mass of X?

b. Make a plot of the probability mass function.
c. What is the CDF of X?
d. Make a plot of the CDF.
Exercise 8.6. Cereal box prizes. A cereal company puts a Star Wars toy
watch in each of its boxes as a sales promotion. Twenty percent of the cereal
boxes contain a watch with Obi Wan Kenobi on it. You are a huge Obi Wan
fan, so you decide to buy 8 boxes of the cereal in hopes that you will find an
Obi Wan watch. Let X denote the number of Obi Wan watches you will have
from these boxes.
a. What are the possible values of X?
b. What is the mass of X?
c. Make a plot of the probability mass function.
d. What is the CDF of X?
e. Make a plot of the CDF.
Exercise 8.7. Number of hearts. Recall that a standard deck of 52 playing
cards has 4 suits (hearts, spades, clubs, and diamonds) and 13 cards in each
suit (labeled 2, 3, 4, 5, 6, 7, 8, 9, 10, J, Q, K, A). The cards are shuffled. You
are dealt the first 5 cards off the top of the deck. Let X be the number of hearts
you get.
a. What are the possible values of X?
b. What is the mass of X?
c. Make a plot of the probability mass function.
d. What is the CDF of X?
e. Make a plot of the CDF.
Exercise 8.8. Mystery constant. Suppose X is a discrete random variable
with a probability mass function pX (x) = c(4 − x) for x in {−1, 2, 3}, and
pX (x) = 0 otherwise.
a. What is the value of c so that pX (x) is a mass?
b. Make a plot of the probability mass function.
c. What is the CDF of X?
d. Make a plot of the CDF.
Exercise 8.9. Discrete Uniform distribution. Suppose X is a discrete
random variable with mass

pX (x) = 1/4 for x = 1, 2, 3, 4,

 8.5. Exercises 99

and pX (x) = 0 otherwise. For future reference, this is a called a Discrete

Uniform distribution. We will study these random variables more in Chapter 20.

a. Make a plot of the probability mass function.

b. What is the CDF of X?
c. Make a plot of the CDF.

Exercise 8.10. Mystery mass. Consider a random variable X that has CDF


 0 if x < 2,
0.3 if 2 ≤ x < 4,




FX (x) = 0.8 if 4 ≤ x < 6,
0.95 if 6 ≤ x < 8,





if 8 ≤ x.

1


What is the mass of X?

Exercise 8.11. Mystery mass. Consider a random variable X that has CDF



0 if x < −10,
0.1 if −10 ≤ x < −5,

FX (x) =


0.8 if −5 ≤ x < 0,
if 0 ≤ x.

1

What is the mass of X?

Exercise 8.12. Butterflies. Alice, Bob, and Charlotte are looking for butter-
flies. They look in three separate parts of a field, so that their probabilities of
success are independent.

• Alice finds 1 butterfly with probability 17%, and otherwise does not find
one.

• Bob finds 1 butterfly with probability 25%, and otherwise does not find
one.

• Charlotte finds 1 butterfly with probability 45%, and otherwise does not
find one.

Let X be the total number of butterflies that they find. Find the probability
mass function of X.

Exercise 8.13. Appetizers. At a restaurant that sells appetizers:

• 8% of the appetizers cost $1 each,

• 20% of the appetizers cost $2 each,

100 Chapter 8. Probability Mass Functions and CDFs

• 32% of the appetizers cost $3 each,

• 40% of the appetizers cost $4 each.
An appetizer is chosen at random, and X is its price. Draw the CDF of X.
Exercise 8.14. Two 4-sided dice. Roll two 4-sided dice and let X denote
the sum.
a. Draw the mass of X.
b. Draw the CDF of X.

8.5.2 Extensions
Exercise 8.15. Poisson distribution. Suppose X has mass
λx e−λ
pX (x) = for x ∈ Z≥0 ,
x!
and pX (x) = 0 otherwise.
a. For λ = 2, make a plot of the probability mass function.
b. What is the CDF of X, when λ = 2?
c. Make a plot of the CDF, when λ = 2.
For future reference, this is called a Poisson random variable. You may want
to refer to the Math Review for help with the summation of the terms of the
x
form λx! . We will study these random variables more in Chapter 18.
Exercise 8.16. Coin flips. Flip a coin three times. Let X denote the number
of heads minus the number of tails. So, for instance, if (H, T, T ) is the outcome,
then X = 1 − 2 = −1.
a. What are the possible values of X?
b. What is the mass of X?
c. Make a plot of the probability mass function.
d. What is the CDF of X?
e. Make a plot of the CDF.

8.5.3 Advanced
Exercise 8.17. Magic. Using a shuffled standard deck of 52 playing cards, a
magician wants to do a trick where he tries to guess which card an audience
member has selected if the audience member chooses a card at random and
doesn’t show the magician. Let X be the number of cards the magician will
guess correctly if he tries the trick with 6 audience members. Each audience
member starts with a complete and well-shuffled deck. The magician decides to
call a trial a success if he guesses the number on the card correctly even if he
doesn’t get the right suit.
 8.5. Exercises 101

a. What are the possible values of X?

b. What is the mass of X?
c. Make a plot of the probability mass function.
d. What is the CDF of X?
e. Make a plot of the CDF.

Exercise 8.18. Wastebasket basketball. Chris tries to throw a ball of paper

in the wastebasket behind his back (without looking). He estimates that his
chance of success each time, regardless of the outcome of the other attempts, is
1/3. Let X be the number of attempts required. If he is not successful within
the first 5 attempts, then he quits, and he lets X = 6 in such a case.

a. Draw the mass of X.

b. Draw the CDF of X.

Exercise 8.19. Pick two cards. (See also Exercise 7.17) Pick two cards at
random from a well-shuffled deck of 52 cards (pick them simultaneously, i.e.,
grab two cards at once—so they are not the same card!). There are 12 cards
which are considered face cards (4 Jacks, 4 Queens, 4 Kings). Let X be the
number of face cards that you get. Draw the CDF FX (x) of X.
Chapter 9

Independence and Conditioning

The most misleading assumptions are the ones you don’t even know you’re
making.
—“Meeting a Gorilla,” by Douglas Adams and Mark Carwardine, from Chap-
ter 2 of The Great Ape Project, edited by Paola Cavalieri and Peter Singer
(St. Martin’s Griffin, 1993)

Seven people stand in a post office lobby. We don’t know who came in first.
What is the probability the first person from that group who walked in the door
was female? How does that probability change if we know the total number of
females in the group of 7? If all 7 people are female, what do we know about
the probability that the first person was a female? If all 7 people are male, what
do we know about the probability that the first person was a female? What
happens to the probabilities for the gender of the first person for total numbers
of women in between 0 and 7?

9.1 Joint Probability Mass Functions

When dealing with two or more random variables at a time, it is usually helpful
to utilize a joint probability mass function (also called a joint PMF, or simply
a joint mass) or a joint cumulative distribution function (also called a joint
CDF). A joint mass of two random variables X, Y specifies, for each pair x, y,
the probability of X, Y taking on these specific values. Similarly, a joint CDF
of X and Y gives, for each x, y, the probability of X ≤ x and Y ≤ y.

102
 9.1. Joint Probability Mass Functions 103

Definition 9.1. Joint probability mass function (a.k.a. joint PMF or

joint mass)
The joint probability mass function of a pair of discrete random variables X
and Y is
pX,Y (x, y) = P ({ω | X(ω) = x and Y (ω) = y}),
i.e.,
pX,Y (x, y) = P (X = x and Y = y).

Definition 9.2. Joint cumulative distribution function (a.k.a. joint

CDF)
The joint CDF of a pair of discrete random variable X and Y is

FX,Y (x, y) = P ({ω | X(ω) ≤ x and Y (ω) ≤ y}),

or equivalently,
FX,Y (x, y) = P (X ≤ x and Y ≤ y).

Example 9.3. Roll two dice. Let X denote the minimum of the two values
that appear, and let Y denote the maximum of the two values that appear.

For instance,

pX,Y (3, 5) = P ({min = 3 and max = 5})

= P ({die values 3,5}) + P ({die values 5,3})
= 2/36.

Also,

pX,Y (3, 3) = P ({min = 3 and max = 3})

= P ({die values 3,3})
= 1/36.

In general, for 1 ≤ x < y ≤ 6,

pX,Y (x, y) = P ({min = x and max = y})

= P ({die values x, y}) + P ({die values y, x})
= 2/36,

and for 1 ≤ x = y ≤ 6,

pX,Y (x, y) = P ({min = y = x = max})

= P ({die values x, x}) since x and y are the same in this case
= 1/36,
104 Chapter 9. Independence and Conditioning

and
pX,Y (x, y) = 0 otherwise.
So the case pX,Y (x, y) = 2/36 with 1 ≤ x < y ≤ 6 corresponds to the two
possible outcomes (x, y) or (y, x), and the case pX,Y (x, y) = 1/36 with 1 ≤ x =
y ≤ 6 corresponds to the one possible outcome (x, x) (here, we emphasize that
y = x, i.e., the maximum and minimum are exactly the same because the die
rolls are the same).
The cumulative distribution function is calculated similarly, e.g.,
FX,Y (2, 4) = P ({min ≤ 2 and max ≤ 4})
= P ({(1, 1), (1, 2), (1, 3), (1, 4), (2, 1), (2, 2),
(2, 3), (2, 4), (3, 1), (3, 2), (4, 1), (4, 2)})
= 12/36.
As with CDFs for one variable, nothing changes if we calculate, for instance,
FX,Y (2.9, 4.1) = P (X ≤ 2.9 and Y ≤ 4.1) = P (X ≤ 2 and Y ≤ 4) = 12/36.
Using the chart below, we can easily obtain all of the values of FX,Y (x, y). As
another example,
FX,Y (5, 2) = P ({min ≤ 5 and max ≤ 2})
= P ({(1, 1), (1, 2), (2, 1), (2, 2)})
= 4/36.
In general, we have
FX,Y (x, y) y=1 y=2 y=3 y=4 y=5 y=6
x=1 1/36 3/36 5/36 7/36 9/36 11/36
x=2 1/36 4/36 8/36 12/36 16/36 20/36
x=3 1/36 4/36 9/36 15/36 21/36 27/36
x=4 1/36 4/36 9/36 16/36 24/36 32/36
x=5 1/36 4/36 9/36 16/36 25/36 35/36
x=6 1/36 4/36 9/36 16/36 25/36 36/36

Example 9.4. Flip a fair coin three times. Let X be the total number of heads
that appear, and let Y be the total number of tails that appear.
First of all, we notice that X + Y = 3 always. So we are certain to have
pX,Y (x, y) = 0 if x + y 6= 3. We can easily make a chart of all of the possible
outcomes—organized strategically into events according to the values of X and
Y that they induce—and the associated values of X:
Event Probability Joint Mass of X and Y
{(H, H, H)} 1/8 pX,Y (3, 0) = 1/8
{(H, H, T ), (H, T, H), (T, H, H)} 3/8 pX,Y (2, 1) = 3/8
{(T, T, H), (T, H, T ), (H, T, T )} 3/8 pX,Y (1, 2) = 3/8
{(T, T, T )} 1/8 pX,Y (0, 3) = 1/8
 9.1. Joint Probability Mass Functions 105

Otherwise, pX,Y (x, y) = 0.

Using the chart below, we can easily obtain all of the values of FX,Y (x, y).
For example,

FX,Y (1, 2) = P ({X ≤ 1 and Y ≤ 2})

= P ({(T, T, H), (T, H, T ), (H, T, T )})
= 3/8.

As another example, FX,Y (2, 2) = FX,Y (1, 2) + FX,Y (2, 1) = 3/8 + 3/8 = 6/8.
In general, we have
FX,Y (x, y) y=0 y=1 y=2 y=3
x=0 0 0 0 1/8
x=1 0 0 3/8 4/8
x=2 0 3/8 6/8 7/8
x=3 1/8 4/8 7/8 8/8

Example 9.5. Roll a 6-sided fair die and let X denote the outcome. Also flip
a spinner that shows “1” with probability 0.30, or shows “2” with probability
0.32, or shows “3” with probability 0.38, and let Y denote the outcome.
Then for each integer j from 1 to 6 inclusive,

pX,Y (j, 1) = (1/6)(0.30);

pX,Y (j, 2) = (1/6)(0.32);
pX,Y (j, 3) = (1/6)(0.38);

and pX,Y (x, y) = 0 otherwise.

Remark 9.6. Since the joint mass and joint CDF are probabilities, then

0 ≤ pX,Y (x, y) ≤ 1 and 0 ≤ FX,Y (x, y) ≤ 1 for all x, y.

The joint mass, summed over all x’s and y’s, takes the probabilities from the
whole sample space into account, so
XX
pX,Y (x, y) = 1.
x y

Example 9.7. As in Example 9.3, roll two dice. Let X denote the minimum
of the two values that appear, and let Y denote the maximum of the two values
that appear.
106 Chapter 9. Independence and Conditioning

We can calculate the mass of X directly, e.g.,

pX (3) = P ({(3, 3), (3, 4), (3, 5), (3, 6), (4, 3), (5, 3), (6, 3)}) = 7/36.

The mass of X is:

x 1 2 3 4 5 6
pX (x) 11/36 9/36 7/36 5/36 3/36 1/36.

Similarly, the mass of Y is

y 1 2 3 4 5 6
pY (y) 1/36 3/36 5/36 7/36 9/36 11/36.

Alternatively, we can compute the mass of X directly from the joint mass of X
and Y , by letting Y take on any value. For instance,
6
X 1 2 2 2
pX (3) = pX,Y (x, y) = 0 + 0 + + + + = 7/36.
36 36 36 36
y=1

A similar concept works much more generally:

Remark 9.8. Calculating the mass of one variable from the joint
mass. The mass of X can be calculated by summing the joint mass over all
possible values of Y : X
pX (x) = pX,Y (x, y).
y

Similarly, the mass of Y is the sum of the joint mass over all possible values
of X: X
pY (y) = pX,Y (x, y).
x

Example 9.9. Again, as in Example 9.3, roll two dice. Let X denote the
minimum of the two values that appear, and let Y denote the maximum of the
two values that appear.

We can calculate the CDF of X directly, e.g.,

FX (2) = P ({(1, 1), (1, 2), (1, 3), (1, 4), (1, 5), (1, 6),
(2, 1), (2, 2), (2, 3), (2, 4), (2, 5), (2, 6)
(3, 1), (3, 2), (4, 1), (4, 2), (5, 1), (5, 2), (6, 1), (6, 2)})
= 20/36.

The CDF of X is:

 9.2. Independent Random Variables 107

x 1 2 3 4 5 6
FX (x) 11/36 20/36 27/36 32/36 35/36 36/36

Similarly, the CDF of Y is

x 1 2 3 4 5 6
FY (y) 1/36 4/36 9/36 16/36 25/36 36/36

As an alternative, we can compute the CDF of X directly from the joint CDF
of X and Y , by letting Y → ∞, e.g.,

FX (3) = lim pX,Y (3, y) = 27/36.

y→∞

Again, we can generalize:

Remark 9.10. Calculating the CDF of one variable from the joint
CDF. The CDF of X can be calculated by taking the limit as y → ∞ in the
joint CDF:
FX (x) = lim FX,Y (x, y).
y→∞

Intuitively, when taking y → ∞ in the joint CDF, we are letting Y < ∞,

i.e., we let Y take on any value, since we only want to extract information
about X.
Similarly, the CDF of Y is the limit as x → ∞ in the joint CDF:

FY (y) = lim FX,Y (x, y).

x→∞

9.2 Independent Random Variables

Just as we discussed the notion of independence of events in Chapter 3, we have
a notion of independent random variables as well.

Example 9.11. As in Example 9.5, the mass of X is pX (j) = 1/6 for j =

1, 2, 3, 4, 5, 6, and pX (x) = 0 otherwise. The mass of Y is pY (1) = 0.30 and
pY (2) = 0.32 and pY (3) = 0.38, and pY (y) = 0 otherwise. So in every possible
case, we have the very nice condition

pX,Y (x, y) = pX (x)pY (y).

108 Chapter 9. Independence and Conditioning

This is one way to characterize two random variables as being independent.

There are several other equivalent formulations of this idea. It should make
intuitive sense that we might call X and Y independent, because the value of
X has no bearing whatsoever on the value of Y , and vice versa. If we think
about each underlying outcome as containing two pieces of information, one
about the die and one about the spinner, then only the piece of information
about the die has any impact on X, and only the piece of information about
the spinner has any impact on Y . As another example:

Example 9.12. Roll a die and flip a fair coin. Let X be the result of the die
roll. Let Y be 0 if the coin shows a “tail” or 1 if the coin shows a “head.” Notice
pX,Y (x, y) = 1/12
for all 1 ≤ x ≤ 6 and 0 ≤ y ≤ 1, since all twelve of these outcomes are equally
likely. Also
pX (x) = 1/6 for 1 ≤ x ≤ 6,
pY (y) = 1/2 for 0 ≤ y ≤ 1.
So pX,Y (x, y) = pX (x)pY (y), so X and Y are independent.

We introduce several equivalent ways to state that two discrete random

variables are independent, and all of these ways are equivalent to each other.
(Exercise 9.11 asks how to prove that these statements are equivalent.)
Definition 9.13. Independent discrete random variables: Several
equivalent formulations
1. Joint mass factors into a function of x times a function of y.
These functions of x and y can also be normalized so they are the masses of X
and Y , respectively:

pX,Y (x, y) = pX (x)pY (y) for all x and y.

2. Joint CDF factors into a function of x times a function of y. These

functions of x and y can also be normalized so they are the CDFs of X and Y ,
respectively:

FX,Y (x, y) = FX (x)FY (y) for all x and y.

3. We will define the conditional mass later in this chapter, but we go ahead
and state a way to use conditional mass for independence.
(a) Mass of X is conditional mass pX|Y (x | y) of X given Y = y,
or (b) Mass of Y is conditional mass pY |X (y | x) of Y given X = x.
 9.2. Independent Random Variables 109

An informal way to summarize is to say that X and Y are independent if

their joint behavior can be completely separated, so that the probability of X
and Y jointly behaving in some way is equal to the probability of X behaving
some way times the probability of Y behaving some way.
In practice, to show that two discrete random variables X and Y are inde-
pendent, we usually either show that

pX,Y (x, y) = pX (x)pY (y),

or (to be discussed in Section 9.4) that

pX|Y (x | y) = pX (x),

or
pY |X (y | x) = pY (y).

Once we get more familiar with these concepts, we will not go through such
tedious calculations. Sometimes we will just go ahead and observe that “X and
Y are independent,” but for now it is good to practice a little bit.

Example 9.14. Let X indicate whether the first baby born to a certain mother
is a girl, i.e., X = 1 if the first baby born is a girl; otherwise, X = 0. Let Y
indicate whether the second baby born to a certain mother is a girl, i.e., Y = 1
if the second baby born is a girl; otherwise, Y = 0. (We are not considering the
birth of twins, in which one baby’s sex might affect the other.) Since

pX,Y (x, y) = 1/4 for 0 ≤ x, y ≤ 1,

and since X and Y each take values in the set {0, 1}, we can factor the joint
mass pX,Y (x, y) = 1/4 into 1/4 = (1/2) · (1/2), and we must have pX (x) = 1/2
for x = 0, 1 and pY (y) = 1/2 for y = 0, 1. So X and Y are independent.

Example 9.15. Roll a die. Let X be 1 if the outcome is 1, 3, or 5; let X be 0

otherwise. Let Y be 1 if the outcome is 5 or 6; let Y be 0 otherwise. Since

pX,Y (0, 0) = P ({outcome is 2, 4}) = 2/6

pX,Y (0, 1) = P ({outcome is 6}) = 1/6
pX,Y (1, 0) = P ({outcome is 1, 3}) = 2/6
pX,Y (1, 1) = P ({outcome is 5}) = 1/6
110 Chapter 9. Independence and Conditioning

Thus pX,Y (x, y) can be factored as pX,Y (x, y) = pX (x)pY (y), by writing

pX (x) = 1/2 for x = 0 or x = 1,

and
pY (1) = 1/3 and pY (0) = 2/3.

So X, Y are independent random variables.

Theorem 9.16. Indicators of independent events are independent

random variables
Consider events A and B. Let X be an indicator for event A, i.e., X = 1 if A
occurs, and X = 0 otherwise. Let Y be an indicator for event B, i.e., Y = 1
if B occurs, and Y = 0 otherwise. Then, A and B are independent events, if
and only if X and Y are independent random variables.

To see that the above box is true, note that

pX,Y (1, 1) = P (A ∩ B), P (A)P (B) = pX (1)pY (1);

pX,Y (1, 0) = P (A ∩ B c ), P (A)P (B c ) = pX (1)pY (0);

pX,Y (0, 1) = P (Ac ∩ B), P (Ac )P (B) = pX (0)pY (1);

pX,Y (0, 0) = P (Ac ∩ B c ), P (Ac )P (B c ) = pX (0)pY (0).

A moment’s thought shows that the commas (in the four lines above) can be
replaced with equalities if and only if A and B are independent events or equiv-
alently if X and Y are independent random variables.

Example 9.17. Flip a coin until a head appears. Let A denote the event that
an even number of flips are required. (The case that a head never appears can
be included in A; it does not matter, since the probability that a head never
appears is 0.) Let B denote the event that 11 or more flips are necessary. Let X
indicate whether A occurs (i.e., X = 1 if A occurs, and X = 0 otherwise). Let
Y indicate whether B occurs (i.e., Y = 1 if B occurs, and Y = 0 otherwise).

In this example, X and Y are independent random variables. To see this, it is

enough to show that A and B are independent events. We notice that B occurs
if the first ten flips are tails, so P (B) = (1/2)10 . Also, A occurs if there are an
 9.2. Independent Random Variables 111

even number of flips until a head, which happens with probability

P (A) = P (T H) + P (T T T H) + P (T T T T T H) + P (T T T T T T T H) + · · ·
= (1/2)2 + (1/2)4 + (1/2)6 + (1/2)8 + · · ·
= 1/4 + (1/4)2 + (1/4)3 + (1/4)4 + · · ·
= (1/4)(1 + (1/4) + (1/4)2 + (1/4)3 + · · · )
  
1 1
=
4 1 − 1/4
  
1 1
=
4 3/4
= 1/3

To get an intuitive idea why P (A) = 1/3, look at the first two flips: We are using
geometric sums.
1. If the first two flips are HH, then Ac occurs, i.e., an odd number of flips
was needed to see the first head.

2. If the first two flips are HT , then Ac occurs, i.e., an odd number of flips
was needed to see the first head.

3. If the first two flips are T H, then A occurs, i.e., an even number of flips
was needed to see the first head.

4. If the first two flips are T T , then we essentially start over; just look at
the next pair of flips.

Thus A occurs in exactly 1 out of the 3 deciding cases (and the 3 deciding cases
are all equally likely).
Finally, we calculate the probability of A and B occurring:

11 13
z }| { z }| {
P (A ∩ B) = P (T T T T T T T T T T T H) + P (T T T T T T T T T T T T T H)
15
z }| {
+ P (T T T T T T T T T T T T T T T H) + · · ·
= (1/2)12 + (1/2)14 + (1/2)16 + (1/2)18 + · · ·
= (1/4)6 + (1/4)7 + (1/4)8 + (1/4)9 + · · ·
= (1/4)6 (1 + (1/4) + (1/4)2 + (1/4)3 + · · · )
 6  
1 1
=
4 1 − 1/4
 6  
1 1
=
4 3/4
= (1/4)5 (1/3)
112 Chapter 9. Independence and Conditioning

So we conclude

P (A ∩ B) = (1/4)5 (1/3) = (1/2)10 (1/3) = P (A)P (B).

So events A and B are independent. Since X and Y are, respectively, indicator

random variables for A and B, this means that X and Y are independent
random variables too.

9.3 Three or More Independent Random Variables

To be independent, as a collection, three or more random variables must satisfy
rules similar to those for collections of three or more events. Possibilities include
the following:

Definition 9.18. Joint probability mass function

If X1 , X2 , . . . , Xn are random variables, then the joint probability mass func-
tion (the joint mass) of X1 , X2 , . . . , Xn is the probability that Xj = xj for
each j, i.e.,

pX1 ,X2 ,...,Xn (x1 , x2 , . . . , xn ) = P (Xj = xj for all j).

Remark 9.19. Independent discrete random variables: Joint mass

equals the product of the masses
A finite number of discrete random variables X1 , X2 , . . . , Xn are independent
if and only if their joint mass is equal to the product of their masses, i.e., if

pX1 ,X2 ,...,Xn (x1 , x2 , . . . , xn ) = pX1 (x1 )pX2 (x2 ) · · · pXn (xn )

for all x1 , x2 , . . . , xn .

Another equivalent way to say that X1 , X2 , . . . , Xn are independent can be

given, after we introduce the concept of a joint cumulative distribution function
for n variables.

Definition 9.20. Joint cumulative distribution function

If X1 , X2 , . . . , Xn are random variables, then the joint cumulative distribution
function (also called joint CDF) of X1 , X2 , . . . , Xn is the probability that Xj ≤
xj for every j, i.e.,

FX1 ,X2 ,...,Xn (x1 , x2 , . . . , xn ) = P (Xj ≤ xj for all j).

 9.4. Conditional Probability Mass Functions 113

Remark 9.21. Independent discrete random variables: Joint CDF

equals the product of the CDFs
A collection of random variables X1 , X2 , . . . , Xn are independent if and only
if their joint CDF is equal to the product of their CDFs, i.e., if

FX1 ,X2 ,...,Xn (x1 , x2 , . . . , xn ) = FX1 (x1 )FX2 (x2 ) · · · FXn (xn )

for all x1 , x2 , . . . , Xn .

Remark 9.22. Indicators of independent events are independent

random variables
Consider a collection of events A1 , A2 , . . . , An . For each j, let Xj be an indi-
cator for event Aj , i.e., Xj = 1 if Aj occurs, and Xj = 0 otherwise. Then,
the Aj ’s are independent events if and only if X1 , X2 , . . . , Xn are independent
random variables.

9.4 Conditional Probability Mass Functions

When two random variables are under consideration at the same time, some-
times it is helpful to know how one of them will impact the other one. In order
to characterize such an effect, we introduce the conditional probability mass
function (also called a conditional PMF, or simply a conditional mass). A con-
ditional mass completely specifies the probability of one random variable, given
that we know the value of the other random variable. Such a conditional mass
of X, given the value of Y , is specified as follows:

Definition 9.23. Conditional probability mass function

The conditional probability mass function of a discrete random variable X,
given the value of another discrete random variable Y , is

pX|Y (x | y) = P (X = x | Y = y).

The conditional probability mass function of X, given the value of Y , is also

referred to as “the conditional PMF of X given Y ” or as “the conditional mass
of X given Y .”
The conditional mass is defined as the ratio of the joint mass divided by the
mass of Y :
P (X = x and Y = y) pX,Y (x, y)
pX|Y (x | y) = = .
P (Y = y) pY (y)

For the conditional mass of X given Y to make sense, we must use Y values
such that P (Y = y) > 0.
114 Chapter 9. Independence and Conditioning

Example 9.24. Roll two dice. Let X denote the value of the first die, and let
Y denote the value of the sum of the two dice. If we are given Y = 4, then we
know that the set of possible outcomes are

{(1, 3), (2, 2), (3, 1)}.

Thus X is either 1, 2, or 3. So the condition mass of X, given Y = 4, is

pX,Y (1, 4) P ({(1, 3)}) 1/36 1

pX|Y (1 | 4) = = = = ,
pY (4) P ({(1, 3), (2, 2), (3, 1)}) 3/36 3
pX,Y (2, 4) P ({(2, 2)}) 1/36 1
pX|Y (2 | 4) = = = = ,
pY (4) P ({(1, 3), (2, 2), (3, 1)}) 3/36 3
pX,Y (3, 4) P ({(3, 1)}) 1/36 1
pX|Y (3 | 4) = = = = ,
pY (4) P ({(1, 3), (2, 2), (3, 1)}) 3/36 3

and pX|Y (x | y) = 0 otherwise.

Example 9.25. Flip the cards from a deck over, one at a time, until the whole
deck has been flipped over. Let X denote the number of cards until the first
ace appears. Let Y denote the number of cards until the first queen appears.
Then X and Y are dependent.

To see that X and Y are dependent, first notice that pY (1) > 0 because it is
possible (i.e., the probability is positive) that the first card is a queen. On the
other hand, if X = 1, i.e., if the first card is an ace, then the first card is not
a queen, so pY |X (1 | 1) = 0. Thus pY (y) 6= pY |X (y | x) when x and y are both
equal to 1. So X and Y are not independent. In fact, X and Y are dependent.
(It is worthwhile to compare and contrast with Example 9.15, to see what is
similar and different in these two examples.)

Example 9.26. Roll a die. Let X be 1 if the outcome is 1, 2, or 3; let X be 0

otherwise. Let Y be 1 if the outcome is even (2, 4, or 6); let Y be 0 otherwise.

To see that X and Y are dependent, first notice that pY (1) = 1/2 because the
outcome is even with probability 1/2. On the other hand, if X = 1, i.e., if
the outcome is 1, 2, or 3, then the outcome is even with probability 1/3. so
pY |X (1 | 1) = 1/3. Thus pY (y) = 1/2 6= 1/3 = pY |X (y | x) when x and y
are both equal to 1. So X and Y are not independent. In fact, X and Y are
dependent.
 9.4. Conditional Probability Mass Functions 115

Example 9.27. As in Example 5.2, in a certain household, 20% of the milk has
two-percent milkfat, and the other 80% of the milk is whole milk. The whole
milk is spoiled 5% of the time; overall, the milk is spoiled 4.7% of the time.

In that example, we let A denote the event that the milk is came from a whole
milk carton, and we let B denote the event that the milk was spoiled. We were
given P (A) = 0.80, and we calculated

P (whole milk | spoiled) = P (A | B) = 0.85.

In this example, let X indicate whether A occurred, i.e.,

(
1 if A occurs,
X=
0 otherwise.

Let Y indicate whether B occurred, i.e.,

(
1 if B occurs (milk was spoiled),
Y =
0 otherwise (milk was not spoiled).

Then X and Y are dependent. To see this, notice P (A) = 0.80 is the same
as pX (1) = 0.80, and P (A | B) = 0.85 is the same as pX|Y (1 | 1) = 0.85. So
pX (x) 6= pX|Y (x | y) when x = 1 and y = 1.

Example 9.28. In Example 9.15, X and Y are independent. (In that example,
we roll a die and let X be 1 if the outcome is 1, 3, or 5; let X be 0 otherwise.
Let Y be 1 if the outcome is 5 or 6; let Y be 0 otherwise.) An alternative way
to see that X and Y are independent is to show that pY |X (y | x) = pY (y).
We have pY (1) = 1/3 and pY (0) = 2/3.
Also, given that X = 1, then the outcome is 1, 3, or 5, so Y = 1 exactly
1/3 of the time, thus pY |X (1 | 1) = 1/3 and pY |X (0 | 1) = 2/3; similarly, given
that X = 0, then the outcome is 2, 4, or 6, so Y = 1 exactly 1/3 of the time,
so pY |X (1 | 0) = 1/3 and pY |X (0 | 0) = 2/3. So, regardless of the value of x, we
have pY |X (1 | x) = 1/3 and pY |X (0 | x) = 2/3.
Thus pY |X (y | x) = pY (y) in all cases. So X and Y are independent.

Example 9.29. As in Exercises 2.21 and 8.2, a sequence of seven people walk
into a post office (one at a time) and only their sexes are noted. Assume that
each of the seven customers is equally likely to be a man or a woman. Let Y
denote the number of customers that are female. Let X = 0 if the first person
is a male, or X = 1 if the first person is a female.
116 Chapter 9. Independence and Conditioning

Method #1 of computing the conditional mass. Before computing the

conditional mass of X given Y , it is usually helpful to compute the mass of Y
separately, and to compute the joint mass of X and Y . This is the method we
follow here. The mass of Y is the following:
pY (0) = 1/128 pY (1) = 7/128 pY (2) = 21/128 pY (3) = 35/128
pY (4) = 35/128 pY (5) = 21/128 pY (6) = 7/128 pY (7) = 1/128
and pY (y) = 0 otherwise.
Now we compute the joint mass of X and Y . We have X = 0 only if the first
person is a male, which happens 1/2 of the time. This leaves Y as an integer
between 0 and 6, so we have
pX,Y (0, 0) = (1/2)(1/64) = 1/128 pX,Y (0, 1) = (1/2)(6/64) = 6/128
pX,Y (0, 2) = (1/2)(15/64) = 15/128 pX,Y (0, 3) = (1/2)(20/64) = 20/128
pX,Y (0, 4) = (1/2)(15/64) = 15/128 pX,Y (0, 5) = (1/2)(6/64) = 6/128
pX,Y (0, 6) = (1/2)(1/64) = 1/128
We have X = 1 only if the first person is a female, which happens 1/2 of the
time. This leaves Y as an integer between 1 and 7, so we have
pX,Y (1, 1) = (1/2)(1/64) = 1/128 pX,Y (1, 2) = (1/2)(6/64) = 6/128
pX,Y (1, 3) = (1/2)(15/64) = 15/128 pX,Y (1, 4) = (1/2)(20/64) = 20/128
pX,Y (1, 5) = (1/2)(15/64) = 15/128 pX,Y (1, 6) = (1/2)(6/64) = 6/128
pX,Y (1, 7) = (1/2)(1/64) = 1/128
and pX,Y (x, y) = 0 otherwise.
Now we are ready to compute pX|Y (x | y) in all cases:
pX,Y (0, 0) 1/128
pX|Y (0 | 0) = = =1
pY (0) 1/128
pX,Y (0, 1) 6/128 6
pX|Y (0 | 1) = = =
pY (1) 7/128 7
pX,Y (0, 2) 15/128 5
pX|Y (0 | 2) = = =
pY (2) 21/128 7
pX,Y (0, 3) 20/128 4
pX|Y (0 | 3) = = =
pY (3) 35/128 7
pX,Y (0, 4) 15/128 3
pX|Y (0 | 4) = = =
pY (4) 35/128 7
pX,Y (0, 5) 6/128 2
pX|Y (0 | 5) = = =
pY (5) 21/128 7
pX,Y (0, 6) 1/128 1
pX|Y (0 | 6) = = =
pY (6) 7/128 7
pX,Y (0, 7) 0/128
pX|Y (0 | 7) = = =0
pY (7) 1/128
 9.4. Conditional Probability Mass Functions 117

and
pX,Y (1, 0) 0
pX|Y (1 | 0) = = =0
pY (0) 1/128
pX,Y (1, 1) 1/128 1
pX|Y (1 | 1) = = =
pY (1) 7/128 7
pX,Y (1, 2) 6/128 2
pX|Y (1 | 2) = = =
pY (2) 21/128 7
pX,Y (1, 3) 15/128 3
pX|Y (1 | 3) = = =
pY (3) 35/128 7
pX,Y (1, 4) 20/128 4
pX|Y (1 | 4) = = =
pY (4) 35/128 7
pX,Y (1, 5) 15/128 5
pX|Y (1 | 5) = = =
pY (5) 21/128 7
pX,Y (1, 6) 6/128 6
pX|Y (1 | 6) = = =
pY (6) 7/128 7
pX,Y (1, 7) 1/128
pX|Y (1 | 7) = = =1
pY (7) 1/128
Notice: The two terms in each row have a sum of 1, i.e., P pX|Y (0 | y) +
pX|Y (1 | y) = 1 for each y, because the sum over x’s, i.e., x pX|Y (x | y)
must be 1 for each fixed value of y, because the first person to enter is either
male (X = 0) or female (X = 1).
Method #2 of computing the conditional mass. In hindsight, we
could have computed pX|Y (x | y) directly, but it was perhaps helpful to go
through such a rigorous exercise. Now we directly compute pX|Y (x | y); this is
much, much shorter but requires some subtle insight. Suppose that it is known
that Y = y. Then, in other words, it is known that exactly y out of the 7
people are female. The first person is equally likely to be any of these 7 people,
of which y are female and 7−y are male. So the probability that the first person
is male is exactly (7 − y)/7; the probability that the first person is female is
exactly y/7. So we always have the following: Given that Y = y is an integer
between 1 and 7, the probability that the first person is a male is exactly
pX|Y (0 | y) = (7 − y)/7;
the probability that the first person is a female is exactly
pX|Y (1 | y) = y/7.

In closing, we note that the joint mass of two random variables X, Y is equal
to the conditional mass of X given Y multiplied by the mass of Y .
Remark 9.30. For all random variables X and Y (regardless of indepen-
dence), we have
pX,Y (x, y) = pX|Y (x | y)pY (y).
118 Chapter 9. Independence and Conditioning

9.5 Exercises
9.5.1 Practice
Exercise 9.1. Random employee hiring. Ten students apply for a job
opening, but only 1 of the students will be selected. The employer chooses
randomly; all ten outcomes are equally likely. If person 3, 5, 7, or 9 gets the
job, let X = 1; otherwise, X = 0. If person 1, 2, 3, 4, or 5 gets the job, let
Y = 1; otherwise, Y = 0. Are X and Y independent random variables? Justify
your answer.

Exercise 9.2. Forgetful morning. Each day, Maude has a 1% chance of

losing her cell phone (her behavior on different days is independent). Each
day, Maude has a 3% chance of forgetting to eat breakfast (again, her behavior
on different days is independent). Her breakfast and cell phone habits are
independent. Let X be the number of days until she first loses her cell phone.
Let Y be the number of days until she first forgets to eat breakfast. (Here, X
and Y are independent.) Find the joint mass of X and Y .

Exercise 9.3. Dependence/independence among coin flips. A student

flips a fair coin until a head appears. Let X be the number of flips until (and
including) this first head. Afterwards, he begins flipping again until he gets
another head. Let Y be the number of flips, after the first head, up to (and
including) the second head. E.g., if the sequence of flips is TTTTTTHTTH
then X = 7 and Y = 3. Are X and Y independent? Justify your answer.

Exercise 9.4. More dependence/independence among coin flips. Same

scenario as problem 9.3. Let Z be the total number of flips until (and including)
the second head. So Z = X + Y ; e.g., in the example given, Z = 10. Are X
and Z independent? Justify your answer.

Exercise 9.5. Butterflies. Alice, Bob, and Charlotte are looking for butter-
flies. They look in three separate parts of a field, so that their probabilities of
success do not affect each other.

• Alice finds 1 butterfly with probability 17%, and otherwise does not find
one.

• Bob finds 1 butterfly with probability 25%, and otherwise does not find
one.

• Charlotte finds 1 butterfly with probability 45%, and otherwise does not
find one.

Let X be the number of butterflies that they find altogether. Let Y be the
number of people who do not find a butterfly.
Find the joint mass pX,Y (x, y) of X and Y .
 9.5. Exercises 119

Exercise 9.6. Dependence/independence among dice rolls. A student

rolls a die until the first “4” appears. Let X be the numbers of rolls required
until (and including) this first “4.” After this is completed, he begins rolling
again until he gets a “3.” Let Y be the number of rolls, after the first “4,” up to
(and including) the next “3.” E.g., if the sequence of rolls is 213662341261613
then X = 8 and Y = 7. Are X and Y independent? Justify your answer.

Exercise 9.7. Pick two cards. Pick two cards at random from a well-shuffled
deck of 52 cards (pick them simultaneously, i.e., grab two cards at once—so they
are not the same card!). There are 12 cards which are considered face cards (4
Jacks, 4 Queens, 4 Kings). There are 4 cards with the value 10. Let X be the
number of face cards in your hand; let Y be the number of 10’s in your hand.
Are X and Y dependent or independent?

9.5.2 Extensions
Exercise 9.8. Dice. Roll two dice, one colored red and one colored blue. Let
Y denote the maximum value that appears on the two dice. Let X denote the
value of the blue die. Find the conditional mass of X given Y .

Exercise 9.9. Wastebasket basketball. Chris tries to throw a ball of paper

in the wastebasket behind his back (without looking). He estimates that his
chance of success each time, regardless of the outcome of the other attempts, is
1/3. Let X be the number of attempts required. If he is not successful within
the first 5 attempts, then he quits, and he lets X = 6 in such a case.
Let Y indicate whether he makes the basket successfully within the first
three attempts. Thus Y = 1 if his first, second, or third attempt is successful,
and Y = 0 otherwise.
Find the conditional mass of X given Y . You will need to list 12 values
altogether, i.e., you need to compute pX|Y (x | y) for 1 ≤ x ≤ 6 and 0 ≤ y ≤ 1.

Exercise 9.10. Two 4-sided dice. Consider some 4-sided dice. Roll two of
these dice. Let X denote the minimum of the two values that appear, and let
Y denote the maximum of the two values that appear.

a. Find the joint mass pX,Y (x, y) of X and Y .

b. Find the joint CDF FX,Y (x, y) of X and Y . It suffices to give the values
FX,Y (x, y) when x and y are integers between 1 and 4. You do not have to list
any other values; these 16 possibilities will suffice.

9.5.3 Advanced
Exercise 9.11. Prove that the statements of independence in Definition 9.13
are equivalent.
Chapter 10

Expected Values of Discrete

Random Variables

Your teacher can open the door, but you must enter by yourself.
—Proverb

You are having your weekly poker game with some of the people from your
dorm. The dealer shuffles the cards and then deals each person five cards. How
many of the cards in your hand do you expect to be hearts?

10.1 Introduction
An expected value (also called a (weighted) average or mean) says something
succinct about a random variable. The expected value is one way to measure the
center of the distribution. It does not tell us everything about the distribution;
many different kinds of random variables can have the same expected value.
Nonetheless, an expected value of a random variable is helpful. It describes the
sum of the values that a random variable takes, in proportion to the mass the
random variable has at each value.
Definition 10.1. Expected value of a discrete random variable
A discrete random variable X that takes on values x1 , x2 , . . . , xn has expected
value
Xn
E(X) = xj pX (xj ).
j=1

If X takes on a countably infinite number of values, x1 , x2 , . . ., then the ex-

pected value of X is
∞
X
E(X) = xj pX (xj ).
j=1

120
 10.2. Examples 121

10.2 Examples

Example 10.2. Flip a coin three times. Let X denote the number of heads.

Since X = 0 with probability 1/8, X = 1 with probability 3/8, X = 2 with

probability 3/8, and X = 3 with probability 1/8, then the expected value of X
is
E(X) = (0)(1/8) + (1)(3/8) + (2)(3/8) + (3)(1/8) = 3/2.

For a second way to view the expected values, notice that we have eight
possible outcomes:
ω1 = T T T,

ω2 = HT T, ω3 = T HT, ω4 = T T H,

ω5 = HHT, ω6 = HT H, ω7 = T HH,

ω8 = HHH,
and we get, with this new interpretation of expected value, the very same result
at the end:

E(X) = (0)(1/8)
+ (1)(1/8) + (1)(1/8) + (1)(1/8)
+ (2)(1/8) + (2)(1/8) + (2)(1/8)
+ (3)(1/8)
= 3/2.

The reason that the two different definitions are equivalent is that we could just
group the 2nd, 3rd, and 4th terms above, which all have value “1” as the value
of X, to get
(1)(1/8) + (1)(1/8) + (1)(1/8) = (1)(3/8),
which was also found in the first definition. Similarly, we could just group the
5th, 6th, and 7th terms above, which all have value “2” as the value of X, and
we get
(2)(1/8) + (2)(1/8) + (2)(1/8) = (2)(3/8),
which was also found in the first definition. So these are just two different ways
of grouping things.

Thus, we can use either of these equivalent formulations:

122 Chapter 10. Expected Values of Discrete Random Variables

Remark 10.3. Two equivalent ways to get the expected value of a

random variable

1. a sum over all of the possible values of X, each weighted by the proba-
bility of X taking on that value, or

2. a sum over all of the possible outcomes, taking the value of X from such
an outcome, weighted by the probability of that outcome
In the second method, where we enumerate the possible outcomes, we have
the following:
Remark 10.4. Expected value of a discrete random variable (always
gives same result as the original definition)
Consider a random phenomenon with possible outcomes ω1 , ω2 , . . . , ωn . Sup-
pose that the outcome ωj causes random variable X to take on value xj . Then
the discrete random variable X has expected value
n
X
E(X) = xj P ({ωj }).
j=1

If the random phenomenon can take on one of infinitely many possible out-
comes ω1 , ω2 , . . ., and we again suppose that the outcome ωj causes random
variable X to take on value xj , then the expected value of X is
∞
X
E(X) = xj P ({ωj }).
j=1

Example 10.5. Consider a class consisting of eight students, who earn the
following scores on an exam: 75, 92, 88, 94, 89, 60, 83, 84. Let X be the
exam score of a randomly chosen student. Each student is equally likely to be
selected. What is the expected value of X?
There are eight possible outcomes in this random phenomenon, each of which
has probability 1/8. So, following the second definition of expected value, we
obtain

E(X) = (75)(1/8) + (92)(1/8) + (88)(1/8) + (94)(1/8)

+ (89)(1/8) + (60)(1/8) + (83)(1/8) + (84)(1/8)
= 83.125.

Example 10.6. Roll a die three times. Let X be the number of times that a
6 appears. Find the expected value of X.
 10.2. Examples 123

We compute:
P (X = 0) = (5/6)3 ;
P (X = 1) = (3)(1/6)(5/6)2
(since there are 3 ways to have 1 occurrence of 6, and each way has probability
(1/6)(5/6)2 );
P (X = 2) = (3)(1/6)2 (5/6)
(since there are 3 ways to have 2 occurrences of 6, and each way has probability
(1/6)2 (5/6));
P (X = 3) = (1/6)3 .
So the expected value of X is

E(X) = (0)(5/6)3 + (1)(3)(1/6)(5/6)2 + (2)(3)(1/6)2 (5/6) + (3)(1/6)3 = 1/2.

If we were to have 100 people each roll a die three times, and then we took
the average of these results (i.e., added the 100 results and divided by 100),
the average from these 100 experiments would likely be very close to 0.5. It
is worthwhile to try it and see. One thousand people would likely give us an
average even closer to 0.5. We will consider what actually happens in the long
run with experiments later in the book, when discussing Normal approximations
and the Central Limit Theorem.

Remark 10.7. Helpful fact for double-checking an answer:

When writing the expected value of a discrete random variable, we always
have a sum of terms, where each term is a value times a probability. We must
consider all possible cases, so the probabilities under consideration had better
add up to 1.
For instance, in the example above, the probabilities are (5/6)3 and
(3)(1/6)(5/6)2 and (3)(1/6)2 (5/6) and (1/6)3 . We can easily check that these
add up to 1:

(5/6)3 + (3)(1/6)(5/6)2 + (3)(1/6)2 (5/6) + (1/6)3 = 1.

Example 10.8. As in Examples 1.13 and 2.17, a student shuffles a deck of cards
thoroughly (one time) and then selects cards from the deck without replacement
until the ace of spades appears. How many cards does the student expect to
draw until the ace of spades appears?

Let X be the number of cards needed until the ace of spades appears. Then

P (X = j) = 1/52 for 1 ≤ j ≤ 52.

124 Chapter 10. Expected Values of Discrete Random Variables

Thus,

E(X) = (1)P (X = 1) + (2)P (X = 2) + (3)P (X = 3) + · · · + (52)P (X = 52)

= (1)(1/52) + (2)(1/52) + (3)(1/52) + · · · + (52)(1/52)
= (1 + 2 + · · · + 52)(1/52)
= 53/2

So the student expects to draw 53/2 cards (i.e., 26.5 cards) to see the ace of
spades appear. Here we used the helpful fact that 1 + 2 + · · · + n = (n)(n + 1)/2.
(We will further study such random variables, which are equally like to be
any one of the values 1, 2, . . . , N , in Chapter 18.)

Example 10.9. A student draws cards from a standard deck of playing cards
until the ace of spades appears for the first time. After every unsuccessful draw,
the student replaces the card and shuffles the deck thoroughly before selecting
a new card. How many cards does the student expect to draw until the ace of
spades appears?

Let X denote the number of cards needed until the ace of spades appears for
the first time. Then P (X = j) = (51/52)j−1 (1/52) for each positive integer j.
So

E(X) = (1)P (X = 1) + (2)P (X = 2) + (3)P (X = 3) + · · ·

X∞
= j(51/52)j−1 (1/52)
j=1
∞
1 X
= j(51/52)j−1
52
j=1

We notice that j(51/52)j−1 is the derivative of xj (with respect to x), evaluated

at x = 51/52. So we rewrite the equation above as follows:

∞ ∞
1 X d j 1 d X j
E(X) = x = x x=51/52
.
52 dx x=51/52 52 dx
j=1 j=1

We use our scrap paper to compute:

∞ ∞
X
j
X x
x =x xj = .
1−x
j=1 j=0
 10.2. Examples 125

So we get

1 d x
E(X) =
52 dx 1 − x x=51/52
1 (1 − x)(1) − (x)(−1)
=
52 (1 − x)2 x=51/52
1 1
=
52 (1 − x)2 x=51/52
1 1
=
52 1 − 51 2


52
1 1
=
52 (1/52)2
= 52

So the student expects to draw 52 cards to get the ace of spades. We will study
more examples like this in Chapter 16, on Geometric random variables.

Example 10.10. A standard deck of 52 cards has 13 hearts. The cards in such
a deck are shuffled, and the top five cards are dealt to a player. What is the
expected number of hearts that the player receives?

Let X denote the number of hearts that the player receives. Then

(39)(38)(37)(36)(35) 2109
P (X = 0) = =
(52)(51)(50)(49)(48) 9520
(13)(39)(38)(37)(36) 27417
P (X = 1) = 5 =
(52)(51)(50)(49)(48) 66640
(13)(12)(39)(38)(37) 9139
P (X = 2) =10 =
(52)(51)(50)(49)(48) 33320
(13)(12)(11)(39)(38) 2717
P (X = 3) =10 =
(52)(51)(50)(49)(48) 33320
(13)(12)(11)(10)(39) 143
P (X = 4) = 5 =
(52)(51)(50)(49)(48) 13328
(13)(12)(11)(10)(9) 33
P (X = 5) = =
(52)(51)(50)(49)(48) 66640

(Double check: P (X = 0) + P (X = 1) + · · · + P (X = 5) = 1; thus, we have

correctly considered all of the possibilities.) So we get
2109 27417 9139 2717 143 33
E(X) = (0) +(1) +(2) +(3) +(4) +(5) = 5/4.
9520 66640 33320 33320 13328 66640
126 Chapter 10. Expected Values of Discrete Random Variables

Example 10.11. Jim and his brother both like chocolate chip cookies best.
They have a jar of cookies with 5 chocolate chip cookies, 3 oatmeal cookies,
and 4 peanut butter cookies. They are each allowed to have 3 cookies. To be
fair, they agree to randomly select their cookies without peeking, and they each
must keep the cookies that they select. How many chocolate chip cookies does
Jim expect to get? (Notice that it does not matter whether Jim or his brother
selects the cookies first—the answer will be the same, either way.)

Let X the number of chocolate chip cookies that Jim selects. Since there are
12 cookies, there are (12)(11)(10) equally likely outcomes for Jim. Exactly
(7)(6)(5) of them have no chocolate chips. So

(7)(6)(5) 7
P (X = 0) = = .
(12)(11)(10) 44
There are exactly 3 ways that one cookie could be chocolate chip; 5 such cookies
could be the chocolate one; the other cookies could be selected in (7)(6) ways.
So
(3)(5)(7)(6) 21
P (X = 1) = = .
(12)(11)(10) 44
There are exactly 3 ways that two cookies could be chocolate chips; 7 cookies
could be the non-chocolate chip one; the other cookies could be chocolate chips,
selected in (5)(4) ways. So

(3)(7)(5)(4) 7
P (X = 2) = = .
(12)(11)(10) 22

Exactly (5)(4)(3) of the outcomes are all chocolate chips. So

(5)(4)(3) 1
P (X = 3) = = .
(12)(11)(10) 22
So the expected value of X is
7 21 7 1
E(X) = (0) + (1) + (2) + (3) = 5/4.
44 44 22 22
Note: The total number of outcomes is (12)(11)(10) = 1320. As a check,
210 + 630 + 420 + 60 = 1320, so all 1320 possible outcomes have been taken into
account.

10.3 Exercises
10.3.1 Practice
Exercise 10.1. Graduation. As in Exercise 3.1, Jack and Jill are indepen-
dently struggling to pass their last (one) class required for graduation. Jack
 10.3. Exercises 127

needs to pass Calculus III, but he only has probability 0.30 of passing. Jill
needs to pass Advanced Pharmaceuticals, but she only has probability 0.46 of
passing. They work independently. Let X = 0 if neither of them graduates,
or X = 1 if exactly one of them graduate, or X = 2 if both of them graduate.
Find the expected value of X.
Exercise 10.2. Japanese pan noodles. As in Exercise 3.2, Four students
order noodles at a certain local restaurant. Their orders are placed indepen-
dently. Each student is known to prefer Japanese pan noodles 40% of the time.
How many of them do we expect to order Japanese pan noodles?
Exercise 10.3. Waiting for folk song. Angelina’s music player, in “shuf-
fle” mode, will play songs without any repetitions, until every song has been
played exactly once. The number of songs of each genre is the following: 330
blues songs; 537 jazz songs; and 1 folk song. So there are 868 songs altogether.
(All possible “shuffles”—i.e., all possible orderings of the 868 songs—are equally
likely.) How many songs does Angelina expect to listen to, until the folk song
finally appears? Please justify your answer. Hint: Use the method in Exam-
ple 10.8.
Exercise 10.4. Bowling strikes. In a bowling alley, 20% of the time when
someone bowls, he or she gets a strike. If there are 3 people in the bowling alley,
and X is the total number of people who get a strike on their current attempt,
what is the expected value of X?
Exercise 10.5. Career fair. You go to a career fair and have some job inter-
views. Based on career fair data, you think you have a 30% chance of getting
an offer for $40,000, a 40% chance of getting an offer of $44,000, a 25% chance
of getting an offer for $51,000, and a 5% chance of an offer for $57,000. What
is your expected offer salary?
Exercise 10.6. Golden ticket. A student was at work at the county am-
phitheater, and was given the task of cleaning 1500 seats. To make the job
more interesting, his boss hid a golden ticket somewhere in the seats. The
ticket is equally likely to be in any of the seats. Let X be the number of seats
cleaned until the ticket is found. Calculate the expected value of X.
Exercise 10.7. Football. A man is looking for a football game to watch on a
Sunday night. He has ten channels, and only one of them shows football, but
he can’t remember which one. Assume they are all equally likely to have it, and
let X be the number of channels he tries until he finds it. What is the expected
number of channels he tries until he finds it?
Exercise 10.8. Cashews. There is a bowl containing 30 cashews, 20 pecans,
25 almonds, and 25 walnuts. I am going to randomly pick and eat 3 nuts. What
is the expected number of cashews I will eat?
Exercise 10.9. Chess. Let X be the number of games of chess I win against
Stephen. Assume that I have a 30% chance of winning in any particular game
128 Chapter 10. Expected Values of Discrete Random Variables

against him, and we play 5 games, with outcomes assumed to be independent.

Find the expected value of X.

Exercise 10.10. Sports fans. The All-Star Pigeons lose 10 fans every time
they lose a game and gain 100 fans every time they win. Assume they have
a 30% chance of losing each game and that their performance in each game is
independent. What is the expected value of X, the number of fans they will
gain or lose over the next 3 games?

Exercise 10.11. Slot machine. People who play a slot machine win a prize
7% of the time. Let X denote the number of times the slot machine is used
until the next winner is found. What is the expected value of X?

10.3.2 Extensions

Exercise 10.12. Butterflies. Alice, Bob, and Charlotte are looking for but-
terflies. They look in three separate parts of a field, so that their probabilities
of success do not affect each other.

• Alice finds 1 butterfly with probability 17%, and otherwise does not find
one.

• Bob finds 1 butterfly with probability 25%, and otherwise does not find
one.

• Charlotte finds 1 butterfly with probability 45%, and otherwise does not
find one.

Let X be the number of butterflies that they find altogether. Find E(X).

Exercise 10.13. Dependence/independence among dice rolls. A student

rolls a die until the first “4” appears. Let X be the numbers of rolls required
until (and including) this first “4.” After this is completed, he begins rolling
again until he gets a “3.” Let Y be the number of rolls, after the first “4,” up to
(and including) the next “3.” E.g., if the sequence of rolls is 213662341261613
then X = 8 and Y = 7.

a. Find the expected value of X.

b. Find the expected value of Y .

Exercise 10.14. Two 4-sided dice. Consider some 4-sided dice. Roll two of
these dice. Let X denote the minimum of the two values that appear, and let
Y denote the maximum of the two values that appear.

a. Find the expected value of X.

b. Find the expected value of Y .
 10.3. Exercises 129

Exercise 10.15. Pick two cards. Pick two cards at random from a well-
shuffled deck of 52 cards (pick them simultaneously, i.e., grab two cards at
once—so they are not the same card!). There are 12 cards which are considered
face cards (4 Jacks, 4 Queens, 4 Kings). There are 4 cards with the value 10.
Let X be the number of face cards in your hand; let Y be the number of 10’s
in your hand.

a. Find the expected value of X.

b. Find the expected value of Y .

Exercise 10.16. Wastebasket basketball. Chris tries to throw a ball of

paper in the wastebasket behind his back (without looking). He estimates that
his chance of success each time, regardless of the outcome of the other attempts,
is 1/3. Let X be the number of attempts required. If he is not successful within
the first 5 attempts, then he quits, and he lets X = 6 in such a case. Find the
expected value of X.

Exercise 10.17. Claw. Jorge has three kids who spotted a Claw machine with
toys they want. In stores the toys costs $10 each, but each play on the claw
only costs $1. The probability of Jorge winning a game on the Claw is 0.12.
Should he use the Claw to get the toys (he needs one toy per kid), or does he
expect it to be cheaper to buy them in the stores?

Exercise 10.18. Super Breakfast Challenge. As in Exercise 8.5, the Super

Breakfast Challenge (SBC) consists of bacon, eggs, oatmeal, orange juice, milk,
and several other foods, and it costs $12.99 per person to order at a local
restaurant. It is known to be very difficult to consume the entire SBC. Only
10% of people are able to eat all of the SBC. The other 90% of people will be
unable to eat the whole SBC (it is too much food!).
A probability student hears about the SBC and goes to the local restaurant.
He observes the number of customers, X, that attempt to eat the SBC, until
the first success. So if there are 4 failures and then 1 success (i.e., the outcome
is (F, F, F, F, T )), then X = 5.
Find the expected value of X, i.e., the number of customers expected to try
the SBC until the first success.

Exercise 10.19. Sum of dice. Two fair dice are rolled. Let X be the sum of
the dice. What is the expected value of X?
Chapter 11

Expected Values of Sums of

Random Variables

I, at any rate, am convinced that [God] does not throw dice.

—Albert Einstein, from a letter to Max Born (December 4, 1926), as quoted
in The Born-Einstein Letters, translated by Irene Born (Walker, 1971)

While procrastinating before starting his probability homework, a student shuf-

fles a deck of cards and deals a randomly selected card to himself once a minute,
putting the used card back in the deck and reshuffling each time. He tells him-
self that when he gets the ace of spades, he will stop procrastinating and get to
work. How many minutes does he expect to procrastinate?

11.1 Introduction
One useful property of expected values of random variables is linearity, i.e.:
1. the expected value of a sum of random variables is equal to the sum of
the expected values, and
2. constants can be factored out of expected values.
Theorem 11.1. Expected value of the sum of discrete random
variables
If X1 , X2 , . . . , Xn are discrete random variables with finite expected values,
and a1 , a2 , . . . , an are constant numbers, then

E(a1 X1 + a2 X2 + · · · + an Xn ) = a1 E(X1 ) + a2 E(X2 ) + · · · + an E(Xn ).

The same property holds for a countably infinite collection of random variables.
This holds even when the Xj ’s are dependent.

130
 11.1. Introduction 131

Justification of Theorem 11.1:

Consider any finite collection of discrete random variables X1 , . . . , Xn with
finite expected values. Let a1 , . . . , an be any constants. Let S denote the
underlying sample space. Let X = a1 X1 +· · ·+an Xn . Then X(ω) = a1 X1 (ω)+
· · · + an Xn (ω) for each outcome ω ∈ S. So
X

E(a1 X1 + a2 X2 + · · · + an Xn ) = a1 X1 (ω) + · · · + an Xn (ω) P ({ω})
ω∈S
Xn X
= aj Xj (ω)P ({ω})
j=1 ω∈S
Xn
= aj E(Xj )
j=1

The same type of justification works if there are a countably infinite number of
random variables and constants.
We often use the property above in just the case where n = 2 and a2 = 1
and X2 is a constant, say “b.” So Theorem 11.1 simplifies to

Corollary 11.2. For any random variable X and any constants a and b, we
have
E(aX + b) = aE(X) + b.

This handy rule holds for any constants a and b.

Another popular use of the property in the box above is when the random
variables are indicators for some events. In the case where X is an indicator
random variable for event A, then E(X) is equal to P (A):

E(X) = 0P (X = 0) + 1P (X = 1) = 0P (Ac ) + 1P (A) = P (A).

This rule is so helpful that we highlight it.

Theorem 11.3. Expected value of an indicator random variable

If X is an indicator random variable for event A, i.e., X = 1 if event A occurs
and X = 0 otherwise, then the expected value of X is equal to the probability
that event A occurs:
E(X) = P (A).

Using Theorem 11.1 and Theorem 11.3 together, we have a very powerful
probability tool that allows us to quickly resolve most of the problems from
Chapter 10. We write a random variable as a sum of indicator random vari-
ables, and we compute the expected value of each indicator random variable by
computing the probability of each corresponding event.
132 Chapter 11. Expected Values of Sums of Random Variables

11.2 Examples

Example 11.4. Flip a coin three times. Let X denote the number of heads.

Let A1 , A2 , A3 be the events that the first, second, third flips (respectively) are
heads. Let X1 , X2 , X3 be indicators for A1 , A2 , A3 , i.e., X1 , X2 , X3 indicate
whether the first, second, third flips (respectively) are heads. So E(Xj ) =
P (Aj ) = 1/2 for each j. Also, X = X1 + X2 + X3 . Thus,

E(X) = E(X1 + X2 + X3 ) = E(X1 ) + E(X2 ) + E(X3 ) = 1/2 + 1/2 + 1/2 = 3/2.

Example 11.5. Consider a class consisting of eight students, who earn the
following scores on an exam: 75, 92, 88, 94, 89, 60, 83, 84. Let X be the
exam score of a randomly chosen student. Each student is equally likely to be
selected. What is the expected value of X?

Let Xj be the jth student’s score if the jth student is selected, and Xj = 0
otherwise. So, e.g., E(X1 ) = (75)(1/8) + (0)(7/8) = (75)(1/8). Then always
X = X1 + X2 + · · · + X8 , because one of the Xj ’s will be the selected student’s
score, and the other Xj ’s will be 0. Therefore,

E(X) = E(X1 + · · · + X8 )
= E(X1 ) + · · · + E(X8 )
= (75)(1/8) + (92)(1/8) + (88)(1/8) + (94)(1/8)
+ (89)(1/8) + (60)(1/8) + (83)(1/8) + (84)(1/8)
= 83.125.

Example 11.6. Roll a die three times. Let X be the number of times that a
6 appears. Find the expected value of X.

Let Aj be the event that the jth roll is a 6, so P (Aj ) = 1/6 for each j. Let Xj
be the indicator for Aj , i.e., Xj indicates whether the jth roll is a 6, so Xj = 1
if the jth roll is a 6, and Xj = 0 otherwise. Thus, E(Xj ) = P (Aj ) = 1/6 for
each j. Also, X = X1 + X2 + X3 . So

E(X) = E(X1 + X2 + X3 ) = E(X1 ) + E(X2 ) + E(X3 ) = 1/6 + 1/6 + 1/6 = 1/2.

More generally:
 11.2. Examples 133

Theorem 11.7. Expected value of the sum of identically distributed

discrete random variables
If X1 , X2 , . . . , Xn are identically distributed discrete random variables, then
the expected value of X1 + X2 + · · · + Xn is equal to one of the expected values
multiplied by n, i.e.,

E(X1 + X2 + · · · + Xn ) = nE(X1 ).

This is true regardless of whether the Xj ’s are dependent or inde-

pendent.

Example 11.8. A standard deck of 52 cards has 13 hearts. The cards in such
a deck are shuffled, and the top five cards are dealt to a player. What is the
expected number of hearts that the player receives?

Let X denote the number of hearts that the player receives. Let Aj be the event
that the jth card is a heart, so P (Aj ) = 1/4 for each j. Let Xj be the indicator
for Aj , i.e., Xj indicates whether the jth card is a heart, so Xj = 1 if the jth
card is a heart, and Xj = 0 otherwise. This yields E(Xj ) = P (Aj ) = 1/4 for
each j. (This is true because we are just focusing momentarily on the jth value;
we are not considering the values of the other cards or conditioning on them.)
Also, X = X1 + X2 + X3 + X4 + X5 . So

E(X) = (5)(1/4) = 5/4.

Example 11.9. Jim and his brother both like chocolate chip cookies best.
They have a jar of cookies with 5 chocolate chip cookies, 3 oatmeal cookies,
and 4 peanut butter cookies. They are each allowed to have 3 cookies. To be
fair, they agree to randomly select their cookies without peeking, and they each
must keep the cookies that they select. How many chocolate chip cookies does
Jim expect to get? (Notice that it does not matter whether Jim or his brother
selects the cookies first—the answer will be the same, either way.)

Let X be the number of chocolate chip cookies that Jim selects. Let Aj be the
event that the jth cookie that Jim selects is chocolate chip, so P (Aj ) = 5/12
for each j. Let Xj be the indicator for Aj , i.e., Xj indicates whether the jth
cookie that Jim selects is chocolate chip, so Xj = 1 if the jth cookie that Jim
selects is chocolate chip, and Xj = 0 otherwise. So E(Xj ) = P (Aj ) = 5/12 for
each j. Also, X = X1 + X2 + X3 , so

E(X) = (3)(5/12) = 5/4.

134 Chapter 11. Expected Values of Sums of Random Variables

Indicator random variables are useful even when they do not necessary have
the same distribution. Some creativity is required for seeing how to apply
them. This creativity can be developed with experience and with persistence.
It is often worthwhile to see if indicator random variables are relevant to ap-
ply in a problem. We present some different ways that they can be used in
the two examples below, to demonstrate that there are often several different
approaches—using different kinds of indicator random variables—to solve the
same problem.

Example 11.10. A student shuffles a deck of cards thoroughly (one time) and
then selects cards from the deck without replacement until the ace of spades
appears. How many cards does the student expect to draw?

Method #1. Let Aj be the event that j or more draws are required, so P (Aj ) =
1− j−1
52 , because Aj occurs if and only if the first j −1 draws are failures. Let Xj
be the indicator for Aj , i.e., Xj indicates whether j or more draws are needed,
so Xj = 1 if j or more draws are needed, and Xj = 0 otherwise. Therefore, we
have E(Xj ) = P (Aj ) = 1 − j−1 52 for each j. Also, X = X1 + X2 + · · · + X52
because (for instance) if exactly 3 draws are needed, then X1 , X2 , X3 are each
equal to 1, and the other Xj ’s are 0, so X = X1 + X2 + X3 , as desired. So

E(X) = E(X1 + X2 + · · · + X52 )

= E(X1 ) + E(X2 ) + · · · + E(X52 )
52  
X j−1
= 1−
52
j=1
P52
j=1 (j − 1)
= 52 −
52
but
52
X
(j − 1) = 0 + 1 + 2 + · · · + 51 = (52)(51)/2,
j=1

since
1 + 2 + · · · + n = (n)(n + 1)/2.

So
(52)(51)/2 51 104 − 51
E(X) = 52 − = 52 − = = 53/2.
52 2 2

Method #2. Assign one indicator random variable to each card in the deck
(except the ace of spades). The indictor is 1 if the corresponding card is drawn
before the ace of spades, or 0 otherwise. The number of cards drawn is X1 +
 11.2. Examples 135

· · · + X51 + 1, where the Xj ’s are the indicators. So the expected number of

cards to draw until the ace of spades appears is

E(X1 + · · · + X51 + 1) = E(X1 ) + · · · + E(X51 ) + E(1)

1 1
= + ··· + + 1
2 2
51
= +1
2
53
= .
2

Method #3. (This method is only included here for readers who know and enjoy
induction; if the reader does not know induction, it is OK to skip this method.)
Let Xj denote the number of flips needed to find a particular card (e.g., the ace
of spades) in a deck of j cards. Then we claim that E(Xj ) = (j + 1)/2 for all j,
and we prove it by induction. If j = 1, then the first card must be the desired
card, so E(X1 ) = (1 + 1)/2 = 1; this shows the base case. Now we handle
the inductive step: If E(Xj−1 ) = j/2, we show that E(Xj ) = (j + 1)/2. Flip
the first card (from the deck of j cards); the first card is the desired card with
probability 1/j. The first card is not the desired card with probability (j −1)/j,
and in such a case, this flip (that just occurred) was used, plus an additional
E(Xj−1 ) flips will be needed, because the problem is essentially starting again
with j − 1 cards. So
 
1 j−1 1 j−1 j j+1
E(Xj ) = + (1 + E(Xj−1 )) = + 1+ = .
j j j j 2 2

In particular, when starting with 52 cards, E(X52 ) = 53/2 flips are expected.

Example 11.11. A student draws cards from a standard deck of playing cards
until the ace of spades appears for the first time. After every unsuccessful draw,
the student replaces the card and shuffles the deck thoroughly before selecting
a new card. How many cards does the student expect to draw until the ace of
spades appears?

Method #1. Let Aj be the event that j or more draws are required, so P (Aj ) =
(51/52)j−1 , because Aj occurs if and only if the first j − 1 draws are failures.
Let Xj be the indicator for Aj , i.e., Xj indicates whether j or more draws are
needed, so Xj = 1 if j or more draws are needed, and Xj = 0 otherwise. So
E(Xj ) = P (Aj ) = (51/52)j−1 for each j. Also, X = X1 + X2 + X3 + · · · because
(for instance) if exactly 3 draws are needed, then X1 , X2 , X3 are each equal to
1, and the other Xj ’s are 0, so X = X1 + X2 + X3 , as desired. Thus, we have

E(X) = E(X1 + X2 + · · ·) = E(X1 ) + E(X2 ) + · · · = 1 + 51/52 + (51/52)2 + · · · .

136 Chapter 11. Expected Values of Sums of Random Variables

It follows that
∞
X 1 1
E(X) = (51/52)j = 51 = 1/52 = 52.
j=0
1 − 52

So the student expects to draw 52 cards to see the ace of spades for the first
time.
Method #2. (Does not use indicators!) Let X be the number of flips that are
necessary. With probability 1/52, the ace of spades appears on the first draw.
With probability 51/52, the ace of spades does not appear on the first draw,
and the problem essentially starts over again: in this case, the original flip was
used, plus E(X) more flips will be needed. So we have

1 51 1 51 51 51
E(X) = (1) + (1 + E(X)) = + + E(X) = 1 + E(X).
52 52 52 52 52 52

Subtracting 51
52 E(X) from both sides, we get 1
52 E(X) = 1, and we conclude that
E(X) = 52.

Example 11.12. As one last example, let X1 , X2 , X3 be independent random

variables that correspond to three consecutive flips of an unbiased coin. Then
define X = X1 + X2 + X3 . We studied X in Example 11.4. In particular, we
know that E(X) = 3/2.
In contrast, consider the random variable Y = 3X1 . We also have E(Y ) =
E(3X1 ) = 3E(X1 ) = 3/2. So X and Y have the same expected value.
On the other hand, X and Y are two very different types of random variables.
For instance, X can take on four different values (0, 1, 2, or 3), and the mass
of X is:

pX (0) = 1/8, pX (1) = 3/8, pX (2) = 3/8, pX (3) = 1/8.

On the other hand, Y can only take on the values 0 or 3, and the mass of Y is:

pY (0) = 1/2, pY (3) = 1/2.

Therefore, although X1 + X2 + X3 and 3X1 have the same expected value, they
have very different distributions. (More generally, X1 + X2 + · · · + Xn and nX1
have different distributions in most similar examples.)
 11.3. Exercises 137

11.3 Exercises
11.3.1 Practice
Exercise 11.1. Graduation. As in Exercise 10.1, Jack and Jill are indepen-
dently struggling to pass their last (one) class required for graduation. Jack
needs to pass Calculus III, but he only has probability 0.30 of passing. Jill
needs to pass Advanced Pharmaceuticals, but she only has probability 0.46 of
passing. They work independently.
Following the notation of Exercise 10.1, let X be 0, 1, or 2, if (respectively)
neither, one, or both of them graduate. Let Y indicate if Jack graduates, so
Y = 1 if Jack graduates, and Y = 0 otherwise. Let Z indicate if Jill graduates,
so Z = 1 if Jill graduates, and Z = 0 otherwise. Notice that X = Y + Z always.
Now find E(X) using E(Y ) and E(Z).
Exercise 11.2. Japanese pan noodles. As in Exercise 10.2, four students
order noodles at a certain local restaurant. Their orders are placed indepen-
dently. Each student is known to prefer Japanese pan noodles 40% of the time.
How many of them do we expect to order Japanese pan noodles?
Let A1 , A2 , A3 , A4 be the events that (respectively) the first, second, third,
or fourth person orders Japanese pan noodles. Let X1 , X2 , X3 , X4 be indicator
random variables for (respectively) A1 , A2 , A3 , A4 . Justify your answer using
the values of the E(Xj )’s.
Exercise 11.3. Yellow ducks. Three hundred little plastic yellow ducks are
dumped in a pond; one of them contains a prize stamped on the bottom.
Leonardo examines each duck until he discovers the prize. He discards each
duck without a prize after he checks it, so that he never needs to check a duck
more than one time. How many ducks does he expect to check until he discovers
the prize?
Exercise 11.4. Super Breakfast Challenge. As in Exercises 8.5 and 10.18,
the Super Breakfast Challenge (SBC) consists of bacon, eggs, oatmeal, orange
juice, milk, and several other foods, and it costs $12.99 per person to order at
a local restaurant. It is known to be very difficult to consume the entire SBC.
Only 10% of people are able to eat all of the SBC. The other 90% of people will
be unable to eat the whole SBC (it is too much food!).
A probability student hears about the SBC and goes to the local restaurant.
He observes the number of customers, X, that attempt to eat the SBC, until
the first success. So if there are 4 failures and then 1 success (i.e., the outcome
is (F, F, F, F, T )), then X = 5.
Find the expected value of X, i.e., the number of customers expected to try
the SBC until the first success.
Exercise 11.5. Weather. A weather forecasting program gets the daily pre-
dictions right about 87% of the time. Assuming each day is independent, what
138 Chapter 11. Expected Values of Sums of Random Variables

is the expected number of days that will pass until the program gets the forecast
wrong?

Exercise 11.6. Random movie picks. A man sits down to watch a movie.
40% of his options are action films, 35% are comedies, and 25% are drama. He
wants to watch a comedy and starts picking movies at random from his box.
If previous picks are put back in the box after marking the choice, how many
movies is he expected to have to pick up until he finds a comedy?

Exercise 11.7. Dice. Robbie rolls a die until he gets a 6. What is the expected
number of rolls?

Exercise 11.8. Free throws. While shooting free throws, you have a 37%
chance of making a basket. How many shots do you expect to have to take until
making a basket?

Exercise 11.9. Computer preferences. You randomly and independently

ask students on campus whether they prefer Mac or PC. Each person prefers
Mac 72% of the time. What is the expected number of students you ask until
the first person prefers PC?

Exercise 11.10. Butterflies. Alice, Bob, and Charlotte are looking for but-
terflies. They look in three separate parts of a field, so that their probabilities
of success do not affect each other.

• Alice finds 1 butterfly with probability 17%, and otherwise does not find
one.

• Bob finds 1 butterfly with probability 25%, and otherwise does not find
one.

• Charlotte finds 1 butterfly with probability 45%, and otherwise does not
find one.

Let X be the number of butterflies that they find altogether.

Write X as the sum of three indicator random variables, X1 , X2 , X3 that
indicate whether Alice, Bob, Charlotte (respectively) found a butterfly. Then
X = X1 + X2 + X3 . Find the expected value of X by finding the expected value
of the sum of the indicator random variables.

Exercise 11.11. Dependence/independence among dice rolls. A student

rolls a die until the first “4” appears. Let X be the numbers of rolls required
until (and including) this first “4.” After this is completed, he begins rolling
again until he gets a “3.” Let Y be the number of rolls, after the first “4,” up to
(and including) the next “3.” E.g., if the sequence of rolls is 213662341261613
then X = 8 and Y = 7.

a. Let Aj be the event containing all outcomes in which “j or more rolls”

are required to get the first “4.” Let Xj indicate whether Aj occurs. Then
 11.3. Exercises 139

X = X1 + X2 + X3 + · · · . Find the expected value of X by finding the expected

value of the sum of the indicator random variables.
b. Let Bj be the event containing all outcomes in which “j or more rolls”
are required, after the first “4,” until he gets a “3.” Let Yj indicate whether Bj
occurs. Then Y = Y1 + Y2 + Y3 + · · · . Find the expected value of Y by finding
the expected value of the sum of the indicator random variables.
Exercise 11.12. Wastebasket basketball. Chris tries to throw a ball of
paper in the wastebasket behind his back (without looking). He estimates that
his chance of success each time, regardless of the outcome of the other attempts,
is 1/3. Let X be the number of attempts required. If he is not successful within
the first 5 attempts, then he quits, and he lets X = 6 in such a case.
Let Aj be the event containing all outcomes in which “j or more attempts”
are required to get the basket. Let Xj indicate whether Aj occurs. Then
X = X1 + X2 + X3 + X4 + X5 + X6 . Find the expected value of X by finding
the expected value of the sum of the indicator random variables.
Exercise 11.13. Two 4-sided dice. Consider some 4-sided dice. Roll two of
these dice. Let X denote the minimum of the two values that appear, and let
Y denote the maximum of the two values that appear.
Let Aj be the event containing all outcomes in which the minimum of the
two dice is “j or greater.” Let Xj indicate whether Aj occurs. Then X =
X1 + X2 + X3 + X4 . Find the expected value of X by finding the expected value
of the sum of the indicator random variables.
Exercise 11.14. Pick two cards. Pick two cards at random from a well-
shuffled deck of 52 cards (pick them simultaneously, i.e., grab two cards at
once—so they are not the same card!). There are 12 cards which are considered
face cards (4 Jacks, 4 Queens, 4 Kings). There are 4 cards with the value 10.
Let X be the number of face cards in your hand; let Y be the number of 10’s
in your hand.
a. Before looking at the cards, put one in your left hand and one in your
right hand. Let X1 and X2 indicate, respectively, whether the cards in your left
and right hands (respectively) are face cards. Then X = X1 + X2 . Find the
expected value of X by finding the expected value of the sum of the indicator
random variables.
b. Before looking at the cards, put one in your left hand and one in your
right hand. Let Y1 and Y2 indicate, respectively, whether the cards in your left
and right hands (respectively) are 10’s. Then Y = Y1 + Y2 . Find the expected
value of Y by finding the expected value of the sum of the indicator random
variables.

11.3.2 Extensions
Exercise 11.15. Oreos. A box of Double Stuf Oreos has a defect. One of
the Oreos is only single-stuffed. There are 6 Oreos in the pack. What is the
140 Chapter 11. Expected Values of Sums of Random Variables

expected number of Oreos you need to check in the pack until the defective one
is found?

Exercise 11.16. Flipping coins. Flip a coin until the second head comes up.
Let X be the number of flips needed to get the second head. What is the E(X)?

Exercise 11.17. Waiting for favorite song. Michael puts his iTunes on
shuffle mode where songs are not allowed to be replayed. He has 2,781 songs
saved in iTunes, and exactly one of these is his favorite.

a. How many songs is he expected to have to listen to until his very favorite
song comes up?
b. Now suppose that he allows songs to be repeated. How many songs does
he expect to listen to until his very favorite song comes up?

Exercise 11.18. Crayons. A little girl has a 96-pack of crayons. She picks up
crayons, at random, to check the color, and she leaves them in a separate pile
after inspecting the color. She pulls crayons out of the pack until she gets the
sea foam green crayon.

a. What is the expected number of crayons she will check until she finds sea
foam green?
b. Assume instead that she puts each crayon back in the box before ran-
domly drawing the next crayon. Now how many crayons does she expect to
check until she finds sea foam green?

Exercise 11.19. Lectures and labs. Consider a group of 120 students. They
are split into two lectures (60 students each). They are also split into six labs
(20 students each). All assignments of students to lectures and labs are equally
likely. How many classmates does Barry expect to be in both his lecture and
his lab too?

Exercise 11.20. Onions. Eighty-seven percent of people tear up when cutting

an onion. People are selected randomly and independently to cut up an onion.
What is the expected number of people who cut up an onion until you find the
third person who does it without crying?

Exercise 11.21. Exiting the elevator. Eight people enter an elevator in the
parking garage below a building. Each person chooses her exit independently of
the other people. The building has floors 1 through 10. What is the expected
number of stops that the elevator makes?

Exercise 11.22. Bears. A total of 30 bears—consisting of 10 red bears, 10

yellow bears, and 10 blue bears—are randomly arranged into 10 groups of 3
bears each. Compute the expected number of groups in which all 3 bears are
different colors.
 11.3. Exercises 141

Exercise 11.23. Rope tying. Consider n pieces of rope. Each piece is colored
blue at one end and red at the other. The blue ends of the ropes are randomly
paired with the red ends of the ropes and tied together, one-to-one, i.e., all n!
such possible methods of joining the ropes this way are equally likely. Let X
be the number of loops that result from this method. Find E(X).
Chapter 12

Variance of Discrete Random

Variables

The probable is what usually happens.

—Aristotle

Two different teachers are teaching probability classes. Before registering, stu-
dents want to know how past students’ grades were distributed for each instruc-
tor. The students have a report of the mean and standard deviation for Exam 1
from last semester. The means are similar, but one instructor has a standard
deviation that is much larger than the other instructor’s. Is it good or bad for
exam scores to be widely spread out from the mean? Why?

12.1 Introduction
The expected value of a function of a random variable is a sum over all values
that a function of a random variable takes; as in Chapter 10, these the values are
taken in proportion to the fraction of the time that the function of the random
variable takes each value.
Definition 12.1. Expected value of a function of a discrete random
variable
If g is any function, and X is a discrete random variable that takes on values
x1 , x2 , . . . , xn , then the expected value of g(X) is
n
X n
X
E(g(X)) = g(xj )P (X = xj ) = g(xj )pX (xj ).
j=1 j=1

If the number of values X takes on is countably infinite, x1 , x2 , . . ., then the

same expressions hold, but the sums are taken over all j ∈ N.

142
 12.1. Introduction 143

Example 12.2. Flip a coin three times. Let X denote the number of heads.
Find E(X 2 ), E(X 3 ), E(eX ), E(2X − 5), and E(17X 3 + 5X 2 − 7X + 22).

Since X = 0 with probability 1/8, X = 1 with probability 3/8, X = 2 with

probability 3/8, and X = 3 with probability 1/8, then the expected value of X 2
is Warning: Note
E(X 2 ) = (02 )(1/8) + (12 )(3/8) + (22 )(3/8) + (32 )(1/8) = 3. E(X 2 ) 6= (E(X))2 .
This is one of the
The expected value of X 3 is most common
mistakes made when
E(X 3 ) = (03 )(1/8) + (13 )(3/8) + (23 )(3/8) + (33 )(1/8) = 27/4. first learning these
concepts.
The expected value of eX is

1 3 3 1
E(eX ) = (e0 )(1/8) + (e1 )(3/8) + (e2 )(3/8) + (e3 )(1/8) = + e + e2 + e3 .
8 8 8 8
The expected value of 2X − 5 is
1 3
E(2X − 5) = ((2)(0) − 5) + ((2)(1) − 5)
8 8
3 1
+ ((2)(2) − 5) + ((2)(3) − 5)
8 8
= −2.

As we learned in Chapter 11, the expected value is linear. Also, by treating the
“5” as a random variable that is always equal to 5, we can use the fact from
Chapter 10 that E(X) = 3/2, and we can simply calculate (as a shorter method
than above),

E(2X − 5) = 2E(X) − E(5) = 2(3/2) − 5 = −2.

We can use the fact that E(X) = 3/2 and E(X 2 ) = 3 and E(X 3 ) = 27/4 to
compute, for instance, E(17X 3 + 5X 2 − 7X + 22), as follows:

E(17X 3 + 5X 2 − 7X + 22) = 17E(X 3 ) + 5E(X 2 ) − 7E(X) + 22

= 17(27/4) + 5(3) − 7(3/2) + 22
= 565/4
= 141.25.

As in Chapter 10, we can also decompose things further, based upon the specific
outcome of the underlying random phenomenon:
144 Chapter 12. Variance of Discrete Random Variables

Definition 12.3. Expected value of a function of a discrete random

variable (always gives same result as the original definition)
Consider a random phenomenon with possible outcomes ω1 , ω2 , . . . , ωn . Sup-
pose that the outcome ωj causes random variable X to take on value xj . Then
g(X) has expected value
n
X
E(g(X)) = g(xj )P ({ωj }).
j=1

If there are a countably infinite number of such ω’s, we just adjust to sums
taken over all j ∈ N.
In the setting of Example 12.2, this means we can just decompose the ran-
dom phenomenon into all eight possible outcomes:
ω1 = T T T,
ω2 = HT T, ω3 = T HT, ω4 = T T H,
ω5 = HHT, ω6 = HT H, ω7 = T HH,
ω8 = HHH,
and we get the same result:
E(X 2 ) = (02 )(1/8) + (12 )(1/8) + (12 )(1/8) + (12 )(1/8)
+ (22 )(1/8) + (22 )(1/8) + (22 )(1/8) + (32 )(1/8) = 3.
(In an analogous way, we had two ways of computing expected values in Chap-
ter 10, which basically amounted to whether we grouped the 2nd, 3rd, and 4th
terms above, which all yield the same value of g(X), and we could also group
the 5th, 6th, and 7th terms above, which again all yield a common value of
g(X). So the two boxed methods above are, as in Chapter 10, just two different
ways of grouping things.)
To summarize: When computing the expected value of a function g of a
random variable X, we can take one of the two following approaches:
1. a sum over all of the possible values of g(X), each weighted by the prob-
ability of X taking on value x, and thus g(X) taking on value g(x), or
2. a sum over all of the possible outcomes, taking the value of g(X) from
such an outcome, weighted by the probability of that outcome

Example 12.4. Consider a class consisting of eight students, who earn the
following scores on an exam: 75, 92, 88, 94, 89, 60, 83, 84. Let X be the
exam score of a randomly chosen student. Each student is equally likely to be
selected. How much do we expect each student’s score to exceed 60, i.e., what
is the expected value of X − 60?
 12.2. Variance 145

There are eight possible outcomes in this random phenomenon, each of which
has probability 1/8. So, following the second definition of expected value, we
obtain

E(X − 60) = (75 − 60)(1/8) + (92 − 60)(1/8) + (88 − 60)(1/8)

+ (94 − 60)(1/8) + (89 − 60)(1/8) + (60 − 60)(1/8)
+ (83 − 60)(1/8) + (84 − 60)(1/8)
= 23.125.

Of course, since we already computed E(X) = 83.125 in Chapter 10, we could

have obtained the same answer using linearity of expected values:

E(X − 60) = E(X) − 60 = 83.125 − 60 = 23.125.

12.2 Variance
One of the reasons that expected values of functions of random variables are
so useful is the concept of variance. Since the variance of a random variable
is positive or zero (as we will see below), we can always compute the square
root of the variance; this quantity is referred to as the standard deviation of
the random variable. The standard deviation measures the spread of a random
variable around the mean, i.e., it describes how widely the mass of a random
variable is spread out. A larger variance (or standard deviation) signifies that
the values of the random variable tend to be spread relatively far from the
expected value. A smaller variance (or standard deviation) signifies that the
values of the random variable tend to be relatively closer to the expected value,
i.e., the values of the random variable do not tend to be very spread out.
We use µX = E(X) and σX 2 = Var(X) to simplify the notation.

Definition 12.5. Variance

The variance σX
2 of a random variable X, with expected value µ = E(X), is
X

Var(X) = E((X − µX )2 ).

One property of the variance is that the variance of a random variable is

always nonnegative, i.e., the variance is always 0 or a positive number. The
reason for this is that (X − µX )2 ≥ 0 always (i.e., the square of a number
cannot be negative). Thus, the variance is the expected value of a nonnegative
number, so the variance itself is nonnegative too.

Theorem 12.6. Variance is always nonnegative

For any random variable X, the variance of X is nonnegative:

Var(X) = E((X − µX )2 ) ≥ 0.
146 Chapter 12. Variance of Discrete Random Variables

We notice that (X − µX )2 = X 2 − 2µX X + µ2X . Using the linearity of

expected values, this gives us

Var(X) = E((X − µX )2 ) = E(X 2 − 2µX X + µ2X ) = E(X 2 ) − 2µX E(X) + µ2X .

Since µX = E(X), this simplifies to

Var(X) = E(X 2 ) − (E(X))2 .

So we have an equivalent formulation of the variance, which is often used in

practice to actually compute the variance.
Remark 12.7. Useful formulation for computing the variance of a
random variable
The variance of a random variable X is

Var(X) = E(X 2 ) − (E(X))2 .

Definition 12.8. Standard Deviation

For a random variable X with variance Var(X), the standard deviation of X
is p
σX = Var(X).

In Example 12.9, we see random variables X and Y that have the same
expected values but different variances.

Example 12.9. Recall Examples 10.5, 11.5, and 12.4, in which the expected
value of a randomly chosen student’s grade in a course is

E(X) = (75)(1/8) + (92)(1/8) + (88)(1/8) + (94)(1/8)

+ (89)(1/8) + (60)(1/8) + (83)(1/8) + (84)(1/8)
= 83.125.

Consider eight other students in a different course, who earn the following scores
on an exam: 82, 82, 81, 84, 89, 80, 83, 84. In this case, let Y be the exam score
of a randomly chosen student. Find Var(X) and Var(Y ).
The expected value of Y is

E(Y ) = (82)(1/8) + (82)(1/8) + (81)(1/8) + (84)(1/8)

+ (89)(1/8) + (80)(1/8) + (83)(1/8) + (84)(1/8)
= 83.125,

so that the average value of the students in the two scenarios is the same, i.e.,

E(X) = E(Y ).
 12.2. Variance 147

On the other hand, the variances of the grades from the two groups of students
are different. We have

E(X 2 ) = (752 )(1/8) + (922 )(1/8) + (882 )(1/8) + (942 )(1/8)

+ (892 )(1/8) + (602 )(1/8) + (832 )(1/8) + (842 )(1/8)
= 7016.875.

and

E(Y 2 ) = (822 )(1/8) + (822 )(1/8) + (812 )(1/8) + (842 )(1/8)

+ (892 )(1/8) + (802 )(1/8) + (832 )(1/8) + (842 )(1/8)
= 6916.375,

So the variance of X is

Var(X) = 7016.875 − (83.125)2 = 107.109375,

and the variance of Y is

Var(Y ) = 6916.375 − (83.125)2 = 6.609375.

The standard deviations of X and Y are, respectively,

√ √
σX = 107.109375 = 10.349366 and σY = 6.609375 = 2.570870.

The variance of Y is much smaller than the variance of X. This signifies the
fact that the values of Y tend to be much closer to the expected value of Y ,
as compared to the values of X, which tend to be very spread out far from the
expected value of X. All of the values of Y are found in the 80’s, while the
values of X are spread from 60 all the way to 94.

Example 12.10. Flip a coin three times. Let X denote the number of heads.
Find the variance of X.

We already computed E(X) = 3/2 and E(X 2 ) = 3. Thus

Var(X) = E(X 2 ) − (E(X))2 = 3 − (3/2)2 = 3/4.

√
The standard deviation is σX = 3/2.

Example 12.11. Roll three dice. Let X denote the number of 6’s that appear
on the dice altogether. Find the variance of X.
148 Chapter 12. Variance of Discrete Random Variables

In Example 10.6, we already computed the mass of X, which is the following:

P (X = 0) = (5/6)3
P (X = 1) = (3)(1/6)(5/6)2
P (X = 2) = (3)(1/6)2 (5/6)
P (X = 3) = (1/6)3

So the expected value of X is

E(X) = (0)(5/6)3 + (1)(3)(1/6)(5/6)2 + (2)(3)(1/6)2 (5/6) + (3)(1/6)3 = 1/2.

The expected value of X 2 is

E(X 2 ) = (02 )(5/6)3 +(12 )(3)(1/6)(5/6)2 +(22 )(3)(1/6)2 (5/6)+(32 )(1/6)3 = 2/3.

So the variance of X is:

Var(X) = E(X 2 ) − (E(X))2 = 2/3 − (1/2)2 = 5/12.

The standard deviation is σX = 5/12 = 0.645497.

p

Example 12.12. Suppose that you are playing a game where you roll the die
three times and count up the number of 6’s you get in those rolls. You have to
pay $3 to play, but you win $4 for every 6 that comes up. So your winnings are
4X − 3 dollars, where X is the total number of 6’s that appear.

Your expected winnings are

E(4X − 3) = 4E(X) − 3 = 4(1/2) − 3 = −1 dollars.

The variance of your winnings is

Var(4X − 3) = E((4X − 3)2 ) − (E(4X − 3))2

= E(16X 2 − 24X + 9) − (−1)2
= 16E(X 2 ) − 24E(X) + 9 − 1
= 16(2/3) − 24(1/2) + 8
= 32/3 − 12 + 8
= 20/3
= 6.6667 dollars2 .

The standard deviation of your winnings is

σX = Var(4X − 3) = 20/3 = 2.581989 dollars.

p p
 12.2. Variance 149

It is worthwhile to note that the dealer wins −4X + 3 during this game. So
dealer’s expected winnings are

E(−4X + 3) = 1 dollar,

and the variance of the dealer’s winnings is

Var(−4X + 3) = E((−4X + 3)2 ) − (E(−4X + 3))2

= E(16X 2 − 24X + 9) − (1)2
= 16E(X 2 ) − 24E(X) + 9 − 1
= 16(2/3) − 24(1/2) + 8
= 32/3 − 12 + 8
= 20/3
= 6.6667 dollars2 .

The standard deviation of the dealer’s winnings is

Var(−4X + 3) = 20/3 = 2.581989 dollars.

p p
σX =

Importantly, we conclude that your expected winnings are the same as the
dealer’s expected winnings, but with the sign reversed (i.e., −1 and 1, respec-
tively). We also conclude that your winnings and the dealer’s winnings have
the same variance and standard deviation. This is true more generally. The
difference in the signs will disappear as the constant terms do not affect the
variance, and the signs of the coefficients are lost during the squaring process
in the variance.

Later in this chapter, we discuss the general remark:

Remark 12.13. For any random variable X and any constants a and b, we
have
Var(aX + b) = a2 Var(X).

This rule will assist in computations like the previous example.

Example 12.14. A student shuffles a deck of cards thoroughly (one time) and
then selects cards from the deck without replacement until the ace of spades
appears. What is the variance of the number of cards that the student draws
until the ace of spades appears?
150 Chapter 12. Variance of Discrete Random Variables

Let X be the number of cards needed until the ace of spades appears. Then,
as we observed in Example 10.8, we have P (X = j) = 1/52 for each j. So the
expected value of X is:

E(X) = (1)P (X = 1) + (2)P (X = 2) + (3)P (X = 3) + · · · + (52)P (X = 52)

= (1)(1/52) + (2)(1/52) + (3)(1/52) + · · · + (52)(1/52)
= (1 + 2 + · · · + 52)(1/52)
= 53/2

The expected value of X 2 is

E(X 2 ) = (12 )P (X = 1) + (22 )P (X = 2) + · · · + (522 )P (X = 52)

= (12 )(1/52) + (22 )(1/52) + · · · + (522 )(1/52)
= (12 + 22 + · · · + 522 )(1/52)
 
(52)(52 + 1)((2)(52) + 1) 1
=
6 52
= 1855/2
(n)(n+1)(2n+1)
We use the fact that 12 + 22 + · · · + n2 = 6 . So the variance of X is

Var(X) = 1855/2 − (53/2)2 = 901/4 = 225.25.

√
The standard deviation of X is σX = 225.25 = 15.01.

Example 12.15. A standard deck of 52 cards has 13 hearts. The cards in such
a deck are shuffled, and the top five cards are dealt to a player. What is the
variance of the number of hearts that the player receives?
Let X denote the number of hearts that the player receives. In Example 10.10,
we computed the mass of X:
(39)(38)(37)(36)(35) 2109
P (X = 0) = =
(52)(51)(50)(49)(48) 9520
(13)(39)(38)(37)(36) 27417
P (X = 1) = 5 =
(52)(51)(50)(49)(48) 66640
(13)(12)(39)(38)(37) 9139
P (X = 2) =10 =
(52)(51)(50)(49)(48) 33320
(13)(12)(11)(39)(38) 2717
P (X = 3) =10 =
(52)(51)(50)(49)(48) 33320
(13)(12)(11)(10)(39) 143
P (X = 4) = 5 =
(52)(51)(50)(49)(48) 13328
(13)(12)(11)(10)(9) 33
P (X = 5) = =
(52)(51)(50)(49)(48) 66640
 12.2. Variance 151

So we get
2109 27417 9139 2717 143 33
E(X) = (0) + (1) + (2) + (3) + (4) + (5) ,
9520 66640 33320 33320 13328 66640
or more simply, E(X) = 5/4; and
2109 27417 9139 2717 143 33
E(X 2 ) = (02 ) +(12 ) +(22 ) +(32 ) +(42 ) +(52 ) ,
9520 66640 33320 33320 13328 66640
so E(X 2 ) = 165/68. Thus, the variance of X is
Var(X) = 165/68 − (5/4)2 = 235/272.

Example 12.16. Jim and his brother both like chocolate chip cookies best.
They have a jar of cookies with 5 chocolate chip cookies, 3 oatmeal cookies,
and 4 peanut butter cookies. They are each allowed to have 3 cookies. To be
fair, they agree to randomly select their cookies without peeking, and they each
must keep the cookies that they select. What is the variance of the number of
chocolate chip cookies that Jim gets? (Notice that it does not matter whether
Jim or his brother selects the cookies first—the answer will be the same, either
way.)
Let X denote the number of chocolate chip cookies that Jim selects. Notice
that there are (12)(11)(10) = 1320 equally likely outcomes for Jim. We already
computed the mass of X in Example 10.11:
(7)(6)(5) 7
P (X = 0) = =
(12)(11)(10) 44
(3)(5)(7)(6) 21
P (X = 1) = =
(12)(11)(10) 44
(3)(7)(5)(4) 7
P (X = 2) = =
(12)(11)(10) 22
(5)(4)(3) 1
P (X = 3) = = .
(12)(11)(10) 22
So, as we already computed in Example 10.11, the expected value of X is
7 21 7 1
E(X) = (0) + (1) + (2) + (3) = 5/4.
44 44 22 22
The expected value of X 2 is
7 21 7 1
E(X 2 ) = (02 ) + (12 ) + (22 ) + (32 ) = 95/44.
44 44 22 22
So the variance of X is
Var(X) = 95/44 − (5/4)2 = 105/176.
152 Chapter 12. Variance of Discrete Random Variables

12.3 Five Friendly Facts with Independence

Caution: We strongly emphasize the fact that the random variables
must be independent, in order to use the formulas in this section!!

Theorem 12.17. The expected value of the product of the functions

of two independent random variables equals the product of the ex-
pected values of the functions of the two random variables
If X and Y are independent random variables, and g and h are any two
functions, then
E(g(X)h(Y )) = E(g(X))E(h(Y )).

To see this, we compute as follows (the second equality holds because X and Y
are independent):
XX
E(g(X)h(Y )) = g(x)h(y)P (X = x and Y = y)
x y
XX
= g(x)h(y)P (X = x)P (Y = y)
x y
X X
= g(x)P (X = x) h(y)P (Y = y)
x y

= E(g(X))E(h(Y ))

Using identity functions g(X) = X and h(Y ) = Y , the following nice fact
follows immediately:

Corollary 12.18. The expected value of the product of two indepen-

dent random variables equals the product of the expected values of
the two random variables
If X and Y are independent random variables, then

E(XY ) = E(X)E(Y ).

When we are working with random variables, the variance of the sum equals
the sum of the variances. When we pull constants outside of the variance, they
get squared.

Theorem 12.19. If X1 , . . . , Xn are independent random variables, and

a1 , . . . , an are constants, then

Var(a1 X1 + · · · + an Xn ) = a21 Var(X1 ) + · · · + a2n Var(Xn ).

 12.3. Five Friendly Facts with Independence 153

To see this, we just compute directly:

n
!
X
Var(a1 X1 + · · · + an Xn ) = Var ai Xi
i=1
n
!2 ! n
!!2
X X
=E ai Xi − E ai Xi
i=1 i=1
n X
n
! n
!2
X X
=E ai aj Xi Xj − ai E(Xi )
i=1 j=1 i=1
n
XX n n
XX n
= ai aj E(Xi Xj ) − ai aj E(Xi )E(Xj ),
i=1 j=1 i=1 j=1

but E(Xi Xj ) = E(Xi )E(Xj ) for i 6= j, since Xi and Xj are independent, so we

get

n
X
Var(a1 X1 + · · · + an Xn ) = ai ai E(Xi2 )
i=1
n X
X
+ ai aj E(Xi )E(Xj )
i=1 j6=i
Xn X n
− ai aj E(Xi )E(Xj ).
i=1 j=1

Now we can cancel all terms with i 6= j, and we get

n
X n
X n
X
Var(a1 X1 +· · ·+an Xn ) = ai ai E(Xi2 )− ai ai E(Xi )E(Xi ) = a2i Var(Xi ).
i=1 i=1 i=1

Now use n = 2, with a1 = a and a2 = 1 and X1 = X and X2 = b (a constant).

Then Var(X2 ) = E(X22 ) − (E(X2 ))2 = E(b2 ) − (E(b))2 = b2 − b2 = 0, so this
gives us immediately:

Corollary 12.20. If X is any random variable, and if a and b are any con-
stants, then
Var(aX + b) = a2 Var(X).
Thus σaX+b = Var(aX + b) = a2 Var(X) = |a| Var(X).
p p p

In the case where ai = 1 for all i, Theorem 12.19 has a simpler form, given
in Corollary 12.21. This corollary is very handy. (Just be sure that the random
variables are independent when using these results; otherwise, these ideas cannot
be applied.)
154 Chapter 12. Variance of Discrete Random Variables

Corollary 12.21. The variance of the sum of independent random

variables equals the sum of the variances
If X1 , . . . , Xn are independent random variables, then

Var(X1 + · · · + Xn ) = Var(X1 ) + · · · + Var(Xn ).

If the Xj ’s are independent and have the same distribution, then

Var(X1 + · · · + Xn ) = n Var(X1 ).

(Remember that the independence was not needed for the analogous method
of computing the expected value of a sum of random variables.)

Corollary 12.21 allows us to recompute several of the variances from this

chapter in a more direct way.

Example 12.22. As in Example 12.10, flip a coin three times. Let X denote
the number of heads. Find the variance of X.

Let Xj indicate whether the jth flip is a head. Then X = X1 + X2 + X3 . Also

Var(Xj ) = E(Xj2 ) − (E(Xj ))2 = 1/2 − (1/2)2 = 1/4.

Since the Xj ’s are independent, then

Var(X) = 3 Var(X1 ) = 3(1/4) = 3/4.

Example 12.23. As in Example 12.11, roll three dice. Let X denote the
number of 6’s that appear on the dice altogether. Find the variance of X.

Let Xj indicate whether the jth roll is a 6. Then X = X1 + X2 + X3 . Also

Var(Xj ) = E(Xj2 ) − (E(Xj ))2 = 1/6 − (1/6)2 = 5/36.

Since the Xj ’s are independent, then

Var(X) = 3 Var(X1 ) = 3(5/36) = 5/12.

It should be evident from these revised versions of the examples that it is

often possible to compute Var(X) very quickly when X can be written as a sum
of independent random variables, i.e., where X = X1 + · · · + Xn and the Xj ’s
are independent.
 12.4. Exercises 155

12.4 Exercises
12.4.1 Practice
Exercise 12.1. Variance of an indicator. Suppose event A occurs with
probability p, and X is an indicator for A, i.e., X = 1 if A occurs, or X = 0
otherwise. We already know E(X) = p. Find Var(X).

Exercise 12.2. Concert tickets. At a concert on campus, 20% of people

purchase Zone A tickets, for $47 each. The other 80% of people purchase Zone
B tickets, for $38 each. If five people are selected at random, what is the variance
of the revenue from these five ticket sales?

Exercise 12.3. Japanese pan noodles. As in Exercises 10.2 and 11.2, four
students order noodles at a certain local restaurant. Their orders are placed
independently. Each student is known to prefer Japanese pan noodles 40% of
the time. Let X be the number of students who order Japanese pan noodles.
What is the variance of X?

Exercise 12.4. Dice. Roll two dice; let X denote the maximum of the two
values that appear.

a. Find E(X).
b. Find E(X 2 ).
c. Find Var(X).

Exercise 12.5. Yellow ducks. Three hundred little plastic yellow ducks are
dumped in a pond; one of them contains a prize stamped on the bottom.
Leonardo examines each duck until he discovers the prize. He discards each
duck without a prize after he checks it, so that he never needs to check a duck
more than one time. Find the variance of the number of ducks he checks until
he discovers the prize.

Exercise 12.6. Dice. Susan rolls a die 4 times. Let X be the number of 1s
that appear.

a. Find E(X).
b. Find E(X 2 ).
c. Find Var(X).

Exercise 12.7. Mystery mass. Suppose that the mass of X is

pX (1) = 0.12
pX (3) = 0.37
pX (27) = 0.42
pX (31) = 0.09
156 Chapter 12. Variance of Discrete Random Variables

a. Find E(X).
b. Find E(X 2 ).
c. Find Var(X).
d. Find E(X 2 − 4X).

Exercise 12.8. Smoke alarms. Let X be the number of smoke alarms in an

apartment. From government data, you find the mass of X is

pX (0) = 0.05
pX (1) = 0.15
pX (2) = 0.35
pX (3) = 0.30
pX (4) = 0.10
pX (5) = 0.05

Assume each alarm requires one battery per year, and batteries cost $2.35.

a. What is the expected cost in batteries for smoke alarms for a randomly
selected apartment?
b. What is the variance of the cost?

Exercise 12.9. Waiting for favorite song. Michael plays a random song on
his iPod. He has 2,781 songs, but only one favorite song. Let X be the number
of songs he has to play on shuffle (songs can be played more than once) in order
to hear his favorite song.

a. Find E(X).
b. Find E(X 2 ).
c. Find Var(X).

Exercise 12.10. Spinners. Two three-partitioned spinners are spun. Each of

the three parts (labeled 1, 2, and 3) have an equally likelihood of occurring. Let
X be the maximum of two spins.

a. Find E(X).
b. Find E(X 2 ).
c. Find Var(X).

Exercise 12.11. Raffle tickets. You purchase 8 raffle tickets at the county
fair. Each ticket costs $5. A ticket is worth $100 with probability 1/400, but is
worthless otherwise. What is the expected value of your purchase? Be sure to
take into account the original purchase price of the tickets.
 12.4. Exercises 157

Exercise 12.12. SAT scores. Suppose that, among a certain group of stu-
dents, SAT scores have a mean value of 1026 and a standard deviation of 209.
Let X denote a randomly chosen student’s score. What is E(X 2 )?

Exercise 12.13. Butterflies. Alice, Bob, and Charlotte are looking for but-
terflies. They look in three separate parts of a field, so that their probabilities
of success do not affect each other.

• Alice finds 1 butterfly with probability 17%, and otherwise does not find
one.

• Bob finds 1 butterfly with probability 25%, and otherwise does not find
one.

• Charlotte finds 1 butterfly with probability 45%, and otherwise does not
find one.

Let X be the number of butterflies that they find altogether. Find the variance
of X.

Exercise 12.14. Appetizers. At a restaurant that sells appetizers:

• 8% of the appetizers cost $1 each,

• 20% of the appetizers cost $2 each,

• 32% of the appetizers cost $3 each,

• 40% of the appetizers cost $4 each.

An appetizer is chosen at random, and X is its price. Each appetizer has 7%

sales tax. So Y = 1.07X is the amount paid on the bill (in dollars). Find the
variance of Y .

Exercise 12.15. Two 4-sided dice. Consider some 4-sided dice. Roll two of
these dice. Let X denote the minimum of the two values that appear, and let
Y denote the maximum of the two values that appear.
Find the variance of X.
(Caution: If Xj is an indicator of whether the minimum of the two dice is “j
or greater”—as in the previous homework—then X = X1 + X2 + X3 + X4 , but
the Xj ’s are dependent.)

Exercise 12.16. Pick two cards. Pick two cards at random from a well-
shuffled deck of 52 cards (pick them simultaneously, i.e., grab two cards at
once—so they are not the same card!). There are 12 cards which are considered
face cards (4 Jacks, 4 Queens, 4 Kings). There are 4 cards with the value 10.
Let X be the number of face cards in your hand; let Y be the number of 10’s
in your hand.
158 Chapter 12. Variance of Discrete Random Variables

a. Find the variance of X.

b. Find the variance of Y .

Exercise 12.17. Die game. In each round of a game, you earn a dollar if a
die shows 1 or 2, you lose a dollar if a die shows 5 or 6, and you neither earn nor
lose anything if a die shows 3 or 4. Let Xj be +1, −1, or 0 according to your
outcome on the jth round. Let X = X1 + X2 + · · · + X10 . Find the variance
of X.

Exercise 12.18. Sleeping schedule. A student gets at least 8 hours of sleep

45% of the nights; the sleeping schedule is independent from night to night.
Let X1 , X2 , X3 , X4 indicate whether the student gets at least 8 hours of sleep
during the next four nights respectively. Let X = X1 + X2 + X3 + X4 . Find
the variance of X.

12.4.2 Extensions
Exercise 12.19. Magic. A magician wants to do a trick where he tries to
guess the value of the card that an audience member has selected (the suit of
the card is not taken into account). He has six decks of cards so he gives one
deck to each of six audience members. Let X denote the number of cards that
he guesses correctly, when each audience member selects one card at random.

a. Find E(X).
b. Find E(X 2 ).
c. Find Var(X).
d. The magician gets paid $500 for performing the show, plus $20 for each
card he guesses correctly. What is the average amount of money that he gets
paid?
e. As in (d), what is the variance of the amount that he gets paid?

Exercise 12.20. Cookies. As in Example 12.16, Jim and his brother both
like chocolate chip cookies best. They have a jar of cookies with 5 chocolate
chip cookies, 3 oatmeal cookies, and 4 peanut butter cookies. They are each
allowed to have 3 cookies. To be fair, they agree to randomly select their cookies
without peeking, and they each must keep the cookies that they select. Suppose
that the cookies are worth $1.20 for each chocolate chip cookie, $0.75 for each
oatmeal cookie, and $1.30 for each peanut butter cookie. find the expected
value and variance of the total value of the cookies Jim gets.

Exercise 12.21. Mystery mass. Consider a random variable X with mass

x 1 2 3 4 5
pX (x) 0.1 0.25 0.5 0.1 0.05
 12.4. Exercises 159

a. Find the expected value of 1/X.

b. Find the standard deviation of 1/X.
Exercise 12.22. Yahtzee. In a dice game, a “Yahtzee” is a result in which all
five dice within a round have the same value. To simplify this problem, assume
that the five dice are just rolled one time per round. (In the actual Yahtzee
game, dice can be re-rolled.) Let X be the number of times that a player gets
a Yahtzee in three separate rounds. Find the variance of X
Exercise 12.23. Die and coin. Roll a die and flip a coin. Let Y be the value
of the die. Let Z = 1 if the coin shows a head, and Z = 0 otherwise. Let
X = Y + Z. Find the variance of X.
Exercise 12.24. Golden ticket. A student was at work at the county am-
phitheater, and was given the task of cleaning 1500 seats. To make the job more
interesting, his boss hid a golden ticket somewhere in the seats. The ticket is
equally likely to be in any of the seats. Let X be the number of seats cleaned
until the ticket is found. Calculate the variance of X.
Exercise 12.25. Dice. Two dice are rolled. Let X be the value of the first die
minus the second. Find Var(X).

12.4.3 Advanced
Exercise 12.26. Mystery mass. Consider the mass

0.15 for x = −1,


0.35 for x = 0,

pX (x) =


0.25 for x = 1,
0.25 for x = 2.


a. Find the average value of X.

b. Find the average value of 4X.
c. Find the average value of X 4 .
d. Find the average value of 4X .
e. Find the variance of X.
f. Find the variance of 4X.
g. Find the variance of X 4 .
h. Find the variance of 4X .
Exercise 12.27. Mystery mass. Consider the mass



0.8 for x = −4,
0.1 for x = −2,

pX (x) =


0.07 for x = 0,
0.03 for x = 2.

160 Chapter 12. Variance of Discrete Random Variables

a. Find E(X).
b. Find E(2/X).
c. Find 2/E(X).
d. Find E(|X|).
e. Find Var(X).
f. Find Var(|X|).

Exercise 12.28. If X and Y have joint density fX,Y (x, y) = 8xy on the triangle
0 ≤ y ≤ x ≤ 1, find E(XY ).

Exercise 12.29. Coins. Consider a pile that has 9 coins altogether. Exactly 3
of the coins are dimes (worth 10 cents each), and the other 6 coins are pennies
(worth 1 cent each). Emily picks up 4 of the coins blindly (all possibilities are
equally likely) without replacement. Find the expected value (in cents) of the
amount she picks up.

Exercise 12.30. Specify a discrete random variable that has the property
E(Xj ) = 0 for all odd integers j, and E(Xj ) = 1 for all even integers j.
Chapter 13

Review of Discrete Random

Variables

13.1 Summary of Discrete Random Variables

What can you do with a probability mass function?

• Find individual probabilities, P (X = x).

• Graph the probability mass function.

• Calculate expected value (average), variance, and standard deviation for

X and functions of X.

• Calculate the cumulative distribution function by adding up each individ-

ual probability step-wise.

• Remember that it is important to check that you have a valid probability

mass function first.

How do you know if your probability mass function is valid?

• The probability mass function table must include every individual value
that the random variable can take.

• Each individual probability must be between 0 and 1.

0 ≤ P (X = x) ≤ 1

• The probabilities—summed over all the possible values of the random

variable—must sum to 1. (We’re 100% sure that we will have an outcome
of some sort!) X
P (X = x) = 1
x

161
162 Chapter 13. Review of Discrete Random Variables

What can you do with a cumulative distribution function?

• Find cumulative probabilities, FX (x) = P (X ≤ x).

• Graph the cumulative distribution function.

• Calculate the probability mass function or individual probabilities by sub-

tracting two consecutive steps on the cumulative distribution function.

How do you find the cumulative distribution function (CDF)?

Just add up each probability mass function step one at a time so that you
can make a list of the FX (x) = P (X ≤ x) accumulated probabilities.
How do you get the mean/(weighted) average/expected value?
X
E(X) = x pX (x)
x
X
E(g(X)) = g(x) pX (x)
x

How do you get the variance and standard deviation?

Var(X) = E(X 2 ) − (E(X))2

p
σX = Var(X)

What do you do if you have a linear combination of X? (Especially

good for money problems.) Here a and b are constants.

E(aX + b) = aE(X) + b
Var(aX + b) = a2 Var(X)
p p
σaX+b = a2 Var(X) = |a| Var(X)

What do you do if you have a sum of random variables and need an

expected value? (This works whether the variables are independent
or dependent.)
X n  X n
E Xi = E(Xi )
i=1 i=1

What if all of these random variables have the same distribution?

(Again, this works whether the variables are independent or depen-
dent.)
Xn 
E Xi = nE(X1 )
i=1
 13.2. Exercises 163

What do you do if you need the variance of n independent random

variables?
n n
!
X X
Var Xi = Var(X)
i=1 i=1
v
u n
uX
σPni=1 Xi =t Var(X)
i=1

What if all of these random variables are independent and also have
the same distribution?
n
!
X
Var Xi = n Var(X)
i=1
p
σPni=1 Xi = n Var(X)

13.2 Exercises
Exercise 13.1. Horse race. There are 5 horses running in a race, and you
own the one named Rosie. Let X be the order Rosie finishes the race (1st, 2nd,
etc.). Below is a table for the probability for each x-value.

x 1 2 3 4 5
P(X = x) 0.4 0.2 0.1 ? 0.02

a. Graph the probability mass function.

b. What place is Rosie expected to finish?
c. What is the standard deviation of the place Rosie will finish?
d. Find the cumulative distribution function.
e. Graph the cumulative distribution function.
Exercise 13.2. Prizes for Rosie. Using the horse race information from
Exercise 13.1, if Rosie’s owner pays $500 for entering but wins $1000 for first
place, $750 for second place, $500 for third place, and nothing for placing any
lower,
a. What are the expected winnings (or losses) for Rosie (i.e., including the
entrance fee)?
b. What is the standard deviation of the winnings (or losses) for Rosie (i.e.,
including the entrance fee)?
Exercise 13.3. Password. A password is to be created at random by selecting
4 characters (with replacement) from the set {A, B, . . . , Z, 0, 1, . . . , 9}. Let X
be the number of letters which are chosen for the password.
164 Chapter 13. Review of Discrete Random Variables

a. Find the probability mass function for the number of letters which are
chosen.
b. Graph the probability mass function.
c. What is the mean number of letters which are chosen?
d. What is the standard deviation of the number of letters which are chosen?
e. Find the cumulative distribution function.
f. Graph the cumulative distribution function.

Exercise 13.4. More passwords. Assume that each password is created

according to Exercise 13.3. If passwords are independent and 10 passwords are
created:

a. What is the expected total number of letters which are used?

b. What is the standard deviation of the total number of letters which are
used?
c. What is the probability that at least 2 of these passwords are identical?
d. If 100 passwords are created, what is the probability that at least 2 of
the passwords are identical?

Exercise 13.5. Math team. There are 20 students competing to be on the

Math Team. Thirteen of them specialize in algebra, 4 of them specialize in
geometry, and 3 of them specialize in calculus. Only 5 spots on the team are
available, and team members will be chosen at random, regardless of specialty.
Let X be the number of calculus specialists who make the team.

a. Find the probability mass function for the number of math team members
who are calculus specialists. (Hint: Can 5 calculus specialists make the team?
Why or why not?)
b. Find the cumulative distribution function for the number of math team
members who are calculus specialists.
c. What is the expected number of calculus specialists who make the team?
d. What is the variance of the number of calculus specialists who make the
team?

Exercise 13.6. Grading. Joan will grade four statistics projects, selected at
random from a large stack, this evening. From past experience, Joan thinks
that 30% of the projects will be “A” quality projects. Let X be the number of
“A” projects she grades tonight. (Assume that all the projects are independent
and that Joan is completely fair and objective with her grading of each one.)

a. Find the probability mass function.

b. Is this a valid probability mass function? Justify your answer.
 13.2. Exercises 165

c. Find the cumulative distribution function.

d. What is the probability that Joan grades at least two “A” projects
tonight?
e. What is the probability that Joan grades at most two “A” projects
tonight?
f. What is the probability that Joan grades exactly four “A” projects given
that she grades at least two “A” projects tonight?
g. What is the probability that Joan grades at most four “A” projects given
that she grades strictly more than two “A” projects tonight?
Exercise 13.7. Mystery CDF. For the following cumulative distribution
function for a discrete random variable X,

0
 if x < −3
0.03 if −3 ≤ x < 1




F (x) = 0.20 if 1 ≤ x < 2.5
0.76 if 2.5 ≤ x < 7





if 7 ≤ x

1


a. Find the probability mass function. Show why it is a valid probability

mass function.
b. What is the probability that X will be more than 2?
c. Given that X is at most 2, what is the probability that it is positive?
d. Given that X is positive, what is the probability that it will be at least
2?
e. Find E(X 3 ).
Exercise 13.8. True/false for masses. Which of the following statements
are true about probability mass functions? Explain why each statement is true
or false.
a. P (X = x) can be 1 for a particular value of x.
b. The graph of P (X = x) must be non-decreasing.
c. P (X = x) can be negative for a particular value of x.
d. P (X = x) can be 1 for more than one particular value of x.
e. P (X = x) can be zero for a particular value of x.
f. P (X = x) can be zero for infinitely many values of x.
g. P (X = x) can be strictly positive for infinitely many values of x.
Exercise 13.9. True/false for CDFs. Which of the following statements are
true about cumulative distribution functions? Explain why each statement is
true or false.
166 Chapter 13. Review of Discrete Random Variables

a. FX (x) can be 1 for a particular value of x.

b. The graph of FX (x) must be non-decreasing.
c. FX (x) can be negative for a particular value of x.
d. FX (x) can be 1 for more than one particular value of x.
e. The graph of FX (x) must be continuous with no jumps.
f. FX (x) can be zero for a particular value of x.
g. FX (x) can be zero for infinitely many values of x.
h. FX (x) can be strictly positive for infinitely many values of x.

Exercise 13.10. True/false for discrete random variables. Which of the

following statements are true for discrete random variables? Explain why each
statement is true or false.

a. The mean can be negative.

b. The mean can be zero.
c. The standard deviation can be negative.
d. The standard deviation can be zero.
e. The standard deviation can be greater than the mean.
f. The variance can be negative.
g. The variance can be zero.
h. The expected value of X 2 is always positive.
i. The expected value of X 2 is always smaller than the square of the mean.

Exercise 13.11. Trick coin game. Suppose an acquaintance has a trick coin
that comes up a head 75% of the time. He wants to play a game where he tosses
the coin 3 times, and for each time the coin comes up a head, he will pay you
$2. He wants to charge you a fee to play the game. If you are willing to play
the game only if the game is fair, how much money would you be willing to pay
(overall) to play the game?

Exercise 13.12. Hidden bones. Chet the dog likes to dig holes in the yard.
He has buried bones in some of the holes, but others he dug just for fun. He
covers them back up with a little mound of dirt when he is done digging. The
yard has 20 holes that have been filled up, and 8 of them contain bones. Chet
decides to randomly search 4 of the holes (without replacement) in hopes of
finding some of his bones. Let X be the number of bones he finds.

a. Give the probability mass function for the number of bones Chet finds.
b. Give the cumulative distribution function for the number of bones he
finds.
 13.2. Exercises 167

c. What is the probability he finds at most 3 bones?

d. What is the probability he finds at least 2 bones?
e. What is the probability he finds exactly 2 bones given that he finds at
most 3 bones?
Exercise 13.13. Mystery CDF. Using the cumulative distribution function
for the discrete random variable X given below:



0 if x < 30,
0.2 if 30 ≤ x < 45,

FX (x) =


0.7 if 45 ≤ x < 60,
if 60 ≤ x.

1

a. What is the probability that X = 45?

b. What is the probability mass function for X?
c. Is your answer to part b a valid probability mass function? Justify your
answer.
d. What is the probability that X is between 25 and 50?
e. What is the probability that X is 70?
f. What is the probability that X is less than 70?
g. What is the probability that X is greater than 70?
h. What is the probability that X is between 60 and 70 (inclusive)?
i. What is the expected value of X?
j. What is the standard deviation of X?
Exercise 13.14. Realtor sales, part 1. Daily sales records for a realtor’s
office show that the realtor will sell 0, 1, or 2 homes in a week with the proba-
bilities listed below:

X 0 1 2
P (X = x) 0.75 0.20 0.05
a. Is this a valid probability mass function? Why or why not?
b. What is the expected number of homes the realtor will sell in a week?
c. What is the standard deviation in the number of homes the realtor will
sell in a week?
d. What is the expected number of homes the realtor will sell in a 12-week
quarter?
e. What is the standard deviation in the number of homes the realtor will
sell in a 12-week quarter? (Assume the weeks are independent.)
168 Chapter 13. Review of Discrete Random Variables

f. Do you think that the assumption about the weeks being independent is
a good assumption? Why or why not? Does independence affect your answer
to part d? To part e?
Exercise 13.15. Realtor sales, part 2. Using the realtor sales information
from Exercise 13.14, the realtor makes a typical commission of $4000 on each
home sold, but he needs to pay an assistant $900 each week.
a. What is the realtor’s expected net profit each week?
b. What is the standard deviation in the realtor’s weekly net profit?
c. During a 12-week quarter, what is the realtor’s expected net profit?
d. During a 12-week quarter, what is the standard deviation in the realtor’s
net profit?
Exercise 13.16. Basketball score. Erika, a basketball player, scores an av-
erage of 10 points per game with a standard deviation of 2.6 points. Let X
denote the number of points that she scores in a game.
a. What is the expected value of X 2 ?
b. What is the variance of 24X − 3?
e. What is the expected value and standard deviation of her total points
over 30 games, in which her performances are independent?
Exercise 13.17. Phone calls. A person makes an average of 3.12 phone calls
a day with a standard deviation of 1.77 phone calls. Let X denote the number
of calls made per day.
a. What is the expected value of X 2 ?
b. What is the expected value of −12X + 10?
c. Find the variance of −12X + 10.
Exercise 13.18. Rental car, part 1. The number of days a driver needs
a rental car after having an accident, X, is a discrete random variable with
probability mass function:
(
8−x
28 for x = 1, 2, . . . , 7,
P (X = x) =
0 otherwise
a. Make a table which shows the values for X and P (X = x).
b. Graph the probability mass function for X.
c. What is the cumulative distribution function for X?
d. Graph the cumulative distribution function for X.
e. What is the expected number of days that a driver will need a rental car?
f. What is the standard deviation of days that a driver will need a rental
car?
 13.2. Exercises 169

Exercise 13.19. Rental car, part 2. An insurance company pays $50 a

day for rental car costs for up to 5 days after an accident and $20 a day after
that until the owner’s car has been repaired. Using this information and the
probability mass function from Exercise 13.18,

a. What is the probability mass function for the insurance payment?

b. What is the expected insurance payment?
c. What is the standard deviation for the insurance payment?

Exercise 13.20. Loss on claims, part 1. An insurance company models the

amount N of loss (in thousands) if a homeowner has at least one small claim.
There is a constant k such that the mass of N is
k
p(N = n) = for n = 1, 2, 3, 4
2n + 1
with no other possible choices for N . Find the value of k to make this a prob-
ability mass function.

Exercise 13.21. Loss on claims, part 2. The probability mass function you
found in Exercise 13.20 was only for homeowners who had at least one small
claim. The insurance company believes the probability that a homeowner will
have at least one small claim is 0.08. What is the probability mass function for
amount of loss for all of their policy holders?

Exercise 13.22. Loss on claims, part 3. The insurance company in Exer-

cise 13.20 and 13.21 has a $2000 deductible. What is the expected net premium
for this policy? (A customer is responsible for any claims up to the $2000
deductible, but a premium is the amount the insurance company pays to the
policy-holder for any costs above that $2000 limit.)

Exercise 13.23. Bed tax. A community institutes a new 5% “bed tax” on

all their hotel rooms to raise money for the local park system. Previously, the
average nightly price of a hotel room was $75 with a standard deviation of $20.
What is the new average price and standard deviation?

Exercise 13.24. Tour group. A tour group is planning to pick a town to stay
in when they get tired of driving. Since they don’t know exactly where they will
be staying, they don’t know exactly how much the hotel rooms will cost. They
will need 12 rooms. They know that individual hotel rooms along their route
have an average nightly price of $75 with a standard deviation of $20. What is
their expected cost and standard deviation for the 12 rooms?

Exercise 13.25. Overbooking. An airline intentionally overbooks its flight

because it knows that that probability an individual passenger will not show
up is 0.05 (assume passengers are independent). The plane can hold 80 people,
and the airline sold 81 tickets for the flight.
170 Chapter 13. Review of Discrete Random Variables

a. What is the probability that there will not be enough seats for the
passengers who show up to fly?
b. Do you think it is reasonable to assume the passengers are independent?
Why or why not?
c. The airline wants to predict their profit from this flight. Each passenger
paid $200 for his or her ticket, and if someone does not show up, he or she is
refunded only $150 of the ticket purchase price. What are the expected value
and the standard deviation of the airline’s income from this flight?
d. Using the story from part c, what is a customer’s expected cost and the
standard deviation of the cost?

Exercise 13.26. Class sizes. A probability mass function for the class sizes
at a small college is given in the table below (assume that these are the only
allowed class sizes):

Class size Probability

10 0.05
20 0.15
30 0.45
40 0.10
50 0.10
60 0.10
70 0.05
a. What is the expected value of the class size for a class chosen at random?
b. What is the standard deviation of the class size for a class chosen at
random?
c. What percentage of classes are within one standard deviation of the mean
class size?
d. What is the cumulative distribution function for class size?

e. What is the probability that a class has between 30 and 60 students?

f. Given that a class has at least 30 students, what is the probability that
it has strictly less than 60 students?
g. Given that a class has at least 30 students, what is the probability that
it has at least 60 students?
h. Given that a class has at least 30 students, what is the probability that
it has strictly more than 60 students?
i. Given that a class has at least 30 students, what is the probability that
it has at most 60 students?
Part III

Named Discrete Random

Variables

In the chapters we have covered so far on discrete random variables, there are
some common themes, and perhaps you have even begun to recognize different
“types” of random variables. In the chapters that follow, we will be even more
precise about the kinds of discrete random variables and what they have in
common. We will describe scenarios that usually lend themselves to a particular
type of random variable, and we will systematically describe formulas to help
make your life easier. In Chapters 14 through 20, we’ll introduce you to these
different “named” distributions. This will culminate with Chapter 21, where we
review some strategies for comparing and contrasting all of these distributions.
This summary gives precise techniques to help you pick the correct distribution
for your situation.
There are many other discrete distributions than the seven we cover here,
but these are the most common ones.
By the end of this part of the book, you should be able to:

1. Distinguish between Bernoulli, Binomial, Geometric, Negative Binomial,

Poisson, Hypergeometric, and Discrete Uniform distributions when read-
ing a story.

2. Identify the variable and the parameters in a story, and state in plain
English what the variable and parameters mean.

3. Use the formulas for the mass, expected value, and variance to answer
questions and find probabilities.

Binomial coefficients (“choose” parentheses), fac-

Math skills you will need: P
torials, summation notation ( ), ex for Chapter 18, and ln x for one particular
type of problem in Chapter 14.3.

171
172 Part III. Named Discrete Random Variables

Additional resources: Computer programs (such as Excel, Maple, Mathe-

matica, Matlab, Minitab, R, SPSS, etc.) and calculators may be used to assist
in the calculations.
Note about the problems in several of the sections: We realize the stories
for some of the problems in these sections are fairly similar. We want you to
notice important differences in wording that set these distributions apart. The
context of the situation yields clues as to which random variable is relevant to
the scenario. Throughout our discussion, we repeatedly emphasize how to build
experience in recognizing which random variable is the right one for you to use.
Chapter 14

Bernoulli Random Variables

A pinch of probability is worth a pound of perhaps.

—“Such a Phrase as Drifts Through Dreams,” reprinted in Lanterns &
Lances by James Thurber (Harper, 1961)

Sometimes the most simple things are some of the most useful as well. Is a
randomly chosen song a rock song or not? What is the probability of winning
one game? Does a cereal box have the preferred prize inside?

14.1 Introduction
In our discussion of discrete random variables, we have already seen many types
of Bernoulli random variables. Some outcomes are considered a “success” and
some outcomes are considered a “failure.” A Bernoulli random variable is
always 1 or 0 to indicate these respective possibilities of success or failure. A
Bernoulli random variable is often called an indicator random variable (or
simply an indicator), because the 1 indicates success, while the 0 indicates fail-
ure. Using the terminology of Chapter 3, a Bernoulli random variable indicates
the success or failure of a trial.
The Bernoulli distribution is the simplest named distribution. It is a building
block for the Binomial, Geometric, and Negative Binomial distributions. We’ll
use the word “success” to mean “the thing that we’re looking for,” regardless
of whether it is good or bad. E.g., when biologists study DNA to find which
markers are associated with a rare disease, a “success” could occur when they
identify such a marker, even though it is bad news for the relevant patient.

“Thumbs up” for success “Thumbs down” for failure

173
 174 Chapter 14. Bernoulli Random Variables

Success probability The notation for a Bernoulli distribution looks like:

is p, so the failure
probability is X ∼ Bernoulli(p),
q = 1 − p.
where p denotes the probability of success.

Bernoulli random variables

The common thread: One success or one failure? For instance, you ask
one person a yes/no question, where “yes” is a “success” and “no” is a “failure.”
Other examples include: flipping a coin (heads vs tails); seeing if one part of
something is defective (yes or no); classifying a pet’s size (large or small); etc.
Things to look for: One trial, which is either a success or a failure.
The variable:
(
0 if the outcome is a failure, or
X=
1 if the outcome is a success.

The parameter:

p = the probability that the outcome is a success

q = 1 − p = the probability that the outcome is a failure

Mass:

P (X = 1) = p
P (X = 0) = q

Expected value formula:

E(X) = p
Variance formula:
Var(X) = pq

14.2 Examples

Example 14.1. Suppose 95% of people put peanut butter on first when making
a peanut butter and jelly sandwich. You select a person at random to ask
whether he or she puts the peanut butter on first.

a. Why is this a Bernoulli distribution situation? What is the parameter?

Single trial with success/failure outcome, p = 0.95
b. What do we consider a “success” for this trial?
 14.2. Examples 175

Finding a person who puts peanut butter on first.

c. What is the probability the person you select puts jelly on first?

q = 1 − p = 0.05

d. What is the expected number of people who put the peanut butter on
first if you ask one person?

E(X) = p = 0.95

e. What is the standard deviation for the number of people who put the
peanut butter on first if you ask one person?

Var(X) = pq = (0.95)(0.05) = 0.0475

√ √
standard deviation = σX = pq = (0.95)(0.05) = 0.0475 = 0.2179
p

f. What does the mass look like for this story?

pX (x)
1
(1, 0.95)
0.8

0.6

0.4

0.2
(0, 0.05)
x
1

g. What does the CDF look like for this story?

FX (x)
1
(1, 1)
0.8

0.6

0.4

0.2
(0, 0.05)
x
1
176 Chapter 14. Bernoulli Random Variables

Example 14.2. Thirty-eight percent of the songs on a student’s music player

are rock songs. A student chooses a song at random, with all songs equally
likely to be chosen. Let X indicate whether the selected song is a rock song.

a. Find the expected number and variance of X.

E(X) = p = 0.38
Var(X) = pq = (0.38)(0.62) = 0.2356

b. How do the mass and CDF of X look?

pX (x) FX (x)
1 1
(1, 1)
0.8 (0, 0.62) 0.8 (0, 0.62)
0.6 0.6
0.4 0.4
(1, 0.38)
0.2 0.2
x x
0.2 0.4 0.6 0.8 1 1.2 −0.2 0.2 0.4 0.6 0.8 1 1.2

Left: Mass of an indicator, and Right: CDF of an indicator.

Remark 14.3. Expected value and variance of a Bernoulli random

variable If X is Bernoulli, then the expected value of the Bernoulli random
variable is
E(X) = (0)(q) + (1)(p) = p
and E(X 2 ), which is also called the second moment of X, is

E(X 2 ) = (02 )(q) + (12 )(p) = p.

Recall that Var(X) = E((X − E(X))2 ) = E(X 2 ) − (E(X))2 , so the variance

of X is

Var(X) = E(X 2 ) − (E(X))2 = p − p2 = p(1 − p) = pq.

Example 14.4. At a certain point in a die game, a roll of 5 or 6 is needed

to win, and any other roll at that point will cause the player to lose. At that
point, the player earns 15 dollars if she wins, but forfeits 9 dollars if she loses.
What is the player’s expected gain or loss at this point?
 14.2. Examples 177

Let X be a Bernoulli random variable that indicates whether the player wins
or loses. The probability of winning is 2/6, so

E(X) = 2/6 = 1/3.

If X = 1, then the winnings are 15; if X = 0, then the winnings are −9. So the
expected winnings are

E(winnings) = (−9)(q) + (15)(p) = (−9)(2/3) + (15)(1/3) = −1

The formal way of writing this is to let f (X) be the gain or loss, depending on
X, so

E(f (X)) = f (0)P (X = 0) + f (1)P (X = 1) = (−9)(2/3) + (15)(1/3) = −1

So the player expects to lose 1 dollar at this point in the game.

Example 14.5. A cereal company puts a Star Wars toy watch in each of its
boxes as a sales promotion. Twenty percent of the cereal boxes contain a watch
with Obi Wan Kenobi on it. You are a huge Obi Wan fan, so you decide to buy
1 box of the cereal in hopes that you will find an Obi Wan watch. You convince
your brother and sister to each buy a box too.

a. How many Obi Wan watches does your family expect to obtain?
Let X = 1 if your box contains an Obi Wan watch, or X = 0 otherwise.
Similarly, let Y be an indicator random variable that indicates whether your
brother gets one; let Z be an analogous indicator for your sister.
Note: we do not even need to know whether our selections are independent
to solve this problem about expected values! We will not rely on independence
in our solution, because independence is not needed when taking the expected
value of the sum of random variables.
The X, Y, Z are indicator variables describing (respectively) whether you,
your brother, and your sister get Obi Wan watches. So the total number of
watches obtained is X + Y + Z. So the expected number of watches obtained is

E(X + Y + Z) = E(X) + E(Y ) + E(Z) = 0.2 + 0.2 + 0.2 = 0.6.

b. What is the variance of the number of Obi Wan watches that your family
obtains?
In this situation, we do need to know about how X, Y, Z depend on each
other. If the number of cereal boxes is sufficiently large that the values of
X, Y, Z do not affect each other, we might be able to assume that X, Y, Z
are independent. (We will consider such situations—with independence—much
178 Chapter 14. Bernoulli Random Variables

more in the next chapter. To handle situations where the X, Y, Z are dependent,
we will have to wait until Chapter 19.) If X, Y, Z are independent, then we can
add their variances:
Var(X + Y + Z) = Var(X) + Var(Y ) + Var(Z)
= (0.2)(0.8) + (0.2)(0.8) + (0.2)(0.8)
= (3)(0.16)
= 0.48.

14.3 Exercises
14.3.1 Practice
Exercise 14.1. Call from home. One out of every eight calls to your house
is from a family member. You will record whether the next call is from a family
member.
a. What do you consider a “success” in this story? What is its probability?
b. What do you consider a “failure” in this story? What is its probability?
c. Why is this a Bernoulli situation? What is the parameter?
d. Define in words what X is in terms of this story. What values can X
take?
e. What is the probability that the next time the phone rings, it will be
from a family member?
f. If the phone calls are independent, what is the probability the 3rd call
today will be from a family member?
g. What is the mean of X?
h. What is the standard deviation of X?
Exercise 14.2. Dice. You roll a fair, six-sided die as part of a game. If you
roll a 5, you will win the game.
a. What do you consider a “success” in this story? What is its probability?
b. What do you consider a “failure” in this story? What is its probability?
c. Why is this a Bernoulli situation? What is the parameter?
d. Define in words what X is in terms of this story. What values can X
take?
e. Your friend will pay you $4 if you win the game. You owe your friend $1
if you lose the game. What are your expected winnings?
f. What is the variance of your expected winnings?
 14.3. Exercises 179

Exercise 14.3. Homework. Hui has a class of 300 students, and only 6 have
done their homework assignment due today. He calls on a student at random to
put a problem on the board to check whether he or she has done the assignment.
a. What does Hui consider a “success” in this story? What is the probability?
b. What does Hui consider a “failure” in this story? What is the probability?
c. Why is this a Bernoulli situation? What is the parameter?
d. Define X in terms of this story. What values can X take?
e. What is the probability that the student Hui selects is one who has done
the assignment?
f. If the students are independent, what is the probability that the 3rd
student Hui checks will have done the assignment?
Exercise 14.4. Blu-rays. Suppose that 1% of Blu-ray discs produced by a
company are defective. You buy one of these discs and check to see if it is
defective.
a. What do you consider a “success” in this story? What is the probability?
b. What do you consider a “failure” in this story? What is the probability?
c. Why is this a Bernoulli situation? What is the parameter?
d. Define X in terms of this story. What values can X take?
e. Draw the labeled graph of the mass for this story.
f. Draw the labeled graph of the CDF for this story.
Exercise 14.5. Shoes. Anne and Jane have shoes spread throughout the dorm
room. Anne has 15 pairs of shoes; twenty percent of her shoe collection consists
of sandals. Jane has 40 pairs of shoes; ten percent of her shoe collection consists
of sandals.
a. A shoe is picked at random from the dorm room belonging to Anne and
Jane; what is the probability that it is a sandal?
b. If a randomly chosen shoe is chosen from the room (with all shoes equally
likely to be chosen), what is the probability that it belongs to Anne?
c. If a randomly chosen shoe is chosen from the room (with all shoes equally
likely to be chosen), and upon examination this shoe is seen to be a sandal, what
is the probability that it belongs to Anne?
Exercise 14.6. Movie date. Chris and Juanita always go to the movies
on Friday night. Before meeting for their date, they each make a decision
(independently, without consulting) of what genre of movie they prefer to see.
Chris prefers an adventure movie with probability 70% and a romance with
probability 30%; Juanita chooses adventure with probability 34% and a romance
with probability 66%.
180 Chapter 14. Bernoulli Random Variables

a. What is the probability that their choices agree?

b. What is the probability that they both choose an adventure movie?

Exercise 14.7. Studying. Let X be the number of nights that you spend
studying in a 30-day month. Assume that you study, on a given night, with
probability 0.65, independent of the other nights. Write X as the sum of thirty
indicators (i.e., as the sum of 30 Bernoulli random variables).

a. Find E(X).
b. Find Var(X).

14.3.2 Extensions
Exercise 14.8. Trucks and cars. On a certain highway, 7% of the vehicles
have 18 wheels, and the other 93% of the vehicles have 4 wheels. (We ignore
motorcycles, etc., for simplicity.) A child looks out the window and counts the
wheels on the next vehicle to pass.

a. What is the expected number of wheels?

b. What is the variance of the number of wheels?
c. Show a labeled graph of the mass for this story.
d. Show a labeled graph of the CDF of this story.

Exercise 14.9. Japanese pan noodles. As in Exercises 3.2, 10.2, 11.2,

and 12.3, at a certain local restaurant, students are known to prefer Japanese
pan noodles 40% of the time (it is a very popular and tasty dish!).

a. Let X be an indicator for whether a randomly selected student orders

Japanese pan noodles. Find E(X) and Var(X).
b. The student is joined by his girlfriend; let Y indicate whether she buys
Japanese pan noodles. Find E(Y ).
c. Japanese pan noodles cost $7, and all of the other noodle dishes at the
restaurant cost $6. What is the expected value of the bill for the couple?

Exercise 14.10. Winning and losing. Suppose that a person wins a game of
chance with probability 0.40, and loses otherwise. If he wins, he earns 5 dollars,
and if he loses, then he loses 4 dollars.

a. What is his expected gain or loss?

b. What is the variance of his gain or loss?
c. Find constants a, b such that if X = 0 when he loses and X = 1 when
he wins, then Y = aX + b is his earnings. (Hint: Solve 5 = a(1) + b and
−4 = a(0) + b. Also: X is a Bernoulli.) Verify your results above by finding
E(Y ) and Var(Y ) with this method.
 14.3. Exercises 181

14.3.3 Advanced
Exercise 14.11. Reciprocal of a random variable. If X is a Bernoulli
random variable, is 1/X a well-defined random variable? If not, why? If so,
what is the mass?
 Chapter 15

Binomial Random Variables

Not everything that can be counted counts, and not everything that counts can
be counted.
—Informal Sociology: A Casual Introduction to Sociological Thinking by
William Bruce Cameron (Random House, 1963)

You are playing a series of one-on-one basketball games with your friend. You
plan on playing 5 games total, and your friend (who is a better player and wins
about 68% of the time you play her), says that she will buy you one lunch for
each game that you win. You will have to buy her lunch for each game she
wins. What is the probability you will have to buy your friend lunch 5 days
next week? More than half the school days next week? Only once?

15.1 Introduction
Binomial random variables are more general than Bernoullis. A Binomial is the
number of successes in n independent trials. Equivalently, a Binomial is the
A Binomial is the sum of n independent Bernoulli random variables with the same probability of
sum of n success p:
independent
Bernoullis. Binomial random variable(n, p) = sum of n independent
Bernoulli(p) random variables
X = X1 + · · · + Xn

(In particular, if n = 1, a Binomial(n, p) is just a Bernoulli(p).)

For Binomial random variables, with parameters n and p, we must know in
advance that there are n independent trials; the number of trials we conduct

182
 15.1. Introduction 183

is not affected by how many are successes. We specify, in advance, how many
trials will take place, regardless of how many succeed or fail.
For instance, if n = 8, we can conduct 8 independent trials and let X1 , . . . ,
X8 be the 8 Bernoulli random variables that show which of the 8 trials succeed
or fail. If X = X1 + · · · + X8 , then X is a Binomial random variable that gives
the total number of successes on these 8 trials. As an example:

“Thumbs up” for success; trials 1, 2, 5, 7 succeed, so X1 = 1, X2 = 1, X5 = 1,

and X7 = 1; the other Xj ’s are 0, and

X = X1 + · · · + X8 = 1 + 1 + 0 + 0 + 1 + 0 + 1 + 0 = 4

is a Binomial that gives the total number of successes.

The notation for the Binomial looks like:

X ∼ Binomial(n, p),

where n is the total number of trials and p is the probability of success on each
trial. We also need the Binomial notation:

Remark 15.1. Binomial coefficients

If there are n trials, the number of ways to have j successes (for 0 ≤ j ≤ n) is:
 
n (n)(n − 1)(n − 2) · · · (n − j + 1) n!
= =
j j! j!(n − j)!

where n! = (n)(n−1) · · · (1); j! = (j)(j−1) · · · (1); and (n−j)! = (n−j) · · · (1).

Note that 0! is defined to be equal to 1.
Also note that, for j < 0 or j > n, we define n


j = 0.

E.g., when n = 8, the number of ways to arrange j successes within the trials
are:      
8 8! 8 8! 8 8!
= =1 = =8 = = 28
0 0!8! 1 1!7! 2 2!6!
     
8 8! 8 8! 8 8!
= = 56 = = 70 = = 56
3 3!5! 4 4!4! 5 5!3!
     
8 8! 8 8! 8 8!
= = 28 = =8 = =1
6 6!2! 7 7!1! 8 8!0!
184 Chapter 15. Binomial Random Variables

Binomial random variables

The common thread: Number of successes in a fixed number of in-
dependent trials with the same probability of success on each trial.
For instance, you ask 10 randomly selected people one yes/no question each.
Other examples include: number of 5’s during 100 die rolls; number of defec-
tive items in a randomly sampled collection of 12 coming off an assembly line;
number of seniors in a survey of 8500 students; etc.
Things to look for: Fixed number of independent trials; same success prob-
abilities.
The variable:

X = # of successes in the n trials, so 0 ≤ X ≤ n

The parameters:

n = the total number of trials (if n = 1, then the Binomial

is just a Bernoulli)
p = the probability that a given trial is a success
q = 1 − p = the probability that a given trial is a failure

These values p and q must be the same for every trial.

Mass:  
n x n−x
P (X = x) = p q , x = 0, 1, . . . , n
x
Expected value formula:
E(X) = np
Variance formula:
Var(X) = npq

Where does the mass formula come from?

 
n
= number of ways to arrange x successes in n trials
x
px = (probability of a success on one trial)# of successes

q n−x = (probability of a failure on one trial)# of failures

We have seen examples of Binomial random variables several times already. See,
for instance, Example 14.5b, in which X, Y, Z are independent Bernoullis, each
with probability of success 0.2, so X + Y + Z is Binomial with parameters n = 3
and p = 0.2. Binomial random variables are especially useful in problems that
have sampling with replacement (the replacement assures us that the relevant
probability of success does not change from trial to trial).
 15.2. Examples 185

15.2 Examples

Example 15.2. In a certain city, we plan to predict each year—during a 31

year period—whether there is snow on New Year’s Day. There is historically a
30% chance of snow that day. We assume the behavior on the New Year’s Days
during this 31 year period are independent from each other. We are interested
in counting up the number of snowy New Year’s Days during the 31 year period.

a. How many trials are there?

We plan to have trials on 31 independent days, so n = 31.
b. What does X represent in terms of this story? What values can it take?
Here, X is the number of snowy New Year’s Days over a period of 31 years,
so X can be any of the values 0, 1, . . . , 31.
c. Why is this a Binomial distribution problem?
We have a specified number (31) of independent trials, with equal probability
of success (0.3) on each trial, and we are counting up the number of successes
(# of snowy days).
d. What is the probability of exactly 5 snowy New Year’s Days?

 
31
P (X = 5) = (0.3)5 (0.7)26 = 0.03876
5

e. What is the probability of at least 1 snowy New Year’s Day?

P (X ≥ 1) = 1 − P (X = 0)
 
31
=1− (0.3)0 (0.7)31
0
= 1 − 0.000016
= 0.999984

Remark 15.3. In general, when computing the probability of “at least one”
of something, it is easier to compute the probability of the complement, i.e.,
the probability that there are zero occurrences of the relevant phenomenon.
For example, in Example 15.2e, it is much easier to use the complement of the
event than to add the probabilities corresponding to 1, 2, . . . , 31 snowy days.

f. How many snowy New Year’s Days do we expect altogether?

E(X) = 31(0.3) = 9.3 days

186 Chapter 15. Binomial Random Variables

g. What is the standard deviation in the number of snowy days? (Re-

call that the standard deviation measures the spread of the distribution of a
random variable. A small standard deviation tells us that a distribution is well-
concentrated around the expected value. A larger standard deviation indicates
that a distribution is more spread out from the expected value.)

Var(X) = 31(0.3)(0.7) = 6.51

√
σX = 6.51 = 2.5515 days

h. What does the mass look like for this story?

pX (x)
0.16
0.12
0.08
0.04
x
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31

i. What does the CDF look like for this story?

FX (x)

1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
x
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31

Example 15.4. Suppose a couple decides to have 4 children. Assume the

probability is the same for having a boy or a girl, and they will not have any
multiple births (so sexes are independent). Let X denote the number of girls
that the couple has.
 15.2. Examples 187

a. How do the mass and CDF of X look?

pX (x) FX (x)
1
(2, 0.375)
0.4 0.8
0.3 (1, 0.25) (3, 0.25) 0.6
0.2 0.4
0.1 (0, 0.0625) (4, 0.0625) 0.2
x x
1 2 3 4 1 2 3 4

Left: Mass of the number of girls. Right: CDF of the number of girls.
b. What is the probability the couple has at least one girl?

P (X ≥ 1) = 1 − P (X = 0)
 
4
=1− (0.5)0 (0.5)4
0
= 1 − 0.0625
= 0.9375

We used the complement, because computing 1 − P (X = 0) is easier than

computing P (X = 1) + P (X = 2) + P (X = 3) + P (X = 4). This concept, of
computing the complement, is helpful in many similar situations.
c. Planning in advance of having any children, how many children would
the couple have to have for there to be at least a 99% chance that at least one
of their children is a girl? (Hint: This is a “backward” Binomial problem, in the
sense that we know the desired probability, and the value of p is known to be
0.5, but we do not know the value of n.)
We need
P (X ≥ 1) ≥ 0.99.
Rewriting the left-hand side using the complement,

1 − P (X = 0) ≥ 0.99,

or equivalently,
0.01 ≥ P (X = 0). (15.1)
We have  
n
P (X = 0) = (0.5)0 (0.5)n = (0.5)n ,
0
since for any positive integer k, k0 = 1 and k 0 = 1. Substituting back into

(15.1) gives
0.01 ≥ (0.5)n .
188 Chapter 15. Binomial Random Variables

Taking a natural log of each side gives ln 0.01 ≥ ln (0.5)n , and then we can pull
the exponent n down to get

ln 0.01 ≥ n ln (0.5).

So −4.605 ≥ n(−0.693), i.e., 6.644 ≤ n. (Remember that, when dividing both

sides by a negative number, the inequality is flipped.)
Our conclusion is that the couple needs to have at least 6.644 children to
have at least a 99% chance of having at least 1 girl. Six children would not
be enough, and the couple cannot have a fraction of a child. So the most
appropriate answer is that the couple needs to have at least 7 children.

Remark 15.5. Expected value and variance of a Binomial random

variable If X1 , . . . , Xn are independent Bernoulli random variables, each with
probability of success p, then

X = X1 + · · · + Xn

is a Binomial(n, p) random variable. So the expected value of X can be com-

puted by using the linearity of the expected value:

E(X) = E(X1 + · · · + Xn )
= E(X1 ) + · · · + E(Xn )
= p + ··· + p
= np.

Since the Xj ’s are independent, then we can also add their variances to obtain
the variance of X:

Var(X) = Var(X1 + · · · + Xn )
= Var(X1 ) + · · · + Var(Xn )
= pq + · · · + pq
= npq.

Remark 15.6. We stress that for the Binomial distribution (and also for
the Geometric and Negative Binomial distributions, which we will see later)
it is important to have independence, or at least relative independence. So
what happens if you have a small population and you are sampling without
replacement, i.e., the probabilities change between trials? We’ll cover that in
Chapter 19, with the Hypergeometric distribution.

Example 15.7. Roll a die n times. Let X denote the total number of 4’s and
5’s that appear.
 15.2. Examples 189

Here X is a Binomial random variable, with each outcome of “4” or “5” treated
as a “success.” So p = 2/6 = 1/3, because 2 out of the 6 equally likely outcomes
are successes. The mass of X is
   j 
1 n−j

n 1
pX (x) = 1−
j 3 3
   j  n−j
n 1 2
=
j 3 3
  n−j
n 2
=
j 3n

for j = 0, 1, 2, . . . , n, and otherwise pX (x) = 0.

The expected value of X is

E(X) = np = n/3,

and the variance of X is

Var(X) = npq = (n)(1/3)(2/3) = 2n/9.

Example 15.8. (Refer to Example 14.1) Suppose 95% of people put peanut
butter on first when making a peanut butter and jelly sandwich. Five peo-
ple are independently asked about their sandwich-making habits. What is the
probability that the majority of them put the peanut butter on first?
Let X be the number of people who put the peanut butter on first, so X is a
Binomial random variable with mass
 
5
P (X = j) = (0.95)j (1 − 0.95)5−j
j
 
5
= (19/20)j (1/20)5−j
j
  j
5 19
= ,
j 205

for j = 0, 1, 2, 3, 4, 5, and P (X = j) = 0 otherwise. The majority of people put

peanut butter on first if X = 3 or X = 4 or X = 5. So the desired probability
is

P (X = 3 or X = 4 or X = 5) = P (X = 3) + P (X = 4) + P (X = 5)
  3   4   5
5 19 5 19 5 19
= 5
+ 5
+
3 20 4 20 5 205
= 0.0214 + 0.2036 + 0.7738
= 0.9988
190 Chapter 15. Binomial Random Variables

Example 15.9. Dominique, Raymond, and Samantha are independently using

an app that randomly creates playlists for them. Forty percent of the songs
this app has access to are rap songs. Dominique’s playlist will have 30 songs,
Raymond’s 70, and Samantha’s 90. What is the distribution for the total num-
ber of rap songs if the three people combine their playlists later? (Assume that
there are no overlaps among their songs.)
The number of rap songs on their playlists are Binomial random variables
X1 , X2 , X3 , each with p = 0.40, and with respective number of trials n1 = 30,
n2 = 70, and n3 = 90. So the total number of rap songs on their playlists is
Y = X1 + X2 + X3 , which is also Binomial, with probability of success p = 0.40
on each trial, and N = 30 + 70 + 90 = 190 trials altogether.

Remark 15.10. Sums of Binomial random variables Consider indepen-

dent Binomial random variables X1 , X2 , . . . , Xm , for which the probabilities
of success “p” are all the same, and for which the number of trials are (re-
spectively) n1 , n2 , . . . , nm . Then X1 + X2 + · · · + Xm is the number of suc-
cesses that occur in n1 + n2 + · · · + nm independent trials. So the sum
Y = X1 +X2 +· · ·+Xm is also Binomial, with parameters N = n1 +n2 +· · ·+nm
and p. So the sum Y has mass
 
N j N −j
pY (j) = p q , for integers 0 ≤ j ≤ N
j

and pY (j) = 0 otherwise.

15.3 Exercises
15.3.1 Practice
Exercise 15.1. Skittles. Skittles candies come in the colors red, orange, yel-
low, green, and purple, with each color having equal probability. You are a
quality control inspector, and your job is to count up the number of purple
candies in a random sample of 25 candies from a large population of candies
coming down the factory line.
a. What is a “success” in this situation? What is the probability of a success
on a single trial?
b. What is a “failure” in this situation? What is the probability of a failure
on a single trial?
c. Explain in words what X is in terms of the story. What values can it
take?
 15.3. Exercises 191

d. Why is this a Binomial distribution situation? What are the parameters?

e. What is the probability that you will find exactly 5 purple candies in
your sample?
f. What is the probability that you will find at least 2 purple candies in
your sample?
g. What is the expected number of purple candies in the sample?
h. What is the standard deviation in the number of purple candies in the
sample?
i. If purple candies cost the company 1.5 cents each, with an additional
general production fee of 25 cents (not related to the number of purple candies),
what is the expected cost of purple candies for this sample?
j. What is the standard deviation of the cost of purple candies in this
sample?

Exercise 15.2. Colorblind. Approximately 8.33% of men are colorblind. You

survey 8 men from the population of a large city and count the number who are
colorblind.

a. What is a “success” in this situation? What is the probability of a success

on a single trial?
b. What is a “failure” in this situation? What is the probability of a failure
on a single trial?
c. Explain in words what X is in this situation and what values it can take.
d. Why is this a Binomial distribution situation? What are the parameters?
e. What is the probability exactly 2 will be colorblind?
f. What is the probability that 6 or fewer will be colorblind?
g. How many men would you have to survey in order to be at least 95%
sure you would find at least 1 who is colorblind?
h. What is the expected value of X?
i. What is the variance of X?
j. Show the labeled graph of the mass for this story.
k. Show the labeled graph of the CDF for this story.

Exercise 15.3. Surgical delivery. Twenty percent of babies in a particular

city are born by a surgical procedure called a Cesarean section (C-section). You
randomly survey 9 parents of babies from the population of the large city and
count the number of babies who are not born by C-section.

a. Explain in words what X is in this situation and what values it can take.
192 Chapter 15. Binomial Random Variables

b. Why is this a Binomial distribution situation? What are the parameters?

c. What is the probability exactly 6 were not born by C-section?
d. What is the probability that between 4 and 6 babies were not born by
C-section?
e. How many babies would you have to survey in order to be at least 90%
sure you would find at least 1 who is not born by C-section?
f. What is the expected value of X?
g. What is the variance of X?
h. Show the labeled graph of the mass for this story.
i. Show the labeled graph of the CDF for this story.
Exercise 15.4. Lost dog. Gracie is looking for her lost dog. She will ran-
domly ask people in her neighborhood if they have seen her dog. Assume that
each neighbor is equally likely to have seen her dog and that the neighbors are
independent. Is this a Binomial situation? Why or why not?
Exercise 15.5. Cereal boxes. A cereal company puts a Star Wars toy watch
in each of its boxes as a sales promotion. Twenty percent of the cereal boxes
contain a watch with Obi Wan Kenobi on it. You are a huge Obi Wan fan, so
you really want one of these watches.
a. You decide to buy 100 boxes of cereal from the warehouse store to be on
the safe side. What is the probability you don’t find any Obi Wan watches?
b. What is the expected number of Obi Wan watches you will find in the
100 boxes?
c. If each box of cereal costs $3.50, and you believe the value of an Obi Wan
watch will be approximately $50 in a few years, are you doing a smart thing by
purchasing 100 boxes? (Think about the expected value for the cost and the
profit.)
d. How many boxes of cereal would you need to buy to be at least 95% sure
of finding at least 1 Obi Wan watch?
Exercise 15.6. Mobile phone survey. A recent survey states that 48% of
mobile devices are iPhones. In order to learn more about how the iPhone works,
a student starts asking random people on campus if they use an iPhone. Assume
that the individuals on campus are independent.
a. If he surveys 20 people, what is the probability he finds exactly 2 people
who use an iPhone?
b. If he surveys 20 people, what is the probability he finds at least 2 people
who use an iPhone?
c. If he surveys 20 people, what is the probability he finds fewer than 2
people who use an iPhone?
 15.3. Exercises 193

d. How many students would he have to ask to be at least 90% certain of

finding at least one person who uses an iPhone?
Exercise 15.7. Spinner. A spinner in a certain game lands on “0” 70% of the
time and lands on “1” 30% of the time. How many times must a player spin so
that the probability of having at least one result of “1” exceeds 95%?
Exercise 15.8. Reshelving in library. There are 7 books needing reshelv-
ing in the undergraduate library. Sixty-five percent of the library’s collection
consists of reference books. Let X be the number of reference books a student
page reshelves out of the 7 on her cart.
a. What is the probability that all 7 of them are reference books?
b. What is the probability that the majority of these 7 books are reference
books?
c. What is the expected number of reference books in this collection of 7
books?
d. What is the standard deviation of the number of reference books in this
collection?
e. Show the labeled graph of the mass for this story.
f. Show the labeled graph of the CDF for this story.
Exercise 15.9. Dice. You have a 4-sided die with colors red, blue, yellow, and
green for each face. You will roll the die 10 times. Let X be the number of rolls
on which blue is the color selected.
a. What is the probability blue is selected exactly 3 times?
b. What is the probability blue is selected at most 3 times?
c. Given that blue is selected at most 3 times, what is the probability blue
is selected at least 1 time?
d. Given that blue is selected at least 1 time, what is the probability blue
is selected at most 3 times?
Exercise 15.10. Blood type. Suppose that 3% of people have AB+ blood
type. How many people need to be sampled so that there is more than a 50%
chance that at least one person in the group has AB+ blood type?
Exercise 15.11. Field goals in football. Jeff typically makes 80% of his field
goals. Steve typically makes 60% of his field goals. Suppose they both have the
opportunity to kick 3 field goals.
a. What is the probability Jeff will succeed in making at least 1 field goal?
b. What is the probability Steve will succeed in making at least 1 field goal?
c. What is the probability that both Steve and Jeff will succeed in making
at least 1 field goal?
194 Chapter 15. Binomial Random Variables

d. What is the probability that either Steve or Jeff will succeed in making
at least 1 field goal?
e. Given that only one field goal total was scored, what is the probability
that Jeff was the one who kicked it?
Exercise 15.12. Tennis. Suppose a tennis player hits an ace once out of every
five serves. Suppose in a match the player performs 80 serves.
a. What is the expected number of aces?
b. What are the variance and standard deviation of the number of aces?
Exercise 15.13. Exam. On a multiple choice exam, a student decides to test
his luck. His exam has 20 questions, each of which has 5 answer choices. The
student decides to roll a die on each question and use the result on the die as
his answer; any time that he rolls a 6, he just discards that roll and tries again.
Let X be the number of questions he gets right on the exam altogether. What
is the probability he gets a grade of A (overall score at least 90%) using this
method?
Exercise 15.14. Dice. Two students decide to make bets about their plans
for lunch during the next work week (Monday through Friday). They roll a
six-sided die five times (once for each day). The agreement is that, for each
day, when a 6 shows up, the first student has to pay for both lunches, and if a
1 shows up, the second student has to pay for both lunches. If neither of these
events occurs, they will bring their lunch from home on that day.
a. How many days does the first student expect to buy lunch for the second
student?
b. How many days do they expect to bring their lunches?
c. What is the mass of the number of days on which the first student buys
lunch for the second student?
d. What is the mass of the number of days on which they bring their lunches?
Exercise 15.15. Winning and losing. Suppose that a person wins a game
of chance with probability 0.40, and loses otherwise. If he wins, he earns 5
dollars, and if he loses, then he loses 4 dollars. Assume that he plays ten games
independently. Let X denote the number of games that he wins. (Hint: His
gain or loss is 5X + (−4)(10 − X) = 9X − 40, since he loses 10 − X games.)
a. What is his expected gain or loss (altogether) during the ten games?
b. What is the variance of his gain or loss (altogether) during the ten games?
c. What is the probability that he wins $32 or more during the ten games?
Exercise 15.16. Telemarketers. Assume that when your phone rings, the
caller is a telemarketer with probability 1/8, and that the probability of a tele-
marketer is independent from call to call. Let X denote the number of telemar-
keters during the next three calls.
 15.3. Exercises 195

a. What is the mass of X?

b. Draw a graph of the mass of X.
c. Draw a graph of the CDF of X.
Exercise 15.17. Hiking. You randomly text 20 of your friends to see who
wants to go on a hiking trip. You think that they all respond to your requests
independently of each other, and you estimate that each one is 7% likely to be
available, interested in going hiking, and will actually text you back to accept
the invitation.
a. Find the probability that at least 3 people would accept the invitation.
b. Find the expected number of people who would accept the invitation.
c. Find the variance of the number of people who would accept the invita-
tion.
Exercise 15.18. Dining hall. Let X, Y, Z be (respectively) the number of
nights that Alice, Bob, and Charlotte eat in the dining hall during a 7-day
week. Assume that X, Y, Z are independent Binomial random variables that
each have n = 7 and p = 0.65.
a. What is the distribution of X + Y + Z, i.e., the total number of meals
eaten by these three people (altogether) during a week?
b. Find Var(X + Y + Z).
Exercise 15.19. Hearts. You draw seven cards, without replacement, from a
shuffled, standard deck of 52 playing cards. Let X be the number of hearts that
are selected.
a. What is the expected number of hearts? Why?
b. Is X a Binomial random variable? Why or why not?
Exercise 15.20. Reserving a room. There is a big exam tonight, and all of
the 400 students are invited to attend the help session. From past experience,
the instructor finds that 60% of the students are likely to attend the help session.
She wants to reserve an appropriately sized room.
a. What is the expected value and variance for the number of students who
will attend?
b. How many seats must be in the reserved room, so that the professor can
be at least 90% sure that everyone who wants to attend will have a seat? (A
computer is necessary to solve this problem numerically.)
Exercise 15.21. Applying to schools. This year 557 students applied to
school A, and only 40% of the students are typically accepted. At the same
time, 903 students applied to school B, and 27% of these students are typically
accepted. Let XA and XB be the number of students accepted to schools A and
B, respectively. Which school do we expect to have more acceptances?
196 Chapter 15. Binomial Random Variables

15.3.2 Extensions
Exercise 15.22. Saving energy. Despite lecturing your roommates on energy
conservation, there is a 60% chance that the lights in a dorm room will be left
on when nobody is home. Each day is independent. Suppose that, every day
the light is left on in a dorm room, there are 1000 Watts of power used. Every
day when the light is turned off, there are 200 Watts of power used. You keep
track of X, the number of days the lights are left on over the next 30 days.

a. What is the expected amount of power used during the 30 days?

b. Find an expression for the probability that 16,400 Watts of power are
used during the 30 days. You do not need to simplify or evaluate the expression.

Exercise 15.23. Selling encyclopedias. Samuel is an encyclopedia salesman.

In order to make his life more interesting, when he encounters an intersection
of two streets, he heads east 30% of the time and north the other 70%. He will
walk 50 blocks (intersection-to-intersection) each day. West-East blocks contain
4 houses each and North-South blocks contain 6.

a. On a certain day, how many houses does he expect to visit?

b. Find an expression for the probability that he visits 290 or more houses.
You do not need to simplify or evaluate the expression.

15.3.3 Advanced
Exercise 15.24. Ants. In the ant world, 98% of the ants are female, and
2% are male. In the queen’s first batch of 100,000 offspring, let X be the
total number of male births. Write an expression for the probability that 2100
or more of the ants are males. You do not need to simplify or evaluate your
expression. Would it be difficult to calculate?

Exercise 15.25. Cereal for breakfast. Sixty percent of students usually eat
cereal for breakfast. If n students eat in the dining halls for breakfast each day,
let Xn be the number who have cereal. Find the limiting probability that at
least one student has cereal, as n grows large, i.e., find limn→∞ P (Xn > 0).
Does your answer make sense intuitively? Why?
Chapter 16

Geometric Random Variables

If at first you don’t succeed, try, try, try again.

—traced to Thomas H. Painter, Frederick Marryat, and perhaps others;
popularized by William Edward Hickson
If at first you don’t succeed, skydiving is not for you.
—Wendy Northcutt

You are playing a series of one-on-one basketball games with your friend. Instead
of the 5-game series you played with your friend last week (described in the
introduction to Chapter 15), this week you have decided that you will only play
as many games as it takes for you to win your first game. You still have to buy
your friend lunch for each game she wins, and then she will buy you lunch when
you win your first game. What is the probability you will have to buy your
friend a lunch exactly 4 days this week, before she buys your lunch? What is
the probability you don’t have to buy lunch for your friend at all? How many A Geometric
lunches do you expect to have to buy for your friend? How is this game different random variable is
from last week’s game? the number of
independent trials
until (and
including) the first
success, when the
16.1 Introduction probability of
success on each trial
A Geometric random variable is the number of independent trials needed until is constant.
the first success occurs. An equivalent interpretation is that a Geometric ran-
dom variable is the number of independent Bernoulli random variables we need
to check until the first one that indicates success. If X1 , X2 , . . . are independent
Bernoulli(p) random variables, then

Geometric random variable = smallest value of M so that XM = 1

(and X1 = . . . = XM −1 = 0)

197
198 Chapter 16. Geometric Random Variables

Geometric random variables

The common thread: Number of independent trials, each with the
same probability of success, until the first success occurs. For instance,
you ask people one yes/no question each, until you get the first “yes” response.
You don’t know how many trials you’ll need in advance because you only stop
asking when you find that 1st success. Other examples include: number of
dice rolls until the first 5; number of parts coming off an assembly line until
the first defective one; number of students surveyed until the first senior is
discovered; etc.
Things to look for: Repeated independent trials until first success; same
success probabilities.
The variable:

X = # of trials until the first success, so X ≥ 1

The parameters:

p = the probability that a given trial is a success

q = 1 − p = the probability that a given trial is a failure

These values p and q must be the same for every trial.

Mass:
P (X = x) = q x−1 p, x = 1, 2, 3, . . .

Expected value formula:

E(X) = 1/p

Variance formula:
Var(X) = q/p2

For instance, if a Geometric random variable X has the value 6, we can view
this as five independent Bernoulli random variables that indicate failure, i.e.,
X1 = X2 = X3 = X4 = X5 = 0, followed by a sixth independent Bernoulli that
indicates success, i.e., X6 = 1.
As an example:

“Thumbs up” for success; trials 1, 2, 3, 4, 5 fail, and trial 6 succeeds. So X = 6

is a Geometric that gives the total number of trials until the first success.
Geometric random variables only have one parameter, p, the probability
of success of the independent trials. An easy way to distinguish Geometric
 16.1. Introduction 199

random variables from Binomials is that a Geometric(p) random variable can

be arbitrarily large (there is no upper limit n), whereas a Binomial(n, p) random
variable is always between 0 and n. A Geometric random variable has to be at
least 1, because at least one trial is needed to get the first success.
For a Geometric random variable with parameter p, the probability of suc-
cess p must remain the same on every one of the independent trials.
We cannot know in advance how many independent trials will be needed to
see the first success.
A key feature of a Geometric is the memoryless property; see Section 16.4.
The notation for the Geometric looks like:

X ∼ Geometric(p),

where p is the probability of success on each trial.

Where does the mass formula come from?

x − 1 = number of trials − 1
= number of failures
q x−1
= (probability of a failure on one trial)# of failures

p = probability of a success on one trial

Example 16.1. Suppose 3% of pet owners give gifts to their pets on Valentine’s
Day. You plan to talk to randomly selected pet owners from a large population
until you find one who gives a gift to his or her pet. Since you talk to them
individually, assume their responses are independent.

a. What does X represent in this story? What values can X take?

The variable X is the number of pet owners we ask until we find the first
one who gives a gift to his or her pet, so X can be 1, 2, . . .
b. Why is this an example of the Geometric distribution?
The trials are independent and have an equal probability of success (0.03).
We are performing the trials—selecting pet owners—until we find our first suc-
cess, i.e., a pet owner who gives his or her pet a Valentine’s gift.
c. How many pet owners do we expect to have to ask until finding one who
gives a Valentine’s gift to his or her pet?

1
E(X) = = 33.33 pet owners
0.03
200 Chapter 16. Geometric Random Variables

d. What is the standard deviation of the number of people you have to talk
to until you find one who gives a Valentine’s gift to his or her pet?
1 − 0.03
Var(X) = 2
= 1077.78
√0.03
σX = 1077.78 = 32.8295 pet owners

e. What does the mass look like for this story?

pX (x)
0.030
0.020
0.010
x
10 20 30 40 50 60 70 80 90 100

f. What does the CDF look like for this story?

FX (x)
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
x
10 20 30 40 50 60 70 80 90 100

g. What is the probability you will have to talk to exactly 40 people total
to find your first pet-gift-giver?

P (X = 40) = (0.97)39 (0.03) = 0.009146

h. If you decide in advance to talk to exactly 40 people, what is the prob-

ability you will have exactly one pet-gift-giver in the sample? How does this
compare to the situation in part e?
 
40
P (X = 1) = (0.97)39 (0.03)1 = 40(0.009146) = 0.3658
1
There are 40 ways that the scenario in part h could happen (any of the 40 people
could be the one pet-gift-giver), each with probability (0.03)1 (0.97)39 , but only
1 way that the scenario in part g could happen. Therefore the probability in
part h is 40 times as large as the probability in part g.
 16.1. Introduction 201

The scenario in part h of the example above is a Binomial situation—not a

Geometric situation—because we fixed the number of trials (40) in advance,
and we are counting up the number of successes anywhere in those 40 trials. In
scenario g we were asking for 39 failures in a row, with 1 success at the very
end, i.e., the success had to occur as the last trial.

More generally, this illustrates why the Geometric formula does not have an
“ nx ” term, because we are not concerned with the number of ways to rearrange
the successes and failures when working with a Geometric random variable. If
you have 3 trials and 1 success, there is only one way to arrange these for the
Geometric, because the success must come last:

but there are 3

= 3 ways to arrange the outcomes for the Binomial:


1

or or

Remark 16.2. A frequent question about Geometric random variables is

whether a Geometric random variable will always be finite. Could a Geometric
random variable be infinite?
For example, if X is the number of fair coin flips until the first head, can X be
infinite? The answer is that X will be finite with probability 1. So the event
that an infinite number of coins are needed has probability 0. (In such a case,
we might be tempted to say that X is infinite, although random variables are
only supposed to take on real values.)
This is true much more generally: If the probability of success p on each
trial is positive (p > 0), then a Geometric random variable will be finite with
probability 1.

One way to see this is to calculate the probability that a Geometric random
variable is finite. If X is a Geometric(p) random variable with p > 0, then
X X X
P (X is finite) = P (X = j) = q j−1 p = p q j = p/(1 − q) = p/p = 1.
j≥1 j≥1 j≥0

Another way to see this is, if X is infinite then for each n we know n < X, so

0 ≤ P (X is infinite) ≤ P (n < X) = q n ,

but limn→∞ q n = 0 since q < 1, so P (X is infinite) = 0.

202 Chapter 16. Geometric Random Variables

Remark 16.3. Expected value of a Geometric random variable If X is

Geometric with probability of success p on each trial, then the expected value
of the Geometric random variable is
X X X d
E(X) = jP (X = j) = jq j−1 p = p qj .
dq
j≥1 j≥1 j≥1

It might seem unusual to differentiate with respect to “q,” but the derivation
lends itself to this change, and now we can switch the order of the summation
and the differentiation, so

d X j d q (1 − q)(1) − (q)(−1) 1
E(X) = p q =p =p 2
= p 2 = 1/p.
dq dq 1 − q (1 − q) p
j≥1

Remark 16.4. Variance of a Geometric random variable We know that

Var(X) = E(X 2 ) − (E(X))2 , but E(X 2 ) will not be as nice for us to work with
in this derivation as E((X)(X − 1)) will be. So we need to do a little algebraic
manipulation: X 2 = X 2 − X + X = (X)(X − 1) + X.
The second moment of X is computed in a similar manner if we observe that
(j)(j − 1)q j−2 is a second derivative of q j with respect to q. We get:
X X
E((X)(X − 1)) = j(j − 1)P (X = j) = j(j − 1)q j−1 p
j≥1 j≥1
X X d2
= pq j(j − 1)q j−2 = pq qj
dq 2
j≥1 j≥1
d2 X d2 q
= pq q j = pq ,
dq 2 dq 2 1−q
j≥1

which we can calculate. This yields

2 2q
E((X)(X − 1)) = pq 3
= 2.
(1 − q) p

This almost gives us the variance. Now we use the fact mentioned above, that
X 2 = (X)(X − 1) + X, and this yields

E(X 2 ) = E((X)(X − 1)) + E(X),

which we now substitute into the equation for the variance:

Var(X) = E(X 2 ) − (E(X))2

= E((X)(X − 1)) + E(X) − (E(X))2
2q 1 1
= 2 + − 2
p p p
2
= q/p .
 16.1. Introduction 203

Example 16.5. Draw a card from a well-shuffled deck until the ace of spades
appears. If a draw is unsuccessful, then replace and reshuffle the deck before
making the next selection. Let X be the number of draws needed until the ace
of spades appears for the first time.

(Note that, in the case where the ace of spades never appears, we have
not specified the value of X, but this omission does not affect the probabilities
associated with the problem, because the probability of the ace of spades never
appearing is zero.)
The mass of X is pX (j) = (51/52)j−1 (1/52) for positive integers j, and
pX (x) = 0 otherwise. So X is a Geometric random variable with p = 1/52. The
expected number of cards until the first ace of spades appears is E(X) = 1/52
1
=
51/52
52. The variance of X is Var(X) = (1/52)2
= 2652.

Example 16.6. Suppose that a basketball player makes 80% of her free throws
successfully. She attempts to make a basket as many times as necessary, until
scoring the first basket. Let X denote the number of necessary attempts.

(Again, the probability that she never scores is 0; we have not specified the
value of X when she never scores.)
The mass of X is pX (j) = (0.20)j−1 (0.80) for positive integers j, and
pX (x) = 0 otherwise. Thus, X is Geometric with p = 0.80. So E(X) =
1/0.80 = 1.25 and Var(X) = 0.20/0.802 = 0.3125.

Example 16.7. People are selected randomly and independently for a drug
test. Each person passes the test 98% of the time. What is the expected
number of people who take the test until the first person fails?

Let X be the number of people tested, until the first person fails the test. Then
P (X = j) = (0.98)j−1 (0.02) for each positive integer j. Thus, X is Geometric
with p = 0.02. So we expect E(X) = 1/0.02 = 50 people to be tested until the
first person fails the test. The variance of X is Var(X) = 0.98/0.022 = 2450.
204 Chapter 16. Geometric Random Variables

16.2 Special Features of the Geometric Distribution

We discuss several common methods for using inequality symbols with the Ge-
ometric distribution. Hint: It’s much easier to work with the Geometric in-
equalities if you rephrase things in terms of a Geometric random variable being
greater than a specific number.
What is the probability we need more than 5 trials to get the 1st success?

P (X > 5) = P (1st 5 rolls all failures) = q 5

What is the probability we need at least 5 trials to get the 1st success?

P (X ≥ 5) = P (1st 4 rolls all failures, 5th roll could be a success or a failure)

= P (X > 4)
= q4

What is the probability we need at most 5 trials to get the 1st success?

P (X ≤ 5) = 1 − P (X > 5) = 1 − q 5

What is the probability we need fewer than 5 trials to get the 1st success?

P (X < 5) = 1 − P (X ≥ 5) = 1 − P (X > 4) = 1 − q 4

16.3 The Number of Failures

Geometric random variables are defined to be the number of trials required until
the first success occurs, and the success itself is included as one of these trials.
So, for instance, if X is Geometric, then X = 10 corresponds to the fact that
the 10th trial is the first success. As an example, X might be the number of
coin flips until the first head occurs, including the head itself.
Sometimes we are more interested in the number of failures that occur until
the first success, i.e., we do not want to take the success itself into account. For
instance, let Y be the number of failures that occur before the first success, not
including the success itself. E.g., if Y is the number of tails that occur before
the first head (not including the head itself), then Y is the number of failures.
Whenever a Geometric random variable X is equal to j, there is an analogous
 16.4. Geometric Memoryless Property 205

random variable Y , the number of failures, that is equal to j − 1 in the same

case.
For the number of failures Y to be exactly j, there must be exactly j failures
in a row, followed immediately by a success, so, the mass of such a random
variable Y is
pY (j) = q j p for j = 0, 1, 2, 3, . . .,

and pY (y) = 0 otherwise.

With these definitions, Y and X − 1 have the same distribution, or equiva-
lently, Y + 1 and X have the same distribution. For instance, pY (3) = q 3 p =
pX (4).
Notice that the number of failures is always one less than the number of
trials until the first success, so the expected values must differ by 1 too. E.g.,
E(Y ) = E(X − 1) = E(X) − 1 = p1 − 1 = 1−p p = q/p.
Also notice that, since the number of failures and the number of trials until
the first success always differ by 1, they have the same variance. (Remember
that the addition or subtraction of a constant from a random variable does not
affect the variance.) Thus Var(Y ) = Var(X − 1) = Var(X).

16.4 Geometric Memoryless Property

If you have conditional probabilities using the “>” symbol on both sides, you
can use this shortcut. (Food for thought: this really works seamlessly only for
“>,” not for “<”; why?)
Given that you need more than 2 trials to get your 1st success, what is the
probability you will need more than 5 trials total?

P (X > 5 | X > 2) = P (X > 3) = q 3

Trial ? ? ?
Outcome No No ? ? ?
Past/future? Past Past Future Future Future

Why does this work? Each trial is independent. Once some trials have
passed, the next trials do not depend on what has already happened in the
past. You already know the first 2 trials were failures. What you don’t know is
what will happen on the next 3 trials.
Here is the same explanation using conditional probability:

P (X > 5 and X > 2)

P (X > 5 | X > 2) = ,
P (X > 2)
206 Chapter 16. Geometric Random Variables

but we only have both X > 5 and X > 2 if we have X > 5. So

P (X > 5)
P (X > 5 | X > 2) = = q 5/q 2 = q 3 = P (X > 3).
P (X > 2)

Example 16.8. Given that more than j trials are needed to get the first success,
what is the probability that more than k trials are needed?
Of course we have P (X > k | X > j) = 0 if k < j. So we now consider k ≥ j.
In such a case, the first j trials are already given to be failures, so only trials
number j + 1, j + 2, . . . , k are needed to be failures for the first k trials to all
be failures. So

P (X > k | X > j) = P (X > k − j) = q k−j .

Here is a more rigorous explanation using conditional probabilities:

P (X > k and X > j)
P (X > k | X > j) =
P (X > j)
P (X > k)
=
P (X > j)
= q k/q j
= q k−j
= P (X > k − j)

Example 16.9. Memoryless Geometric property. As in Exercises 8.5,

10.18, and 11.4, the Super Breakfast Challenge (SBC) is known to be very
difficult to consume. Only 10% of people are able to eat all of the SBC.
Let X be the number of people required until a customer is able to eat the
whole SBC.
a. How many people are needed, on average, until the first successful cus-
tomer?
Since X is Geometric with p = 0.10, then E(X) = 1/0.10 = 10.
b. What is the variance of the number of people needed?
The variance of the number of customers needed is Var(X) = 0.90/0.102 =
90.
c. What is the probability that more than 4 customers are needed, until the
first success?

P (X > 4) = P (1st 4 were unsuccessful) = q 4 = (0.90)4 = 0.6561

 16.5. Random Variables That Are Not Geometric 207

d. Given that the first 4 are unsuccessful, what is the probability at least 8
are needed?

P (X ≥ 8 | X > 4) = P (X > 7 | X > 4)

= P (X > 3)
= P (1st 3 are unsuccessful)
= (0.90)3
= 0.729

Remark 16.10. A note of caution:

The memoryless property does not work with any of the other discrete random
variables that we have learned. The memoryless property will also work with
one of the continuous random variables (the Exponential random variable,
which will be covered in the second half of the book). Do not try to use the
memoryless property with any other discrete distributions.

16.5 Random Variables That Are Not Geometric

One of the things to notice in particular about Geometric random variables is
that the probability of success p must not change during the series of indepen-
dent trials. So, as an example, consider a shuffled deck of cards, in which we
draw cards repeatedly—without replacement—until the ace of spades appears.
Let X denote the number of draws necessary. Then X is not a Geometric ran-
dom variable. There are a few ways to see this. One way to see this immediately
is that X ≤ 52 always, so P (X = 53) = 0, and thus X cannot be Geometric.
Another way to see this is P (X = j) = 1/52, because X = j if and only if the
jth card in the deck was the ace of spaces. So X does not have the right kind
of mass to be a Geometric random variable.

16.6 Exercises
16.6.1 Practice
Exercise 16.1. Skittles. Skittles candies come in the colors red, orange, yel-
low, green, and purple, with each color having equal probability. Due to a dye
mix-up there are a few rainbow-striped candies coming down the line. There is a
5% chance that a candy is rainbow striped. You are a quality control inspector,
and your job is to find the first rainbow-striped candy coming down the line.
(The population is so big that we can assume relative independence.)

a. Explain in words what X is in terms of the story. What values can it

take?
208 Chapter 16. Geometric Random Variables

b. Why is this a Geometric distribution situation? What is the parameter?

c. What is the probability that you will need to check exactly 10 candies to
find the first rainbow one?
d. What is the probability that you will have to check fewer than 12 candies
to find the first rainbow one?
e. What is the probability you will have to check more than 8 candies to
find the first rainbow one?
f. Given that you have to check at least 3 candies, what is the probability
that you will need to check a total of more than 9 for the first rainbow one?
g. What is the expected number of candies you will have to check until you
find the first rainbow one?
h. What is the standard deviation in the number of candies you will have
to check until you find your first rainbow one?

Exercise 16.2. Colorblind. Approximately 8.33% of men are colorblind. You

survey men from a large population until you find one who is colorblind.

a. Explain in words what X is in this situation and what values it can take.
b. Why is this a Geometric distribution situation? What is the parameter?
c. What is the probability you will have to survey at most 16 men until you
find the first one who is colorblind?
d. What is the probability you will have to ask exactly 12 men until you
find the first who is colorblind?
e. What is the expected value of X?
f. What is the variance of X?
g. Show the labeled graph of the mass for this story.
h. Show the labeled graph of the CDF for this story.

Exercise 16.3. Surgical delivery. Twenty percent of babies in a particular

city are born by a surgical procedure called a Cesarean section (C-section). You
are interested in gathering information about hospital experiences of parents
who did not have their baby by C-section. You survey randomly parents of
babies from the population of the large city until you find the first one who had
a baby not born by C-section.

a. Explain in words what X is in this situation and what values it can take.
b. Why is this a Geometric distribution situation? What is the parameter?
c. What is the probability that you have to survey parents of at least 5
babies until you find the first one not born by C-section?
 16.6. Exercises 209

d. Given that you have to survey parents of more than 1 baby to find the
first one not born by C-section, what is the probability that you have to survey
at least 5 parents of babies?
e. What is the probability that you must survey between 4 and 6 (inclusive)
parents of babies to find one that was not born by C-section?
f. What is the expected value of X?
g. What is the variance of X?
h. Show the labeled graph of the mass for this story.
i. Show the labeled graph of the CDF for this story.

Exercise 16.4. Mobile phone survey. (Refer to Exercise 15.6) A recent

survey states that 48% of mobile devices are iPhones. In order to learn more
about how the iPhone works, a student starts asking his friends if they use an
iPhone. Let X be the number of friends he will need to ask until he finds one
who uses an iPhone.

a. What is the expected value of X?

b. What is the standard deviation of X?
c. Show the labeled graph of the mass for this story.
d. Show the labeled graph of the CDF for this story.

Exercise 16.5. Radio airplay. A certain radio station plays songs from the
1970’s, 80’s, and 90’s. We know that 20% of the songs on the station are from
the 70’s; 37% are from the 80’s; and 43% of the songs are from the 90’s. Let
X be the number of songs you listen to until you hear the first one from the
1990’s.

a. What is the expected value of X?

b. What is the standard deviation of X?
c. Show the labeled graph of the mass for this story.
d. Show the labeled graph of the CDF for this story.

Exercise 16.6. Homecoming. A young woman realizes that Homecoming is

quickly approaching, and she needs to find a date. She estimates that 72% of the
male students (in a large population) would be willing to accept her invitation.
She plans to start randomly asking men for a date until someone accepts.

a. What is the expected number of men she will need to invite until she has
a date for Homecoming?
b. Show the labeled graph of the mass for this story.
c. Show the labeled graph of the CDF for this story.
210 Chapter 16. Geometric Random Variables

Exercise 16.7. Chores. Sarah and Thomas play a card game to determine
who will have to take out the trash (the loser gets this unpopular chore) on
Monday. They use a standard 52-card deck by taking turns randomly drawing
cards from a shuffled deck, with replacement, until somebody draws an ace.
Whoever gets the ace does not have to do the chore.
a. How long do they expect to have to play the game?
b. What is the probability that it takes at least four cards for them to find
an ace?
c. If they play this game for six weeks, what is the probability it takes at
least four cards for them to find an ace on each of those 6 Mondays? What
distribution is being used here? What are the parameters?
Exercise 16.8. Random guessing. On a certain online math assignment, a
student is allowed to submit an unlimited number of answers before moving on
to the next problem. The problem is a free response question, and the student
believes that his guessed answer is correct about 7% of the time. (There are
so many possible guesses that he thinks this ratio stays the same every time.)
Under these assumptions, let X be the number of times he needs to submit a
different random guessed answer until he gets the question correct.
a. How many tries should he expect if he wants to get this question right
without actually learning the material?
b. What is the probability it takes him at least 20 tries to get the question
right?
c. If there are 10 homework questions of a similar nature on this assignment,
and he uses the same random technique with all of the questions, what is the
probability it takes him at least 20 tries to get each of these questions right?
What distribution is being used here? What are the parameters?
d. If each homework question takes him 2 minutes to read (once) and 30
seconds for each random entry attempt, what is the expected time it will take
him to get one question correct? What about the 10 question assignment? (Do
you think he would have been better off just learning the material and doing
the homework properly?)
Exercise 16.9. Shopping before Thanksgiving. Studies have shown that
28% of people do all of their Christmas shopping before Thanksgiving each year.
Let X be the number of people you have to ask until you find somebody who
has finished all of their Christmas shopping before Thanksgiving, assuming each
person is independent of the others.
a. What is the expected number of people you will have to ask until you
find somebody who has finished her Christmas shopping before Thanksgiving?
b. If each interview takes approximately 5 minutes, how much time should
you expect to have to spend total?
 16.6. Exercises 211

c. What is the probability that you will have to ask more than 10 people?
Exercise 16.10. Lucky Charms. Michael reaches into a very large box and
pulls out Lucky Charms. If percentages of the pieces are: 50% regular cereal,
6.25% each for hearts, stars, horseshoes, clovers, blue moons, pots of gold,
rainbows, and red balloons, and he only wants blue moons. Let X be the
number of individual pieces he has to pull out until he gets a blue moon.
a. What is the probability that X is more than 8?
b. What is the probability that X is at least 8?
c. What is the probability that X is less than 8?
d. What is the probability that X is at most 8?
Exercise 16.11. Vampire difficulties. Edward and his friend cannot go to
school when it’s sunny outside because they are vampires. Forks, Washington
is experiencing a sunny period these days, and there is an 80% chance each day
that it will be sunny. Assume that the weather is independent from day to day.
Let X be the number of days until the first day Edward can go outside.
a. What is the probability that Edward will have to wait between 10 to 12
days (inclusive) to go back to school?
b. Given that he has already waited 5 sunny days, what is the probability
that his waiting time altogether is between 10 to 12 days (inclusive) to go back
to school?
c. What is the expected number of days that he will have to wait to go back
to school?
Exercise 16.12. Blindfolded basketball. A basketball player shoots free
throws until he makes one. However, because his coach wants him to practice
his technique and the feel of the motions, his coach has the player blindfold
himself, which makes each throw independent from the others. When he is
blindfolded, he has only a 2% chance of making a free throw on any particular
try.
a. What is the probability that he will have to throw over 100 balls until he
makes his first basket?
b. What are the expected value and standard deviation for the number of
throws he will need to make until he gets his first basket?
c. What would you have to change about this story to turn it into a Binomial
question?
Exercise 16.13. Winning and losing. Suppose that a person wins a game
of chance with probability 0.40, and loses otherwise. He plays the game until
he wins for the first time, and then he stops. Assume that the games are
independent of each other. Let X denote the number of games that he must
play until (and including) his first win.
212 Chapter 16. Geometric Random Variables

a. How many games does he expect to play until (and including) his first
win?
b. What is the variance of the number of games he plays until (and includ-
ing) his first win?
c. What is the probability that he plays 4 or more games altogether?

Exercise 16.14. Winning and losing (continued) Continue to use the sce-
nario from the previous problem. As before, let X denote the number of games
that he must play until (and including) his first win. Assume that, if he wins,
he earns 5 dollars, and if he loses, then he loses 4 dollars. (Also assume that he
is allowed to borrow money, i.e., having a negative amount of money is not a
problem here.)

a. Find a formula for his gain or loss, in terms of X, i.e., if Y denotes his
gain or loss in dollars, write Y in terms of X.
b. What is his expected gain or loss (altogether) during the X games, i.e.,
what is E(Y )?
c. What is the variance of his gain or loss (altogether) during the X games,
i.e., what is Var(Y )?

Exercise 16.15. Dating. You randomly call friends who could be potential
partners for a dance. You think that they all respond to your requests indepen-
dently of each other, and you estimate that each one is 7% likely to accept your
request. Let X denote the number of phone calls that you make to successfully
get a date.

a. Find the expected number of people you need to call, i.e., E(X).
b. Find the variance of the number of people you need to call, i.e., Var(X).
c. Given that the first 3 people do not accept your invitation, let Y denote
the additional number of people you need to call (Y does not include those
first 3 people); i.e., suppose X > 3 is given, then let Y = X − 3. Under these
conditions, what is the mass of Y ?

Exercise 16.16. Hearts. You draw cards, one at a time, with replacement
(i.e., placing them randomly back into the deck after they are drawn), from a
shuffled, standard deck of 52 playing cards. Let X be the number of cards that
are drawn to get the first heart that appears.

a. How many cards do you expect to draw, to see the first heart?
b. Now suppose that you draw five cards (again, with replacement), and
none of them are hearts. How many additional cards (not including the first
five) do you expect to draw to see the first heart?
 16.6. Exercises 213

16.6.2 Extensions
Exercise 16.17. Telemarketers. Assume that when your phone rings, the
caller is a telemarketer with probability 1/8, and that the probability of a tele-
marketer is independent from call to call. Let X denote the number of tele-
marketers during the next three calls. If n is a nonnegative integer, what is
P (X > n)?

Exercise 16.18. Looking for a wife. Prince Charming has to go around town
asking if the glass slipper fits until he finds a woman whose foot fits properly in
the slipper so that he will know who to marry. (We do not endorse this technique
for finding a wife.) The probability of a glass slipper fitting a randomly selected
woman is 0.12, and the probabilities are independent for the various women.
Let X be the number of women he has to visit until he finds a woman who fits
the slipper. Assume that there are an unlimited supply of eligible women in the
town.

a. Given that he has already checked with 4 women without success, what
is the probability he will still need to check with at least 5 more? (He’s getting
impatient.)
b. Given that he has already checked with 4 women without success, what
is the probability he will succeed with the very next woman?
c. What is the probability he will succeed on his first try?
d. What is the expected number of women he will have to try?
e. If he takes an entourage with him everywhere he goes, and he has to pay
the entourage $100 for the day plus $10 for every visit he makes, how much
does he expect to pay if he does all his visits on one day?

Exercise 16.19. Geometric versus Binomial. Why does the memoryless

property work for the Geometric distribution but not for the Binomial distri-
bution?

16.6.3 Advanced
Exercise 16.20. Consider two independent Geometric random variables, X
and Y , with parameters p1 and p2 , respectively. Give a general formula (in terms
of p1 and p2 ) for the probability that X and Y are equal, i.e., for P (X = Y ).

Exercise 16.21. Use the probability mass function to justify the fact that, if
X is a Geometric random variable, then P (X > x) = q x . (In Section 16.2, we
justified this intuitively, but we did not use the PMF to prove it.)
 Chapter 17

Negative Binomial Random

Variables

The toughest thing about success is that you’ve got to keep on being a success.
—Irving Berlin

A coach wants to put together an intramural basketball team, from people

living in a large dorm. She estimates that 12% of people in the dorm like to
play basketball. She goes door to door to ask people if they would be interested
in playing on the team. How many dorm residents does she expect to interview
before finding 5 people to create the team? How is this expected value different
than if she was only trying to find one person to join a team? What is the
probability that she needs to talk to 20 people, in order to find 5 people who
will join the team?

17.1 Introduction
A Negative Binomial random variable is the number of independent trials re-
A Negative quired until a certain number of successes have occurred. For instance, a Nega-
Binomial random tive Binomial random variable could be the number of independent trials until
variable is the the 3rd success occurs. The successes do not have to be consecutive (they
number of usually are not). A Negative Binomial random variable can be interpreted as
independent trials
the sum of several independent Geometric random variables. For example, if
until r successes
have occurred. X, Y, Z are independent Geometric random variables with the same parameter,
then X + Y + Z is a Negative Binomial random variable for the number of trials
until the third success.

214
 17.1. Introduction 215

Negative Binomial random variables

The common thread: Number of independent trials, each with prob-
ability p of success, until the rth success occurs. For instance, you
randomly ask people one yes/no question each, until you get the rth “yes” re-
sponse. The successes do not need to be consecutive. Other examples include:
number of dice rolls until the third 5; number of cars observed until the tenth
black car; etc.
Things to look for: Repeated independent trials until rth success; same
success probabilities.
The variable:

X = # of trials until the rth success, so X ≥ r

X can’t be less than r because you need at least r trials to get r successes.
The parameters:

r = the desired number of successes

p = the probability that a given trial is a success
q = 1 − p = the probability that a given trial is a failure

These values p and q must be the same for every trial.

Mass:
 
x − 1 x−r r
P (X = x) = q p , x = r, r + 1, r + 2, . . .
r−1

Expected value formula:

E(X) = r/p

Variance formula:
Var(X) = qr/p2

Negative Binomial random variables have two parameters: p, the probability

of success of the independent trials, and r, the desired number of successes. The
probability of success p must stay the same on each trial. Negative Binomial
random variables are always r or larger, because it takes at least r trials to have
r successes.

Negative Binomial random variable = j if exactly r − 1 of X1 , . . . , Xj−1

are 1 (i.e., successes) and also Xj = 1

After each individual success, the search for the next success starts again,
independent of what came before. For this reason, a Negative Binomial with
parameters r and p is exactly the sum of r independent Geometric random
216 Chapter 17. Negative Binomial Random Variables

variables, each with parameter p. Therefore, the expected value and variance of
Negative Binomial random variables are easy to determine from the expected
value and variance of Geometric random variables.
For instance, if a Negative Binomial random variable X with r = 3 has the
value 13, this means that the 13th trial is a success, and exactly 2 of the earlier
12 trials are a success too. As an example:

The notation for the Negative Binomial looks like:

X ∼ NegBinomial(r, p)

where p is the probability of success on each trial, and we want r successes to

occur.
Where does the mass formula come from?
 
x − 1 r x−r
P (X = x) = p q
r−1
x − r = number of failures
q x−r = (probability of a failure on one trial)# of failures

pr = (probability of a success on one trial)# of successes

 
x−1
= number of possible arrangements of the successes
r−1
= # of ways to arrange r − 1 successes in x − 1 trials

We use x−1 x
r−1 instead of r because the last success is always the last trial,

so we only need to consider the ways to arrange r − 1 successes within the first
x − 1 trials.

17.2 Examples

Example 17.1. A coach needs to build a team of 5 basketball players quickly

for a campus league. She decides to ask randomly selected people whether they
played basketball in high school. Assume that 12% of the people have played
basketball in high school, that all people with basketball experience would be
willing to play on the team, and that the school is very large, so there is near
independence among the responses she gets.
 17.2. Examples 217

a. What does X represent in terms of this story? What values can it take?
The variable X is the number of people she has to ask until finding the 5th
member of the team, so X can be 5, 6, . . . (She needs to ask at least 5 people
since she is looking for 5 successes.)
b. Why is this a Negative Binomial situation? What are the parameters?
She is counting up number of trials until getting the 5th success, which is
a person who played basketball in high school (r = 5). Each trial (person) is
independent and each person has the same probability of success (p = 0.12).
c. What is the probability that she finds the 5th member of the team on
the 20th person she asks?

 
19
P (X = 20) = (0.12)5 (0.88)15 = 0.01418
4

d. What is the expected number of people she needs to ask in order to form
her team?

5
E(X) = = 41.6667 people
0.12

e. What is the standard deviation in the number of people she will need to
ask in order to form the team?

5(1 − 0.12)
Var(X) = 2
= 305.5556
√ 0.12
σX = 305.5556 = 17.4801 people

Example 17.2. Comparing Binomial and Negative Binomial. In the

basketball team example above, we compare two different situations:

What is the probability there will be

question 5 basketball players in the first 20
people she asks?
distribution Binomial (n = 20, p = 0.12)

meaning of X X = # of basketball players = 5

probability 20
(0.12)5 (0.88)15 = 0.0574


P (X = 5) = 5
218 Chapter 17. Negative Binomial Random Variables

What is the probability the 20th

question person she asks will be the 5th
player?
distribution Negative Binomial (r = 5, p = 0.12)

meaning of X X = # of people you ask = 20

probability 19
(0.12)5 (0.88)15 = 0.0143


P (X = 20) = 4

Remark 17.3. Expected value and variance of a Negative Binomial

random variable If X1 , X2 , . . . , Xr are independent Geometric random vari-
ables, each with parameter p, then

X = X1 + X2 + · · · + Xr

is a Negative Binomial random variable with parameters r and p. So

1 1 r
E(X) = E(X1 + · · · + Xr ) = E(X1 ) + · · · + E(Xr ) = + ··· + = ,
p p p
and
q q qr
Var(X) = Var(X1 +· · ·+Xr ) = Var(X1 )+· · ·+Var(Xr ) = 2
+· · ·+ 2 = 2 .
p p p

This makes sense when you think that a Negative Binomial random variable
is just the sum of r independent Geometric random variables. For the basketball
example, you have the sum of 5 Geometric random variables since you need 5
basketball players.

X1 X2 X3 X4 X5
z }| { z}|{ z }| {z }| { z }| {
person 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
answer N N Y Y N N N N N Y N N N N N N N Y N Y

Example 17.4. Roll a die until the third “5” appears.

X1 X2 X3
z }| { z}|{ z }| {
die roll 1 2 3 4 5 6 7 8
result 2 3 5 5 4 3 3 5
If we let X denote the number of rolls needed until the third “5” appears, then
X is a Negative Binomial random variable with parameters 3 and 1/6. We
could write X = X1 + X2 + X3 , where X1 is the number of rolls until the first
 17.3. Exercises 219

“5” appears, and X2 is the number of rolls after that point until the second
“5” appears, and X3 is the number of rolls after that point until the third “5”
appears, etc. Then each Xj is Geometric with parameter 1/6, and the Xj ’s are
independent.

Example 17.5. Dominique, Raymond, and Samantha are independently using

an app that randomly creates playlists for them. Forty percent of the songs this
app has access to are rap songs. Dominique randomly adds songs to her playlist
until she gets 5 rap songs. Raymond randomly adds songs until he gets 12 rap
songs. Samantha randomly adds songs until she gets 10 rap songs. What is
the distribution for the total number of songs if the three people combine their
playlists later?
The number of songs on each of their playlists are Negative Binomial random
variables X1 , X2 , X3 , each with probability of success (i.e., probability of a rap
song) p = 0.40, and with respective numbers of successes r1 = 5, r2 = 12, and
r3 = 10. So the total number of songs on their playlists is Y = X1 + X2 + X3 ,
which is also Negative Binomial, with probability of success p = 0.40 on each
trial, and R = 5 + 12 + 10 = 27 successes altogether.

Remark 17.6. Sums of Negative Binomial random variables

Suppose X1 , X2 , . . . , Xm are independent Negative Binomial random variables,
each with a common probability of success “p” on each trial, and for which
the desired numbers of successes are (respectively) r1 , r2 , . . . , rm . Then X1 +
X2 + · · · + Xm is the number of trials needed to reach r1 + r2 + · · · + rm
successes. Thus, the sum Y = X1 + X2 + · · · + Xm is also Negative Binomial,
with parameters R = r1 + r2 + · · · + rm and p. So the sum Y has mass
 
y − 1 y−R R
pY (y) = q p , for integers y ≥ R
R−1

and pY (y) = 0 otherwise.

17.3 Exercises
17.3.1 Practice
Exercise 17.1. Skittles. Skittles candies come in the colors red, orange, yel-
low, green, and purple, with each of these colors having equal probability. Due
to a dye mix-up there are a few rainbow-striped candies coming down the line.
There is a 5% chance that a candy is rainbow striped. You are a quality control
inspector, and your job is to find the 3rd rainbow-striped candy. (The popula-
tion is so big, and we randomly sample the Skittles, so we can assume relative
independence.)
220 Chapter 17. Negative Binomial Random Variables

a. Explain in words what X is in terms of the story. What values can it

take?
b. Why is this a Negative Binomial distribution situation? What are the
parameters?
c. What is the probability that you will need to check exactly 20 candies to
find the 3rd rainbow one?
d. What is the probability that you will have to check fewer than 6 candies
to find the 3rd rainbow one?
e. What is the probability you will have to check more than 6 candies to
find the 3rd rainbow one?
f. What is the expected number of candies you will have to check to find
the 3rd rainbow one?
g. What is the standard deviation of the number of candies you will have
to check to find your 3rd rainbow one?

Exercise 17.2. Colorblind. Approximately 8.33% of men are colorblind.

You randomly survey men from a large population until you find 2 who are
colorblind.

a. Explain in words what X is in this situation and what values it can take.
b. Why is this a Negative Binomial distribution situation? What are the
parameters?
c. What is the probability you will have to ask exactly 12 men to find the
2nd who is colorblind?
d. What is the probability that you will have to ask between 12 and 14 men
to find the 2nd who is colorblind?
e. What is the expected value of X?
f. What is the variance of X?
g. Show the labeled graph of the mass for this story. Draw an arrow to
indicate where the mean is.
h. Show the labeled graph of the CDF for this story.

Exercise 17.3. Surgical delivery. Twenty percent of babies in a particular

city are born by a surgical procedure called a Cesarean section (C-section). You
are interested in gathering information about hospital experiences of parents
who did not have their baby by C-section. You randomly survey parents of
babies from the population of the large city until you find 7 babies not born by
C-section.

a. Explain in words what X is in this situation and what values it can take.
 17.3. Exercises 221

b. Why is this a Negative Binomial distribution situation? What are the

parameters?
c. What is the probability that you will have to survey exactly 10 parents
to find the seventh one not born by C-section?
d. What is the probability that you have to survey parents of at least 10
babies to find the seventh one not born by C-section? (Hint: Think about the
complement and what X is allowed to be.)
e. What is the expected value of X?
f. What is the variance of X?
g. Show the labeled graph of the mass for this story.
h. Show the labeled graph of the CDF for this story.
Exercise 17.4. Consumer panel. A public relations intern realizes that she
forgot to assemble the consumer panel her boss asked her to do. She panics and
decides to randomly ask (independent) people if they will work on the panel for
an hour. Since she is willing to pay them for their work, she believes she will
have a 75% chance of people agreeing to work with her.
a. What is the expected number of people she will need to ask until she
finds 5 who will agree to participate?
b. What is the standard deviation in the number of people she will need to
ask until she finds 5 who will agree to participate?
c. Show the labeled graph of the mass for this story.
d. Show the labeled graph of the CDF for this story.
Exercise 17.5. Surfing. Harry is a surfer who can successfully ride about
70% of waves. Assume that his wave-riding ability is independent, from wave
to wave. Let X be the number of waves that pass until he successfully catches
his 8th wave. What is the probability that X is between 10 and 12 (inclusive)?
Exercise 17.6. Dart board. You throw darts at a board with 20 equally
likely spaces. You throw until you hit “1,” then you throw until you hit “2,”
then you throw until you hit “3,” etc., and finally you throw until you hit “20.”
Let X be the number of throws you make to achieve this goal.
a. What are the expected value and standard deviation of X?
b. Let Y be the number of throws needed until the 20th time hitting 1. Do
X and Y have the same distribution? Why or why not?
Exercise 17.7. Interviewing students. In the very large lecture class CHEM
115, 85% of students passed the first exam. I want to start a study group for
students who failed the first test, with 10 students in the group. Because of
FERPA privacy laws, I’m not allowed to get this information from the instruc-
tor. In surveying CHEM 115 students, let X be the number I have to interview
222 Chapter 17. Negative Binomial Random Variables

to find 10 who failed the first exam. What are the expected value and standard
deviation of X?

Exercise 17.8. Still looking for a wife. Twelve percent of single women
in the kingdom have feet which will fit into a glass slipper. Prince Charming
thinks he must continue finding women who fit such a slipper, so that he has
a collection to choose from. He would like 10 women who fit the slipper to
compete on a “Bachelor”-type show for his hand in marriage.

a. How many women should he expect to have to check until he finds 10

women who fit the glass slipper?
b. What is the probability he has to check 75 women?
c. Given that he has to check between 75 and 80 women, what is the
probability he has to check exactly 75 women?

Exercise 17.9. Replaying a video game. You vow to replay a tough level in
Super Mario World until you win 3 times. Assume that you have a 25% chance
of winning each time you play, and each round is independent (your skill does
not improve from game to game, because there is a lot of luck involved). Let X
denote the number of times you have to play. Each attempt takes 5 minutes.

a. What is the probability it will take you more than an hour to win 3
times?
b. How many minutes do you expect to need to win 3 times?

17.3.2 Extensions
Exercise 17.10. Missing an early class. My alarm clock wakes me up only
64% of the time. My probability class meets at 8:30 am, and if I don’t hear my
alarm, I’ll miss class. My teacher takes off one percent of our grade for every
class that we miss. Let X be the number of class days that pass until I have
lost 10% of my grade.

a. What is the expected value of X?

b. What is the probability that X is 30?
c. If there are 48 class days over the semester, what is the probability that
I will have lost only 9% of my grade? What distribution is this? How do you
know?
d. Let Y be the number of days until I lose my first percentage off my grade.
What distribution is this? How do you know?

Exercise 17.11. Zombies. During a zombie apocalypse, one human finds that
about 1 out of every 3 shots he makes actually kills a zombie. Let X be the
number of shots he has to take until he kills his fifth zombie.
 17.3. Exercises 223

a. Given that it takes between 14 and 16 shots (inclusive) to kill his fifth
zombie, what is the probability that it takes at least 15 shots?
b. Given that it takes him 12 tries to get his 4th success, what is the
probability that he will need exactly 3 more tries to get his 5th success? What
distribution is this? How do you know?

Exercise 17.12. Monopoly. Philip and Callum are playing Monopoly. Cal-
lum has an 80% chance of winning whenever he plays Philip. If they play
Monopoly until somebody wins 3 games (assuming no ties and that the games
are independent):

a. What is the probability that Callum wins the series in exactly 3 games?
In 4 games? In 5 games?
b. What is the probability Philip wins the series in exactly 3 games? In 4
games? In 5 games?
c. Given that the series took 5 games, what is the probability that Callum
won?
d. Given that Callum wins the series, what is the expected number of games
that the boys play?
 Chapter 18

Poisson Random Variables

Math is like war, people! If you fall behind in your unit, you will die!
—attributed to Mr. Williams, math teacher from Poland Seminary High
School in Poland, Ohio, as remembered by Catharine Patrone, Director of Stu-
dent Services for the Honors College at Purdue University

You are an epidemiologist trying to find people who have a rare disease you
would like to study. The disease is so rare that only 1 out of every 20,000
people have it. In a city of 100,000 people, what is the probability that exactly
4 people have the disease?

18.1 Introduction
How do you Instead of counting up the number of trials or the number of successes, Poisson
pronounce Poisson? random variables have a different motivation. When you know the average
Pwah-sow(n), with rate of occurrences of some event, the Poisson distribution is often correct for
the accent on the describing the number of events that actually occur. For instance, there might
second syllable (the
be a Poisson number of cars passing a building during a 1-hour period, or a
n is implicit).
Poisson is French for Poisson number of raindrops landing on a sidewalk square in five minutes, or a
“fish,” but it is Poisson number of shoppers in a store during a given 3.5-hour afternoon, etc.
actually named after
We use λ as the average number of occurrences of an event during a fixed
Siméon Poisson.
time period. We can think of λ as an average rate, because it depends on the
time period. For instance, suppose that there are an average of 10 cars passing
by a building in an hour. We use λ = 10 if we want the average rate of cars
passing by the building in an hour, but we use λ = 5 if we want to know the
average number of cars passing in a 30-minute period, or λ = 40 for the average
number of cars passing in a 4-hour period. To use Poisson random variables in
these simple settings, we must assume that the average rate is proportional to

224
 18.1. Introduction 225

the length of time we observe. This is a very different use of random variables
than we have encountered so far.
Poisson random variables can occur in other related ways too. For instance,
while reading a novel, the number of errors per page can be treated as a Poisson
random variable. In this case, the flow of words corresponds to the flow of time,
and an error on a page corresponds to an event.
The notation for a Poisson distribution looks like: The λ is the average
rate of events.
X ∼ Poisson(λ),
where λ denotes the average rate of events.
Poisson random variables
The common thread: How many events in a given period? You know
the average rate of events, i.e., number of events that occur in a period (this
average rate must be proportional to the length of the period). You count the
actual number of events that occur in such a time period. For instance, you
measure the number of cars that pass in a 40-minute period, the number of
raindrops that fall in one minute on a particular sidewalk square, the number
of times that cell phones in a room are ringing, etc.
Things to look for: A number of events, for which the average rate is known.
The variable:

X = # of events that occur during the specified period

The parameter:

λ = the average rate of events that occur during the specified period

Mass:
e−λ λx
P (X = x) = , x = 0, 1, 2, 3, . . .
x!
Expected value formula:
E(X) = λ
Variance formula:
Var(X) = λ
Poisson random
variables are not
poison.
Example 18.1. Let X denote the number of errors on a randomly selected
page of a book. Suppose that X has mass
e−0.2 (0.2)j
pX (j) = P (X = j) = .
j!
Then X is a Poisson random variable.
226 Chapter 18. Poisson Random Variables

In this case, for instance, the expected number of errors on a randomly selected
page is E(X) = 0.2, and the variance of the number of errors on a randomly
selected page is also Var(X) = 0.2.
For example, the probability of exactly 3 errors on a randomly selected page
of the book is
e−0.2 (0.2)3
pX (3) = = 0.00109.
3!

Example 18.2. Customers arrive at a checkout counter at an average rate of

8 per hour. The cashier will count how many come in a particular time frame.
a. Why is this story a Poisson situation? What is the parameter?
We are not looking for successes in individual trials. We are given a rate
(customers per hour), and we are asked to count up the number of customers
who come in a particular time frame. The average rate is λ = 8 customers/hour.
b. What is the probability that exactly 2 customers arrive in the next hour?
82 e−8
P (X = 2) = = 0.01073
2!
c. What is the probability that no more than 3 customers arrive in the next
hour?

P (X ≤ 3) = P (X = 0) + P (X = 1) + P (X = 2) + P (X = 3)
80 e−8 81 e−8 82 e−8 83 e−8
= + + +
 0!0 1! 2! 3!
8 81 82 83
= + + + e−8
0! 1! 2! 3!
= 126.333e−8
= 0.04238
d. Given that at least 1 customer arrives in the next hour, what is the
probability that more than 3 arrive? (Notice also that the Poisson is not mem-
oryless.)

P (X > 3 ∩ X ≥ 1)
P (X > 3 | X ≥ 1) =
P (X ≥ 1)
P (X > 3)
= (X > 3 & X ≥ 1 only if X > 3)
P (X ≥ 1)
1 − P (X ≤ 3)
= computing the complement
1 − P (X = 0)
1 − 0.04238
=
1 − 0.0003355
= 0.9579
 18.1. Introduction 227

In general, for Poisson random variables, if we need to compute P (X > a), and
we only have a calculator, then we need to compute the complement:

P (X > a) = 1 − P (X ≤ a) = 1 − P (X = 0) − P (X = 1) − · · · − P (X = a).
P∞
A direct calculation is infeasible because P (X > a) = j=a+1 P (X = j) is
an infinite sum that we cannot simplify (except by using the complement, as
suggested above).
e. What does the mass look like for this story?

pX (x)
0.14
0.12
0.10
0.08
0.06
0.04
0.02
x
10 20 30

f. What does the CDF look like for this story?

FX (x)
1.0
0.8
0.6
0.4
0.2
x
5 10 15 20 25 30

Example 18.2 (continued) Now we continue to study customer arrivals

that have a Poisson distribution, but we suppose that the length of time changes.
It is important to make sure that the expected number of arrivals in the time
interval is correct.
g. What is the probability that there will be exactly 14 customer arrivals in Be careful: Make
the next 2 hours? sure your units for λ
match the question
First, we have changed our time interval, so we need to change our λ to 16 you are being asked.
customers/2-hour interval.

1614 e−16
P (X = 14) = = 0.09302
14!
h. How many customers do you expect in the next 2 hours?

E(X) = λ = 16 customers
228 Chapter 18. Poisson Random Variables

i. What is the probability that there will be exactly 6 customer arrivals in

the next 20 minutes?

Since 20 minutes is 1/3 of an hour, we need to change the scale: we change

λ to 2.6667 customers/20-minute interval.

2.6676 e−2.6667
P (X = 6) = = 0.03472
6!

Example 18.3. What is the probability that there will be exactly 6 customer
arrivals in exactly one out of the next three 20-minute intervals? (Hint: You
have already done some of the work for this problem in the previous question.)

This is a Binomial problem now with n = 3 and with

p = the answer to Example 18.2i = 0.03472.

The random variable X is the number of 20-minute intervals in which exactly

6 customers arrive.

 
3
P (X = 1) = (0.03472)1 (0.96528)2 = 0.09705
1

Remark 18.4. Expected value of a Poisson random variable If X is

Poisson with average rate λ of events, then the expected value of the Poisson
random variable is
X X e−λ λj
E(X) = jP (X = j) = j .
j!
j≥0 j≥0

The j = 0 term is 0, so drop it, and this leaves

X e−λ λj X e−λ λj−1
E(X) = j =λ .
j! (j − 1)!
j≥1 j≥1

Then we shift the indices by 1, to obtain

X e−λ λj
E(X) = λ = λe−λ eλ = λ.
j!
j≥0
 18.2. Sums of Independent Poisson Random Variables 229

Remark 18.5. Variance of a Poisson random variable As with Geo-

metric random variables, we use X 2 = (X)(X − 1) + X to make the second
moment easier. We compute

Var(X) = E(X 2 ) − (E(X))2

= E((X)(X − 1)) + E(X) − (E(X))2
= E((X)(X − 1)) + λ − λ2 .

Now we think about how the simplification of j/j! = 1/(j − 1)! was performed
for the mean. We will use (j)(j − 1)/j! = 1/(j − 2)! below. We write
X X e−λ λj
E((X)(X − 1)) = (j)(j − 1)P (X = j) = (j)(j − 1) .
j!
j≥0 j≥0

The j = 0 and j = 1 terms are 0, so we drop them, and get

X e−λ λj X e−λ λj−2
E((X)(X − 1)) = (j)(j − 1) = λ2 .
j! (j − 2)!
j≥2 j≥2

Finally, we shift the indices by 2, to conclude

X e−λ λj
E((X)(X − 1)) = λ2 = λ2 e−λ eλ = λ2 .
j!
j≥0

Thus
Var(X) = λ2 + λ − λ2 = λ.

18.2 Sums of Independent Poisson Random Variables

Remark 18.6. The sum of independent Poisson random variables is Poisson
too. If X and Y are both Poisson random variables with parameters λ1 and λ2 ,
respectively, then we can show X + Y is also Poisson with parameter λ1 + λ2 .
We define Z = X + Y and λ = λ1 + λ2 , and it suffices to prove that Z has
mass
e−λ λj
pZ (j) = for j = 0, 1, 2, . . ..
j!
and pZ (z) = 0 otherwise.

To prove this, observe Z can only take on nonnegative integer values, since
X and Y are each known to take on only nonnegative integer values. Also, in
order to have Z = j, we need to have X = i for some i with 0 ≤ i ≤ j, and
then we also need Y = j − i. So we just compute
j
X j
X
pZ (j) = P (X = i and Y = j − i) = P (X = i)P (Y = j − i),
i=0 i=0
230 Chapter 18. Poisson Random Variables

where the last equality comes from the independence of X and Y . Moreover,
we know the masses of X and Y and can use them, as follows
j
X e−λ1 λi1 e−λ2 λj−i
2
pZ (j) = .
i! (j − i)!
i=0

Next we factor out e−λ1 −λ2 , and multiply and divide by j!. Using λ = λ1 + λ2 ,
this gives
j j  
e−λ1 −λ2 X j! i j−i e−λ X j i j−i
pZ (j) = λ λ = λ λ .
j! i!(j − i)! 1 2 j! i 1 2
i=0 i=0

By the Binomial theorem, we conclude

e−λ e−λ λj
pZ (j) = (λ1 + λ2 )j = .
j! j!
Thus, we have established the following:

Theorem 18.7. Sum of independent Poisson random variables

If X and Y are independent Poisson random variables with parameters λ1 and
λ2 , then X + Y is also Poisson with parameter λ1 + λ2 .

Applying this reasoning repeatedly, we get the following:

Corollary 18.8. Sum of independent Poisson random variables

If X1 , X2 , . . . , Xn are independent Poisson random variables with parameters
λ1 , λ2 , . . . , λn , then X1 + X2 + · · · + Xn is also Poisson with parameter λ1 +
λ2 + · · · + λn .

Example 18.9. With the knowledge that the sum of two independent Poisson
random variables is also a Poisson, we can return to Example 18.2g, and let Y
be the number of customers during hour 1, and Z be the number of customers
during hour 2. Then Y, Z are independent Poisson random variables, each with
mean 8. So X = Y + Z is the total number of customers during the two hours,
and X is also Poisson, with mean 8 + 8 = 16.

18.3 Using the Poisson as an Approximation to the

Binomial
The Poisson distribution can also be used as an approximation to the Binomial
distribution in certain cases. Here’s how it works.
 18.3. Using the Poisson as an Approximation to the Binomial 231

When does this work? If n is really big (say, for instance, n ≥ 1000) and p
is really close to 0 or really close to 1. How “big” and how “close” are needed?
If n is large and npq is near 1, or perhaps (as a rule of thumb) within a factor
of 10 away from 1, i.e.,
1/10 ≤ npq ≤ 10,
then the Poisson approximation to the Binomial is usually very appropriate.
For instance, if n = 10,000 and p = 10,000
1
, and q = 10,000
9999
, then nqp = 0.9999,
so the Poisson approximation to the Binomial should be good.
What parameters are involved? The expected value of a Poisson is λ, which
is an average rate. The expected value of a Binomial is n (number of trials)
times p (probability of success on a single trial). So we must set the expected
values equal.
How do you make the switch?

λ = np

We’re setting

E(X) for a Poisson = E(X) for a Binomial.

Remark 18.10. Poisson approximation to Binomial random variables

If X is a Binomial random variable with parameters n and p, and if n is a
really large positive integer, and if npq is somewhat close to 1, then many
of the attributes of X are approximately the same as the attributes of the
Poisson random variable with parameter λ = np.
(The reasoning behind Poisson approximation is explained in Remark 18.13.)

Example 18.11. Poisson approximation to the Binomial. In a city with

a population of 100,000, a person has a 0.00005 chance of developing a rare
disease. What is the probability that exactly 4 people have this disease?
a. If X is Binomial with n = 100,000 and p = 0.00005, then

 
100,000
P (X = 4) = (0.00005)4 (0.99995)99,996 = 0.175470002.
4

Our calculators don’t want the calculation above. Does yours? On the other
hand, we can do the approximation in the next part on any hand-held calculator,
and it agrees with the actual answer to several decimal places of accuracy.
b. If Y is Poisson with λ = np = (100,000)(0.00005) = 5, then

54 e−5
P (Y = 4) = = 0.175467370.
4!
232 Chapter 18. Poisson Random Variables

pX (x) FY (y)
0.18 0.18
0.16 0.16
0.14 0.14
0.12 0.12
0.10 0.10
0.08 0.08
0.06 0.06
0.04 0.04
0.02 0.02
x y
5 10 15 20 5 10 15 20

Figure 18.1: Left: The mass of X, a Binomial with n = 100,000 and p =

0.00005. Right: The mass of Y , a Poisson with λ = 5.

c. What is the expected number of people in this city who have this disease?
Using Binomial or Poisson, E(X) = E(Y ) = 5.
d. What do the masses look like for this story? Show for both the Binomial
and for the Poisson.
See Figure 18.1. As you can see from these graphs, the masses look practi-
cally identical for the Poisson and the Binomial with these parameters, and the
differences are extremely small.

Example 18.12. If there are n = 10000 words in a short story, and if each
word has a probability of p = 1/3000 of getting misspelled, then the number
of misspelled words in the entire short story is approximately Poisson, with
expected value np = 10/3 = 3.3333.

In this example, if we let X be the number of misspelled words, then X

is a Binomial random variable with n = 10000 and p = 1/3000. So the exact
probability
that there is at most

1 misspelled word on the page is pX (0) +
n 0 n 1
pX (1) = 0 p (1 − p) 10000 + 1 p (1 − p) 9999 = 0.15454. The number of errors,
however, is approximately a Poisson random variable Y with parameter λ =
np = 10/3. So the probability that there is at most 1 misspelled word on the
page in the story is approximately

e−10/3 (10/3)0 e−10/3 (10/3)1

pY (0) + pY (1) = + = e−10/3 (13/3) = 0.15459.
0! 1!
 18.4. Exercises 233

Remark 18.13. We briefly explain why a Poisson random variable Y with

parameter λ = np has many of the same properties as a Binomial random
variable X with parameters n and p, when n is large and npq is moderate.
The exact mass of X at j is
 
n j n−j (n)(n − 1)(n − 2) · · · (n − j + 1) j
pX (j) = p q = p (1 − p)n−j .
j j!

When n is much larger than j, then n and n − 1 and n − 2 and . . . and n − j + 1

are all approximately near n, and thus (n)(n − 1)(n − 2) · · · (n − j + 1) ≈ nj .
Also
n−j
 np n−j
(1 − p) = 1− .
n

n
Now we substitute λ = np. From calculus, we learned that 1 − nc → e−c
as n → ∞. Again, if j is much smaller than n, this means that

λ n−j
 
 np n−j
(1 − p) n−j
= 1− = 1− ≈ e−λ
n n

Putting all of this together, and again using np = λ, we have

 
n j n−j
pX (j) = p q
j
(n)(n − 1)(n − 2) · · · (n − j + 1) j
= p (1 − p)n−j
j!
nj e−λ λj
≈ pj e−λ = = pY (j)
j! j!
So the mass of X at j is approximately equal to the mass of Y at j.

18.4 Exercises
18.4.1 Practice

Exercise 18.1. Marriage licenses. According to the Guinness Book of World

Records (2005), there are 280 marriage licenses issued in Las Vegas per day. Let
X be the number of marriage licenses issued in the next 5 minutes.

a. What values can X take? Why is this a Poisson situation? What is the
parameter?
b. What is the exact probability that exactly 300 marriage licenses will be
issued tomorrow (a decimal approximation is not needed)?
c. What is the expected number of marriage licenses issued during the next
hour ?
234 Chapter 18. Poisson Random Variables

d. What is the standard deviation of the number of marriage licenses issued

in the next hour?
e. What is the probability that between 10 and 12 (inclusive) marriage
licenses will be issued in the next hour?
f. What is the probability that between 10 and 12 marriage licenses will be
issued in 12 out of the next 24 hours? What distribution are you using now and
what are the parameters?
Exercise 18.2. Left-handed. Approximately 6.85 left-handed people are
killed each day by using an object or machinery designed for right-handed peo-
ple. Let X be the number of left-handed people killed this way in one day.
a. What values can X take? Why is this a Poisson situation? What is the
parameter?
b. What is the probability that exactly 7 left-handed people will be killed
using a right-handed object tomorrow?
c. What is the expected number of left-handed people killed using a right-
handed object over the next week ?
d. What is the standard deviation of the number of left-handed people killed
using a right-handed object over the next week ?
e. What is the probability that at least 2 left-handed people will be killed
using a right-handed object tomorrow?
f. What is the probability that you will have to check more than 3 days
until you find the first day which has at least 2 left-handed people killed using
a right-handed object? What distribution are you using now and what is the
parameter?
Exercise 18.3. Laughing. Children laugh an average of 16.67 times an hour
and adults laugh an average of 0.6667 times an hour. Let X be the number
of times a child laughs in an hour, and let Y be the number of times an adult
laughs in an hour.
a. What is the probability that a child will laugh exactly 20 times in the next
hour? What is the corresponding probability that an adult will laugh exactly
20 times in the next hour?
b. Why are these situations Poisson?
c. Given that an adult laughs more than 1 time in an hour, what is the
probability that the adult laughs at least 3 times in the next hour?
d. Given that an adult laughs less than 3 times in the next hour, what is
the probability that the adult laughs once or less?
e. Show the labeled graph of the mass for adults in this story.
f. Show the labeled graph of the mass for children in this story.
 18.4. Exercises 235

g. Show the labeled graph of the CDF for adults in this story.
h. Show the labeled graph of the CDF for children in this story.
(The graphs for e and g should look relatively clear after just a few points,
but perhaps two dozen points will be needed to get a good understanding of the
graphs for f and h. A graphing calculator or computer might make the graphs
in f and h quicker to computer.)
Exercise 18.4. Pumpkin carving. According to the Guinness Book of World
Records (2005), the fastest pumpkin-carver on record, Steven Clarke, carved 42
pumpkins an hour. Assume this is his average rate. Let X be the number of
pumpkins Steven carves in an hour. (We suppose that the carver can steadily
maintain work at his record rate.)
a. Assume that this a Poisson situation. What is the parameter?
b. What is the probability Steven will carve exactly 40 pumpkins in the
next hour?
c. Given that he has carved at least 3 pumpkins in a 5-minute interval, what
is the probability that he will carve at least 4 pumpkins during that 5-minute
interval?
d. Given that he carves fewer than 4 pumpkins in a 5-minute interval,
what is the probability he carves fewer than 2 pumpkins during that 5-minute
interval?
e. What is the expected number of pumpkins he can carve in 3 minutes at
this pace?
f. Show the labeled graph of the mass for pumpkins carved in a 3-minute
interval.
g. Show the labeled graph of the CDF for pumpkins carved in a 3-minute
interval.
Exercise 18.5. Quadruplets. The probability of a mother giving birth to
quadruplets is 1 in 729,000. An obstetrician checks the records of 1,000,000
mothers, to see how many mothers of quadruplets are in her database.
a. Which distribution is technically appropriate in this situation? What are
the parameters?
b. Write a formula (using the distribution in part a) for the probability that
there are exactly 3 mothers of quadruplets, but do not solve.
c. What is the expected number of mothers of quadruplets in this sample?
d. Which distribution would be a good approximation to the distribution
in part a? What is the parameter?
e. Approximate the probability in part b by using the approximate distri-
bution in part d, and solve.
236 Chapter 18. Poisson Random Variables

Exercise 18.6. Roseate spoonbills. Roseate spoonbills are common along

the Gulf Coast of the U.S. See nationalzoo.si.edu/Animals/Birds/Facts/
fact-rosespoonbill.cfm for more information. Suppose that the number of
roseate spoonbills that fly overhead in 1 hour (at a nature preserve) has a Poisson
distribution with mean 2. Also suppose that the number of roseate spoonbills
is independent from hour to hour. For example, the number of birds between
noon and 1 PM does not affect the number of birds between 1 PM and 2 PM.

a. A bird watcher sits and looks for the birds for 3 hours. How many of
these birds does she expect to see?
b. What is the probability that she sees exactly 5 of these birds during a
3-hour time period?
c. What is the probability that she sees exactly 1 bird in each of 3 separate
1-hour time periods? Which distribution is this, and what are the parameters?

Exercise 18.7. Lottery. In a person’s lifetime, he enters a local daily raffle

10,000 times, but his chances of winning each time are only 1 in 5000.

Approximate the probability that he wins at least one time during his life-
time.

Exercise 18.8. Rare disease. There is a certain disease in America which

is very rare. Each person has a probability of 1 in 100,000,000 of having the
disease. Assume that there are 310,000,000 people in America.

a. Approximate the probability that nobody in America actually has the

disease.
b. Approximate the probability that at most 3 people (i.e., 3 or less) in
America have the disease.

Exercise 18.9. Cars. Cars pass an intersection at an average rate of 4 cars

per 5-minute interval. Let X be the number of cars that pass in the next 5
minutes.

a. Show the labeled graph of the mass for X.

b. Show the labeled graph of the CDF for X.

Exercise 18.10. Customers. Pedestrians walk by your store front frequently,

but they only come into your store about once every 10 minutes on average.
Let X be the number of pedestrians who enter the store in the next hour.

a. Show the labeled graph of the mass for X.

b. Show the labeled graph of the CDF for X.

Exercise 18.11. Beetles. Beetles walk across a lunch counter at an average

rate of 5 beetles per hour.
 18.4. Exercises 237

a. What is the probability that more than 2 beetles will cross the counter
in the next half hour?
b. What is the probability that more than 2 beetles will cross the counter
in each of the next two half hours?
c. What is the probability that more than 4 beetles will cross the counter
in the next hour?

Exercise 18.12. New car contest. People are competing to win a new car.
The task, which only 1 in 5000 people can achieve, is to hit a golf ball into a
cup 250 feet away. One hundred thousand people sign up to compete.

a. How many people will win, on average? What is the standard deviation
of the number of people who will win?
b. What is the probability that exactly 18 people win?
c. What is the probability that between 18 and 20 people win?

Exercise 18.13. Bakery. Customers arrive at a bakery at an average rate of

6 per half-hour. For simplicity, suppose each customer spends $2.50 on their
purchase.

a. What is the probability that exactly 3 customers will arrive in the next
10 minutes?
b. What is the probability that at least 3 customers will arrive in the next
10 minutes?
c. Given that at least 3 customers will arrive in the next 10 minutes, what
is the probability that exactly 3 will arrive?
d. What is the expected amount of money the bakery will make in the next
10 minutes?
e. What is the standard deviation in the amount of money the bakery will
make in the next 10 minutes?

Exercise 18.14. Toy defects. Workers at a factory produce a toy with a defect
about once every 4 hours on average. Each toy costs the factory approximately
$7 in labor and supplies.

a. What is the expected number of toys with defects at the end of a 40-hour
work week?
b. What is the standard deviation in the number of toys with defects at the
end of a 40-hour work week?
c. What is the expected cost to the factory for toys with defects at the end
of a 40-hour work week?
d. What is the standard deviation in cost to the factory for toys with defects
at the end of the 40-hour work week?
238 Chapter 18. Poisson Random Variables

Exercise 18.15. Toy defects (continued) Workers at a factory produce a

toy with a defect about once every 4 hours on average.

a. What is the probability there will be exactly 5 defects in 24 hours?

b. What is the probability that there will be exactly 5 defects in each of the
next seven 24-hour periods?
c. What is the probability there will be a total of 5 × 7 = 35 defects in the
next week (168 hours)? Why is this answer different from part b?

Exercise 18.16. Albino fish. There are approximately 11,000 fish in a lake.
Each fish has a 1 in 5500 chance of being albino. Let X be the number of albino
fish.

a. What is the expected number of albino fish in the lake?

b. What is the approximate probability that there will be exactly 4 albino
fish in the lake?
c. Given that there are at least 2 albino fish in the lake, what is the proba-
bility that there are exactly 4?

Exercise 18.17. Telemarketing. A telemarketing firm uses a computer con-

nected to an online phone book for Metropolis. A large number of messages are
sent, but on average only 10 people per hour listen to such messages. Let X be
the number of people who listen to these messages during a given hour.

a. What is the probability that there will be exactly 8 successful calls in the
next hour?
b. How many hours are expected to pass until the first hour with exactly 8
people listening to the messages? What distribution is this?

Exercise 18.18. Hungry customers. At a certain hot dog stand, during the
working day, the number of people who arrive to eat is Poisson, with an average
of 1 person every 2 minutes.

a. What is the probability that exactly 3 people arrive during the next 10
minutes?
b. What is the probability that nobody arrives during the next 10 minutes?
c. What is the probability that at least 3 people arrive during the next 10
minutes?

Exercise 18.19. Errors in a book. An author has carefully edited his book,
but as all careful readers know, all books have some errors. Suppose that the
number of errors per page is Poisson, with an average of 0.04 errors per page.

a. How many errors are expected in the next 100 pages?

b. What is the probability of exactly 5 errors in the next 100 pages?
 18.4. Exercises 239

Exercise 18.20. Telemarketers. Suppose that, on average, 3 telemarketers

call your house during a 7-day period.

a. What is the mass of the number of telemarketers calling your house

during one day?
b. What is the probability that no telemarketers will call your house tomor-
row?
c. What is the probability that exactly 2 telemarketers will call your house
tomorrow?

Exercise 18.21. Baseball fans. The number of Yankees fans shopping at a

sports store in a given hour is Poisson with mean 8. The number of Red Sox
fans shopping at the same store is Poisson with mean 6 per hour. Assume that
the numbers of fans of the two types are independent. In particular, there is no
person who is simultaneously a fan of both teams.

a. In a 3-hour period, how many Yankees and Red Sox fans do we expect
altogether?
b. Find the probability that exactly one person enters the store during the
next 20 minutes who likes the Yankees or Red Sox.

Exercise 18.22. Website visitors. Suppose that the number of men who
visit a website is Poisson, with mean 12 per minute, and the number of women
who visit the same site is also Poisson, with mean 15 per minute. Assume that
the number of men and women are independent.

a. During the next 10 seconds, what is the probability that 1 man and 2
women visit the site?
b. What is the expected number of people who visit the site in the next
5 minutes?
b. What is the variance of the total number of people who visit the site in
the next 5 minutes?

Exercise 18.23. Superfans. At the local stadium, there are 60,000 fans at-
tending a football game. It is well known that only a few people at the game
will be impartial (i.e., will not care about the outcome of the game). Suppose
that each person at the game has probability 10,000
1
of being impartial (and that
the impartiality of a fan has no bearing on the other fans).

a. Give an exact formula for the probability that 8 of the people at the game
are impartial. You do not have to evaluate the formula on your calculator.
b. Use a Poisson estimation for the probability above.
c. Use your calculator to evaluate the Poisson expression that you gave in
part b.
240 Chapter 18. Poisson Random Variables

Exercise 18.24. Shoppers. During the holiday rush, there are 100,000 shop-
pers in a certain region. Each of these shoppers is extremely likely to make a pur-
chase. Suppose that a person makes a purchase with probability 49,999/50,000
and declines to make a purchase with probability 1/50,000. Let X be the num-
ber of people who decline to make a purchase.

a. Give an exact formula for the probability that P (X ≤ 3). You do not
have to evaluate the formula on your calculator.
b. Use a Poisson estimation for the probability above.
c. Use your calculator to evaluate the Poisson expression that you gave in
part b.

18.4.2 Extensions
Exercise 18.25. Bakery (continued) If it costs the bakery owner $20 to keep
the bakery open each hour (staff, electricity, etc.), each customer spends $2.50,
and customers arrive at the bakery at an average rate of 6 per half-hour, will
the bakery be able to stay in business (i.e., will the bakery make more money
than it spends in an hour)?

Exercise 18.26. Verifying a mass. Verify that the mass of a Poisson random
variable is really a mass, i.e., verify that the terms of the mass sum to 1.

18.4.3 Advanced
Exercise 18.27. If X is a Poisson random variable with λ = 3, find E(5X ).
Chapter 19

Hypergeometric Random
Variables

And will you succeed? Yes indeed, yes indeed! Ninety-eight and three-quarters
percent guaranteed.
—Oh, the Places You’ll Go! by Dr. Seuss (Random House, 1990)

In a Lotto game, there are 40 balls labeled 1 to 40. You pick 5 different numbers
for your lottery ticket. That night on TV, the state lottery office will randomly
select 5 different balls (each with a unique number) to be the winning ticket.
You could win $1 million if you picked all 5 winning numbers, but you can win
smaller prizes for picking 3 or 4 correct numbers. What is the probability that
you pick all 5 winning numbers? What is the probability that you win a prize
of some sort? How many numbers do you expect to get right?

19.1 Introduction
A Hypergeometric random variable is the number of desirable items we pick
when selecting some items without replacement from a mixed collection of de-
sirable and undesirable items.
Hypergeometric random variables have three parameters, N, M, n:

N total items are available

M of the items are desirable
N − M of the items are undesirable
n items are selected

241
242 Chapter 19. Hypergeometric Random Variables

Hypergeometric random variables

The common thread: We choose some items from a collection, some
of which are desirable and the others are undesirable. We see how
many desirable items we get. For instance, we see how many peanut butter
cookies we get when we select 5 cookies from a jar containing peanut butter
and lemon cookies. Other examples include: We randomly choose some shoes
and see how many are blue; we see how many parts on an assembly line are
working correctly; etc.
Things to look for: We choose a collection of items without replacement and
see how many of the chosen items are desirable. The selections are dependent.
The variable:
X = # of desirable items selected

The parameters:

N total items are available

M of the items are desirable
N − M of the items are undesirable
n items are selected

Mass:
M N −M



x n−x
P (X = x) = N

.
n

Expected value formula:

M
E(X) = n
N
Variance formula:
 
M M N −n
Var(X) = n 1−
N N N −1

Where does the mass formula come from?

 
M
= number of ways to select x out of the M desirable items
x
 
N −M
= number of ways to select n − x out of the N − M undesirables
n−x
 
N
= number of ways to select n out of the N items altogether
n

Hypergeometric random variable = number of desirable items we received

= X1 + X2 + · · · + Xn ,
 19.2. Examples 243

where Xj = 1 if the jth item we choose is desirable. We emphasize that the

Xj ’s are dependent. This is the first named random variable that we have
seen that has dependent trials. Binomial, Geometric, and Negative Binomial
random variables all use independent trials with the same probability of success
on each trial.
The notation for the Hypergeometric looks like:

X ∼ Hypergeometric(M, N, n),

where X is the number of desirable items we select.

19.2 Examples

Example 19.1. In a box of 64 crayons, 60 are in good condition, and 4 are

broken. If a child randomly selects two crayons from the box, what is the
probability that both are in good condition? What is the probability that
exactly one of the two selected crayons is broken? What is the probability that
both are broken?

Let X be the number of crayons selected which are in good condition. Then X
is a Hypergeometric random variable with N = 64 crayons altogether, M = 60
crayons in good condition, and N − M = 4 crayons that are broken. The child
selects n = 2 crayons. So the probability that both crayons selected are in good
condition is
60 4



(60)(59)/2 (60)(59) 1770
pX (2) = P (X = 2) = 64
0 =
2
= = = 0.877976.
2
(64)(63)/2 (64)(63) 2016

The probability that exactly one of the two selected crayons is broken is
60 4



(60)(4) 240
pX (1) = P (X = 1) = 64
1 =
1
= = 0.119048.
2
(64)(63)/2 2016

The probability that both crayons selected are broken is

60 4



(4)(3)/2 6
pX (0) = P (X = 0) = 64
2 =
0
= = 0.002976.
2
(64)(63)/2 2016

Notice that these 3 probabilities add up to 1 because they cover every possible
way of selecting 2 crayons.
244 Chapter 19. Hypergeometric Random Variables

Example 19.2. A college student is running late for his class and does not
have time to pack his backpack carefully. He has 12 folders on his desk, 4 of
which include homework assignments due today. Without taking time to look,
he accidentally grabs just 3 folders from his stack. When he gets to class, he
counts how many of them contain his homework assignments. Assume that all
of the outcomes are equally likely.

a. Explain in words what X is in this story. What values can it take?

The random variable X is the number of homework assignment folders he
grabs. Thus, X can be 0, 1, 2, or 3 since only three folders are sampled. (Even
if more folders were to be sampled (i.e., if n was larger) the X could not—even
under these new assumptions—be larger than 4, since there are only 4 desirable
folders.)
b. Why is this a Hypergeometric distribution? What are the parameters?
Sampling without replacement; M = 4, N = 12, n = 3.
c. What is the probability 2 of them contain his homework assignments?
4

8

2 (6)(8)
P (X = 2) = 12

1 = = 0.2182
3
220

d. What is the probability at least 2 of the 3 folders contain his homework

assignments?

P (X ≥ 2) = P (X = 2) + P (X = 3)
4 8 4 8





2
=
1 +
12
3

0
12
3 3
= 0.21818 + 0.01818
= 0.2364

e. What is the probability that at least 1 of the folders out of the 3 he

grabbed contain his homework?
4

8

0 56
P (X ≥ 1) = 1 −
3 = 1 −
12 = 1 − 0.2545 = 0.7455
3
220

f. On a different day, in the same situation (this student should really invest
in a reliable alarm clock), the student grabbed 8 folders at random. What is
the probability he got all 4 homework assignment folders?
Now n = 8, but there are only 4 possible “successes” available.
4

8

4 (1)(70)
P (X = 4) = 12

4 = = 0.1414
8
495
 19.2. Examples 245

g. When grabbing the 8 folders, what is the probability he grabbed at least

4 homework assignment folders?
This will be the same as the answer for part f since there are only 4 successes
possible.
h. When grabbing 8 folders, what is the average number of homework
assignments he will get?
 
4
E(X) = 8 = 2.6667 homework assignments
12

i. What is the standard deviation in the number of homework assignments

he will get when he grabs 8 folders?

   
4 8 12 − 8
Var(X) = 8 1− = 0.3232
12 12 12 − 1
√
σX = 0.3232 = 0.5685 homework assignments

Example 19.3. Suppose that there are 100 roseate spoonbills altogether in
Indiana. Also suppose that 40 of them have been observed in 2009 by an or-
nithologist (a computerized tagging system allows the observer to know when
the observances are unique). In 2010 he observes 32 birds. What is the proba-
bility that j out of these 32 birds were already seen previously, in 2009?

In this case, there are N = 100 roseate spoonbills altogether. The “desirable”
birds are the ones that were previously seen in 2009, i.e., M = 40; the undesir-
able birds are the ones that were not seen in 2009, i.e., N − M = 60. We use X
to denote the number of the birds out of the n = 32 seen in 2010 which had
been previously seen in 2009. So the probability that exactly j of the 32 birds
this year were seen in 2009 is
40 60



j 32−j
pX (j) = P (X = j) = 100

.
32

Remark 19.4. Expected value and variance of a Hypergeometric ran-

dom variable If X is Hypergeometric, then X can be viewed as the sum of
Bernoulli random variables that are dependent, where Xj = 1 if the jth
item selected is desirable, and Xj = 0 otherwise. Thus

X = X1 + X2 + · · · + Xn .
246 Chapter 19. Hypergeometric Random Variables

Any of the N items is equally likely to be chosen on the jth draw, and exactly
M of them are desirable, so E(Xj ) = M/N for each j. (This is worrisome to
some students at first, but just keep in mind that we are momentarily only
focused on the jth draw.) Even though the Xj ’s are dependent, the expected
value is still linear, i.e., the expected value of the sum is equal to the sum of
the expected values:

M M M
E(X) = E(X1 + · · · + Xn ) = E(X1 ) + · · · + E(Xn ) = + ··· + =n .
N N N

For the second moment:

E(X 2 ) = E((X1 + · · · + Xn )(X1 + · · · + Xn ))

Xn X
= E(Xj Xj ) + 2 E(Xi Xj )
j=1 1≤i<j≤n

Each Xj is either 0 or 1, so Xj Xj = Xj always. Thus E(Xj Xj ) = E(Xj ) =

M/N .
Also, for i 6= j, we see Xi Xj = 1 if and only if Xi = 1 and Xj = 1. So

E(Xi Xj ) = P (Xi Xj = 1)
= P (Xi = 1 and Xj = 1)
= P (Xi = 1)P (Xj = 1 | Xi = 1).

As before, P (Xi = 1) = M/N . Given that Xi = 1, there are M − 1 desirable

items out of N − 1 other items that could be chosen on the jth draw, so

P (Xj = 1 | Xi = 1) = (M − 1)/(N − 1).

There are n terms of the form E(Xj Xj ). The other n2 − n terms are of the form
E(Xi Xj ). So

M M M −1
E(X 2 ) = n + (n2 − n) .
N N N −1

Thus

M 2
 
2 M 2 2 M M −1
Var(X) = E(X ) − (E(X)) = n + (n − n) − n ,
N N N −1 N

which simplifies to
 
M M N −n
Var(X) = n 1− .
N N N −1
 19.3. Binomial Approximation to the Hypergeometric 247

19.3 Binomial Approximation to the Hypergeometric

The Binomial distribution can be used as an approximation to the Hypergeo-
metric distribution in certain cases.
When does this work? If N is really big (e.g., in the hundreds or thousands,
such an approximation works great) and n is relatively small, compared to N .
Remember that N is the population size and n is the sample size.
Why does this work? If your population size is really big, then even though
you are sampling without replacement for a Hypergeometric, your population
doesn’t really notice that you’ve taken a sample. The probability of getting a
success on your second selection is approximately the same as your probability
of getting a success on your first selection. For a big population, sampling with
replacement and sampling without replacement give similar results because it
is very unlikely that the same person or item would be selected twice.
What parameters are involved?
Binomial needs an n (number of trials) and a p (probability of success on a
single trial).
Hypergeometric has the number of trials n already, and

M # successes in population
= ≈p
N total # in population

for a large population.

Example 19.5. Binomial approximation to the Hypergeometric. In a

state lottery, one million tickets are issued, out of which only 50,000 are winning
tickets. If a man buys 10 tickets, what is the probability exactly one of them is
a winning ticket?

a. As a Hypergeometric, n = 10, N = 1,000,000, M = 50,000,

50,000 950,000



1 9
P (X = 1) = 1,000,000

.
10

Our calculators do not want to do this! Does yours?

50,000
b. As a Binomial, n = 10, p = MN = 1,000,000 = 0.05,
 
10
P (X = 1) = (0.05)1 (0.95)9 = 0.3151.
1

c. As a Hypergeometric, what is the expected number of winning tickets

this man will have? What is the variance?
248 Chapter 19. Hypergeometric Random Variables

 
50,000
E(X) = 10 = 0.5
1,000,000

   
50,000 50,000 1,000,000 − 10
Var(X) = 10 1−
1,000,000 1,000,000 1,000,000 − 1
= 10(0.05)(0.95)(0.999991)
= 0.4749957.

d. As a Binomial, what is the expected number of winning tickets this man

will have? What is the variance?

E(X) = 10(0.05) = 0.5

Var(X) = 10(0.05)(0.95) = 0.475

The answers for Binomial and Hypergeometric are almost exactly the same!
The masses for the Hypergeometric and for the Binomial approximation in
this story agree to five decimal places of accuracy.

19.4 Exercises
19.4.1 Practice
Exercise 19.1. Vending machine. You are hungry and decide to go to
patronize your office’s vending machine. There are 6 bags of potato chips, 7
bags of pretzels, and 5 packs of chocolate chip cookies. Unfortunately there is
something wrong with the buttons on the vending machine, and it will not let
you type in your selection. However, it will drop 3 snacks out at random. You
are hoping for 2 bags of cookies, and you don’t care what the other snack will
be.
a. What is a success in this story? What is a failure?
b. Explain in words what X is in terms of this story. What values can it
take?
c. Why is this a Hypergeometric situation? What are the parameters?
d. What is the probability you get the 2 bags of cookies?
e. What is the probability that you get at most 1 bag of cookies?
f. What is the expected number of bags of cookies you will get?
g. What is the standard deviation in the number of bags of cookies you will
get?
 19.4. Exercises 249

h. What is the expected number of bags of pretzels you will get?

i. What is the expected number of bags of potato chips you will get?
Exercise 19.2. SUV. You are going on a trip with some friends, and you need
to rent 4 large cars to get all of you to the beach. There are 15 SUVs and 10
minivans on the lot. The rental car dealer will pick 4 cars at random. You are
hoping to get SUVs because you think they look more stylish.
a. What is a success in this story? What is a failure?
b. Explain in words what X is in terms of this story. What values can it
take?
c. Why is this a Hypergeometric situation? What are the parameters?
d. What is the probability you get 3 SUVs?
e. Given that you get at least 1 SUV, what is the probability you get at
least 3 SUVs?
f. What is the expected number of SUVs you will get?
g. What is the standard deviation in the number of SUVs you will get?
h. What is the expected number of minivans you will get?
i. What is the standard deviation in the number of minivans you will get?
Exercise 19.3. Skittles. Skittles candies come in the colors red, orange, yel-
low, green, and purple. Green is your favorite color. Assume that the colors are
evenly distributed in a bag of 45 candies. You will eat 10 candies, counting the
number of green ones as you eat.
a. What is a success in this story? What is a failure?
b. Explain in words what X is in terms of this story. What values can it
take?
c. Why is this a Hypergeometric situation? What are the parameters?
d. What is the probability you eat more than 2 green candies?
e. Given that you eat more than 2 green candies, what is the probability
you eat exactly 9 green candies?
f. What is the expected number of green Skittles you will get?
Exercise 19.4. Bouquet. You are picking a bouquet of 20 flowers for your
mother at random from a garden with 25 dahlias, 35 daisies, and 42 bachelor’s
buttons. Your mother likes daisies the best, and you want to end up with 15 of
those in your bouquet.
a. What is a success here? What is a failure?
b. Explain in words what X is in terms of this story. What values can it
take?
250 Chapter 19. Hypergeometric Random Variables

c. Why is this a Hypergeometric situation? What are the parameters?

d. What is the probability you pick exactly 15 daisies in your bouquet?
e. What is the probability that you end up with all daisies in your bouquet?
f. What is the expected number of daisies you will pick?
Exercise 19.5. Mislabeled DVDs. Vera wants to buy 20 copies of the DVD
of “Bridget Jones’ Diary” as a graduation gift for the women in her sorority. She
buys them from a large internet company with 50,000 of the DVDs in stock.
However 10,000 of the DVDs labeled as “Bridget Jones” were really “Terminator”
and just mislabeled, leaving only 40,000 correctly labeled DVDs.
a. Explain in words what X is in terms of this story. What values can it
take?
b. Why is this a Hypergeometric situation? What are the parameters?
c. What are the expected number of movies she receives which are truly
“Bridget Jones”?
d. What is the probability that 18 of the DVDs she purchases are actually
“Bridget Jones” and not “Terminator”? Just set up the equation, but don’t try
to solve it.
e. What would be a good approximation to use in this situation? What are
the parameters?
f. What is the approximate probability that 18 of the DVDs she purchases
are actually “Bridget Jones” and not “Terminator”? Set up the equation and
solve.
Exercise 19.6. Cat fur vaccine. Dr. I.M.A. Genius has developed an exper-
imental vaccine for an allergy to cat fur. He needs to test it out on some people
to make sure it works. He has a list of 4000 people from his clinic, and 1250 of
those have an allergy to cat fur. His receptionist randomly calls in 50 different
people from his list, and he checks how many of them have cat fur allergies.
a. Explain in words what X is in terms of this story. What values can it
take?
b. Why is this a Hypergeometric situation? What are the parameters?
c. What are the expected number of people in his sample who have cat fur
allergies?
d. What is the probability that exactly 20 of the people in his sample have
cat fur allergies? Just set up the equation, but don’t try to solve it.
e. What would be a good approximation to use in this situation? What are
the parameters?
f. What is the approximate probability that 20 of the people in his sample
have cat fur allergies? Set up the equation and solve.
 19.4. Exercises 251

Exercise 19.7. Student volunteers. You have 5 freshmen, 6 sophomores, 10

juniors, and 2 seniors in your probability class. The teacher randomly selects 3
students to go to the board to do problems.

a. What is the probability all 3 students will be juniors?

b. What is the probability all 3 students will be from the same year?
c. What is the probability that at least one student will be a junior?
d. Given that at least one student will be a freshman, what is the probability
that all 3 students will be freshmen?
e. What is the expected number of juniors selected each day?
f. What is the standard deviation of the number of juniors selected each
day?
g. Assume that, every day, the teacher uses the same random method of
selecting 3 students from the whole class. It is possible for the same student to
be called on 2 days in a row but not twice in one class. What is the probability
that the teacher will select only juniors for 2 days in a row? What distribution
is this?

Exercise 19.8. Popsicles. In your freezer, there is a box of popsicles with 5

grape, 5 cherry, and 5 lime. Your freezer light isn’t working, so you reach in and
randomly pull out 4 popsicles for you and your friends on a hot afternoon. You
are hoping to get all 4 the same flavor so that there won’t be fighting. What is
the probability of this happening?

Exercise 19.9. Ramen noodles. There are 20 packs of ramen noodle pack-
ages in a variety pack box, with 10 chicken and 10 beef flavored. What is the
probability that you grab one of each when you reach in to pull out two packages
of noodles without looking?

Exercise 19.10. Playlist. Suppose you have a playlist called “I Love the 90s”
which contains 35 songs, including 10 tracks from the Spin Doctors’ “Pocket
Full of Kryptonite.” If I shuffle the songs, what is the probability that 3 songs
from the album (no repeats) shuffle to the top 5?

Exercise 19.11. Random beverages. At a barbecue there is a cooler with

50 beverages inside: 30 bottles of water and 20 cans of soda pop. I grab 4
beverages out of the cooler at random to share with some friends.

a. What is the probability of 3 of those 4 beverages being water?

b. What is the expected number of water bottles that I grab out of 4 total
bottles?

Exercise 19.12. Granola bars. I have 6 chocolate chunk granola bars, 10

raspberry granola bars, and 8 chocolate chip granola bars. I grab 3 without
looking.
252 Chapter 19. Hypergeometric Random Variables

a. What is the probability that I grab exactly 2 that are chocolate (either
chunk or chip)?
b. What is the probability that I grab fewer than 2 that are chocolate?
c. What is the expected number of chocolate granola bars out of the 3 that
I grab?

Exercise 19.13. Lucky Charms. Michael reaches into a bowl of 100 pieces of
Lucky Charms cereal. He desires blue moons, which make up 5% of the charms.
If his cereal bowl contains 40 charms, what is the probability he won’t get any
blue moons? What number of blue moons should he expect to get?

Exercise 19.14. Random pants. Henry has 10 pants: 4 dress slacks and 6
jeans. In a hurry while getting ready for a trip, he asks his kids to throw 3
pants in a suitcase for him without specifying which kind of pants he needs.

a. What is the probability the kids correctly throw in 2 pairs of dress slacks
and 1 pair of jeans?
b. What is the probability the kids throw in 2 pairs of jeans and 1 pair of
dress slacks?
c. Are the probabilities in parts a and b the same? Do they add up to 1?
Should they add up to 1? Why or why not?

Exercise 19.15. Corn and beans. As a prank, your roommate removed all
the labels from all the cans in your pantry and shuffled them around. Now you
have no idea what is in a can when you get ready to cook dinner! You know
that you have 8 cans of corn and 5 cans of beans. You don’t want to waste any
of the cans, so the best thing to do is to randomly open 2 cans every night and
just eat whatever is in those cans.

a. What is the probability that on the first night you get one of each type
of can?
b. If you got one of each type of can on the first night, what is the probability
you got one of each type of can on the second night?
c. Are these probabilities in parts a and b the same? Should they be? Why
or why not?

Exercise 19.16. Hungry customers. At a certain restaurant, during the

working day, there are 12 customers. Seven of them have pizza, and the other
five have burgers. Suppose that a person is conducting a survey of three cus-
tomers at the restaurant, and he conducts the survey without replacement, i.e.,
he does not talk to the same customer more than once. All possible selections
of three people for the survey are equally likely.
Let X be the number of people in the survey who are eating pizza.
 19.4. Exercises 253

a. What is the mass of X?

b. Evaluate the mass at the four values where the mass is strictly positive.
(Make sure that your four numbers sum to 1.)
c. What is the average number of people in the survey who are eating pizza?

Exercise 19.17. Harmonicas. Carlos “Coyote” Jones owns quite a few har-
monicas. In particular, he has 7 professional harmonicas and 12 cheaper har-
monicas. They have relatively similar shapes, so when he reaches into his har-
monica container without looking, he does not notice a difference between them.
Suppose that he grabs 8 harmonicas, without replacement, and all selections are
equally likely.

a. How many professional harmonicas does he expect to select?

b. What is the variance of the number of professional harmonicas that he
selects?
c. What is the probability that exactly 5 out of the 8 harmonicas are pro-
fessional?
d. What is the probability that he selects exactly 5 professional harmonicas,
given that he selects at least 5 professional harmonicas?

19.4.2 Extensions
Exercise 19.18. Capture-recapture sampling. A wildlife biologist is trying
to estimate population size of a pack of hyenas. She tags 20 hyenas and then
releases them back into the wild. She revisits the same area 1 year later, and
examines 50 hyenas. She notes that 10 of the hyenas have the tags from 1 year
earlier. Estimate the population size, and explain your reasoning. Why would
this be considered a form of Hypergeometric distribution?

Exercise 19.19. Spades. Given a standard deck of cards, you are dealt a
hand of 13.

a. What is the probability that more than half the spades in the deck wind
up in your hand?
b. How many spades are you expected to get in this hand?
Chapter 20

Discrete Uniform Random

Variables

Uncertainty and expectation are the joys of life.

—Love for Love, Act IV, Scene XX by William Congreve (play that pre-
miered on April 30, 1695)

When did you first start thinking about probability when you were young? Was
it while playing a board game that involved rolling a die? A die roll is a fairly
simple use of probability because each side is equally likely to come up. The
values 1 to 6 are equally likely to appear.

20.1 Introduction
Think about rolling a die 200 times. We did this, and we show the results below.
Our results should show a fairly similar number of 1s, 2s, 3s, 4s, 5s, and 6s.
die value 1 2 3 4 5 6
number of occurrences 36 36 28 31 38 31
percent of occurrences 0.18 0.18 0.14 0.155 0.19 0.155
Using the computer to simulate rolling a million dice, the results might be
the following:
die value 1 2 3 4 5 6
occurrences 166863 166786 166513 166890 166753 166195
percent 0.1669 0.1668 0.1665 0.1669 0.1668 0.1662
(The more we roll, the more evenly distributed the results would be.) The
result of a single die toss follows the Discrete Uniform distribution because each
of the 6 possible outcomes is equally likely to occur.

254
 20.1. Introduction 255

For our purposes, we’ll assume that any Discrete Uniform distribution, if
not numbered consecutively with integers already, could be relabeled this way.
The Discrete Uniform is one of the simplest distributions.
Notation for the Discrete Uniform looks like:

X ∼ D.Uniform(N )

Discrete Uniform random variables

The common thread: We choose one item out of a collection, to see
which one we get. All possibilities are equally likely. For instance,
the number of tickets (out of a fixed set of 157 shuffled tickets) that we need
to check until we find the unique one with the special bonus prize. Other
examples include: We choose exactly one student in a classroom; one book
from a bookshelf; one employee in a company; etc.
Things to look for: We know the number of possible outcomes, all equally
likely. We record which outcome is selected.
The variable:
X = the choice of the outcome
You usually order your outcomes and number them 1, 2, . . . , N .
It is also possible to have a Discrete Uniform distribution with X having values
between a and b (an example would be raffle tickets numbered between 120
and 185). You could work with the numbers as they are or you could re-label
these tickets as 1 to 66.
The parameters:

N = the number of possible outcomes.

Mass:
1
P (X = x) = x = 1, 2, . . . , N
N
Expected value formula:
N +1
E(X) =
2

Variance formula:
N2 − 1
Var(X) =
12

Where does the mass formula come from? All of the pX (x) values are the
same, and we know that they also sum to 1:

pX (1) + pX (2) + · · · + pX (N ) = 1.
256 Chapter 20. Discrete Uniform Random Variables

So we must have pX (x) = 1/N for each 1 ≤ x ≤ N .

Note: For some of these Discrete Uniform distribution situations, the ex-
pected value and the variance won’t make much sense to the story. The Gilbreth
family example below is one of those, but the die tossing example does make
sense.

20.2 Examples

Example 20.1. The Gilbreth family (from Cheaper by the Dozen fame) has
12 children (in order from oldest to youngest): Anne, Ernestine, Mary, Martha,
Frank, Bill, Lillian, Fred, Dan, Jack, Bob, and Jane. Mr. Gilbreth needs some
help with a project, so he whistles for a child to come help him. Each of the
children is equally likely to appear. We are numbering the children so that 1 is
for Anne (the oldest) and 12 is for Jane (the youngest).

a. What is the probability Anne comes to help?

P (X = 1) = 1/12 = 0.08333
b. What is the probability Bob comes to help?
P (X = 11) = 1/12 = 0.08333
c. What is the probability Martha does not come to help?
P (X 6= 4) = 1 − P (X = 4) = 11/12 = 0.91667
d. Why is this a Discrete Uniform distribution, and what is the parameter?
All 12 outcomes are equally likely, and we are only selecting one at random.
N = 12, the total number of possible outcomes.
e. Explain in words what X is in terms of the story. What values can it
take on?
The random variable X is the age order of the person who comes to help.
So X can be 1, 2, . . . , 12.

Example 20.2. For a die toss on a fair, 6-sided die, the value of the roll is a
Discrete Uniform, between 1 and 6. Let X be the value of a die roll.

a. E(X) = (N + 1)/2 = (6 + 1)/2 = 3.5

b. Var(X) = (N 2 − 1)/12 = (62 − 1)/12 = 35/12 = 2.9167
c. The mass and CDF of X were already shown back in Figures 8.1 and 8.2.
 20.3. Exercises 257

d. The experimental results above give very similar results to the theoretical
distribution. As the number of experiments (here, number of rolls) grows, the
closer the experimental results will get to the theoretical results (this will be
investigated further in Chapter 37, when we study the Central Limit Theorem).

Example 20.3. As in Example 10.8, a student shuffles a deck of cards thor-

oughly (one time) and then selects cards from the deck without replacement
until the ace of spades appears. If we let X denote the number of cards that the
student expects to draw, until the ace of spades appears, then X is a Discrete
Uniform with n = 52, and P (X = j) = 1/52 for 1 ≤ j ≤ 52.

20.3 Exercises
20.3.1 Practice
Exercise 20.1. Skittles. Skittles candies come in 5 different colors: red,
orange, yellow, green, and purple. You have a bowl of the candies, so you reach
in and grab one. Each of the five candies is equally likely to appear.

a. Why is this a Discrete Uniform distribution situation? What is the

parameter?
b. What is the probability the candy you grab is a purple one?
c. Explain in words what X is in terms of the story. What values can it
take if you number the candies in the order listed above?
d. What is the probability the candy you grab is not a purple one?
e. Show the labeled graph of the mass for the colors of candies.
f. Show the labeled graph of the CDF for the colors of candies.

Exercise 20.2. New ID. A computer is randomly selecting a 4-digit ID num-

ber for you, and there are 10,000 possibilities, each equally likely. (0000 would
be the smallest possible number, and 9999 would be the biggest.)

a. What is the probability your new ID number happens to match your

birthdate (e.g., 0822 for August 22nd)?
b. Why is this a Discrete Uniform distribution situation? What is the
parameter?
c. Explain in words what X is in terms of the story. What values can it
take?
d. What is the average ID number?
258 Chapter 20. Discrete Uniform Random Variables

e. What is the standard deviation of the ID numbers?

f. Describe in words what the graph of the mass for the ID numbers would
look like.
g. Describe in words what the graph of the CDF for the ID numbers would
look like.

20.3.2 Extensions
Exercise 20.3. Let X be a Bernoulli random variable with p = 1/2. Let Y
be Discrete Uniform on {1, 2}. Explain (intuitively) why they have different
expected values but the same variance. No calculation should be needed.

20.3.3 Advanced
Exercise 20.4. Prove that the expected value of a Discrete Uniform random
variable on the set {1, 2, . . . , N } is E(X) = (N + 1)/2. Also prove that the
variance is Var(X) = (N 2 − 1)/12.
Chapter 21

Review of Named Discrete

Random Variables

21.1 Summing-up: How To Tell Random Variables

Apart

These distributions are all starting to sound confusingly similar, aren’t they? To
help you sort them all out, we present Frequently Confused Distributions:
Bernoulli vs. Geometric vs. Binomial
• Bernoulli is a single yes/no trial. We use X = 1 if you get a success, or
X = 0 if you get a failure.
• Geometric is continuing to do more independent Bernoulli trials until
you get your 1st success. The X is the number of trials you have to do.
• Binomial is doing a pre-set number of independent Bernoulli trials and
then counting up the number of successes. The X is the number of suc-
cesses.
• Below is a comparison of results for 1 success in 5 trials for the Geometric
and Binomial distributions.
Binomial Geometric
Y N N N N
N Y N N N
N N Y N N
N N N Y N
N N N N Y NNNNY
5
= 5 arrangements only 1 arrangement since the


1
success must come on the last trial

259
 Summary of Named Discrete Random Variables
Name Mass Expected value Variance Parameters What X is When used
Bernoulli pX (1) = p p pq p = prob. 0 or 1 1 success
pX (0) = q succ./trial (no or yes) or failure
n = # trials; 0, 1, 2, . . . , n successes
Binomial n
px q n−x p = prob. (successes) in n trials


x np npq
succ./trial
Geometric q x−1 p 1/p q/p2 p = prob. 1, 2, 3, . . . # trials
succ./trial (trials) to 1st succ.
p = prob. r, r + 1, . . .
Negative x−1
q x−r pr qr/p2 succ./trial; # trials


r−1 r/p
Binomial r = # of (trials) to rth succ.
260

succ. needed
Poisson e−λ λx/x! λ λ λ rate 0, 1, 2, 3, . . . # events
(events) in period
Hyper- M good, 1, 2, . . . , M # of good
−M
(Mx )(Nn−x )
geometric nM M N −n
N − M bad; selected


N
(n) N 1− N N −1 nM/N
n selected
Discrete 1/N (N + 1)/2 (N 2 − 1)/12 N outcomes 1, 2, . . . , N equally
Uniform likely
 21.1. Summing-up: How To Tell Random Variables Apart 261

Geometric vs. Negative Binomial

• Geometric is doing trials until you get your 1st success.
• Negative Binomial is doing trials until you get your 2nd or 4th or 15th
or rth success. A Negative Binomial is the sum of several Geometrics.
Binomial vs. Negative Binomial
• Binomial has the number of trials set in advance, and you wait until all
the trials are over to count your successes. Successes can come in any
order.
• Negative Binomial has the number of successes set in advance, and you
do as many trials as you necessary in order to get that last success. The
final success has to be on the last trial.
• Below is a table comparing possible arrangements of outcomes for 6 trials
with 2 successes for both the Binomial and the Negative Binomial.

Binomial Negative Binomial

Y Y N N N N
Y N Y N N N
Y N N Y N N
Y N N N Y N
Y N N N N Y YNNNNY
N Y Y N N N
N Y N Y N N
N Y N N Y N
N Y N N N Y NYNNNY
N N Y Y N N
N N Y N Y N
N N Y N N Y NNYNNY
N N N Y Y N
N N N Y N Y NNNYNY
N N N N Y Y NNNNYY
6
= 15 arrangements 5
= 5 arrangements since 2nd



2 1
since any 2 of 6 can be success success must come on last trial

Binomial vs. Hypergeometric

• The X is the number of successes in n trials for both.
• Binomial is sampling with replacement. You will know the sample size
and the probability of success on a single trial. The probability of success
will be the same for each trial, and the trials are independent.
262 Chapter 21. Review of Named Discrete Random Variables

• Hypergeometric is sampling without replacement. You will know the

population size, the sample size, and the population number of successes.
The probability of success will not be the same for each trial because for
each trial, the population size will be decreased by 1.

• If N (population size) is really big for a Hypergeometric and n (sample

size) is much smaller, you may want to use the Binomial approximation
to the Hypergeometric. In a large population, even if you do sampling
without replacement, the results will be very similar to sampling with
replacement. Use the same n, but set p = M/N to make the switch. The
X stays the same.

Binomial vs. Poisson

• The X is the number of successes (or arrivals or counts) for both.

• Binomial is used when you know the number of trials (n) and the prob-
ability of success on each trial (p). The probability of success will be the
same for each trial, and the trials are independent from each other.

• Poisson is used when you know the average rate of arrival for the counts
(λ). You have a set interval instead of a set number of trials.

• If your sample size (n) is really big and your probability of success on a
single trial (p) is really small, you may want to use the Poisson approx-
imation to the Binomial. To make this switch, find the average for the
Binomial E(X) = np, and set that equal to λ. The X stays the same.
Bernoulli vs. Discrete Uniform
• Bernoulli is a single yes/no trial with a defined probability of success.
The probability of success and the probability of failure do not have to be
the same, but they do need to add up to 1.

• Discrete Uniform has one or more possible outcomes, and all of the
outcomes are equally likely.

• A Bernoulli trial with a probability of success p = 0.5, like a coin toss, is

similar to a Discrete Uniform distribution with N = 2. The only thing
different is that the outcomes are shifted to {0, 1} for the Bernoulli, instead
of the {1, 2} in a Discrete Uniform. So the expected value is different
(because of the shift), but the variance does not change.
Poisson vs. Discrete Uniform
• Poisson counts up the number of successes in an interval when you know
the average rate of arrival.

• Discrete Uniform has only one success, and it must come from one of
the N possible outcomes.
 21.2. Exercises 263

21.2 Exercises
For each of the following situations, state which distribution (and approximate
distribution, if applicable) would be most appropriate, and why you think so.

Exercise 21.1. Let X be the number of ice cream cones in your sample which
are broken if you sample 50 of them from a large, independent population, and
12% of the cones in the entire population are broken.

Exercise 21.2. Let X be the number of ice cream cones you need to sample
(again, from a large population) in order to find your 4th broken one, if they
come from a large, independent population, and 12% of the cones in the entire
population are broken.

Exercise 21.3. Let X be the number of broken cones you would find in the
next hour if broken cones come down the assembly line at a rate of 2 broken
cones per minute.

Exercise 21.4. Let X indicate whether the next ice cream cone is broken if
12% of the cones in a large, independent population are broken.

Exercise 21.5. Let X be the number of broken ice cream cones in your sample
if you check 30 from a box (without replacement). Twelve are broken out of the
100 in the box total.

Exercise 21.6. Let X be the number of broken ice cream cones you need
to sample in order to find your first broken one if they come from a large,
independent population, and 12% of the cones in the entire population are
broken.

Exercise 21.7. Let X be the number on the box you randomly select, if you
are choosing 1 box from 7 numbered boxes of ice cream cones.

Exercise 21.8. Let X be the number of broken ice cream cones in your sample
if you check 30 from a shipment (without replacement). The lot has 1200 broken
cones out of a total of 10,000.

Exercise 21.9. Let X be the number of undercooked ice cream cones in your
shipment of 10,000 if you sample from a large population. Undercooked ice
cream cones have a 0.005% chance of occurring in general.

21.3 Review Problems

As you answer each question, also state which distribution you are using and
what the parameters are.

Exercise 21.10. Chinese checkers. Philip and Callum play a game of Chi-
nese checkers. Each time they play a game, Philip has a 0.7 chance of winning.
Assume the games are independent.
264 Chapter 21. Review of Named Discrete Random Variables

a. In a tournament, Philip and Callum will play 8 games. What is the

probability Philip wins exactly 6 games?
b. Philip and Callum decide to keep playing until Callum wins 3 times.
What is the probability they play a total of 12 games?
c. How many 8-game tournaments will they expect to have to play until
Philip wins exactly 6 games?

Exercise 21.11. Car dealer. According to past experience, 20% of customers

coming into Frank’s car dealership will buy one of his cars. Assume customers
are independent and sampled from a very large population. Assume his dealer-
ship is open 7 days a week.

a. What is the probability the next person who walks in the door will buy
a car?
b. If 10 customers come in to the dealership today, what is the probability
at least 2 of them will buy cars?
c. What is the probability that the 4th customer coming in today is the first
one who will buy a car?
d. What is the expected number of customers he needs to come into the
dealership, to sell his 3rd car?
e. If 10 customers come to his dealership each day, what is the probability
that he will sell at least 2 cars in each of 3 days out of the next week?
f. If customers are equally likely to want to buy cars painted red, brown,
blue, black, or white, what is the probability that the next customer who buys
a car picks a black car?

Exercise 21.12. Babies. On average, 9 babies are born per 24-hour day at
the local hospital.

a. What is the probability that at least one baby will be born today on the
8 am to 4 pm shift?
b. What is the probability that at least one baby will be born in the next
hour?
c. What is the probability that exactly 4 babies will be born on the 8 am
to 4 pm shift?
d. What is the probability that exactly 4 babies will be born on each of the
next three 8-hour shifts?
e. What is the probability that you would have to wait for four 8-hour shifts
until you got the first one with exactly 4 babies born?
f. What is the probability that exactly 12 babies will be born total in the
next 24 hours?
 21.3. Review Problems 265

Exercise 21.13. Scholarship. There are 5 juniors and 10 seniors (one of which
is Amelia), trying to win a scholarship to a summer music program. Only 3
students can win, and the winners will be selected randomly by pulling names
out of a hat.

a. If no one can win more than once, what is the probability that all 3
winners will be seniors?
b. If students can win more than once and the students are independent
from each other, what is the probability that all 3 winners will be seniors?
c. If students can win more than once and the students are independent
from each other, what is the probability that the name-puller will call the first
senior on the third name?
d. What is the probability that the first person to win a scholarship is a
senior?
e. What is the probability that the third person to win a scholarship is a
senior?

Exercise 21.14. Cookies. Chocolate chips are randomly distributed onto on

a bakery’s cookies with a rate, on average, of 11 chips per cookie. Assume that
the cookies are independent.

a. What is the probability that a randomly selected cookie contains exactly

12 chips?
b. On average, how many cookies would you have to check until found a
cookie with exactly 12 chips?
c. On average, how many cookies would you have to check until you found
the 5th cookie with exactly 12 chips?
d. If you sample 30 cookies (with replacement) what is the probability there
will be at least one cookie with exactly 12 chips?
e. You have a batch of 100 cookies, and 10 of these cookies have exactly 12
chips. If you randomly eat 5 cookies from this 100-cookie batch, what is the
probability that at least 1 of the cookies you eat will have exactly 12 chips?

Exercise 21.15. Donuts. The probability that a student has a donut for
breakfast is 0.08.

a. If I ask students (at random) from a large university campus whether

they had a donut for breakfast, what is the probability the twentieth person I
ask will be the first one who had a donut for breakfast?
b. How many students do I expect to have to ask until I find the fourth one
who had a donut for breakfast?
c. If I ask 150 students whether they had a donut for breakfast, what is the
expected number of students who did?
266 Chapter 21. Review of Named Discrete Random Variables

d. What is the probability the next student I ask did not have a donut for
breakfast?
Exercise 21.16. Cards. Stacey is playing a game of cards using a standard
52-card deck (13 each of hearts, clubs, diamonds, and spades).
a. What is the probability the first card she is dealt will be the ace of
spades?
b. She is dealt 7 cards at the beginning of the game to hold in her hand.
What is the probability that all 7 will be hearts?
c. What is the expected number of hearts she will have in her 7 cards?
What is the standard deviation?
Exercise 21.17. Rain. Suppose rain is falling at an average rate of 30 drops
per square inch per minute.
a. What is the probability that a particular square inch is hit by exactly 4
drops in the next 10-second period?
b. How many 10-second intervals do you expect to observe this square inch
until you find a 10-second interval with exactly 4 raindrops?
c. What is the probability you would have to observe ten 10-second intervals
to find three of them with exactly 4 raindrops?
Exercise 21.18. Guessing on an exam. An exam consists of 20 multiple-
choice questions with 5 possible answer choices per question. You haven’t stud-
ied, so you decide to make random guesses for each answer choice. Assume each
question’s answer is independent from the others.
a. What is the probability that a person who randomly guesses on each
question gets exactly 5 questions correct given that they got at least 1 question
correct?
b. What is the expected number of correct answers on the exam? What is
the standard deviation?
c. What is the probability I will have to grade more than 12 exams to find
one with exactly 5 correct answers, assuming all of my students were randomly
guessing on all the questions?
d. What is the expected number of correct answers if a person takes 6
exams? What is the standard deviation in the number of correct answers if a
person takes 6 exams?
e. If you have to pay $2 to take the exam, but your parents pay you 50
cents for every correct answer, what is your expected net profit? What is the
standard deviation in your net profit?
Exercise 21.19. Socks. You are doing laundry, and you are trying to find any
small new red socks which may have fallen into the pile of white clothes in the
 21.3. Review Problems 267

laundry basket by mistake. It’s really dark in your basement laundry room, so
you have to randomly sample items of clothing.

a. In a single laundry basket, there are 35 white socks and 10 red socks. If
you sample 5 socks at random (without replacement), what is the probability
that at least one of them will be a red sock?
b. Now suppose instead of being in your basement laundry room, you are
dealing with this problem at a huge cotton clothing manufacturer’s pre-wash
room, where there are 35,000 white socks and 1000 red socks. You sample 5
items of clothing at one time. What is the approximate probability at least one
of them will be a red sock?
c. Using your answer to part b, how many 5-item samples of clothing would
you expect to have to take to find the first 5-item sample with at least one red
sock?

Exercise 21.20. Coffee beans. You are in charge of coffee bean quality
control. In a very large batch of coffee beans, you estimate that the chance
a bean is roasted incorrectly is 0.008. You have your employees sample 1000
beans from the seemingly endless supply, to see how many incorrectly roasted
beans they find. (The roast of the beans is assumed to be independent, since
the quantity of beans is large, and the beans are well-mixed before they get
inspected.)

a. What is the expected number of incorrectly roasted beans in the 1000-

item sample? What is the standard deviation?
b. Set up the exact probability (but do not solve) that at least 2 beans from
the sample are incorrectly roasted.
c. What is the approximate probability that at least 2 beans from the sample
are incorrectly roasted.

Exercise 21.21. Dogs. Frédérique has 7 dogs: Molly, Ted, Fido, Rocket, Max,
Sandy, and Nellie. Assume each dog is independent from the other.

a. One of the dogs has stolen her sandwich off her plate at lunch, and she
wants to figure out which one. They are each equally likely to have done it, so
she is going to randomly check a dog to see if he or she has bread crumbs on
his or her fur. What is the probability Rocket was the one who did it?
b. What is the expected number of sandwiches that would be stolen until
Rocket was the one who did it, assuming each dog is equally likely to steal her
lunch, and that the sandwich thieves are working independently, from day to
day (only 1 dog can steal the sandwich of the day).
c. Each time she leaves the table, there is a 0.25 chance that a dog will
steal her lunch. Out of 20 lunchtimes in which she is interrupted, what is the
probability that her lunch will be stolen exactly 4 times?
268 Chapter 21. Review of Named Discrete Random Variables

d. If Frédérique stays at the table for the entire lunchtime, there is only
a 0.0001 chance a dog will steal her lunch. Out of 10 years worth of lunches
(3650) what is the approximate probability a dog will steal her lunch exactly 4
times?
Exercise 21.22. Slug. The number of vehicles crossing a line on an interstate
is Poisson with an average of 5 per hour. A slug arrives at the interstate and
will wait until no vehicles have crossed the line for 5 minutes until it attempts
to get to the other side of the interstate. (This particular slug can count and
tell time.) Assume the cars are independent.
a. For a given 5-minute period, what is the expected number of cars to cross
the line?
b. What is the probability that no vehicles will pass in a given 5-minute
period?
c. How many 5-minute intervals will the slug have to wait until the first one
with no vehicles?
d. It will take the slug 10 minutes to cross the road. What is the probability
the slug will be run over by at least one car? (Start the clock and the car-
counting when the slug starts crossing the road.)
Exercise 21.23. Mice experiment. There are 5 white mice, 6 brown mice,
and 7 gray mice in a cage. For your psychology experiment, you reach in and
randomly select 3 to run in a maze.
a. What is the expected and standard deviation in the number of white
mice which are selected?
b. What is the probability that at least 2 of the mice selected are white?
c. What if you had 5000 white mice, 6000 brown mice, and 7000 gray mice to
choose from? You still need 3 mice to run in the maze. What is the approximate
probability that at least 2 of the mice selected are white?
Exercise 21.24. Left-handed desk. Fred is a left-handed person who walks
into a large lecture room for a 4-hour exam. He’s a little absent-minded at the
moment because he’s worried about the exam. He needs to find a left-handed
desk to sit in for the exam or else his back will get cramped up before the exam
is over, but he won’t know if the desk is left-handed until he actually sits in
one, because the desk part folds underneath the chair. Ten percent of the desks
in the exam room are for left-handed people, and the choice of left-or-right
handedness is independent from desk to desk, throughout the room. Assume
Fred is the first student to walk in the door.
a. What is the probability that Fred finds his first left-handed desk on his
5th try?
b. If Fred keeps sampling desks to find the left-handed desk which “feels
lucky,” what is the probability he finds his 3rd left-handed desk on his 8th try?
 21.3. Review Problems 269

c. What is the average number of desks Fred has to try until finding his 2nd
left-handed desk?
d. If the first 9 desks tested are not left-handed, what is the probability that
he has to keep looking for more than 12 desks total to find his first left-handed
desk?

Exercise 21.25. Ming and Shaheed each roll a dice (repeatedly, in rounds)
until their results match, and then they stop. Let X be the number of rounds
in which the sum of the two dice was 3, and let Y be the number of rounds in
which the sum of the two dice was 9. Find the conditional PMF of X, given
X + Y = 10.

Exercise 21.26. Bethany has a fair six-sided die (with sides 1, . . . , 6). Angelica
has a fair coin, with “1” on one side and “4” on the other side. Each day, Bethany
rolls the die one time, and Angelica flips the coin one time (the results are
independent). Let N denote the first day on which Bethany’s die has a strictly
larger result than Angelica’s coin. Let X denote the value of Bethany’s die on
day N . Find the expected value of X.

Exercise 21.27. Let X be a Poisson random variable with mean λ. Let Y be

a Geometric random variable with parameter p, i.e., Y has mass P (Y = j) =
p(1 − p)j−1 for integers j ≥ 1. (Here, λ > 0 and 0 < p < 1.) Also X and Y are
independent. Find P (Y > X).
this page left intentionally blank
Part IV

Counting

We have talked about basic probability ideas in the previous chapters, and
we have discussed the difference between sampling with or without replacement
and whether order does/doesn’t matter. However, in this part, we want to focus
purely on counting problems, including rearrangements. You will see familiar
ideas from the Binomial and Hypergeometric discrete distributions, but we will
expand on those ideas in more complex situations. These problems can be both
challenging and fun. We suggest you draw pictures or even create little models
for yourself if you are having trouble seeing the story clearly in your mind. Much
of counting is common sense once you can visualize the rules for a particular
story.
By the end of this part of the book, you should be able to:

1. Characterize the differences between sampling with/without replacement.

2. Characterize the differences between sampling in which order matters, or

in which order does not matter.

3. Perform probability calculations for sampling with/without replacement.

4. Perform probability calculations for sampling, either depending on the

order of selection, or when the order of selection doesn’t matter.

5. Calculate the number of possible rearrangements of dice, playing cards,

seating, passwords, etc.

Math skills you will need: Binomial coefficients (“choose” parentheses), fac-
torials.
Additional resources: Calculators may be used to assist in the calculations.

271
Chapter 22

Introduction to Counting

The hardest arithmetic to master is that which enables us to count our blessings.
—Reflections On The Human Condition by Eric Hoffer (Harper, 1973)

How many unique results can appear on 3 differently colored dice? How many
unique results are possible if the dice are indistinguishable?

22.1 Introduction
The concept of “counting,” as it relates to probability, is usually performed in
the context of a sample space S with finitely many outcomes. Moreover, the
outcomes are usually all equally likely. We have seen a few examples of such
situations, during our study of outcomes, events, and discrete random variables.
For instance, if we shuffle a deck of 52 cards, and we remove one card, then there
are 52 possible outcomes, and they are each equally likely to occur. Each event
that contains just one outcome has probability 1/52. If the event contains 13 of
the outcomes (for instance, the 13 outcomes in which a heart is selected), then
the event has probability 13/52. More generally, if the event has j outcomes,
then the event has probability j/52. Recall that the general situation was
established in Corollary 2.10:

Corollary 2.10 If sample space S has n equally likely outcomes, and A is

an event with j outcomes, then event A has probability j/n of occurring, i.e.,
P (A) = j/n.

We explore some ways this corollary can be applied. Counting (combina-

torics) is an area of study, with courses, books, and deep theorems of its own.
We admittedly only scratch the surface. If the study of counting interests you,
we encourage you to pursue further study in combinatorics. Combinatorics has
a long history of development, often in tandem with probability theory.

272
 22.1. Introduction 273

Example 22.1. Roll two dice. The sample space of all 36 equally likely out-
comes is given in Figure 22.1. Some students have difficulty distinguishing how
often each outcome happens. One remedy for this is to imagine that one of the
dice is painted red and one is painted green, or to think of one die as getting
rolled first and the other to get rolled second. The probability that the sum of
the dice is 8 or larger is 15/36 = 5/12.

1 2 3 4 5 6
1 2 3 4 5 6 7
2 3 4 5 6 7 8
3 4 5 6 7 8 9
4 5 6 7 8 9 10
5 6 7 8 9 10 11
6 7 8 9 10 11 12

Figure 22.1: Bubbles appear around outcomes corresponding to sums of 8

or larger on two dice. E.g., all outcomes with a sum of 10 are in a bubble.

Example 22.2. What is the probability that the results on the two dice differ
by 2 or less, i.e., if the two results are X and Y , what is P (|X − Y | ≤ 2)?

Since 24 of the 36 outcomes differ by 2 or less, the probability is 24/36 = 2/3.

See Figure 22.2.

1 2 3 4 5 6
1 0 1 2 3 4 5
2 1 0 1 2 3 4
3 2 1 0 1 2 3
4 3 2 1 0 1 2
5 4 3 2 1 0 1
6 5 4 3 2 1 0

Figure 22.2: A loop around 24 outcomes on 2 dice that differ by 2 or less.

274 Chapter 22. Introduction to Counting

Example 22.3. Let X and Y denote the results on two dice. Define Z =
max(X, Y ). Find the mass of Z.

The value of Z is between 1 and 6. We can just add the number of possible
outcomes that correspond to each value of Z. For instance, Z = 5 if the outcome
is contained in the event

{(1, 5), (2, 5), (3, 5), (4, 5), (5, 1), (5, 2), (5, 3), (5, 4), (5, 5)}.

Thus P (Z = 5) = 9/36. Figure 22.3 shows the values of Z for each outcome.
In general, the mass of Z is

z 1 2 3 4 5 6
pZ (z) 1/36 3/36 5/36 7/36 9/36 11/36

1 2 3 4 5 6
1 1 2 3 4 5 6
2 2 2 3 4 5 6
3 3 3 3 4 5 6
4 4 4 4 4 5 6
5 5 5 5 5 5 6
6 6 6 6 6 6 6

Figure 22.3: The maximum of two dice. A loop is drawn around each set of
outcomes for which the maximum is the same.

One of the most helpful rules of thumb in counting problems is the method of
multiplying the number of outcomes when two or more things are happening
simultaneously.

Example 22.4. If there are 4 colors of pants that you could pick from, and 3
colors of shirts, then there are 12 ways that you can put on both your pants
and shirt.
 22.1. Introduction 275

More generally, if there are two things happening simultaneously, and the first
thing has n possibilities, and (for each of the first things) the second thing
always has m possibilities, then there are nm possibilities altogether.

Theorem 22.5. Multiplying possibilities to get the total number of

outcomes (two simultaneous processes)
Suppose two processes are happening, and the first one has n possibilities,
and for each such possibility, there are m possibilities for the second process.
Then there are nm possibilities altogether for the pair of results from the two
processes.

Continuing to generalize, suppose now that we not only must put on our
pants and shirt, but also our shoes.

Example 22.6. If there are 4 pants, and for each such choice, 3 shirts, and for
each pair of pants/shirt, there are 5 styles of shoe, then there are (4)(3)(5) = 60
possible outfits altogether.

In the most general view, with j processes happening all at once:

Theorem 22.7. Multiplying possibilities to get the total number of

outcomes (j simultaneous processes)
Suppose j processes are happening, in which there are n1 possibilities for the
first process, and for each such possibility, there are n2 possibilities for the
second process, and etc., etc., there are always nj possibilities for the jth
process. Then there are n1 × n2 × · · · × nj possibilities altogether.

Example 22.8. A playlist has 10 rock songs, 3 blues songs, and 7 hip-hop
songs. The mp3 player is working in a “shuffle” mode in which each of the 20
songs are equally likely to appear each time, and none of the song choices affect
any of the other song choices. In particular, repetitions are certainly allowed.
What is the probability that the listener hears two consecutive hip-hop songs
followed by a rock song?

There are 20 × 20 × 20 = 8000 equally likely outcomes for the sequence of

three songs. Exactly 7 × 7 × 10 = 490 have the form (hip-hop, hip-hop, rock).
So the probability of hip-hop, hip-hop, rock is 490/8000 = 0.06125.
276 Chapter 22. Introduction to Counting

Example 22.9. While building a loft in your dorm room, you find that there
are 8 holes remaining but only 6 screws available. The 8 holes are arranged in a
line, from top to bottom. You randomly pick 6 holes to fill, with all possibilities
equally likely. What is the probability that you do not fill the bottom hole?

There are 86 = 6!2!

8!
= 28 ways to pick which 6 holes will be filled. In exactly


7
1 = 7 of these ways, the last hole will remain empty. So the probability that

the last hole remains empty is 7/28 = 1/4.

Example 22.10. Chris has an 18-pack of Gatorade sports drink: 6 orange,

6 lemon-lime, and 6 fruit punch. What is the probability that he pulls out
six bottles that match (i.e., have the same type), followed by another six that
match (i.e., have the same type), followed by another six that match?

There are 18! equally likely ways that he can choose the flavors. To be
successful (i.e., to pull them out in the way that is described), there are 3! = 6
ways that the types can be ordered. There are 6! ways that the bottles of the
first type chosen can be arranged, and then there are 6! ways that the second
type chosen can be arranged, and then there are 6! ways that the third type
chosen can be arranged. So the desired probability is

3!6!6!6! 1
= = 3.4979 × 10−7 .
18! 2858856

22.2 Order and Replacement in Sampling

In this section, we describe four related general scenarios for handling various
selection methods. For each method, we give the number of “unique” ways that
the selection can be performed. Note that the uniqueness of the ways depends
on whether we are regarding the order of selection.
The distinction of “with” or “without” replacement is analogous to Binomial
versus Hypergeometric random variables. For a Binomial random variable, the
probability p does not change between selections; this is analogous to sampling
with replacement of each selection after it is done, so that the ratios for the
selections stay the same from trial to trail. For a Hypergeometric random
variable, on the other hand, there are a fixed number of possible good and bad
outcomes; once a trial is done, the selected item is not restored; this is sampling
without replacement. Throughout this section, for consistency, we use

n = # of objects available to choose from,

r = # of objects we actually choose.
 22.2. Order and Replacement in Sampling 277

Sampling with replacement Sampling without replacement

nr n!
(n−r)! (permutation)
If I roll one red die, If I have 10 students,
one green die, and how many unique ways
one blue die, can 4 different
how many possible students go to
unique results the board for 4 different
Order can I get? homework problems?
matters (r = 3 dice, (r = 4 chosen,
n = 6 possibilities) n = 10 possibilities)
General: you have an General: you care about
unlimited supply and “when” not just “if”
you care about “when” something is selected.
objects are selected Nothing can be selected
(and how many times). more than once.

n+r−1 n n!
(combination)



r r = r!(n−r)!
If I roll 3 identical If I have 10 students,
dice one time each, how many ways can
how many possible 4 different students
unique results go to the board
Order does can I get? for a problem?
not matter (r = 3 dice, (r = 4 chosen,
n = 6 possibilities) n = 10 possibilities)
General: you have an General: you only care
unlimited supply, and “if” not “when”
you only care “if” objects are selected.
not “when” something Nothing can be chosen
is selected. more than once.

The upper left box has nr ways because we have r choices, each of which
has n possibilities, so there are (n)(n) · · · (n) = nr ways. The upper right box
has n first choices, n − 1 second choices, . . . , n − r + 1 for the rth choice, so
(n)(n − 1) · · · (n − r + 1) = n!/(n − r)! ways altogether. In the lower right box,
the count is the same, but the r! different arrangements of each are viewed as
the same, so the count is n!/((n − r)!r!). Finally, in the lower left box, we
have n types of items to choose from r types, so we can just put r − 1 dividing
lines between n unlabeled balls. The number of balls before the first line are
of type 1; the number of balls between the first and second lines are of type 2;
etc., the number of balls between the (r − 2)nd and (r − 1)st lines are of type
r − 1, and the number of balls after the (r − 1)st line are of type r. It is helpful
to try this on your own, with some small n’s and r’s.
278 Chapter 22. Introduction to Counting

Sampling with replacement, order matters: nr

Example 22.11. If I roll one red die, one green die, and one blue die, how
many possible unique results can I get?

Each die has 6 possible values (n = 6), and the 3 dice have distinguishable
colors (r = 3), which means a “Red 1, Blue 2, Green 3” would be considered
a different result than “Red 3, Blue 2, Green 1,” for example. Also, dice use
sampling with replacement because each time you roll a die, you have the exact
same options with the same probabilities that you had the first time. The red
die result has no effect on the blue or green dice results. Therefore the number
of possible unique results I can get will be 63 = 216.

Example 22.12. If I choose 5 cards from a deck, with replacement in be-

tween the draws, keeping track of the order, then I am choosing 5 cards, each
1 through 52 (numbers are reusable). How many different combinations are
possible, keeping track of the order?

The 52 numbers are reusable, so for example, I could choose {A♠, A♠, A♠, A♠
,2♥}, or {2♥, A♠, A♠, A♠, A♠}; the order distinguishes these two. Therefore
I have n = 52 for my options for each card, and r = 5 because I need to select
five cards. There are 525 = 380204032 possible selections of the cards, with
replacement, and keeping track of the order of selection.

n+r−1

Sampling with replacement, order does not matter: r

Example 22.13. If I roll 3 indistinguishable dice, each one time, how many
possible unique results can I get?

This time we still have 6 options on each die (n = 6), and we still have 3 dice
(r = 3), but since the dice are identical, a result of “1, 2, 3” would be the same
as “2, 3, 1” or “3, 1, 2” because

we8
cannot tell the dice apart. The number of
unique results will be 6+3−13 = 3 = 56. Compare this answer to what we
found in the example with the red, blue, and green distinguishable dice (216
unique outcomes). Losing the distinction of the ordering causes there to be
fewer unique sets of values.
 22.2. Order and Replacement in Sampling 279

Example 22.14. If I choose 5 cards from a deck, with replacement in between

the draws, but this time I do not keep track of the order, then I am choosing
r = 5 cards from the n = 52 available (numbers are reusable). How many
different combinations are possible, without keeping track of the order?

The 52 numbers are reusable, so for example, I could choose {A♠, A♠, A♠,
A♠, 2♥}, or {2♥, A♠, A♠, A♠, A♠}, but these are considered as the same
possibility; the order does not distinguish these two. Therefore I have n = 52 for
my options
for each card, and r = 5 because I need to select five cards. There
are 52+5−1
5 = 3819816 possible selections of the cards, with replacement, but
without keeping track of the order of selection.

n!
Sampling without replacement, order matters: (n−r)!

Example 22.15. If I have 10 students, how many unique ways can 4 different
students go to the board to do 4 different homework problems?

There are n = 10 students in the room, and each student can only be selected
once (sampling without replacement). I need r = 4 different students to do
4 different homework problems, so the order that the students are chosen are
distinct. The number of unique ways I can select these 4 students is (10−4)!
10!
=
10!
6! = (10)(9)(8)(7) = 5040.

Example 22.16. If I choose 5 cards from a deck, without replacement in be-

tween the draws, keeping track of the order, then I am choosing 5 cards, each 1
through 52 (numbers are not reusable). How many different combinations are
possible, keeping track of the order?

The 52 numbers are not reusable, e.g., I could choose {A♠, 7♠, 10♠, 3♥, 7♥},
or {10♠, 3♥, A♠, 7♥, 7♠}; the order distinguishes these two. Therefore I have
n = 52 for my options for each card, and r = 5 because I need to select five
cards. There are (52−5)!
52!
47! = (52)(51)(50)(49)(48) = 311875200 possible
= 52!
selections of the cards, without replacement, and keeping track of the order of
selection. This is a smaller number of possibilities, as compared to the situation
“with replacement.”
280 Chapter 22. Introduction to Counting
n

Sampling without replacement, order does not matter: r

Example 22.17. If I have 10 students, how many unique ways can 4 different
students go to the board to work on the same problem?
There are n = 10 students in the room, and each student can only be selected
once (sampling without replacement). I need r = 4 different students to do the
same problem, so the order that the students are chosen makes no difference. All
that matters is which 4 of the 10 students are selected. The number of unique
(10)(9)(8)(7)
ways I could choose 4 of these 10 students is 10 10!


4 = 4!(10−4)! = 4! =
210. This is closely related to the previous situation, in which the order of the
students mattered. Grouping together the selections from before according to
the students, we see that 4! = 24 of the previous selections corresponds to just
1 selection here. So there is a factor of 4! fewer distinct possibilities when we
ignore the order in which the students are chosen.

Example 22.18. If I choose 5 cards from a deck, without replacement in be-

tween the draws and without keeping track of the order, then I am choosing
r = 5 cards from the n = 52 available (numbers are not reusable). How many
different combinations are possible, when we do not keep track of the order?
The 52 numbers are not reusable, e.g., I could choose {A♠, 7♠, 10♠, 3♥, 7♥},
or {10♠, 3♥, A♠, 7♥, 7♠}; but these are considered as the same choice, because
the order does not distinguish these two. Therefore I just choose 5 out of the
52 cards, so there are 5!(52−5)!
52! 52!
= 5!47! = (52)(51)(50)(49)(48)
5! = 2598960 possible
selections of the cards, without replacement, and without keeping track of the
order of selection. Again, this is a factor of 5! fewer possibilities than the
previous situation, because we ignore the orderings of the cards.

22.3 Counting: Seating Arrangements

Example 22.19. If Alice and Alan (a couple) and Barbara and Bob (another
couple) and Christine and Charlie (another couple) sit in a row of chairs, what
is the probability that each of the 3 couples sit together?
There are 6! = 720 ways that the 6 people can be seated altogether. (This
is true in all of the questions with 6 people, so we won’t repeat this fact.)
Grouping the couples together again, there are 3! ways that the couples can
be seated (i.e., either A’s/B’s/C’s, or A’s/C’s/B’s, or B’s/A’s/C’s, etc... 3! ways
 22.3. Counting: Seating Arrangements 281

total). For each such way, there are 2 ways that the A’s can be arranged among
themselves, and 2 ways that the B’s can be arranged among themselves, and 2
ways that the C’s can be arranged among themselves. So the total probability
that the couples are seated together is 3!(2)(2)(2)/720 = 1/15.

Example 22.20. If n couples sit in a row of chairs, what is the probability

that each of the n couples sits together?

There are (2n)! ways that the 2n people can be seated altogether. (This is
true in all of the questions with 2n people, so we won’t repeat this fact.)
Grouping the couples together again, there are n! ways that the couples can
be seated. For each such way, there are 2 ways within each couple that the men
and woman can be arranged in their two reserved seats. So the total probability
that the couples are seated together is n!2n /(2n)!.

Example 22.21. If Alice, Barbara, and Christine (three women) and Alan,
Bob, and Charlie (three men) sit in a row of chairs, what is the probability that
the women all sit together (the men may or may not be in a group)?

There are 3! ways that the men can be arranged. The women, as a group, can
be collectively put into any of the 4 gaps between the men, including the spaces
on the left- or right-hand ends. Once the women are placed as a group, there
are 3! arrangements among just the women themselves. So the total probability
that the women sit together as a group is (3!)(4)(3!)/6! = 1/5.

women sofa man man man

man women sofa man man
man man women sofa man
man man man women sofa

Figure 22.4: Ways to arrange 3 men and 3 women, if the women should sit
together in a group.

This can be visualized by thinking about the women sitting together on a

sofa, and the men on individual chairs. There are four places where the sofa for
the women can go, as in Figure 22.4.
282 Chapter 22. Introduction to Counting

Example 22.22. If n women and n men sit in a row of chairs, what is the
probability that the women all sit together?

There are n! ways to arrange the men. The women, as a group, can be
put into any of the n + 1 gaps between the men, including the left- or right-
hand ends. There are n! arrangements among the women themselves. So the
probability is n!(n + 1)n!/(2n)!. (Compare with Example 22.21, when n = 3.)

Example 22.23. If Alice, Barbara, and Christine (three women) and Alan,
Bob, and Charlie (three men) sit in a row of chairs, what is the probability that
the women all sit together and all the men sit together?

Either the women sit collectively on the right or on the left (i.e., 2 ways).
Once the women are placed as a group, there are 3! arrangements among just
the women themselves, and there are 3! arrangements among just the men
themselves. So the probability is (2)(3!)(3!)/6! = 1/10.

Example 22.24. If n women and n men sit in a row of chairs, what is the
probability that the women all sit together and all the men sit together?

Either the women sit collectively on the right or on the left (i.e., 2 ways).
Once the women are placed as a group, there are n! arrangements among just
the women themselves, and there are n! arrangements among just the men
themselves. So the probability is (2)(n!)(n!)/(2n)!.

Example 22.25. If Alice, Barbara, and Christine (three women) and Alan,
Bob, and Charlie (three men) sit in a row of chairs, what is the probability that
none of the women are adjacent and none of the men are adjacent?

Either the leftmost chair is for a woman or a man (i.e., 2 ways). After-
wards, the sexes of the people are determined. This leaves 3! arrangements
among the women, and 3! arrangements among the men. So the probability is
(2)(3!)(3!)/6! = 1/10.

Example 22.26. If n women and n men sit in a row of chairs, what is the
probability none of the women are adjacent and none of the men are adjacent?
 22.3. Counting: Seating Arrangements 283

Either the leftmost chair is for a woman or a man (i.e., 2 ways). Afterwards,
the sexes of the people are determined. Once they are determined, there are n!
arrangements among just the women themselves, and there are n! arrangements
among just the men themselves. So the probability is (2)(n!)(n!)/(2n)!.

Example 22.27. If n couples sit in a row of chairs, what is the expected number
of couples that sit together?

Method #1. Let X denote the number of couples that sit together. Let Xj
indicate whether the jth man sits next to his wife, so that

Xj = 1 if the jth man sits next to his wife, or Xj = 0 otherwise.

Then we always have X = X1 + X2 + · · · + Xn , so

E(X) = E(X1 + · · · + Xn ) = E(X1 ) + · · · + E(Xn ).

We know E(Xj ) = P (Xj = 1), i.e., E(Xj ) is just equal to the probability that
the jth man sits next to his wife. This equals the probability that he sits on
the either end of the row and his wife sits in the unique seat next to him, plus
the probability that he sits in the interior of the row and his wife sits in either
of the seats next to him. The probability that the man sits at either end of the
row is 2/(2n), and, given that he sits at either end of the row, the probability
that his wife sits next to him is 1/(2n − 1). On the other hand, the probability
that the man sits in the interior of the row is (2n − 2)/(2n), and, given that he
sits in the interior of the row, the probability that his wife sits next to him is
2/(2n − 1). So the probability that he sits next to his wife is
     
2 1 2n − 2 2
E(Xj ) = + ,
2n 2n − 1 2n 2n − 1

which simplifies to
2 + 4n − 4 4n − 2 1
E(Xj ) = = = .
(2n)(2n − 1) (2n)(2n − 1) n
So the expected number of couples that sit together is
1 1 1
E(X) = + + · · · + = n(1/n) = 1.
n n n
Method #2. Let X denote the number of couples that sit together. Let Xj
indicate whether the person in the jth chair has her/his partner to the right,
so that
(
1 if the person in the jth chair has her/his partner to the right,
Xj =
0 otherwise,
284 Chapter 22. Introduction to Counting

Then we always have X = X1 + X2 + · · · + X2n−1 , so

E(X) = E(X1 + · · · + X2n−1 ) = E(X1 ) + · · · + E(X2n−1 ).

We know E(Xj ) = P (Xj = 1), i.e., E(Xj ) is just equal to the probability that
the jth person has her or his partner to the right. No matter who is in the jth
chair, the probability that the person’s partner is found to the right is 1/(2n−1).
So
E(Xj ) = 1/(2n − 1).
Thus, the expected number of couples that sit together is
1 1 1 2n − 1
E(X) = + + ··· + = = 1.
2n − 1 2n − 1 2n − 1 2n − 1

Example 22.28. If n couples sit in a circle of chairs (as opposed to a row, in the
previous example), what is the expected number of couples that sit together?
Method #1. Let X denote the number of couples that sit together. Let Xj
indicate whether the jth man sits next to his wife, so that

Xj = 1 if the jth man sits next to his wife, or Xj = 0 otherwise.

Then we always have X = X1 + X2 + · · · + Xn , so

E(X) = E(X1 + · · · + Xn ) = E(X1 ) + · · · + E(Xn ).

We know E(Xj ) = P (Xj = 1), i.e., E(Xj ) is just equal to the probability that
the jth man sits next to his wife. Regardless of where the jth man sits, his wife
will sit next to him with probability 2/(2n − 1). So

E(Xj ) = 2/(2n − 1).

Thus, the expected number of couples that sit together is

2 2 2n
E(X) = + ··· + = .
2n − 1 2n − 1 2n − 1
Method #2. Let X denote the number of couples that sit together. Let Xj
indicate whether the person in the jth chair has her/his partner to the right,
so that
(
1 if the person in the jth chair has her/his partner to the right,
Xj =
0 otherwise.

Then we always have X = X1 + X2 + · · · + X2n (we include the (2n)th chair,

since this scenario includes people in a circle, so the partner could be in the 1st
chair; this was not the case when the chairs were in a row), so

E(X) = E(X1 + · · · + X2n ) = E(X1 ) + · · · + E(X2n ).

 22.4. Exercises 285

We know E(Xj ) = P (Xj = 1), i.e., E(Xj ) is just equal to the probability that
the jth person has her or his partner to the right. No matter who is in the jth
chair, the probability that the person’s partner is found to the right is 1/(2n−1).
So
E(Xj ) = 1/(2n − 1).
So the expected number of couples that sit together is
1 1 1 2n
E(X) = + + ··· + = .
2n − 1 2n − 1 2n − 1 2n − 1
Note that it is slightly more likely for a man to sit next to his wife when the
row of chairs wraps around into a circle. If the man and the woman were on
opposite ends of a row, they will not be next to each other, but if the row is
wrapped into a circle, then they are next to each other.

22.4 Exercises
22.4.1 Practice
Exercise 22.1. Raffle tickets. There are 30 raffle tickets in a bowl. Three
winning tickets will be selected. Each ticket can win at most one prize. How
many ways can the prizes be distributed if the following additional information
is known?
a. All 3 winners receive goldfish (the goldfish are indistinguishable).
b. The 1st winner receives a car, the 2nd a bicycle, and the 3rd a goldfish.
Exercise 22.2. Defective robots. Among 7 robots produced by a factory
one day, 3 are defective. If 3 robots are purchased by the local toy store, find
the probability that at least one will be nondefective.
Exercise 22.3. Rearrangements. Consider the ways that the letters in the
word “Mississippi” can be rearranged (the 4 i’s are indistinguishable, the 4 s’s
are indistinguishable, and the 2 p’s are indistinguishable).
a. What is the probability that the S’s are grouped together?
b. What is the probability that the S’s are grouped together and the P’s
are grouped together?
c. What is the probability S is the 1st letter and I is the 5th letter?
d. What is the probability S or P is in the 1st spot?
Exercise 22.4. Cloudy days. In a certain city, the weather is cloudy on a
given day with probability 0.55, and is sunny with probability 0.45. The weather
is measured on several consecutive days (which are deemed to be sufficiently
independent for this problem).
286 Chapter 22. Introduction to Counting

a. What is the probability of 3 consecutive cloudy days, followed by a sunny

day?
b. What is the probability that exactly 1 out of 4 consecutive days will be
sunny?
c. What is the probability that at least 1 out of 4 consecutive days will be
sunny?
d. Given that the 1st 3 days were cloudy, what is the probability the 4th
day is sunny?

Exercise 22.5. Mochas. Suppose that typically 3/10 of the customers order
“grande” mocha (made with skim milk and extra whipped cream). Assume that
customers are independent.

a. What is the probability that exactly 2 of the next 5 customers will order
a “grande” mocha?
b. What is the probability that the 1st 3 customers do not order “grande”
mochas and the 4th and 5th customers do order “grande” mochas?

Exercise 22.6. Jelly beans. A student buys a bag of jelly beans at the store.
She eats most of them, but 20 remain at the end of the day. Exactly 13 of these
jelly beans are fruity, and the other 7 are root beer favored. Her boyfriend grabs
5 of the 20 jelly beans.

a. What is the probability that exactly 3 of the 5 jelly beans that he grabs
are fruity?
b. What is the expected number of fruity jelly beans that he grabs?

Exercise 22.7. Stamps. A professor buys postage stamps and puts them in
an envelope. Exactly 30 of the stamps are for letters, and exactly 10 of the
stamps are for postcards. The professor’s wife grabs two stamps. All outcomes
are equally likely.

a. What is the probability that both stamps she selects are for letters?
b. What is the probability that both stamps she selects are for postcards?
c. What is the probability that she gets one of each type?
(Hint: These three answers should sum to 1 altogether.)

Exercise 22.8. Rolling dice. Roll five dice. What is the probability that all
five of the values that appear are distinct, i.e., there are no repetitions among
the five dice?

Exercise 22.9. Letters. A player is given ten letters in a game:

“B,E,G,M,N,P,Q,R,S,U”
 22.4. Exercises 287

He randomly arranges all 10 of the letters on the table in front of him, in a

row, without paying attention (e.g., with his eyes closed). All of the possible
orders are equally likely. What is the probability the first 5 letters in the row
are “S,U,P,E,R,” in that exact order?

Exercise 22.10. Hiring. An employer has a very large pool of applicants for
8 jobs. The employer needs to report only the sex of the 8 people who are hired
(not the names or any other distinguishing features), to a gender-equity review
board. How many ways of hiring men and women for these positions are there
if:

a. The jobs are all identical?

b. Each of the 8 jobs are unique?

Exercise 22.11. Socks. Running late for class, you grab 2 socks out of your
drawer without looking at what color they are. In the drawer you have one pair
each of black, red, green, brown, blue and white socks, but they are not folded
as pairs—it’s a big jumbled mess in your drawer! What is the probability that
you grab 2 socks of the same color?

Exercise 22.12. Gumballs. You fill a mini gumball machine with 60 gumballs:
red cherry, pink bubblegum, green lime, and orange orange, 15 of each. Over
time you eat all of the gumballs, two at a time. What is the probability of being
left with one pink gumball and one green gumball at the end?

Exercise 22.13. Rock block. On a certain radio station, 70% of the songs
are rock songs, and 30% of the songs are pop songs. The songs are selected
independently. Each “block” of songs (a “block” is a set of songs between com-
mercials) contains 10 songs. The DJ says that the next “block” of songs has at
least 8 rock songs. Given this information, what is the probability that all 10
songs in that “block” will be rock songs?

Exercise 22.14. Ramen noodles. There are 20 ramen noodle packages in a

bag. There are 10 beef flavored, and the other 10 are chicken flavored. What is
the probability of getting at least one chicken and at least one beef flavor when
you grab 3 packages randomly?

Exercise 22.15. Badminton. A coach is choosing 4 students to go to a

badminton competition (the order of selection does not matter). There are 20
boys and 15 girls to choose from, and the coach chooses randomly so everyone
has an equal chance of being picked, but nobody can be picked more than once.

a. What is the probability that all 4 chosen will be boys?

b. What is the probability that all 4 chosen will be girls?
c. Should these two probabilities add up to 1? Why or why not?
288 Chapter 22. Introduction to Counting

Exercise 22.16. Family photos. Your computer has 437 family photos. You
decide to take 30 random photos (without replacement) from the collection
(each is equally likely), to use as a slideshow on the TV when guests come over.
Of your family photos, 42 are from your favorite vacation. What is the expected
number of photos from your favorite vacation in the slideshow?

Exercise 22.17. Action figures. In a box there are 31 Power Ranger action
figures. Seven of them are red and the other 24 are blue. Timmy closes his eyes
and randomly picks 8 action figures out of the box. What is the probability 6
of them are red?

Exercise 22.18. Cereal. On the cereal shelf of the house you share with
several roommates, you have 4 types of cereal with chocolate (Reese’s Puffs,
Cocoa Puffs, Count Chocula, and Cookie Crisp) and 3 types of cereal without
chocolate (Corn Chex, Cheerios, and Captain Crunch). You haven’t had your
coffee yet, so you will blindly choose 3 boxes of cereal to mix together for
breakfast. What is the probability you will get at least one chocolate cereal
mixed in?

Exercise 22.19. Deck of cards. In a game of chance, you are allowed to

shuffle a standard deck of cards and then choose 3 cards randomly. If two or
more of them are hearts, then you win.

a. What is the probability of winning the game?

b. What is the expected number of hearts drawn?

Exercise 22.20. Pizza parlor. You make pizzas for a local pizza parlor. On a
busy Friday night, you’ve had 12 orders in the last hour. The orders are placed
independently. Customers are known to prefer cheese pizza 37% of the time,
meaty pizza 47% of the time, and veggie pizza the other 16% of the time. What
is the probability that exactly 3 of the pizzas are veggie, 4 are cheese, and 5 are
meaty?

22.4.2 Extensions
Exercise 22.21. Daughters and mothers. If you randomly assign daughters
a, b, c, d, e to mothers A, B, C, D, E, and you let X be the number of daughters
who are correctly assigned to their mother, find the mass of X.

Exercise 22.22. Balls in boxes. Consider 6 identical balls and 3 distinguish-

able boxes.
a. In how many ways can we put the 6 identical balls into the 3 distinguish-
able boxes?
b. If each box gets at least 1 ball?
c. If boxes can be empty?
 22.4. Exercises 289

Exercise 22.23. Bears. A little girl randomly arranges 30 bears—consisting

of 10 red bears, 10 yellow bears, and 10 blue bears—into 10 bowls of 3 bears
each. (All outcomes are equally likely.)
Let X denote the number of bowls that have at least one blue bear in the
bowl. (The little girl loves blue bears.) Find the expected value of X.
Exercise 22.24. More bears. A total of 30 bears—consisting of 10 red bears,
10 yellow bears, and 10 blue bears—are randomly arranged in a bucket. A child
begins grabbing the bears at random, with all selections equally likely. The
bears are selected “without replacement,” i.e., she never puts the bears back
after she grabs them.
How many bears does the child expect to grab before the first red bear
appears? (Please do not include the red bear itself; only include the bears that
appear strictly before the first red bear appears.)
Exercise 22.25. Picking letters at random. Five friends, named Ali, Bob,
Charlie, Daniel, and Ebony, each pick a letter from the alphabet on their own
(i.e., independently), with all possible outcomes equally likely. For instance,
they might pick (starting with Ali’s choice) “M,D,P,Z,P.”
a. What is the probability that they choose 5 distinct letters?
b. What is the probability that they choose 5 distinct letters and that the
letters are in increasing alphabetic order starting from Ali’s choice and ending
at Ebony’s choice? For instance, starting with Ali’s choice, “D,H,P,T,X” would
be such an alphabetical ordering.
Exercise 22.26. Bridge. If you are playing bridge with a standard 52-card
deck, what is the probability that a player will get 5 clubs, 4 diamonds, 3 hearts,
and 1 spade in a 13-card deal?
Exercise 22.27. Getting dressed. Running late on Wednesday for proba-
bility class, Shaheed realizes that snow is forecast, thus he decides to wear two
T-shirts layered together. He has 10 T-shirts that are basically identical except
for color. He has 5 colors available, with 2 shirts of each color. One of these
colors is black. What is the probability that he grabs at least one black shirt?
Exercise 22.28. Fire dance. There are 48 candles for the ring of fire dance.
Exactly half are orange, and the other half are black. Each dancer can choose
two candles to hold during the dance. What is the probability that 5 dancers
choose 2 black candles each? The candles are not replaced after a dancer
chooses, so this would leave 24 untouched orange candles and only 14 black
candles after the selections are made.
Exercise 22.29. Work clothes. Pierre works 5 days a week. He has 12 shirts,
8 pants, 8 ties, and 4 jackets that he can wear to work. Of these, 4 shirts, 3
pants, 2 ties, and 2 jackets are blue. Each day he randomly selects one of each
item to wear. Assume the selections are independent, and assume his butler
launders his clothes every night so he has a full closet each morning.
290 Chapter 22. Introduction to Counting

a. What is the probability Pierre’s entire outfit next Monday will be blue?
b. What is the probability Pierre will wear at least 1 entirely blue outfit
during his next 5-day work week?
c. What is the probability that Pierre will wear entirely blue outfits on
Monday and Friday while wearing outfits which are not entirely blue on Tuesday
through Thursday?
d. What is the probability Pierre will wear an entirely blue outfit on exactly
2 of the 5 days next week?

Exercise 22.30. Lunch choices. According to an ad in the newspaper for

the local café, their Bagel Sandwiches can be “made your way” with “100s of
choices for you!” They advertise 7 flavors of bagels, 5 kinds of meat, 3 kinds of
cheese, 9 dressings, and 3 veggie selections.

a. If you select a sandwich made with one of each: bagel, meat, cheese,
dressing, and veggie (and you don’t care about the order), how many different
combinations of sandwich do you really have to choose from?
b. Now assume you only eat whole wheat bagels. The same 20 toppings
listed above can be used. How many combinations do you have?
c. If you can choose 3 items to go on your bagel, how many different types of
sandwiches can you have if the different items are reusable (tomato and double
ham would be ok)?
d. If you can choose 3 items but they are not reusable, how many are
possible?

Exercise 22.31. Pizza toppings. A pizza place offers the choice of the fol-
lowing toppings: extra cheese, mushrooms, pepperoni, ham, sausage, onions,
and green peppers. Assume that each pizza must have at least 1 topping and
that order of toppings is irrelevant. Also assume that toppings cannot be reused
(double sausage is impossible).

a. How many 3-topping pizzas are possible?

b. If I randomly order a 3-topping pizza, what is the probability that sausage
is a topping?
c. If I randomly order a 3-topping pizza, what is the probability that sausage
and pepperoni are toppings?
d. Now suppose that you allow toppings to be reused. For instance, double
sausage is now possible. (We have the possibility to double, triple, or even
quadruple a topping, if desired.) How many different 4-topping pizzas are there?

Exercise 22.32. Bookshelf. A student has 3 math books, 4 history books, 2

chemistry books, and 1 Latin book. He wants to arrange them on a bookshelf.
 22.4. Exercises 291

a. If all the books have different titles, in how many distinct ways can he
arrange them?
b. Throughout parts b–e, assume that all the books from a particular topic
have the same title (for example, 3 indistinguishable copies of “Calculus” by
Carey). In how many distinct ways can he arrange his books?
c. If he groups the identical math books together, and he groups the identical
history books together, and he groups the identical chemistry books together,
in how many distinct ways can he arrange his books?
d. If he groups the identical history books together (but isn’t picky about
the other books), in how many distinct ways can he arrange his books?
e. Given that the identical history books are grouped together, what is the
probability that he has grouped each of the identical books (the situation in
part c)?
Exercise 22.33. Movie theater. Ten people are going to a private screening
of a movie in a small theater. There are two rows in the theater. The first row
has 4 seats, and the second row has 6 seats.
a. How many ways can you arrange the seating of the 10 people altogether?
b. How many seating arrangements are possible if 3 of the 10 people are in
a family, and that family wants to sit together?
Exercise 22.34. Beach books. You have a row of 20 books on your desk.
You’re about to go on vacation and want to grab some good reading material,
preferably a good novel or some other work of fiction. Of the 20 books, 12 are
fiction and 8 are nonfiction. You grab 5 books at random.
a. What is the expected number of fiction books you will have?
b. What is the probability that 2 or more of your 5 books are fiction?
Exercise 22.35. Chair circle. A group of men and women sit in a circle.
There are 20 chairs in the circle, and 10 pairs of married individuals. What is
the probability that a man will sit directly across from his wife if everyone sits
randomly?
Exercise 22.36. License plates. On a “Save The Wetlands” license plate,
there are always two letters (e.g., “SW”) followed by four digits, and thus 10,000
combinations of plates are available for each pair of letters. If the four digits
are selected randomly, and all 10,000 possibilities are equally likely, what is the
probability that the four digits are distinct and in ascending order?
Exercise 22.37. Socks. In my sock drawer there are 21 white socks, 8 black
socks, and 4 brown socks.
a. If I randomly pull out 6 socks to take with me on a trip, what is the
probability that I pull out one pair of each color?
292 Chapter 22. Introduction to Counting

b. What is the probability all the socks are the same color?
c. What is the probability that I pull out 2 socks of one color and 4 socks
of a second color?
Exercise 22.38. Coins. My friend Alejandro has 35 coins, of which 26 are
quarters and the other 9 are pennies. If he gives me 4 coins at random, what is
the probability that I will have enough money to buy my favorite 89 cent candy
bar? How much money do I expect to get from Alejandro?

22.4.3 Seating Arrangement Problems

In this chapter we covered many questions about couples who are sitting in a
row of chairs. Now we consider some questions in which the couples sit in a
circle of chairs.
Exercise 22.39. Rodrigo and Edna are divorced. They are sitting at a circular
table with n ≥ 2 chairs altogether, and all seating arrangements are equally
likely. What is the probability that they are not sitting next to each other?
Exercise 22.40. If Alice and Alan (a couple) and Barbara and Bob (another
couple) sit in a circle of chairs, what is the probability that each of the couples
sit together?
Exercise 22.41. If Alice and Alan (a couple) and Barbara and Bob (another
couple) and Christine and Charlie (another couple) sit in a circle of chairs, what
is the probability that each of the 3 couples sit together?
Exercise 22.42. If n couples sit in a circle of chairs, what is the probability
that each of the n couples sits together?
Exercise 22.43. If Alice and Barbara (two girls) and Alan and Bob (two boys)
sit in a circle of chairs, what is the probability that the girls both sit together
and both boys sit together?
Exercise 22.44. If Alice, Barbara, and Christine (three girls) and Alan, Bob,
and Charlie (three boys) sit in a circle of chairs, what is the probability that
the girls all sit together and the boys all sit together?
Exercise 22.45. If n girls and n boys sit in a circle of chairs, what is the
probability that the girls all sit together and the boys all sit together?
Exercise 22.46. If Alice and Barbara (two girls) and Alan and Bob (two boys)
sit in a circle of chairs, what is the probability that none of the girls are adjacent
and none of the boys are adjacent?
Exercise 22.47. If Alice, Barbara, and Christine (three girls) and Alan, Bob,
and Charlie (three boys) sit in a circle of chairs, what is the probability that
none of the girls are adjacent and none of the boys are adjacent?
Exercise 22.48. If n girls and n boys sit in a circle of chairs, what is the
probability that none of the girls are adjacent and none of the boys are adjacent?
Chapter 23

Two Case Studies in Counting

23.1 Poker Hands

In all of the “hands” below, we are describing the probability of an event associ-
ated with

a52!selection of 5 cards from a well-shuffled deck, so that in every case,
all 52
5 = 5!47! = 2,598,960 possible hands are equally likely. We do not take the
order of the dealing of the cards into consideration. (If the reader wants to take
the order in account, then multiply the numerator by 5! and the denominator
by 5!, for each probability in this section, and these 5!’s will cancel out.)

23.1.1 Straight Flush

A straight flush consists of 5 cards from the same suit that can be placed in
order. There are four possible suits for a straight flush. Within each suit, there
are ten types of straights:

(A, 2, 3, 4, 5) (2, 3, 4, 5, 6) (3, 4, 5, 6, 7) (4, 5, 6, 7, 8) (5, 6, 7, 8, 9)

(6, 7, 8, 9, 10) (7, 8, 9, 10, J) (8, 9, 10, J, Q) (9, 10, J, Q, K) (10, J, Q, K, A)

Since there are 4 suits, there are (4)(10) = 40 types of possible straight flushes.
So the probability of getting a straight flush in one 5-card hand is 40/2598960.

23.1.2 Four of a Kind

A four of a kind consists of four cards of the same rank (a quadruple, e.g., four
9’s) and also one other card from another rank. There are 13 possible ranks for
the quadruple:
A, 2, 3, 4, 5, 6, 7, 8, 9, 10, J, Q, K;
for each such choice, there remain 12 possible ranks for the single card. Once
the rank of the single card is chosen, it can have any of four possible suits.
So there are (13)(12)(4) = 624 types of possible four of a kinds. Thus, the
probability of getting a four of a kind in one 5-card hand is 624/2598960.

293
294 Chapter 23. Two Case Studies in Counting

23.1.3 Full House

A full house consists of a triple of cards from one rank and a pair of cards from
another rank. There are 13 possible ranks for the triple; for each such choice,
there remain
12 possible ranks for the pair. Once the

ranks have been selected,
4 4
there are 3 = 4 ways to pick the triple, and 2 = 6 ways to pick the pair.
So there are (13)(12)(4)(6) = 3744 types of possible full houses. Therefore, the
probability of getting a full house in one 5-card hand is 3744/2598960.

23.1.4 Flush
A flush consists of five cards

from the same suit. There are 4 possible suits;
for each suit, there are 135 = 1287 possible cards from that suit. So there
are (4)(1287) = 5148 types of possible flushes. Usually the straight flushes are
removed from this classification (because they have a classification all on their
own); therefore, there are 5148−40 = 5108 types of flushes that are not straight
flushes. so the probability of getting a flush (but not a straight flush) in one
5-card hand is 5108/2598960.

23.1.5 Straight
A straight consists of five cards that can be placed in order. There are 10
possible sets of values for the cards in the straight, as seen

in the straight flush
4
example above; for each type of straight, there are 1 = 4 ways to pick a
card from each of the ranks, so there are (4)(4)(4)(4)(4) = 45 ways to pick the
cards altogether. So there are (10)(45 ) = 10,240 types of possible straights.
Usually the straight flushes are removed from this classification (because they
have a classification all on their own), which leaves 10,240 − 40 = 10,200 types
of straights that are not straight flushes. Thus, the probability of getting a
straight (but not a straight flush) in one 5-card hand is 10200/2598960.

23.1.6 Three of a Kind

A three of a kind consists of a triple of cards from one rank and two other cards
from two different, other ranks. There are 13 possible ranks for the triple; for
each such choice, there remain 12 possible ranks for the other two cards.


2 = 66
Once the ranks have been selected, there are 43 = 4 ways to pick the triple,

and 41 = 4 ways to pick the greater of the two non-matched cards, and 41 = 4

ways to pick the lesser of the two non-matched cards. Combining these, we have
(13)(66)(4)(4)(4) = 54,912 types of possible three of a kinds. So the probability
of getting three of a kind in one 5-card hand is 54912/2598960.

23.1.7 Two Pair

Two pair consists of a pair of cards from one rank, another pair of cards

from
13
a different rank, and one other cards from a third rank. There are 2 = 78
possible ranks for the two pairs; and then there always remains 11 ways to pick
 23.2. Yahtzee 295

the rank for the non-matched card. for each such choice, there are 42 = 6 ways

to pick the pair of cards in the greater-valued pair, and 42 = 6 ways to pick

the pair of cards in the lesser-valued pair. There are 41 = 4 ways to pick the

non-matched card. This gives us (78)(11)(6)(6)(4) = 123,552 types of possible

two pairs, and therefore the probability of getting two pairs in one 5-card hand
is 123552/2598960.

23.1.8 One Pair

One pair consists of a pair of cards from one rank, and then three other un-
matched cards, all from unique ranks. There

are 13 possible ranks for the one
12
pair; and then there always remains 3 = 220 ways to pick the ranks for
the three non-matched cards. For each such choice, there are 42 = 6 ways to

pick the pair of cards in the pair, and 41 = 4 ways to pick each of the non-

matched cards. So there are (13)(220)(6)(4)(4)(4) = 1,098,240 types of possible

one pairs. Finally, the probability of getting one pair in one 5-card hand is
1098240/2598960.

23.2 Yahtzee
In Yahtzee, five dice (each is a standard die, with 6 sides) are rolled, so that
there are 65 = 7776 equally likely outcomes. It is usually helpful to think of the
five dice separately, to keep the outcomes clear in one’s mind. For instance, one
might think of five differently colored dice, or five dice that are rolled by five
numbered players 1, 2, 3, 4, 5, etc. Thus, the scenario is a bit different than in
the poker hands case study above.
In Yahtzee, we are allowed to roll the dice up to three times, but
for simplicity here, we will only roll each die one time.

23.2.1 Upper Section

Fix a desired value, for instance, “3,” that a player wants to appear on a die.
The number of dice X that show a specific value (in our case, “3”) is Binomial
with parameters n = 5 and p = 1/6. So

5!55−j
 
5
P (X = j) = (1/6)j (5/6)5−j = .
j j!(5 − j)!65

For another method of computing this probability, note that all 65 outcomes
are equally likely. In order to have
exactly j dice show a certain value (for
instance, the value “3”), there are 5j ways to pick which dice will show that
value. The 5 − j dice that will not equal the desired value can appear in 5
ways each (e.g., the other

5 − j dice can each be anything except “3”). This
means that there are 5j 55−j possible outcomes in which there are exactly j
296 Chapter 23. Two Case Studies in Counting

occurrences of a specific desired value. The desired probability is

5 5−j


j 5 5!55−j
= .
65 j!(5 − j)!65

23.2.2 Three of a Kind

To obtain three of a kind, there must be exactly 3 occurrences of some value.
Fix a desired value, for instance, “1,” that a player wants to get for the three
of a kind. The player needs 3 of the 5 dice to equal value “1.” Then the other
two dice need to have two different, distinct values. Thus, the probability of
getting a three of a kind that consists of exactly three occurrences of value “1”
(without being a full house, i.e., without the other pair matching) is
 
5 5!(5)(4)
(1/6)3 (5/6)(4/6) = = 25/972 = 0.02572.
3 3!(5 − 3)!65
There are 6 possible values for the three of a kind, and these are all disjoint, so
the total desired probability is
 
25 25
(6) = = 0.1543.
972 162

For another method of computing this probability, note that all 65 outcomes
are equally likely. There are six ways to choose the value for the three of a kind,

e.g., “1.” To get exactly three values equal to this specific value, there are 53
ways to pick the dice that must show this value; there are (5)(4) ways that the
other two dice can show two
other values that are distinct. Combining these,
we see that there are (6) 53 (5)(4) possible outcomes in which there are exactly
3 occurrences of one of the values, without the other pair matching. Therefore,
the desired probability is
(6) 53 (5)(4)


25
5
= = 0.1543.
6 162

23.2.3 Four of a Kind

To obtain four of a kind, there must be exactly 4 occurrences of some value.
Fix a desired value, for instance, “3,” that a player wants to get for the four
of a kind. The number of dice X that show that specific value (in our case, “3”)
is Binomial with parameters n = 5 and p = 1/6. So we have
 
5 5!5
P (X = 4) = (1/6)4 (5/6)5−4 = = 25/7776 = 0.003215.
4 4!(5 − 4)!65
There are 6 possible values for the four of a kind, and these are all disjoint;
thus, the total desired probability is
 
25 25
(6) = = 0.01929.
7776 1296
 23.2. Yahtzee 297

For another method of computing this probability, note that all 65 outcomes
are equally likely. There are six ways to fix a desired value of j. In order to
have exactly four values equal to this specific j, there are ways to pick which
5
of the 4 dice show this value; the remaining 5 − 4 = 1 die can show any of

the other 5 values. So there are (6) 54 5 possible outcomes in which there are

exactly 4 occurrences of one of the values. So the chance of a four of a kind is

(6) 54 5


25
5
= = 0.01929.
6 1296

23.2.4 Full House

To obtain a full house, there must be three occurrences of one value, and two
occurrences of a different value.
There are 6 ways to pick the value for the triple (e.g., “5”s), and 5 remaining
ways to pick
the value for the pair (e.g., “1”s). Once these values are chosen,
there are 53 = 10 ways to decide which dice will get the triple (and the other
two will get the pair). So the desired probability is

(6)(5) 53


300 25
= = = 0.03858.
65 7776 648

23.2.5 Small Straight

To obtain a small straight, which is four dice in consecutive order, there are
3 ways to choose which kind of small straight, i.e., either “1 through 4,” or “2
through 5,” or “3 through 6.”
For the type “1 through 4,” the other die cannot be a 5, or then we have a
large straight. We either have 1–4, along with 6; or we have 1–4, with one value
duplicated. There are 5! ways to have 1–4 along with 6. To have 1–4, with one
value duplicated, there are 4 ways to choose which value is duplicated, and then
there are 52 = 10 ways to choose which dice get that repeated value, and then

there are 3! ways to choose how the other 3 values are arranged. So there are
 
5
5! + (4) (3!) = 360 ways.
2
For the type “2 through 5,” the other die cannot be 1 or 6, or then we have
a large straight. We have 2–5, with one value duplicated. Thus,
there are 4
ways to choose which value is duplicated, and then there are 52 = 10 ways to
choose which dice get that repeated value, and then there are 3! ways to choose
how the other 3 values are arranged. So, altogether, there are
 
5
(4) (3!) = 240 ways.
2
For the type “3 through 6,” the other die cannot be a 2, or then we have a
large straight. We either have 3–6, along with 1; or we have 3–6, with one value
298 Chapter 23. Two Case Studies in Counting

duplicated. There are 5! ways to have 3–6 along with 1. To have 3–6, with one
value duplicated, there are 4 ways to choose which value is duplicated, and then
there are 52 = 10 ways to choose which dice get that repeated value, and then

there are 3! ways to choose how the other 3 values are arranged. This yields
 
5
5! + (4) (3!) = 360 ways.
2

Altogether, there are 360 + 240 + 360 = 960 ways that the results can be
arranged on the dice. So the desired probability is
960 10
5
= = 0.1235.
6 81

23.2.6 Large Straight

To obtain a large straight, which is five dice in consecutive order, there are
2 ways to choose which kind of large straight, i.e., either 1 through 5, or 2
through 6. Once the kind has been chosen, there are 5! ways that the results
can be arranged among the 5 distinct dice. Thus, the desired probability is

(2)(5!) 5
5
= = 0.03086.
6 162

23.2.7 Yahtzee
To obtain a “yahtzee,” which is just five of a kind, there must be exactly 5
occurrences of some value.
Fix a desired value (e.g., “3”) that a player wants to get for the yahtzee, i.e.,
for the Yahtzee. The number of dice X that show that specific value (in our
case, “3”) is Binomial with parameters n = 5 and p = 1/6. So
 
5 5!
P (X = 5) = (1/6)5 (5/6)5−5 = = 1/65 = 1/7776 = 0.0001286.
5 5!(5 − 5)!65

There are 6 possible values for the yahtzee, and these are all disjoint. So we
conclude that the total desired probability is
 
1 1
(6) = = 0.0007716.
7776 1296

For another method of computing this probability, note that all 65 outcomes
are equally likely. There are six ways to fix a desired value of j. Once the desired
value of the yahtzee is chosen, then all of the dice must equal the value. So the
desired probability is
6 1
5
= = 0.0007716.
6 1296
Part V

Continuous Random Variables

You have learned how to calculate probabilities for discrete random variables.
Earlier in the text we discussed the differences between discrete and continuous
random variables. In the chapters that follow, you will learn how to calculate
probabilities for continuous random variables. Afterwards, we will apply these
rules and formulas in the next part of the book with five different types of named
continuous random variables.
By the end of this part of the book, you should be able to:

1. Distinguish between discrete and continuous random variables when read-

ing a story.

2. Calculate probabilities for continuous random variables.

3. Calculate and graph a density (i.e., probability density function, PDF).

4. Calculate and graph a CDF (i.e., a cumulative distribution function)

5. Calculate the mean, variance, standard deviation, and median of the con-
tinuous random variable.

Math skills you will need: integrals of one and two variables (in particular,
integration by parts), derivatives, ex , and ln x.
Additional resources: Computer programs (such as Excel, Maple, Mathe-
matica, Matlab, Minitab, R, SPSS, etc.) and calculators may be used to assist
in the calculations.

299
Chapter 24

Continuous Random Variables

and PDFs

The true logic of this world is in the calculus of probabilities.

—James Clerk Maxwell

The number of passengers on a randomly chosen bus is an integer, but the speed
of the bus is a real-valued number. The height of a tree is real-valued, but the
number of leaves is discrete.

24.1 Introduction
In Chapter 7, you already learned the difference between a discrete random
variable and a continuous random variable. Here’s a review:
Discrete means that you could list list the specific possible outcomes that
the variable can take (there are either a finite number of them, or a countably
infinite number). An example is X ∈ {0, 1, 2, 3, 4, 5, . . .}, for the number of
4-leaf clovers you will find in a 1-acre pasture this afternoon.
Continuous means that the variable takes on a range of values. You could
usually state the beginning and end points, but you would have infinitely many
possibilities of answers within that range. An example is 62.8 ≤ X ≤ 67.0, for
how many ounces of soda will actually be in the next 64-ounce bottle you open.
We give a table with an overview of the differences between discrete and
continuous random variables. As you can see, many of the formulae are very
similar in structure. A key difference is that for discrete random variables you
use summations, and for continuous random variables you use integrals.

300
 24.1. Introduction 301

Discrete Continuous
probability mass (probability mass density (probability density
function function; PMF) function; PDF)
0 ≤ pX (x) ≤ 1 0 ≤ fX (x)
(not necessarily ≤ 1)
P R∞
x pX (x) =1 −∞ fX (x) dx = 1
P (0 ≤ X ≤ 2) P (0 ≤ X ≤ 2)
R2
= P (X = 0) + P (X = 1) + = 0 fX (x) dx
P (X = 2)
if X is integer valued
P (X ≤ 3) 6= P (X < 3) P (X ≤ 3) = P (X < 3)
when P (X = 3) 6= 0 since P (X = 3) = 0 always
cumulative FXP(a) = P (X ≤ a) R a = P (X ≤ a)
FX (a)
distribution = x≤a P (X = a) = −∞ fX (x) dx
function graph of CDF is a graph of CDF is
(CDF) step function with jumps nonnegative and
FX (x) of the same size as continuous, rising
the mass, from 0 to 1 up from 0 to 1
examples counting: defects, hits, lifetimes, waiting times,
die values, coin heads/tails, height, weight, length,
people, card arrangements, proportions, areas, volumes,
trials until success, etc. physical quantities, etc.
named Bernoulli, Binomial, Continuous Uniform,
distributions Geometric, Negative Exponential, Gamma,
Binomial, Poisson, Beta, Normal
Hypergeometric,
Discrete Uniform
R∞
expected
P
E(X) = x xpX (x) E(X) = −∞ xfX (x) dx
R∞
value
P
E(g(X)) = x g(x)pX (x) E(g(X)) = −∞ g(x)fX (x) dx
R∞
E(X 2 ) E(X 2 ) = x x2 pX (x) E(X 2 ) = 2
P
−∞ x fX (x) dx
variance Var(X) = Var(X) =
E(X 2 ) − (E(X))2 E(X 2 ) − (E(X))2
std. dev.
p p
σX = Var(X) σX = Var(X)

A continuous random variable takes on values within an interval of finite

or infinite length. For instance, X could be a waiting time, such as the time
until the phone rings, or the time until the next email arrives, or the time until
you win a lottery game. In all of these cases, X could take on any nonnegative
real number. As another example, X could be the height of a randomly chosen
person, in which case X could be any real number from (say) 0 to 9 (height
given in feet). Another example is that X is the horizontal distance to the left
or right of the location where a dart lands, in comparison to the center of a dart
board.
302 Chapter 24. Continuous Random Variables and PDFs

The probabilities associated with a continuous random variable are obtained

from the probability density function, often abbreviated as “PDF” or just
called the density. Densities have some things in common with masses, but
some things are different too. A density is only used with a continuous random
variable. A mass is only used with a discrete random variable. Every density
has the following properties:

Definition 24.1. Density of a Continuous Random Variable

If X is a continuous random variable, and fX (x) is the density of X, then:
1. The density is always nonnegative, i.e., fX (x) ≥ 0 for all real numbers x.
2. The density, integrated from a to b, gives the probability that X is between
a and b, i.e.,
Z b
P (a ≤ X ≤ b) = fX (x) dx.
a

To calculate probabilities from a density, we just integrate the density over

the range of interest. E.g., we find the probability that X is between 3.2 and
5.82 by integrating the density over [3.2, 5.82],
Z 5.82
P (3.2 ≤ X ≤ 5.82) = fX (x) dx.
3.2

Remark 24.2. Integral of a Density over the entire real line is 1

If X is a continuous random variable, and if fX (x) is the density of X, then

P (−∞ < X < ∞) = 1,

so the integral over the whole real line must be 1:

Z ∞
fX (x) dx = 1.
−∞

How are the mass and density alike?

• Both are building blocks for probabilities, the CDF, the expected value,
and the variance.

• Both are always nonnegative:

– Mass (discrete): 0 ≤ pX (x) ≤ 1

– Density (continuous): 0 ≤ fX (x)

• The sum of the mass over all possible x values is 1. The integral of the
mass over all possible x values is 1.

– Mass (discrete): x pX (x) = 1

P
R∞
– Density (continuous): −∞ fX (x) dx = 1
 24.1. Introduction 303

How are the mass and density different?

• Continuous versus isolated points:

– The mass consists of positive probabilities for individual points (cor-

responding to the possible outcomes)
– The densities we consider are piecewise continuous on the real line.1

• Adding versus integrating:

– Summing values of the mass gives probabilities.

– Integrating the density over an interval gives probabilities.

• Maximum values:

– The mass has to be between 0 and 1 for every x value.

– The density has no upper limit. E.g., if fX (x) = 10 for 0 ≤ x ≤ 1/10,
and fX (x) = 0 otherwise, then fX (x) is a density.

• At a single point:

– The mass P (X = x) = pX (x) is strictly positive if x corresponds to

a possible outcome of the random variable (pX (x) = 0 otherwise).
– For continuous X, the probability P (X = x) is 0 for any specific
value x. This corresponds to integrating a random variable over just
one point, i.e.,
Z x
P (X = x) = P (x ≤ X ≤ x) = fX (x) dx = 0.
x

Remark 24.3. Probability that a Continuous Random Variable is

equal to a specific value
If X is a continuous random variable, and if a is any particular value, then X
has probability 0 of being equal to that particular value. To see this, we just
write Z a
P (X = a) = P (a ≤ X ≤ a) = fX (x) dx = 0.
a

Does it matter whether we include the endpoints of an interval

when computing probabilities with continuous random variables? No!
Since P (X = a) = 0 and P (X = b) = 0 for any values a, b, then all four of the
1
In more advanced courses, densities can be defined using the notion of measure, but we
do not consider such generalities at this level of presentation. See books such as Billingsley [1]
or Durrett [2] for a more advanced treatment.
304 Chapter 24. Continuous Random Variables and PDFs

following values are the same:

Z b
P (a ≤ X ≤ b) = fX (x) dx
a
Z b
P (a ≤ X < b) = fX (x) dx
a
Z b
P (a < X ≤ b) = fX (x) dx
a
Z b
P (a < X < b) = fX (x) dx
a
Any nonnegative function for which “the integral of the function
over the whole real line is 1” is the density of a continuous random
variable.
Remark 24.4. Creating a Continuous Random R ∞Variable
If g(x) is any integrable, nonnegative function, and −∞ g(x)dx = 1, then g(x)
is the density for a continuous random variable. We can write g(x) = fX (x)
and then the associated random variable X has the property that
Z b
P (a ≤ X ≤ b) = fX (x) dx.
a

The density and CDF are related just like derivatives and integrals are re-
lated.
Remark 24.5. Densities and CDFs
The CDF is the integral of the density, i.e.,
Z a
FX (a) = fX (x) dx,
−∞

so the density is the derivative of the CDF:

d
fX (x) = FX (x).
dx
The second equation only holds at x for which the CDF is differentiable.

24.2 Examples

Example 24.6. If X is a continuous random variable with density

(
1
(4x + 1) if 2 ≤ x ≤ 4,
fX (x) = 26
0 otherwise,
what is P (1 ≤ X ≤ 3)?
 24.2. Examples 305

The density of X is shown on the left of Figure 24.1.

Remember that P (1 ≤ X ≤ 3) is the region under the curve of the density.
So we need to integrate the density over the range 1 ≤ x ≤ 3. Since the density
is 0 in the left-hand-portion of this range, i.e., for 1 ≤ x ≤ 2, then it suffices to
integrate the density only over the x’s for which the density is nonzero, i.e., for
2 ≤ x ≤ 3. Thus

P (1 ≤ X ≤ 3) = P (2 ≤ X ≤ 3)
Z 3
1
= (4x + 1) dx
2 26
3
1
= (2x2 + x)
26 x=2
1
((2)(32 ) + 3) − ((2)(22 ) + 2)


=
26
= 11/26
= 0.4231

1 fX (x) 1 FX (x)

0.5 0.5

x x
1 2 3 4 5 1 2 3 4 5

Figure 24.1: Left: The density fX (x) = 26 1

(4x + 1). Right: The CDF
FX (x) = 26 (2x + x − 10), of a random variable defined on [2, 4].
1 2

Example 24.6 (continued) If X is a continuous random variable with density

(
1
(4x + 1) if 2 ≤ x ≤ 4,
fX (x) = 26
0 otherwise,

what is the CDF?

Remember that the CDF of a random variable X is

FX (x) = P (X ≤ x).
306 Chapter 24. Continuous Random Variables and PDFs

Since X is always between 2 and 4, then the CDF is

FX (x) = P (X ≤ x) = 0 for x < 2,

and
FX (x) = P (X ≤ x) = 1 for 4 < x.
The most interesting values of the CDF happen for 2 ≤ x ≤ 4. We compute
the density from −∞ to a, where a is in the interval [2, 4], but the density is 0
for x < 2. Thus, it suffices to integrate over the interval [2, a].
Z a
1
P (X ≤ a) = (4x + 1) dx
2 26
a
1
= (2x2 + x)
26 x=2
1
((2)(a ) + a) − ((2)(22 ) + 2)
2


=
26
1
= (2a2 + a − 10)
26
Thus the CDF of X is:

if x < 2,

0

FX (x) = 1
(2x2 + x − 10) if 2 ≤ x ≤ 4,
 26
1 if 4 < x.


The CDF of X is shown on the right of Figure 24.1.

Notice that the CDF for a continuous random variable—just like the CDF for a
discrete random variable—is a nondecreasing function that is between 0 and 1.
On the other hand, the CDF for a discrete random variable has a jump at each
point where the random variable takes its mass, but the CDF for a continuous
random variable is continuous, with a limit of 0 as x → −∞ and a limit of 1 as
x → ∞.

Example 24.6 (continued) If X is a continuous random variable with density

(
1
(4x + 1) if 2 ≤ x ≤ 4,
fX (x) = 26
0 otherwise,

we calculate some probabilities using the CDF.

The probability that X is 3 or smaller is
1
P (X ≤ 3) = FX (3) = ((2)(32 ) + 3 − 10) = 11/26 = 0.4231.
26
 24.2. Examples 307

The probability that X is between 2.8 and 4.3 is

P (2.8 ≤ X ≤ 4.3) = P (X ≤ 4.3) − P (X ≤ 2.8)

= FX (4.3) − FX (2.8)
1
= 1 − ((2)(2.82 ) + 2.8 − 10)
26
= 0.6738

The probability that X is at least 3 is

1
P (X ≥ 3) = 1 − P (X ≤ 3) = 1 − FX (3) = 1 − ((2)(32 ) + 3 − 10) = 0.5769.
26
Given that X is less than 3, what is the probability X is greater than 2.8?

P (X > 2.8 and X < 3)

P (X > 2.8 | X < 3) =
P (X < 3)
P (X ≤ 3) − P (X ≤ 2.8)
=
P (X ≤ 3)
FX (3) − FX (2.8)
=
FX (3)

We already know FX (3) = 0.4231. Also, FX (2.8) = 1 2

26 ((2)(2.8 ) + 2.8 − 10) =
0.3262. Thus
0.4231 − 0.3262
P (X > 2.8 | X < 3) = = 0.229.
0.4231

Example 24.7. If X is a continuous random variable with CDF

(
1 − e−x/4 if x > 0,
FX (x) =
0 otherwise,

what is the density fX (x)?

The CDF of X is shown on the right of Figure 24.2.

As before, a key thing to remember is that, since the CDF is the integral of
the density, then the density is the derivative of the CDF. Thus, for x > 0,
d d 1
fX (x) = FX (x) = (1 − e−x/4 ) = e−x/4 ,
dx dx 4
and fX (x) = 0 otherwise.
Now calculate some probabilities using the CDF.
The probability that X is less than 4 is

P (X < 4) = FX (4) = 1 − e−4/4 = 1 − e−1 = 0.6321.

308 Chapter 24. Continuous Random Variables and PDFs

0.3
1
fX (x) FX (x)

0.2

0.5
0.1

x x
2 4 6 8 10 2 4 6 8 10

Figure 24.2: Left: The density fX (x) = 41 e−x/4 . Right: The CDF FX (x) =
1 − e−x/4 , of a random variable with nonnegative density for x > 0.

The probability that X is at least 1 is

P (X > 1) = 1 − P (X ≤ 1) = 1 − FX (1) = 1 − (1 − e−1/4 ) = e−1/4 = 0.7788.
Given that X is less than 4, what is the probability that X is also less than 1?
P (X < 1 and X < 4) P (X < 1) FX (1)
P (X < 1 | X < 4) = = = .
P (X < 4) P (X < 4) FX (4)
Using the closed form for the CDF, we conclude:
1 − e−1/4
P (X < 1 | X < 4) = = 0.3499.
1 − e−4/4

Example 24.8. Suppose a random variable X has a density given by:

(
k(x3 + 1)(3 − x) for 1 ≤ x ≤ 3,
fX (x) =
0 otherwise.
Find the constant k so that this function is a valid density.
The integral of fX (x) over all x needs to evaluate to 1, for fX (x) to be a density.
In this case, the density is zero for x outside the range [1, 3]. So we just integrate
over [1, 3] and solve for k, as follows:
Z 3
1= k(x3 + 1)(3 − x) dx
1
Z 3
= k(−x4 + 3x3 − x + 3) dx
1
 5 3
x 3x4 x2
=k − + − + 3x
5 4 2 x=1
 5
3(3 ) 32
4
  5
3(14 ) 12

3 1
=k − + − + (3)(3) − − + − + (3)(1)
5 4 2 5 4 2
= (k)(68/5)
 24.2. Examples 309

Thus k = 5/68 = 0.0735.

Example 24.9. Sometimes a density can have more than two pieces. As an
example:
3/4 if 0 ≤ x ≤ 1,


fX (x) = 1/4 if 3 ≤ x ≤ 4,
otherwise.

0


The density is shown on the left of Figure 24.3. Find the CDF.
We integrate to find the CDF in five disjoint regions:
For a < 0, we have Z a
FX (a) = 0 dx = 0.
−∞
For 0 ≤ a ≤ 1, we have
Z 0 Z a
3
FX (a) = 0 dx + 3/4 dx = a.
−∞ 0 4
For 1 ≤ a ≤ 3, we have
Z 0 Z 1 Z a
3
FX (a) = 0 dx + 3/4 dx + 0 dx = .
−∞ 0 1 4
For 3 ≤ a ≤ 4, we have
Z 0 Z 1 Z 3 Z a
3 1
FX (a) = 0 dx + 3/4 dx + 0 dx + 1/4 dx = + (a − 3).
−∞ 0 1 3 4 4
For 4 ≤ a, we have
Z 0 Z 1 Z 3 Z 4 Z a
3 1
FX (a) = 0 dx + 3/4 dx + 0 dx + 1/4 dx + 0 dx = + = 1.
−∞ 0 1 3 4 4 4
Thus, the CDF of X is


0 if x < 0,
if

3
4x 0 ≤ x ≤ 1,



FX (x) = 43
if 1 < x < 3,
4 + 4 (x − 3) if
 3 1
3 ≤ x ≤ 4,




if

1 4 < x.


The CDF is shown on the right of Figure 24.3. By the way, the median can be
read directly from figure for this CDF. We have FX (x) = 1/2 when x = 2/3,
since (looking at the region where 0 ≤ x ≤ 1 and FX (x) = 43 x), we can solve
FX (x) = 34 x = 1/2 to get x = 2/3.
310 Chapter 24. Continuous Random Variables and PDFs

1 fX (x) 1 FX (x)

0.5 0.5

x x
2 4 6 2 4 6

Figure 24.3: Left: The density fX (x). Right: The CDF FX (x), of the
random variable described in Example 24.9, in which the density is nonnegative
on two disjoint regions.

Example 24.10. Check to make sure that the following function is a valid
density:
3
fX (x) = (x)(2 − x) for 0 ≤ x ≤ 2,
4
and fX (x) = 0 otherwise. See Figure 24.4. Also, find P (X ≤ 1).

1
fX (x)

0.5

x
0.5 1 1.5 2

Figure 24.4: The density function fX (x) = 34 (x)(2 − x) for 0 ≤ x ≤ 2.

The only time when fX (x) is not equal to zero is for 0 < x < 2. In
this region, x > 0 and 2 − x > 0 so fX (x) > 0 too. Thus, fX (x) is always
nonnegative. Since X is always between 0 and 2, we calculate
Z ∞ Z 2
3
fX (x) dx = (x)(2 − x) dx = 1,
−∞ 0 4

so fX (x) is a density (nonnegative, and integrates to 1).

 24.2. Examples 311

Now we find the probability that X ≤ 1. Since fX (x) is 0 when −∞ < x < 0,
we compute
Z 1 Z 1
3
P (X ≤ 1) = fX (x) dx = (x)(2 − x) dx = 1/2.
−∞ 0 4

Initially we might believe that the probability is 1/2 since we included half
of the interval from 0 to 2 in the previous computation, but the situation is not
always so simple. For instance, the following example does not have a symmetric
density:

Example 24.11. Suppose that a random variable X has density

3
fX (x) = (x)(2 − x)(3 − x) for 0 ≤ x ≤ 2,
8
and fX (x) = 0 otherwise. Find P (X ≤ 1). See Figure 24.5.

1
fX (x)

0.5

x
0.5 1 1.5 2

Figure 24.5: The density function fX (x) = 38 (x)(2 − x)(3 − x) for 0 ≤ x ≤ 2.

We compute
Z 1 Z 1
3
P (X ≤ 1) = fX (x) dx = (x)(2 − x)(3 − x) = 19/32.
−∞ 0 8

Example 24.12. One of the most frequently used types of continuous random
variables are constant on an interval, and zero otherwise. For instance, consider
the density
fX (x) = 1/6 for 4 < x < 10,
and fX (x) = 0 otherwise.
312 Chapter 24. Continuous Random Variables and PDFs

We notice that fX (x) ≥ 0 for all x, and also

Z ∞ Z 10
fX (x) dx = 1/6 dx = (1/6)(6) = 1.
−∞ 4

So this function really is a density. In fact, for any constants a and b inside the
interval, i.e., with 4 ≤ a ≤ b ≤ 10, we have
Z b
P (a ≤ X ≤ b) = 1/6 dx = (b − a)/6.
a

For example,
Z 5
P (2 ≤ X ≤ 5) = 1/6 dx = (5 − 2)/6 = 1/2.
2

Such nice facts work because, when we integrate a constant value over an inter-
val, the result is just the constant times the length of the interval. For instance,
when we integrate 1/6 on the interval [2, 5], the result is just 1/6 times the
length of the interval [2, 5] (i.e., 3). So the result is (1/6)(3) = 1/2. This is an
example of the Uniform density, to be considered more in Chapter 31.

Example 24.13. Let

fX (x) = 3e−3x for x > 0,

and fX (x) = 0 otherwise. See Figure 24.6.

3
fX (x)

x
1 2 3 4

Figure 24.6: The density fX (x) = 3e−3x for x > 0.

In such a case, X assumes only nonnegative values. The density fX (x) of

 24.3. Exercises 313

X is always nonnegative, i.e., fX (x) ≥ 0 for all x. Also, we compute

Z ∞ Z ∞
fX (x) dx = 3e−3x dx
−∞ 0
∞
e−3x
=3 dx
−3 x=0
= −e−3x |∞
x=0
= lim (−e−3x ) − (−e−(3)(0) )
x→∞
=0+1
=1
So fX (x) is a density.
The probability that X ≤ 2 is
Z 2
P (X ≤ 2) = fX (x) dx
−∞
Z 2
= 3e−3x dx
0
2
e−3x
=3
−3 x=0
= −e−3x |2x=0
= (−e−(3)(2) ) − (−e−(3)(0) )
= 1 − e−6
= 0.997521
The density considered here is Exponential, to be revisited in Chapter 32.

24.3 Exercises
24.3.1 Practice
Exercise 24.1. Identify from the information below whether X is a discrete or
continuous random variable.
a. Let X be the height of a randomly selected 8-year-old child.
b. Let X be a random variable that takes values on the nonnegative integers,
i.e., X ∈ {0, 1, 2, 3, . . .}.
c. Let X be a random variable that takes values on all nonnegative real
numbers, i.e., X ≥ 0.
d. Let X be the number of 8-year-old children who will attend the 1:30
showing of the movie “Shrek 3.”
314 Chapter 24. Continuous Random Variables and PDFs

Exercise 24.2. Identify from the information below whether X is a discrete or

continuous random variable.

a. Let X be a random variable that takes values on the set {3, 4, 5, 6, 7}

b. Let X be the number of defective light bulbs in a shipment.
c. Let X be a random variable that takes values on the real interval [3, 7]

Exercise 24.3. If you know P (X > 2) = 0.3, fill in the chart for the other
values you know. Write “???” if there is not enough information to figure out a
value.

X is discrete X is continuous
P (X ≥ 2)
P (X < 2)
P (X ≤ 2)
P (X = 2)

Exercise 24.4. If you know P (X ≤ 5) = 0.9, fill in the chart for the other
values you know. Write “???” if there is not enough information to figure out a
value.

X is discrete X is continuous
P (X < 5)
P (X > 5)
P (X ≥ 5)
P (X = 5)

Exercise 24.5. Let X have density fX (x) = kx2 (1 − x)2 for 0 ≤ x ≤ 1, and
fX (x) = 0 otherwise, where k is constant.

a. Find the value of k.

b. Find P (X ≥ 3/4).
√
3(x+2)
Exercise 24.6. Let X have density fX (x) = 6 for −2 ≤ x ≤ 1, and
fX (x) = 0 otherwise. Find the probability that X is positive.

Exercise 24.7. Let X have density

1
fX (x) = , for 2 ≤ x ≤ 7,
5
and fX (x) = 0 otherwise.

a. Find P (X > 3).

b. Find the CDF FX (x).

c. Find P (X < 5).

 24.3. Exercises 315

d. Find P (X > 8).

e. Find P (5 ≤ X ≤ 6.5).

f. Find P (X < 1).

g. Graph the density fX (x).

h. Graph the CDF FX (x).

Exercise 24.8. Let X be the waiting time (in minutes) until a student’s friend
arrives. Suppose that X has density
1
fX (x) = e−x/3 , for 0 < x,
3
and fX (x) = 0 otherwise.

a. Find P (3 ≤ X ≤ 6).

b. Find the CDF FX (x).

c. Find P (X ≥ 24).

d. Find P (X ≤ −3).

e. Find P (3 ≤ X ≤ 12).

f. Find P (X < 300).

g. Graph the density fX (x).

h. Graph the CDF FX (x).

Exercise 24.9. Consider the function from Example 24.12, i.e., fX (x) = 1/6
for 4 < x < 10, and fX (x) = 0 otherwise.

a. What is the cumulative distribution function FX (x)?

b. Draw a picture of the cumulative distribution function FX (x).

Exercise 24.10. Consider the function from Example 24.13, i.e., fX (x) =
3e−3x for x > 0, and fX (x) = 0 otherwise.

a. What is the cumulative distribution function FX (x)?

b. Draw a picture of the cumulative distribution function FX (x).

Exercise 24.11. Let

√
fX (x) = x/4, for 0 ≤ x ≤ 8,

and fX (x) = 0 otherwise. What is the probability that X exceeds 1?

316 Chapter 24. Continuous Random Variables and PDFs

Exercise 24.12. Let

3 2
fX (x) = x , for 0 ≤ x ≤ 4,
64
and fX (x) = 0 otherwise. Find P (X > 2).
Exercise 24.13. Let

fY (y) = 12y 2 (1 − y), for 0 ≤ y ≤ 1,

and fY (y) = 0 otherwise. Find the CDF of Y .

Exercise 24.14. Suppose that X has CDF

if x < 3,

0

FX (x) = 171 (x − 6x − 9) if 3 ≤ x ≤ 6,
1 3

if 6 < x.

1


Find the density fX (x).

24.3.2 Extensions
Exercise 24.15. Suppose that X has density
1
fX (x) = sin x, for 0 ≤ x ≤ π,
2
and fX (x) = 0 otherwise.

a. Find P (X ≤ π/6).

b. Find P (π/3 ≤ X ≤ 2π/3).

c. Find P (π/3 ≤ X).

Exercise 24.16. Let

ln x
fX (x) = , for e ≤ x ≤ e2 ,
e2
and fX (x) = 0 otherwise. Find the CDF of X.
Exercise 24.17. Let c > 0 be a fixed constant. Let

fX (x) = ce−cx , for 0 < x,

and fX (x) = 0 otherwise. Find a general expression for the CDF FX (x), for
any x > 0.
Exercise 24.18. Let
√
FY (y) = y, for 0 ≤ y ≤ 1;

and FY (y) = 0 for y < 0, and FY (y) = 1 for y > 1. Find the density of Y .
 24.3. Exercises 317

Exercise 24.19. Suppose a density fX (x) increases linearly from (16, 0) to

(24, 1/4).

a. Find the CDF of X.

b. What is the value of a so that P (X > a) = 0.75?

Exercise 24.20. What is the constant k that makes the following function a
valid density? (
kx9 (1 − x)2 if 0 ≤ x ≤ 1,
fX (x) =
0 otherwise,

Exercise 24.21. What is the constant k that makes the following function a
valid density? (
k x1 − 3x2 if 1 ≤ x ≤ 20,


fX (x) =
0 otherwise,

Exercise 24.22. Suppose that X has CDF


0
 if x < 0,
if

7 2
8x 0 ≤ x ≤ 1,



FX (x) = 78 if 1 < x < 7,
if
 7 1
8 + 8 (x − 7) 7 ≤ x ≤ 8,




if

1 8 < x.


a. Find the density fX (x).

b. Find the median of X.

Exercise 24.23. Again consider the function from Example R ∞24.13, i.e., let
fX (x) = 3e−3x for x > 0, and fX (x) = 0 otherwise. Evaluate −∞ xfX (x) dx.

Exercise 24.24. Let X be the lifetime (in years) of a carbon-14 atom before
it decays. Then X has density
 
ln 2 ln 2
fX (x) = exp − x , for 0 < x,
5730 5730

and fX (x) = 0 otherwise. Find the length of time a (in years) so that

P (X ≤ a) = 1/2 and P (X > a) = 1/2.

24.3.3 Advanced
Exercise 24.25. Let X have density
1
fX (x) = x2 e−x , for x > 0,
2
and fX (x) = 0 otherwise. Find P (X < 2).
318 Chapter 24. Continuous Random Variables and PDFs

Exercise 24.26. Let k be a fixed, positive integer. Let X have density

fX (x) = xk e−x , for x > 0,

and fX (x) = 0 otherwise. Find the value of k so that fX (x) is a valid density
function.

Exercise 24.27. Comparing two CDFs. Is it possible to define two random

variables X and Y in such a way that the CDF for X is always strictly larger
than the CDF for Y ? In other words, is it possible to have FX (x) > FY (y)
always? If yes, then give an example. If not, then explain why not.

Exercise 24.28. Moments of a Continuous Uniform random variable.

Let X be a Continuous Uniform random variable on the interval [a, b]. Compute
E(X n ).
Chapter 25

Joint Densities

To expect the unexpected shows a thoroughly modern intellect.

—An Ideal Husband, Act 3 by Oscar Wilde (play that premiered on Jan-
uary 3, 1895)

You use a social networking application that sends out notifications for personal
messages and notifications for your friends posting new pictures. You know the
distribution for wait time for a personal message and the distribution of wait
time for a new-picture notification. How long do you expect to have to wait for
either a personal message or a picture notification?

25.1 Introduction
Just as with discrete random variables, we can handle more than one random
variable at a time. We can have a joint probability density function, also called
a joint density, associated with two continuous random variables:
Definition 25.1. Joint probability density function
The joint probability density function—also called a joint density—of a pair of
continuous random variables X and Y is fX,Y (x, y), and it has the following
properties:

1. The joint density is always nonnegative, i.e.,

fX,Y (x, y) ≥ 0 for all x, y.

2. The joint density can be integrated to get probabilities, i.e., if A and B

are sets of real numbers, then
Z Z
P (X ∈ A and Y ∈ B) = fX,Y (x, y) dy dx.
A B

319
320 Chapter 25. Joint Densities

We also have a notion of a joint cumulative distribution function, also called

a joint CDF:
Definition 25.2. Joint cumulative distribution function
If X and Y are random variables, then the joint cumulative distribution func-
tion (also called joint CDF) of X and Y is the probability that X does not
exceed x and Y does not exceed y, i.e.,

FX,Y (x, y) = P (X ≤ x and Y ≤ y).

If we are given the joint density fX,Y (x, y), we know how to compute the
joint CDF.
Remark 25.3. Obtaining the joint CDF from the joint density of a
pair of continuous random variables
If X and Y are a pair of continuous random variables, with joint density
fX,Y (x, y), then X and Y have joint CDF
Z a Z b
FX,Y (a, b) = fX,Y (x, y) dy dx.
−∞ −∞

The observation in the box above can also be viewed as follows: Since
FX,Y (a, b) = P (X ≤ a and Y ≤ b),
then to get −∞ < x ≤ a and −∞ < y ≤ b, we can use A = (−∞, a] and
B = (−∞, b], and we can integrate the joint density in each of the variables:
Z a Z b
FX,Y (a, b) = fX,Y (x, y) dy dx.
−∞ −∞
We can create pairs of random variables by using joint densities that satisfy
properties like those in the previous chapter. Any nonnegative function of x
and y, whose integral over all x and all y is 1, will describe a pair of continuous
random variables.
Remark 25.4. Creating a Pair of Continuous R ∞ R ∞Random Variables
If f (x, y) is any nonnegative function, and −∞ −∞ f (x, y) dy dx = 1, then
f (x, y) is the joint density for a pair of continuous random variables. We can
write f (x, y) = fX,Y (x, y) and then the associated random variables X and Y
have the property that
Z bZ d
P (a ≤ X ≤ b and c ≤ Y ≤ d) = fX,Y (x, y) dy dx.
a c

Note: The reason f (x, y) has to be a nonnegative function in the box above is
that densities are always nonnegative. This doesn’t mean that the associated
random variables have to be nonnegative. For instance, if f (x, y) = 1/4 for
−1 ≤ x ≤ 1 and −1 ≤ y ≤ 1, and f (x, y) = 0 otherwise, then f (x, y) is
nonnegative, but the random variables X and Y associated with this density
could be positive or negative, i.e., −1 ≤ X ≤ 1 and −1 ≤ Y ≤ 1.
 25.2. Examples 321

25.2 Examples
First we present a series of four examples that all focus on the same pair of
random variables. In Example 25.5, we verify that fX,Y (x, y) is a joint density.
In Example 25.6, we consider the joint CDF and we calculate a probability. In
Example 25.7, we study another random variable, Z, that is defined to be the
minimum of X and Y . Finally, in Example 25.8, we compute the probability
that Y is bigger than X, and we do this in two different ways.

Example 25.5. Let X be the time (in minutes) that Maxine waits for a traffic
light to turn green, and let Y be the time (in minutes, at a different intersection)
that Daniella waits for a traffic light to turn green. Suppose that X and Y have
joint density

fX,Y (x, y) = 15e−3x−5y , for x > 0 and y > 0,

and fX,Y (x, y) = 0 otherwise. See Figure 25.1.

10

0 0
0 1
1 2
2 x
y 3 3

Figure 25.1: The joint density fX,Y (x, y) = 15e−3x−5y , for x > 0 and y > 0.

We first check that fX,Y (x, y) is a joint density. We see that fX,Y (x, y) is
nonnegative. Next, we check that the integral over all x’s and y’s is 1:
Z ∞ Z ∞ Z ∞Z ∞
fX,Y (x, y) dy dx = 15e−3x−5y dy dx
−∞ −∞ 0 0

We can factor the 15 into 3 times 5, anticipating the division by 3 and 5 that
322 Chapter 25. Joint Densities

will occur in the respective divisions. We get

Z ∞Z ∞ Z ∞ Z ∞
fX,Y (x, y) dy dx = 3e−3x dx 5e−5y dy
−∞ −∞ 0 0
−3x ∞ −5y ∞



= −e x=0
−e y=0
=1

So we have verified that fX,Y (x, y) is a joint density.

Example 25.6. Now we compute FX,Y (1/2, 1/4) = P (X ≤ 1/2 and Y ≤ 1/4),
i.e., the probability that Maxine waits 1/2 of a minute or less, and Daniella
waits 1/4 of a minute or less. In the computation, we take advantage of the
fact that the integration in x does not affect the integration in y, so we split
the integral into two parts.
Z 1/2 Z 1/4
FX,Y (1/2, 1/4) = fX,Y (x, y) dy dx
−∞ −∞
Z1/2 Z 1/4
= 15e−3x−5y dy dx
0 0
Z 1/2  Z 1/4 
= 3e−3x dx 5e−5y dy
0 0
1/2
−5y 1/4
−e−3x x=0


= −5e y=0
−3/2 −5/4
= (1 − e )(1 − e )
= 0.5543

We can compute much more than just joint CDFs using joint densities. For in-
stance, we could compute P (X > 1/10 and Y > 2/7), i.e., the probability that
Maxine waits more than 1/10 of a minute and Daniella waits more than 2/7 of
a minute, as follows:
Z ∞ Z ∞
P (X > 1/10 and Y > 2/7) = fX,Y (x, y) dy dx
1/10 2/7
Z ∞ Z ∞
= 15e−3x−5y dy dx
1/10 2/7
Z ∞ Z ∞
−3x
5e−5y dy



= 3e dx
1/10 2/7
∞ ∞
−e−3x x=1/10 −5e−5y



= y=2/7
−3/10

−10/7

= e e
= e−121/70
= 0.1775
 25.2. Examples 323

Example 25.7. Using the same X and Y as above—the waiting times for
Maxine and Daniella at their respective traffic lights—we let Z = min(X, Y ).
Notice that Z is a waiting time too, because Z is the waiting time until the first
one of the pair (Maxine or Daniella) has her own light turn green. So Z is the
waiting time until Maxine or Daniella gets to depart. We find the cumulative
distribution function of Z.
First of all, since X > 0 always and Y > 0 always, then Z > 0 always too.
So, for a ≤ 0, we have FZ (a) = P (Z ≤ a) = 0.
Now we turn our attention to a > 0. Since Z = min(X, Y ), then FZ (a) is
equal to P (min(X, Y ) ≤ a). This is a difficult probability to calculate, because
we could either have X ≤ a or Y ≤ a or both. It is much easier to calculate the
complementary probability,

P (Z > a) = P (min(X, Y ) > a) = P (X > a and Y > a).

We compute, for a > 0,

FZ (a) = P (Z ≤ a)
= 1 − P (Z > a)
= 1 − P (min(X, Y ) > a)
= 1 − P (X > a and Y > a)
Z ∞Z ∞
=1− 15e−3x−5y dy dx
Za ∞ a
∞
=1− −3e−3x−5y y=a dx
Za ∞
=1− 3e−3x−5a dx
a
∞
=1+ e−3x−5a x=a
−8a
=1−e

So, in summary, the cumulative distribution function of Z is

FZ (z) = 1 − e−8z for z > 0,

and FZ (z) = 0 otherwise.

Example 25.8. With joint densities, we can calculate probabilities about all
kinds of relationships between X and Y . For instance, we could calculate the
probability that Daniella waits longer at her light than Maxine, e.g., P (Y > X).
We have two ways to set up the integral:
324 Chapter 25. Joint Densities

y y
8 8
7 7
6 6
5 5
4 4
3 3
2 2
1 1
x x
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8

Figure 25.2: Left: Setting up the integral for P (Y > X), with y as the outer
integral and x as the inner integral. The y is fixed (e.g., y = 3.2), and x ranges
from 0 to y. Right: Setting up the integral for P (Y > X), with x as the outer
integral and y as the inner integral. The x is fixed (e.g., x = 2.6), and y ranges
from x to ∞.

Method #1: We can integrate for the outer integral over all y’s; for the inner
integral, y is fixed, and we integrate over all of the x’s that are smaller than y,
i.e., 0 ≤ x ≤ y, as shown on the left of Figure 25.2. We get
Z ∞Z y Z ∞
y
P (Y > X) = 15e−3x−5y dx dy = −5e−3x−5y x=0 dy
0 0 0

which yields
Z ∞
∞
P (Y > X) = (5e−5y − 5e−8y ) dy = (−e−5y + (5/8)e−8y ) y=0
= 3/8.
0

Method #2: We can integrate for the outer integral over all x’s; for the inner
integral, x is fixed, and we integrate over all of the y’s that are larger than x,
i.e., x ≤ y, as shown on the right of Figure 25.2. We get
Z ∞Z ∞ Z ∞
−3x−5y ∞
P (Y > X) = 15e dy dx = −3e−3x−5y y=x dx
0 x 0

which yields
Z ∞ Z ∞
−3x−5x ∞
P (Y > X) = 3e dx = 3e−8x dx = −(3/8)e−8x x=0
= 3/8.
0 0

This agrees with the result computed using Method #1.

 25.2. Examples 325

Example 25.9. Suppose that a person throws a dart at a square dart board.
Let X and Y denote, respectively, the x- and y-coordinates (in feet) of the
location where the dart lands; the middle of the dart board is at (0, 0). Suppose
that the dart board is two feet wide and two feet tall, so that the dart only
lands on the dart board if −1 ≤ X ≤ 1 and −1 ≤ Y ≤ 1. Also suppose that the
person always hits the dart board, and moreover, X and Y have joint density:
9
fX,Y (x, y) = (1 − x2 )(1 − y 2 ) for −1 ≤ x ≤ 1, −1 ≤ y ≤ 1,
16
and fX,Y (x, y) = 0 otherwise. See Figure 25.3.

0.6

0.4

0.2

0 −1
−1 0
−0.5 0
0.5 x
1 1
y

Figure 25.3: Joint PDF fX,Y (x, y) = 9

16 (1 − x2 )(1 − y 2 ), for −1 ≤ x ≤ 1,
−1 ≤ y ≤ 1.

To check that fX,Y (x, y) is a joint density, we see that fX,Y (x, y) ≥ 0 always;
the function is 0 except possibly when −1 ≤ x ≤ 1, −1 ≤ y ≤ 1, in which case
1 − x2 and 1 − y 2 are both nonnegative. Also
Z ∞Z ∞ Z 1Z 1
9
fX,Y (x, y) dy dx = (1 − x2 )(1 − y 2 ) dy dx
−∞ −∞ −1 −1 16
Z 1 Z 1
9 2
= (1 − x ) dx (1 − y 2 ) dy
16 −1 −1
9 1 1
= (x − x3/3) x=−1 (y − y 3/3) y=−1
16
= (9/16)(4/3)(4/3)
=1

So fX,Y (x, y) is a joint density. Now we find the probability that the x- and
y-coordinates of the location where the dart lands are each within a 1/2 foot
326 Chapter 25. Joint Densities

from the center of the board. We compute

 Z 1/2 Z 1/2
 9
P − 12 ≤ X ≤ 12 , − 21 ≤ Y ≤ 12 = (1 − x2 )(1 − y 2 ) dy dx
−1/2 −1/2 16
Z 1/2 Z 1/2
9 2
= (1 − x ) dx (1 − y 2 ) dy
16 −1/2 −1/2
9 1/2 1/2
= (x − x3 /3) x=−1/2 (y − y 3 /3) y=−1/2
16
= 121/256
= 0.4727

Example 25.10. Consider X and Y with joint density

fX,Y (x, y) = 1/9 for 0 ≤ x ≤ 3, 0 ≤ y ≤ 3,

and fX,Y (x, y) = 0 otherwise. See Figure 25.4.

3
2
1
x
1 2 3

Figure 25.4: The joint density fX,Y (x, y) = 1/9 for 0 ≤ x ≤ 3, 0 ≤ y ≤ 3,

and fX,Y (x, y) = 0 otherwise.

To see that fX,Y (x, y) is a joint density, we note that fX,Y (x, y) ≥ 0 for all
x, y, and we could compute that
Z 3Z 3
1/9 dy dx = 1.
0 0

For two-variable densities, a probability is calculated as the area under the

curve, i.e., the density is the volume of a solid. If the density is constant, such
and integration is easy to perform. In such a case, the integral is equal to the
integrand times the area over which we are integrating. In this example, the
integrand is just the constant 1/9, and the area over which we integrate is 9, so
the integral is (1/9)(9) = 1. So fX,Y (x, y) is a joint density.
Now we find the probability that the minimum of X and Y is less than 1. In
other words, we find P (min(X, Y ) ≤ 1). The area for which min(X, Y ) ≤ 1 is 5,
 25.2. Examples 327

and seen in Figure 25.5. So the desired probability is just (1/9)(5) = 5/9. There
is no need to perform the integration using complicated methods. Anytime that
the integrand is constant, we just multiply the integrand (in this case, 1/9) by
the area of the integration region (in this case, 5).
If a reader insists on performing the integration using the tools learned in
calculus, we caution that the integral will need to be broken into at least two
parts. See Figure 25.5. We could integrate over the region in the middle of
Figure 25.5, and then integrate over the region on the right side of Figure 25.5,
and add the results together, since these two regions are disjoint, and their
union is the entire region over which we need to integrate:
Z 1Z 3 Z 3Z 1
P (min(X, Y ) ≤ 1) = 1/9 dy dx + 1/9 dy dx
0 0 1 0
= (1)(3)(1/3) + (2)(1)(1/9)
= 5/9

The final answer, as expected, is 5/9.

We emphasize the utility of the first method, which can always be used
easily when the integrand is constant and the area of the region of integration
can be easily determined.

y y y

3 3 3
2 2 2
1 1 1
x x x
1 2 3 1 2 3 1 2 3

Figure 25.5: Left: The region where min(X, Y ) ≤ 1. The region can be
split into two disjoint parts. Middle: The region where min(X, Y ) ≤ 1 and
0 ≤ X ≤ 1. Right: The region where min(X, Y ) ≤ 1 and 1 ≤ X ≤ 3.

Example 25.11. Consider the joint density

(
9e−3x for 0 < y < x;
fX,Y (x, y) =
0 otherwise.

Find the density of X and the density of Y .

The CDF of X is
Z a Z ∞
FX (a) = P (X ≤ a) = fX,Y (x, y) dy dx,
−∞ −∞
328 Chapter 25. Joint Densities

Differentiating on both sides with respect to a yields

Z ∞
fX (a) = fX,Y (a, y) dy,
−∞

or equivalently, Z ∞
fX (x) = fX,Y (x, y) dy.
−∞

This idea works in general—nothing here is specific to the X and Y

in the example:

Theorem 25.12. Finding the density of one random variable, using

the joint density
If two continuous random variables X and Y have joint density fX,Y (x, y), the
density of X can be retrieved by integrating over all y’s:
Z ∞
fX (x) = fX,Y (x, y) dy.
−∞

Similarly, the density of Y is found by integrating the joint density over all
x’s: Z ∞
fY (y) = fX,Y (x, y) dx.
−∞

Returning to Example 25.11, we see that the density of X is, for x > 0,
Z ∞ Z x
fX (x) = fX,Y (x, y) dy = 9e−3x dy = 9xe−3x .
−∞ 0

Similarly, the density of Y is, for y > 0,

Z ∞ Z ∞
fY (y) = fX,Y (x, y) dx = 9e−3x dx = 3e−3y .
−∞ y

25.3 Exercises
25.3.1 Practice
Exercise 25.1. Consider a pair of random variables X, Y with constant joint
density on the triangle with vertices at (0, 0), (3, 0), and (0, 3). Find P (X +Y >
2).

Exercise 25.2. Consider a pair of random variables X, Y with constant joint

density on the quadrilateral with vertices (0, 0), (2, 0), (2, 6), (0, 12). Find
P (Y ≥ 3X).
 25.3. Exercises 329

Exercise 25.3. Let X, Y have joint density

(
14e−2x−7y if x > 0 and y > 0,
fX,Y (x, y) =
0 otherwise.

Find P (X > Y ).
Exercise 25.4. Suppose X, Y have joint density
(
1/16 if −2 ≤ x ≤ 2 and −2 ≤ y ≤ 2,
fX,Y (x, y) =
0 otherwise.

Find P (|X − Y | ≤ 1).

Exercise 25.5. Suppose X, Y have joint density
(
1
(3 − x)(2 − y) if 0 ≤ x ≤ 3 and 0 ≤ y ≤ 2,
fX,Y (x, y) = 9
0 otherwise.

Find P (Y > X).

Exercise 25.6. Consider random variables X and Y with joint density
1
fX,Y (x, y) = (1 − x2 )(3 − y) for −1 ≤ x ≤ 1 and −1 ≤ y ≤ 1,
8
and fX,Y (x, y) = 0 otherwise. Find the probability that X and Y are both
negative.
Exercise 25.7. Let X, Y have joint density
3
fX,Y (x, y) = (x3 + y 2 ), for 0 ≤ x ≤ 10 and 0 ≤ y ≤ 5,
38750
and fX,Y (x, y) = 0 otherwise. Find P (X ≤ 5, Y ≤ 2).
Exercise 25.8. Suppose X, Y are jointly distributed with joint density
1
fX,Y (x, y) = (x + 1)(y 2 + 1), for 0 ≤ x, y ≤ 4,
304
and fX,Y (x, y) = 0 otherwise.
a. Find P (1 ≤ X ≤ 2, 3 ≤ Y ≤ 4).
b. Find the density of X.
c. Find the density of Y .
Exercise 25.9. Consider X, Y with joint density
9 2 2
fX,Y (x, y) = x y , for 0 ≤ x ≤ 2, 0 ≤ y ≤ 2,
64
and fX,Y (x, y) = 0 otherwise.
330 Chapter 25. Joint Densities

a. Find P (X < 1, Y < 1).

b. Find P (X > 1, Y < 1).
c. Convince yourself that, by symmetry, P (X < 1, Y > 1) is the same as
the value from part b.
d. Find P (X > 1, Y > 1).
e. Check that the answers from parts a–d sum to 1.
Exercise 25.10. Consider X, Y with joint density
1
fX,Y (x, y) = (10 − x)(10 − y), for 0 ≤ x ≤ 10, 0 ≤ y ≤ 10,
2500
and fX,Y (x, y) = 0 otherwise.
a. Find P (X < 9, Y < 9).
b. Find P (X < 9, Y > 9).
c. Convince yourself that, by symmetry, P (X > 9, Y < 9) is the same as
the value from part b.
d. Find P (X > 9, Y > 9).
e. Check that the answers from parts a–d sum to 1.
Exercise 25.11. Let X, Y have joint density
fX,Y (x, y) = x + y, for 0 ≤ x, y ≤ 1,
and fX,Y (x, y) = 0 otherwise. Find the joint CDF, i.e., FX,Y (x, y) = P (X ≤
x, Y ≤ y).
Exercise 25.12. Let X, Y have joint density
6
fX,Y (x, y) = (x + y)2 , for 0 ≤ x, y ≤ 1,
7
and fX,Y (x, y) = 0 otherwise. Find P (X ≤ 1/2, Y ≤ 1/2).
Exercise 25.13. Consider a pair of random variables X, Y with joint density
fX,Y (x, y) = 8e−2x−4y , for 0 < x, 0 < y,
and fX,Y (x, y) = 0 otherwise. Find P (X ≤ 4, Y ≤ 8).
Exercise 25.14. Let X, Y have constant density on the square where 0 ≤ X ≤
4, 0 ≤ Y ≤ 4. Find P (X + Y < 4).
Exercise 25.15. If X and Y have joint density
9 2 2
fX,Y (x, y) = x y , for |x| ≤ 1 and |y| ≤ 2,
32
and fX,Y (x, y) = 0 otherwise, find the probability that X and Y are both
positive.
 25.3. Exercises 331

Exercise 25.16. Consider X, Y with joint density

fX,Y (x, y) = 10e−2x−5y , for 0 < x, 0 < y,

and fX,Y (x, y) = 0 otherwise. Find P (X < 2, Y < 1).

25.3.2 Extensions
Exercise 25.17. Consider the random variables X and Y defined in Exam-
ple 25.5, i.e., X is Maxine’s waiting time, and Y is Daniella’s waiting time.
Let W = max(X, Y ), i.e., W is either Maxine’s or Daniella’s waiting time,
whichever is larger! Find FW (w) = P (W ≤ w), the cumulative distribution
function of W . This is equal to P (max(X, Y ) ≤ w), i.e., the probability that
Maxine waits less than w minutes and Daniella waits less than w minutes.

Exercise 25.18. Freddy and Jane have entered a game in which they each win
between 0 and 2 dollars. If X is the amount Freddy wins, and Y is the amount
that Jane wins, they believe that the joint density of their winnings will be
1
fX,Y (x, y) = xy for 0 ≤ x ≤ 2 and 0 ≤ y ≤ 2,
4
and fX,Y (x, y) = 0 otherwise. Find the probability that their combined win-
nings exceed 2, i.e., find P (X + Y > 2).

Exercise 25.19. If X and Y have joint distribution

fX,Y (x, y) = sin x cos y, for 0 ≤ x, y ≤ π/2,

and fX,Y (x, y) = 0 otherwise, then find the probability that X and Y are both
smaller than π/6.

Exercise 25.20. If X and Y have joint distribution

1 1
fX,Y (x, y) = cos x sin y + , for −1 ≤ x, y ≤ 1,
4 4
and fX,Y (x, y) = 0 otherwise, then find the probability that 0 ≤ X ≤ 1/2 and
Y ≤ 2X.

Exercise 25.21. Suppose that X, Y are jointly distributed with fX,Y (x, y) =
1/10, for (x, y) in the triangle with vertices at the origin, (2, 0), and (0, 10), and
fX,Y (x, y) = 0 otherwise. Find the probability that Y ≤ 2.

25.3.3 Advanced
Exercise 25.22. Consider X, Y with joint density

sech2 x
fX,Y (x, y) = , for x ≥ 0 and y ≥ 0,
(y + 1)2
332 Chapter 25. Joint Densities

and fX,Y (x, y) = 0 otherwise. Find P (X ≤ 2, Y ≤ 2). (The function “sech” is

the hyperbolic secant, which should be familiar to many readers who recently
completed a course in calculus. Other readers may choose to skip this exercise,
because sech will make very few appearances in the densities that we consider
in this text.)

Exercise 25.23. Consider X, Y with joint density

1
fX,Y (x, y) = , for x2 + y 2 ≤ 4,
4π
and fX,Y (x, y) = 0 otherwise. Find the probability that (X, Y ) is at most 1
unit from the origin and is located in the first quadrant, i.e., that X ≥ 0 and
Y ≥ 0 and X 2 + Y 2 ≤ 1.

Exercise 25.24. Suppose that the joint PDF of X and Y is f (x, y) = 2e−x−y
for 0 < y < x < ∞, and f (x, y) = 0 otherwise.

a. Find fX (x), the density of X.

b. Find fY (Y ), the density of Y .
Chapter 26

Independent Continuous
Random Variables

I’m trying to handle all this unpredictability

In all probability
—“Long Shot,” sung by Kelly Clarkson, from the album All I Ever Wanted ;
song written by Katy Perry, Glen Ballard, Matt Thiessen

Are a person’s height and weight independent or dependent random variables?

How about a person’s height and GPA?

26.1 Introduction
Two continuous random variables are independent if they satisfy the continuous
versions of Definition 9.13, using densities instead of masses.
Definition 26.1. Independent continuous random variables: Several
equivalent formulations
1. Joint density can be factored into a function of x times a function of y.
These functions of x and y can also be normalized so that they are the densities
of X and Y , respectively: fX,Y (x, y) = fX (x)fY (y) for all x and y.
2. Joint CDF can be factored into a function of x times a function of y. Again,
we can normalize these, so that they are the CDFs of X and Y , respectively:
FX,Y (x, y) = FX (x)FY (y) for all x and y.
3. We will define conditional densities in the next chapter, but we will go
ahead and state the way they will be used for independence: (3a) Density
of X equals conditional density fX|Y (x | y) of X given Y = y, or (3b) Density
of Y equals conditional density fY |X (y | x) of Y given X = x.

333
334 Chapter 26. Independent Continuous Random Variables

26.2 Examples
One of the trickiest things about the above statement of independence is re-
membering to check that the factoring works for all x’s and y’s. For instance,
we briefly reconsider Exercise 25.6:

Example 26.2. Consider random variables X and Y with joint density

1
fX,Y (x, y) = (1 − x2 )(3 − y) for −1 ≤ x ≤ 1 and −1 ≤ y ≤ 1,
8
and fX,Y (x, y) = 0 otherwise.

The region where the density is nonzero is a square, and thus the factorization
of fX,Y (x, y) into the “x stuff” and the “y stuff” works for all x’s and y’s. So,
in this case, X and Y are independent.
Since the X and Y in this example are independent, then we can find the
density of X by itself; all we need to do is determine the constant at the front
of the density of X. We compute
Z 1
(1 − x2 ) dx = (x − x3/3)|1x=−1 = 4/3.
−1

So we need a constant of 3/4 at the front of the density of X, so that the integral
will turn out to be 1. Thus, the density of X is

fX (x) = (3/4)(1 − x2 ) for −1 ≤ x ≤ 1,

and fX (x) = 0 otherwise. Similarly, we compute

Z 1
(3 − y) dy = (3y − y 2/2)|1y=−1 = 6.
−1

So, as before, we need a constant at the front of the density of Y ; in this case,
we need a 1/6 in the density of Y , so that the integral will turn out to be 1.
Thus, the density of Y is

fY (y) = (1/6)(3 − y) for −1 ≤ y ≤ 1,

and fY (y) = 0 otherwise. Finally, we check that

1
fX,Y (x, y) = (1 − x2 )(3 − y)
8
= (3/4)(1 − x2 )(1/6)(3 − y)
= fX (x)fY (y) for all −1 ≤ x ≤ 1 and −1 ≤ y ≤ 1,

and otherwise fX,Y (x, y) = 0, fX (x) = 0, and fY (y) = 0.

 26.2. Examples 335

Example 26.3. In Exercise 25.18, we have two continuous random variables

X and Y with joint density
1
fX,Y (x, y) = xy for 0 ≤ x ≤ 2 and 0 ≤ y ≤ 2,
4
and fX,Y (x, y) = 0 otherwise.
In this example, the region where the density is nonzero is again square, so the
factorization of fX,Y (x, y) into the “x stuff” and the “y stuff” again works for
all x’s and y’s. Therefore, again in this case, X and Y are independent.
We easily compute that
Z 2
x dx = x2/2|2x=0 = 2.
0

So we need a correction factor of 1/2 at the front to make this into a density.
So the density of X must be

fX (x) = (1/2)(x) for 0 ≤ x ≤ 2,

and fX (x) = 0 otherwise. Since the entire problem is symmetric in terms of X

and Y , the density of Y must also be

fY (y) = (1/2)(y) for 0 ≤ y ≤ 2,

and fY (y) = 0 otherwise.

Notice that fX,Y (x, y) = xy/4 = (1/2)(x)(1/2)(y) = fX (x)fY (y).

Example 26.4. Just because the joint density appears to factor, we must use
caution. Consider, for instance, the joint density
3
fX,Y (x, y) = xy for 0 ≤ x and 0 ≤ y and x + y ≤ 2,
2
and fX,Y (x, y) = 0 otherwise.
Important: We might initially be inclined to say that fX,Y (x, y) can be factored
in a way that works for all x and y, but this is not the case. For instance, if
x = 1/2, then y can be between 0 and 3/2; but, if x = 1, then y can be between
0 and 1. So Y is not independent of X in this example. Nonetheless, we can
still use the technique of finding the density of one random variable, using the
joint density (given in the box at the end of the previous chapter), i.e.,
Z ∞
fX (x) = fX,Y (x, y) dy.
−∞
336 Chapter 26. Independent Continuous Random Variables

In this example, we know fX (x) = 0 for x < 0 or x > 2, since X is always

between 0 and 2. Now we consider 0 ≤ x ≤ 2, and we get
Z 2−x
3 3 3
fX (x) = xy dy = xy 2/2|2−x 2
y=0 = x(2 − x) .
0 2 2 4
So we conclude that the density of X is
3
fX (x) = x(2 − x)2 for 0 ≤ x ≤ 2,
4
and fX (x) = 0 otherwise. In particular, note that X and Y are dependent,
since fX (x) is not a factor of fX,Y (x, y).

y v
4 4
3 3
2 2
1 1
x u
1 2 3 4 5 6 7 1 2 3 4 5 6 7

Figure 26.1: Left: A region S containing six rectangles. Right: A region T

containing four rectangles.

For two continuous random variables to be independent, the domains where

the random variables are defined must be rectangles (this is a necessary—but
not sufficient—condition). If X and Y are defined in a range that is a cir-
cle, or a triangle, or some other non-rectangular shape, then X and Y must be
dependent. Also, X and Y can be defined on several rectangles and be indepen-
dent. For instance, consider a region S with six rectangles, as on the left side
of Figure 26.1.
Let fX,Y (x, y) be constant on the region S, so

fX,Y (x, y) = 1/6 for (x, y) ∈ S,

and fX,Y (x, y) = 0 otherwise. Then X and Y are independent. On the other
hand, consider another pair of random variables, say U and V , which are defined
on a region T with four rectangles, as on the right side of Figure 26.1. Let
fU,V (u, v) be constant on the region T , so

fU,V (u, v) = 1/10 for (u, v) ∈ T ,

and fU,V (u, v) = 0 otherwise. Then U and V are dependent. For instance, if
v = 0.5 then either u is in the range [0, 1] or [3, 4] or [6, 7], but on the other
 26.2. Examples 337

hand, if v = 3.5, then u can be anywhere in the range [0, 7]. So U and V are
dependent. So, the random variables must be defined in rectangles, and if more
than one rectangle is present, the rectangles must be in a grid shape.
One more note: We stated earlier that, “For two continuous random variables
to be independent, the domains where the random variables are defined must
be rectangles (this is a necessary—but not sufficient—condition).” We are just
emphasizing that having a rectangle-shaped domain of definition is not enough
for independence either. The density must (of course) also factor, for the random
variables to be independent. Consider, for instance, Exercise 25.11, in which X
and Y are defined in the unit square 0 ≤ x, y ≤ 1, with density fX,Y (x, y) =
x + y. Despite the fact that the domain of definition is a rectangle (moreover,
a square), the joint density does not factor, so X and Y are dependent.
Remark 26.5. How to check for independence, using a joint density:
1. The joint density fX,Y (x, y) must be defined on a rectangle, or on several
rectangles arranged in a grid shape.
2. The joint density fX,Y (x, y) must factor into a product, with all of the x’s
in one part and all of the y’s in the other part. Properly normalized, the x
part is the density fX (x) of X in such a case, and the y part is the density
fY (y) of Y .

Example 26.6. Each time a pitcher delivers a fastball, the speed is distributed
between 90 and 100 miles per hour, with density 1/10. Assume that the speeds
of pitches are independent.
Consider two fastballs thrown by the pitcher. Find the probability that at
least one of them is 93 miles per hour or faster.
Let X and Y denote the speeds of the two pitches. Then
fX (x) = 1/10 for 90 ≤ x ≤ 100,
and fX (x) = 0 otherwise. Similarly,
fY (y) = 1/10 for 90 ≤ y ≤ 100,
and fY (y) = 0 otherwise. Since X and Y are independent, it follows that the
joint density of X and Y is
fX,Y (x, y) = 1/100 for 90 ≤ x ≤ 100, 90 ≤ y ≤ 100,
and fX,Y (x, y) = 0 otherwise.
So the desired probability is
P (X ≥ 93 or Y ≥ 93) = 1 − P (X ≤ 93 and Y ≤ 93)
Z 93 Z 93
=1− 1/100 dy dx
90 90
= 1 − 9/100
= 91/100
338 Chapter 26. Independent Continuous Random Variables

Another method for computing the desired probability is the following: Let Z
denote the number of the two fastballs that exceed 93 miles per hour. Then Z is
either 0 or 1 orR2, and moreover, Z is Binomial with n = 2 and with probability
100
of success p = 93 1/10 dx = 7/10 on each throw. So the probability of at least
one fastball over 93 miles per hour is

P (Z ≥ 1) = 1 − P (Z = 0)
 
2 0
=1− p (1 − p)2
0
= 1 − (1 − 7/10)2
= 1 − (3/10)2
= 1 − 9/100
= 91/100.

Example 26.7. The lifetime (in years) of a music player—before it permanently

fails—is a random variable X with density
1
fX (x) = e−x/3 for x > 0,
3
and fX (x) = 0 otherwise.

a. Consider 2 music players that are assumed to have independent lifetimes,

X and Y , respectively. Then the joint density of X and Y is
1 1 1
fX,Y (x, y) = e−x/3 e−y/3 = e−(x+y)/3 for x > 0 and y > 0,
3 3 9
and fX,Y (x, y) = 0 otherwise.
b. The probability that a music player permanently fails within the first
ten years is
Z 10
1 −x/3 10
e dx = −e−x/3 = 1 − e−10/3 = 0.9643260.
0 3 x=0

c. Now suppose that each of the 23 students in a class has a music player.
What is the probability that all 23 of the music players permanently fail within
the first ten years? The probability is
Z 10 23  23
1 −x/3 −x/3
10  23
e dx = e = 1 − e−10/3 = 0.4336599.
0 3 x=0

Another way to see this is the following: The number of music players that
permanently fail within the first ten years is a Binomial random variable with
 26.3. Exercises 339

n = 23 and p = 1 − e−10/3 = 0.9643260. So the probability that all 23 fail

within the first ten years is
 
23 23  23
p (1 − p)0 = 1 − e−10/3 = 0.4336599.
23

d. Now suppose that we check music players until we find one that dies
within only one year! The number of music players that we need to check is
R1 1
Geometric with probability of success 0 13 e−x/3 dx = e−x/3 x=0 = 1 − e−1/3 =
0.2834687. So the expected number of music players that we would need to
check, until we find one that dies within only one year, is
1
= 3.527726.
1 − e−1/3

26.3 Exercises
26.3.1 Practice
Exercise 26.1. Suppose X, Y have joint density
(
1
(3 − x)(2 − y) if 0 ≤ x ≤ 3 and 0 ≤ y ≤ 2,
fX,Y (x, y) = 9
0 otherwise.

a. Are X and Y independent? Why or why not?

b. Find the density fX (x) of X.
c. Find the density fY (y) of Y .
Exercise 26.2. If two random variables X and Y have joint density
12
fX,Y (x, y) = (xy + x2 ), for 0 ≤ x ≤ 1, 0 ≤ y ≤ 1,
7
and fX,Y (x, y) = 0 otherwise, are X and Y independent? Why or not?
Exercise 26.3. When X, Y have joint density
x2
fX,Y (x, y) = , for 0 ≤ x ≤ 1, 1/2 ≤ y ≤ 4,
y ln 2
and fX,Y (x, y) = 0 otherwise, find:
a. The density fX (x) of X.
b. The density fY (y) of Y .
Exercise 26.4. Let X, Y have joint density

fX,Y (x, y) = 2x/45, for 0 ≤ x ≤ 3, 0 ≤ y ≤ 5

and fX,Y (x, y) = 0 otherwise.

340 Chapter 26. Independent Continuous Random Variables

a. Are X and Y independent?

b. Find the density fX (x) of X.
c. Find the density fY (y) of Y .
Exercise 26.5. Let X, Y have joint density

fX,Y (x, y) = 3xy/1250, for 0 ≤ x, 0 ≤ y, x + y ≤ 10

and fX,Y (x, y) = 0 otherwise.

a. Are X and Y independent?
b. Find the density fX (x) of X.
c. Find the density fY (y) of Y .

26.3.2 Extensions
Exercise 26.6. When an emergency occurs, the response time (in hours) of
the first police car is a random variable X with density

fX (x) = 12e−12x for x > 0,

and fX (x) = 0 otherwise. The response time (in hours) of the first fire engine
is a random variable Y with density

fY (y) = 10e−10y for y > 0,

and fY (y) = 0 otherwise. Assume X and Y are independent.

Find the probability that the first police car arrives before the first fire
engine.
Exercise 26.7. Police cars are randomly stationed throughout the town. When
an emergency occurs, the distance a police car must travel north or south is a
random variable X with density

fX (x) = 1/6 for 0 ≤ x ≤ 6,

and fX (x) = 0 otherwise. The distance a police car must travel east or west is
a random variable Y with density

fY (y) = 1/6 for 0 ≤ y ≤ 6,

and fY (y) = 0 otherwise. Assume that X and Y are independent. So X + Y is

the total distance traveled. Find the probability that X + Y ≤ 4.

(Hint: First find the joint density fX,Y (x, y) of X and Y , and draw a picture
for where the joint density is defined. If you write an integral for the probability,
the integrand is constant. So you can compute the desired area in your picture,
divided by the total area.)
 26.3. Exercises 341

Exercise 26.8. Let

9√
 
√
fX,Y (x, y) = 2 (x2 )( y), for 0 ≤ x ≤ 2, 0 ≤ y ≤ 2,
64

and fX,Y (x, y) = 0 otherwise.

a. Are X and Y independent? Why?

b. Find P (X ≤ 1).
c. Find P (Y ≤ 1).
d. Find P (X + Y ≤ 1).

Exercise 26.9. Suppose X and Y have joint density

fX,Y (x, y) = e−x−y , for 0 < x and 0 < y,

and fX,Y (x, y) = 0 otherwise.

a. Are X and Y independent?

b. Find P (X < 2, Y < 2).
c. Compare with the situation in which X and Y have joint density

fX,Y (x, y) = 2e−x−y , for 0 < x < y,

and fX,Y (x, y) = 0 otherwise. Find P (X < 2, Y < 2) in this case. Are X and
Y independent in this new case? Why or why not? Can you see whether X and
Y are independent by only looking at the joint density of X and Y ?

Exercise 26.10. Suppose that X, Y have constant joint density on the triangle
with corners at (4, 0), (0, 4), and the origin.

a. Find P (X < 3, Y < 3).

b. Are X and Y independent?

Exercise 26.11. Suppose X and Y have joint density

fX,Y (x, y) = 18x2 y 5 , for 0 < x < 1, 0 < y < 1

and fX,Y (x, y) = 0 otherwise.

a. Are X and Y independent?

b. Find P (X < 1/2).
c. Find P (Y < 1/2).
d. Find P (X + Y < 1).
342 Chapter 26. Independent Continuous Random Variables

Exercise 26.12. Suppose X and Y have joint density

fX,Y (x, y) = 1512x2 y 5 , for 0 ≤ x, 0 ≤ y, x + y < 1

and fX,Y (x, y) = 0 otherwise.

a. Are X and Y independent?
b. Find P (X < 1/2).
c. Find P (Y < 1/2).
d. Find P (X + Y < 1/2).
Exercise 26.13. Let X and Y have constant joint density on the parallelogram
with corners at the origin, (1, 1), (1, 2), and (0, 1).
a. Find the joint density fX,Y (x, y).
b. Find P (Y < 3/2).
Exercise 26.14. Suppose X and Y have joint density

8 − 2x2
fX,Y (x, y) = , for 1 ≤ x ≤ 2, 1/2 ≤ y ≤ 1
3x2 y 3
and fX,Y (x, y) = 0 otherwise.
a. Are X and Y independent?
b. Find P (X < 3/2, Y < 3/4).
Exercise 26.15. Let X, Y, Z have joint density

fX,Y,Z (x, y, z) = 6, for 0 < x < y < z < 1

and fX,Y,Z (x, y, z) = 0 otherwise.

a. Are X and Y and Z independent?
b. Find the density fX (x) of X.
c. Find the density fY (y) of Y .
d. Find the density fZ (z) of Z.
Exercise 26.16. Consider two random variables X and Y with joint density

fX,Y (x, y) = x/8 + y/8, for 0 < x < 2, 0 < y < 2

and fX,Y (x, y) = 0 otherwise.

a. Are X and Y independent?
b. Find P (X > 1) and P (Y > 1).
c. Find P (X + Y < 2).
 26.3. Exercises 343

Exercise 26.17. If X and Y have joint density

fX,Y (x, y) = 75xe−5x−3y , for 0 < x, 0 < y

and fX,Y (x, y) = 0 otherwise:

a. Are X and Y independent?

b. Find P (X < Y ).

Exercise 26.18. Suppose X and Y have constant density on the region in

Figure 26.2.

a. Are X and Y independent?

b. Find P (X + Y < 2).
c. Find P (X + Y < 2.5).

x
1 2

Figure 26.2: Region where X and Y have positive density.

Exercise 26.19. If the pair of random variables X and Y have

fX,Y (x, y) = 180x2 y 2 , for 0 < x, 0 < y, and x + y < 1

and fX,Y (x, y) = 0 otherwise, find P (X + Y > 1/2).

Exercise 26.20. Let X, Y have joint density

9√
fX,Y (x, y) = xy, for 0 ≤ x ≤ 1, 0 ≤ y ≤ 4
32
and fX,Y (x, y) = 0 otherwise.

a. Are X and Y independent?

b. Find P (X < Y ).
c. Find P (X < 2Y ).
d. Find P (X < 4Y ).
344 Chapter 26. Independent Continuous Random Variables

Exercise 26.21. Let X, Y have joint density

3
fX,Y (x, y) = (x − y), for 0 ≤ y ≤ x ≤ 4
32
and fX,Y (x, y) = 0 otherwise.
a. Are X and Y independent?
b. Find P (X ≤ 2).
c. Find P (Y ≤ 2).
Exercise 26.22. Let X, Y have joint density
(8 − 2x)(2 − y)
fX,Y (x, y) = , for −2 ≤ x ≤ 2, −2 ≤ y ≤ 2
256
and fX,Y (x, y) = 0 otherwise.
a. Are X and Y independent?
b. Find the density fX (x) of X.
c. Find the density fY (y) of Y .

26.3.3 Advanced
Exercise 26.23. Let
1
fX,Y (x, y) = sin x sec2 (y/4) for 0 ≤ x ≤ π, 0 ≤ y ≤ π,
8
and fX,Y (x, y) = 0 otherwise. Find:
a. The density fX (x) of X.
b. The density fY (y) of Y .
c. P (X < π/2, Y < π/2).
Exercise 26.24. Suppose X and Y have joint density
2sech2 (1/x)
fX,Y (x, y) = , for 0 < x < y,
y3
and fX,Y (x, y) = 0 otherwise.
a. Show that fX,Y (x, y) is a joint density, i.e., the integral over all x’s and
y’s is 1.
b. Are X and Y independent?
Exercise 26.25. When X and Y have joint density
2x2 y 2
fX,Y (x, y) = , for 0 < x, 0 < y
ex+2y
and fX,Y (x, y) = 0 otherwise, find P (X < Y ).
Chapter 27

Conditional Distributions

The probability of meeting someone you know increases when you are with
someone you don’t want to be seen with.
—Anonymous

A dancer is twirling randomly on a semi-circular stage with a radius of 3 feet.

When the music stops, she is 2 feet from the edge of the stage closest to the
audience. The spotlight is set up to shine on the middle of the stage, but if the
dancer is too far stage left or stage right (more than 2.5 feet from the middle),
she will be poorly lit. What is the probability that the dancer will end up in
the dark?

27.1 Introduction

Conditional densities are defined similarly to conditional masses. The condi-

tional density of X given Y = y tells us how X behaves when Y = y. We can
use the conditional density to get conditional probabilities. Later, once we delve
into expected values, we will use conditional densities to get expected values for
one variable that are conditional on knowing the value of the other variable.

The conditional density of X given Y , written fX|Y (x | y), is only defined

when fY (y) > 0. The conditional density is defined as follows:

345
346 Chapter 27. Conditional Distributions

Definition 27.1. Conditional density

The conditional density of a continuous random variable X, given Y = y, is
defined as the joint density of X and Y divided by the density of Y :

fX,Y (x, y)
fX|Y (x | y) = .
fY (y)

For the conditional density of X given Y to make sense, we must use Y values
such that fY (y) > 0.
Equivalently, when fY (y) > 0, then fX|Y (x | y) is the unique function that
gives
fX,Y (x, y) = fY (y)fX|Y (x | y).

To explain how the conditional density can actually be used to obtain con-
ditional probabilities, consider the following example:
The probability
Z
P (X ∈ A | Y = 3) = fX|Y (x | 3) dx
A
only depends on A. For instance, if A is the interval from 0 to 10, we have
Z 10
P (0 ≤ X ≤ 10 | Y = 3) = fX|Y (x | 3) dx.
0
More generally,
Z
P (X ∈ A | Y = y) = fX|Y (x | y) dx.
A
The value “y” should be thought of as “fixed,” and then we are interested in
whether X is in some interval A. For instance, if we think of “y” as fixed, we
might ask whether X is found in the interval from 0 to 10:
Z 10
P (0 ≤ X ≤ 10 | Y = y) = fX|Y (x | y) dx.
0

Some specific examples should help to clarify this idea.

27.2 Examples

Example 27.2. A bird lands in a grassy region described as follows: x ≥ 0,

and y ≥ 0, and x + y ≤ 10. This region is shown in Figure 27.1.
Let X and Y be the coordinates of the bird’s landing. Assume that X and
Y have the joint density
fX,Y (x, y) = 1/50 for 0 ≤ x and 0 ≤ y and x + y ≤ 10,
and fX,Y (x, y) = 0 otherwise.
 27.2. Examples 347

y
10

x
5 10

Figure 27.1: The grassy region where a bird lands.

If we know that the bird lands somewhere on the line where Y = 2, we can find
the conditional density of X given that Y = 2. First we need the density of
Y . For 0 ≤ y ≤ 10, we integrate over all relevant x’s—here, over x from 0 to
10 − y. After the integral with respect to x is performed, and we substitute in We can think of this
10 − y and 0 for x, there will be no x’s remaining. In other words, as integrating x out
Z ∞ of the picture.
fY (y) = fX,Y (x, y) dx
−∞
Z 10−y
= 1/50 dx
0
= (10 − y)/50

We know that 0 ≤ X ≤ 10 − 2 = 8 when Y = 2. So, for 0 ≤ x ≤ 8, we compute

fX,Y (x, 2)
fX|Y (x | 2) =
fY (2)
1/50
=
(10 − 2)/50
= 1/8

Given that the bird’s landing y-coordinate is 2, what is the probability that the
x-coordinate is between 0 and 5?
The probability is
Z 5 Z 5
P (0 ≤ X ≤ 5 | Y = 2) = fX|Y (x | 2) dx = 1/8 dx = 5/8.
0 0

This makes sense, because in this problem, all of the integrands are constant. We
are asking where the x-coordinate will be, and we know that the x-coordinate
is somewhere from 0 to 8, and all of the parts of the line are treated equally, in
a sense.
348 Chapter 27. Conditional Distributions

Example 27.3. Consider a dancer who is twirling around a semicircle shaped

stage. Suppose that her location on the stage has x- and y-coordinates X and
Y , respectively, with the constraints that X 2 +Y 2 ≤ 32 (i.e., X and Y are found
in a circle of radius 3), and also Y ≥ 0 (i.e., the dancer is in a semicircle at the
front of the stage). See Figure 27.2 for an illustration of this region. Suppose
that the joint density of her location is
2
fX,Y (x, y) = for x, y in the semicircle,
9π
and fX,Y (x, y) = 0 otherwise.

Given that the dancer’s y-coordinate is exactly 2, what is the density of the
x-coordinate of her location?
We first compute that, for 0 ≤ y ≤ 3, we have − 9 − y 2 ≤ x ≤ 9 − y 2 .
p p

The density of Y for 0 ≤ y ≤ 3 is thus:

Z ∞ Z √9−y2 p
2 4 9 − y2
fY (y) = fX,Y (x, y) dx = √ dx =
−∞ − 9−y 2 9π 9π

and fY (y) = 0 otherwise.

√Given that √
the dancer’s
√ y-coordinate
√ is 2, her√x-coordinate
√ must be between
− 9 − 2 = − 5 and 9 − 2 = 5. So, for − 5 ≤ x ≤ 5, we have
2 2

fX,Y (x, 2) 2/(9π) 1

fX|Y (x | 2) = = √ = √
fY (2) 2
4 9 − 2 / (9π) 2 5

y
3

1
x
−3 −2 −1 1 2 3

Figure 27.2: A stage where a dancer is twirling around.

We give more examples in Chapter 31 about constant joint densities and condi-
tional distributions (constant densities correspond to Continuous Uniform ran-
dom variables).
 27.2. Examples 349

Example 27.4. Consider the joint density from Example 26.4:

3
fX,Y (x, y) = xy for 0 ≤ x and 0 ≤ y and x + y ≤ 2,
2

and fX,Y (x, y) = 0 otherwise.

Suppose we want to compute the density of X given Y . Then we need to know

the density of Y . For 0 ≤ y ≤ 2, we integrate x out of the picture, as follows:
Z ∞
fY (y) = fX,Y (x, y) dx
−∞
Z 2−y
3
= xy dx
0 2
2−y
3 x2
= y
2 2 x=0
3 (2 − y)2
= y
2 2

It follows that the conditional density of X given Y , for 0 ≤ y ≤ 2 is

3
2 xy 2x
fX|Y (x | y) = 2 = for 0 ≤ x ≤ 2 − y,
3 (2−y)
y (2 − y)2
2 2

and fX|Y (x | y) = 0 otherwise.

As a specific example, suppose that we know Y = 1/2. In such a case,
we also know that 0 ≤ X ≤ 3/2. (See Figure 27.3.) What is the conditional

y
2

x
1 2

Figure 27.3: Region in which Y = 1/2 is given; so X must be between 0

and 2 − Y = 3/2.
350 Chapter 27. Conditional Distributions

probability that X ≤ 1, given Y = 1/2? We can compute

Z 1
P (X ≤ 1 | Y = 1/2) = fX|Y (x | 1/2) dx
−∞
Z 1
2x
= dx
0 (2 − 1/2)2
1
x2
=
(3/2)2 0
12
=
9/4
= 4/9

As another specific example, suppose again that we know Y = 1/2. In such a

case, we also know that 0 ≤ X ≤ 3/2. What is the conditional probability that
X ≥ 4/5, given Y = 1/2? We can compute
Z ∞
P (X ≥ 4/5) = fX|Y (x | 1/2) dx
4/5
Z 3/2
2x
= dx
4/5 (2 − 1/2)2
3/2
x2
=
(3/2)2 4/5
(3/2)2
− (4/5)2
=
9/4
= 161/225
= 0.7156

Finally, we can verify that fX|Y (x | y) = 2x

(2−y)2
is a density for 0 ≤ y ≤ 2 and
for 0 ≤ x ≤ 2 − y. To see this, we note that (2−y)2
2x
≥ 0 and also

Z ∞ Z 2−y
2x
fX|Y (x | y) dx = dx
−∞ 0 (2 − y)2
2−y
x2
=
(2 − y)2 0
(2 − y)2
=
(2 − y)2
= 1.

So fX|Y (x | y) is a density.
 27.3. Exercises 351

Remark 27.5. Conditional density for independent random variables

As discussed in the previous chapter, X and Y are independent random vari-
ables if and only if the conditional density of X given Y is exactly the same
as the density of X.
To see this: The conditional density of a continuous random variable X, given
Y = y, is defined as:
fX,Y (x, y)
fX|Y (x | y) = ,
fY (y)
but X and Y are independent if and only if

fX,Y (x, y) = fX (x)fY (y),

and in this case, the equation for the conditional density simplifies to

fX,Y (x, y) fX (x)fY (y)

fX|Y (x | y) = = = fX (x).
fY (y) fY (y)

Thus, X and Y are independent if and only if the conditional density

fX|Y (x | y) of X given Y is equal to the density fX (x) of X.

27.3 Exercises
27.3.1 Practice
For the joint densities in Exercises 27.1 to 27.10, find (a) the density
fY (y) of Y , and (b) the conditional density fX|Y (x | y) of X given Y .

Exercise 27.1. Let

15 2
fX,Y (x, y) = x y, for 0 ≤ x, 0 ≤ y, x + y ≤ 4,
256
and fX,Y (x, y) = 0 otherwise.

Exercise 27.2. Let fX,Y (x, y) be constant on the region shown in Figure 27.4.

Exercise 27.3. Let

3 2
fX,Y (x, y) = (x + y), for 0 ≤ x ≤ 2, 0 ≤ y ≤ 4,
80
and fX,Y (x, y) = 0 otherwise.

Exercise 27.4. Let

3
fX,Y (x, y) = (4 − x)y, for 0 < y < x < 4,
32
and fX,Y (x, y) = 0 otherwise.
352 Chapter 27. Conditional Distributions

x
1 2 3 4

Figure 27.4: Region where X and Y have positive density.

Exercise 27.5. Let

fX,Y (x, y) = 5e−x−3y , for 0 < y < x/2,
and fX,Y (x, y) = 0 otherwise.
Exercise 27.6. Let
9 −x/4−y/5
fX,Y (x, y) = e , for 0 < x < y,
100
and fX,Y (x, y) = 0 otherwise.
Exercise 27.7. Let
6
fX,Y (x, y) = (x + y)2 , for 0 ≤ x ≤ 1, 0 ≤ y ≤ 1,
7
and fX,Y (x, y) = 0 otherwise.
Exercise 27.8. Let
12
fX,Y (x, y) = (1 − x2 ), for 0 ≤ x, 0 ≤ y, x + y ≤ 1,
5
and fX,Y (x, y) = 0 otherwise.
Exercise 27.9. Let fX,Y (x, y) be constant on the region where x and y are
nonnegative and x + y ≤ 30.
Exercise 27.10. Let
3
fX,Y (x, y) = xy, for 0 ≤ x, 0 ≤ y, x + y ≤ 2,
2
and fX,Y (x, y) = 0 otherwise.
 27.3. Exercises 353

27.3.2 Extensions
Exercise 27.11. Consider a pair of random variables X, Y with constant joint
density on the triangle with vertices at (0, 0), (3, 0), and (0, 3).
a. For 0 ≤ y ≤ 3, find the conditional density fX|Y (x | y) of X, given Y = y.
b. Find the conditional probability that X ≤ 1, given Y = 1.
c. Find the conditional probability that X ≤ 1, given Y ≤ 1.
Exercise 27.12. Consider a pair of random variables X, Y with constant joint
density on the quadrilateral with vertices (0, 0), (2, 0), (2, 6), (0, 12).
a. For 0 ≤ y ≤ 6, find the conditional density fX|Y (x | y) of X, given Y = y.
b. For 6 ≤ y ≤ 12, find the conditional density fX|Y (x | y) of X, given
Y = y.
c. Find the conditional probability that X ≤ 1, given 3 ≤ Y ≤ 9.
Exercise 27.13. Let X, Y have joint density fX,Y (x, y) = 14e−2x−7y for x > 0
and y > 0; and fX,Y (x, y) = 0 otherwise.
a. For y > 0, find the conditional density fX|Y (x | y) of X, given Y = y.
b. Find the conditional probability that X ≥ 1, given Y = 3.
c. Find the conditional probability that Y ≤ 1/5, given X = 2.7.
Exercise 27.14. Let X, Y have joint density fX,Y (x, y) = 18e−2x−7y for 0 <
y < x; and fX,Y (x, y) = 0 otherwise.
a. For y > 0, find the conditional density fX|Y (x | y) of X, given Y = y.
b. For x > 0, find the conditional density fY |X (y | x) of Y , given X = x.
Exercise 27.15. Suppose X, Y have joint density
(
1
(3 − x)(2 − y) if 0 ≤ x ≤ 3 and 0 ≤ y ≤ 2,
fX,Y (x, y) = 9
0 otherwise.

a. For 0 ≤ y ≤ 2, find the conditional density fX|Y (x | y) of X, given Y = y.

b. Find the conditional probability that X ≤ 2, given Y = 3/2.
c. Find the conditional probability that Y ≥ 1, given X = 5/4.
Exercise 27.16. Every day, a student calls his mother and then (afterwards)
calls his girlfriend. Let X be the time (in hours) until he calls his mother, and
let Y be the time (in hours) until he calls his girlfriend. Since he always calls
his mother first, then X < Y . So let the joint density of the time be
fX,Y (x, y) = 10e−3x−2y for 0 < x < y,
and fX,Y (x, y) = 0 otherwise. Given that X = 1/2, find the conditional proba-
bility that Y > 2/3. In other words, find P (Y > 2/3 | X = 1/2).
354 Chapter 27. Conditional Distributions

Exercise 27.17. Assume that Wyoming is shaded exactly like a rectangle, and
that a person’s location in Wyoming is 0 ≤ X ≤ 350 and 0 ≤ Y ≤ 276. Assume
that the joint density of a person’s location is constant on this region.
Assume that I-80 runs perfectly straight along the east-to-west line Y =
30, as in Figure 27.5. Given that the person is located on I-80, what is the
conditional probability that he is within 10 miles of a State border? (He could
be either west or east; please take both into account.)

250
200
150
100
50
x
100 200 300

Figure 27.5: State of Wyoming.

Exercise 27.18. Consider the semicircle with radius 2 seen in Figure 27.6. A
dancer is randomly located in the semicircle, with the joint density of their
location to be
2
fX,Y (x, y) = if x, y is in the semicircle,
9π
and fX,Y (x, y) = 0 otherwise. Find the probability that the person is in the
region where −1 ≤ X ≤ 1 and 0 ≤ Y ≤ 1.

y
3

1
x
−3 −2 −1 1 2 3

Figure 27.6: Picture of a semicircle-shaped stage.

Exercise 27.19. Show that a conditional PDF satisfies the conditions of a

PDF.
Chapter 28

Expected Values of Continuous

Random Variables

A thing long expected takes the form of the unexpected when at last it comes.
—Mark Twain

How long should we expect to have to wait at a traffic light? What is the
expected length of a randomly chosen song on our playlist? What is the expected
distance from the center, when a dart is thrown at a dartboard?

28.1 Introduction
To a great extent, the way that expected values are calculated for continuous
random variables should not be surprising: the method is analogous to how
expected values are calculated for discrete random variables. For continuous
random variables, we cannot use the probabilities of values of random variables
as a method of scaling, because the probability of any particular value is 0,
i.e., P (X = x) = 0 for each x. Instead, we use the density fX (x) for scaling.
In other words, we integrate x over all possible x’s (just as we summed over
all possible x’s with discrete random variables), and we give a weight of fX (x)
to each x (just as we gave a weight of pX (x) to each x with discrete random
variables).

Definition 28.1. Expected value of a continuous random variable

A continuous random variable X with density fX (x) has expected value
Z ∞
E(X) = xfX (x) dx.
−∞

355
356 Chapter 28. Expected Values of Continuous Random Variables

(This should remind us of thePformula for the expected value of a discrete

random variable X, i.e., E(X) = x xpX (x) dx.)
We revisit some earlier examples and calculate the expected values of the
associated random variables:

Example 28.2. If X is a continuous random variable with density

(
1
26 (4x + 1) if 2 ≤ x ≤ 4,
fX (x) =
0 otherwise,

find the expected value of X.

The expected value of X is

Z ∞
E(X) = xfX (x) dx,
−∞

but fX (x) = 0 for x’s outside the range 2 ≤ x ≤ 4. Thus, we have

Z 4
1
E(X) = (x) (4x + 1) dx
2 26
Z 4
1
= (4x2 + x) dx
2 26
4
1 4 3 x2

= x +
26 3 2 x=2
4 3 42 4 3 22
   
1
= 4 + − 2 +
26 3 2 3 2
= 121/39
= 3.1026

Note that, if a continuous random variable is always found in a finite-length

interval, then the expected value of X must be found in this interval too:

Theorem 28.3. Bounds on the expected value of a continuous ran-

dom variable
If a ≤ X ≤ b, then we must also have a ≤ E(X) ≤ b too.
 28.1. Introduction 357

To see this:
Z b
E(X) = xfX (x) dx
a
Z b
≤ bfX (x) dx since x ≤ b on the interval [a, b]
a
Z b
=b fX (x) dx
a
= (b)(1) dx since the density must integrate to 1
= b,

and similarly
Z b Z b Z b
E(X) = xfX (x) dx ≥ afX (x) dx = a fX (x) dx = (a)(1) dx = a.
a a a

So, as we claimed, if we have a ≤ X ≤ b, we must have a ≤ E(X) ≤ b too.

Example 28.4. Suppose X has density

3
fX (x) = (x)(2 − x)(3 − x) for 0 ≤ x ≤ 2,
8
and fX (x) = 0 otherwise.

The expected value of X is

Z ∞
E(X) = xfX (x) dx
−∞
Z 2  
3
= (x) (x)(2 − x)(3 − x) dx
0 8
3 2
Z
= (6x2 − 5x3 + x4 ) dx
8 0
3 2
= (6x3/3 − 5x4/4 + x5/5) x=0
8
3
= (6(2)3/3 − 5(2)4/4 + 25/5)
8
3
= (960 − 1200 + 384)/60
8
= 9/10
358 Chapter 28. Expected Values of Continuous Random Variables

Example 28.5. Let X be a random variable with

fX (x) = 3e−3x for x > 0,

and fX (x) = 0 otherwise.

The expected value of X is

Z ∞ Z ∞
E(X) = xfX (x) dx = (x)(3)e−3x dx
−∞ 0

Now we use integration by parts with u = 3x and dv = e−3x dx, so du = 3 dx

−3x
and v = e−3 . So we get
∞ ∞ ∞
e−3x e−3x e−3x
  Z
∞
E(X) = (3x) − (3) dx = −x e−3x x=0
+
−3 x=0 0 −3 −3 x=0

We note that −xe−3x → −0e−(3)(0) = 0 as x → 0. Also −xe−3x = − ex3x → − ∞

∞
as x → ∞, so, using L’Hospital’s Rule, − ex3x → − 3e13x → 0 as x → ∞. Also
e−3x e−3x
−3 → 0 as x → ∞ and −3 → −1/3 as x → 0. Thus

E(X) = 1/3.

28.2 Some Generalizations about Expected Values

Suppose that a and b are any constants, with a < b. Suppose that fX (x) is
constant on the interval [a, b] and is 0 otherwise. Then the “constant” must be
1/(b − a) so that the integral of the density yields 1. Thus

1
fX (x) = for a ≤ x ≤ b,
b−a
and fX (x) = 0 otherwise.

Theorem 28.6. Expected value of a continuous random variable that

is constant on an interval and zero otherwise
If X is a continuous random variable with density
1
fX (x) = for a ≤ x ≤ b,
b−a
and fX (x) = 0 otherwise, then

a+b
E(X) = .
2
 28.3. Some Applied Problems with Expected Values 359

To derive the expected value in the previous box, we compute

Z ∞ Z b     b
1 2 1
E(X) = xfX (x) dx = (x) dx = (x /2)
−∞ a b−a b−a x=a
Substituting in the values a and b for x, we conclude
   
2 2 1 1
E(X) = ((b − a )/2) = ((a + b)(b − a)/2) = (a + b)/2.
b−a b−a
Theorem 28.7. Expected value of a continuous random variable with
exponentially decreasing density
If, for some constant λ > 0, the random variable X has density

fX (x) = λe−λx for x > 0,

and fX (x) = 0 otherwise, then

E(X) = 1/λ.
To derive this expected value, we follow the earlier example, in which λ was
equal to 3. The same argument works here. The expected value of X is
Z ∞ Z ∞
E(X) = xfX (x) dx = xλe−λx dx
−∞ 0

Now we use integration by parts with u = λx and dv = e−λx dx, so du = λ dx

−λx
and v = e−λ . So we get
 −λx  ∞ Z ∞ ∞
e e−λx −λx
∞ e−λx
E(X) = (λx) − (λ) dx = −x e +
−λ x=0 0 −λ x=0 −λ x=0
We note that −xe−λx → −0e−(λ)(0) = 0 as x → 0. Also −xe−λx = − eλx
x
→ −∞
∞
as x → ∞, so, using L’Hospital’s Rule, − eλx → − λeλx → 0 as x → ∞. Also
x 1

e−λx e−λx
−λ → 0 as x → ∞ and −λ → −1/λ as x → 0. Thus
E(X) = 1/λ.

28.3 Some Applied Problems with Expected Values

Example 28.8. Let X be the time (in minutes) that Maxine waits for a traffic
light to turn green, and let Y be the time (in minutes, at a different intersection)
that Daniella waits for a traffic light to turn green. Suppose that X and Y have
joint density
fX,Y (x, y) = 15e−3x−5y , for x > 0 and y > 0,
and fX,Y (x, y) = 0 otherwise. Find the expected time that each of them wait
at their lights.
360 Chapter 28. Expected Values of Continuous Random Variables

In this example, fX (x) = 0 for x ≤ 0, and fY (y) = 0 for y ≤ 0. We could notice

that 15e−3x−5y factors into the product of two densities, (3e−3x )(5e−5y ), or we
could get the density of X by integrating Y out of the picture, as follows:
Z ∞
fX (x) = fX,Y (x, y) dy
Z−∞
∞
= 15e−3x−5y dy
0
∞
= −3e−3x−5y y=0
= 3e−3x

Thus
fX (x) = 3e−3x for x > 0,

and fX (x) = 0 otherwise.

Using the boxed result in the previous section, it follows that

E(X) = 1/3.

We could also have integrated from the start:

Z ∞Z ∞
E(X) = (x)(15e−3x−5y ) dy dx.
0 0

Similarly, for y > 0,

fY (y) = 5e−5y for y > 0,

and fY (y) = 0 otherwise. Again using the boxed result in the previous section,
it follows that
E(Y ) = 1/5.

Example 28.9. Suppose that a person throws a dart at a square dart board.
Let X and Y denote, respectively, the x- and y-coordinates (in feet) of the
location where the dart lands; the middle of the dart board is at (0, 0). Suppose
that the dart board is two feet wide and two feet tall, so that the dart only
lands on the dart board if −1 ≤ X ≤ 1, −1 ≤ Y ≤ 1. Also suppose that the
person always hits the dart board, and moreover, X and Y have joint density:

9
fX,Y (x, y) = (1 − x2 )(1 − y 2 ) for −1 ≤ x ≤ 1, −1 ≤ y ≤ 1,
16

and fX,Y (x, y) = 0 otherwise. Find the expected values of X and Y .

 28.4. Exercises 361

In this example, fX (x) = 0 for x > 1 and for x < −1.

For −1 ≤ x ≤ 1, we can get the density of X by integrating Y out of the
picture, as follows:
Z ∞
fX (x) = fX,Y (x, y) dy
−∞
Z1
9
= (1 − x2 )(1 − y 2 ) dy
−1 16
1
9
= (1 − x2 )(y − y 3/3)
16 y=−1
1
9
= (1 − x2 )((2/3) − (−2/3))
16 y=−1
3
= (1 − x2 )
4
Thus
1
3 1
Z   Z
3 2 3 1
E(X) = (x) (1−x )dx = (x−x3 )dx = (x2/2 − x4/4) x=−1
= 0.
−1 4 4 −1 4

Since the entire problem is symmetric in X and Y , it follows also that E(Y ) = 0.

28.4 Exercises
28.4.1 Practice
For the densities in Exercises 28.1 to 28.8, find the expected value of
the random variable.

Exercise 28.1. A student models the number of revolutions X on a pencil

sharpener that are needed to fully sharpen his pencil, by using density
(
2
(x − 19) for 19 ≤ x ≤ 20;
fX (x) = 3 1
− 3 (x − 22) for 20 ≤ x ≤ 22;

and fX (x) = 0 otherwise.

Exercise 28.2. The time (in minutes) it takes until Julie’s boyfriend calls is a
random variable X with density
1 −x/10
fX (x) = e , for 0 < x,
10
and fX (x) = 0 otherwise.
362 Chapter 28. Expected Values of Continuous Random Variables

Exercise 28.3. Let X have density

1
fX (x) = e−x/3 , for 0 < x,
3
and fX (x) = 0 otherwise.
Exercise 28.4. Let X have density
√
fX (x) = x, for 0 ≤ x ≤ 2,

and fX (x) = 0 otherwise.

Exercise 28.5. Suppose that the length X of a tungsten wire is
3 2
fX (x) = x , for 0 ≤ x ≤ 10,
1000
and fX (x) = 0 otherwise.
Exercise 28.6. Edward’s trip to school is three miles long, down a straight
highway. Let X denote his location at a randomly chosen point on his drive to
school, with density
1
fX (x) = , for 0 ≤ x ≤ 3,
3
and fX (x) = 0 otherwise.
Exercise 28.7. Bruno goes fishing every day in the summer for 5 hours a day.
Let X be the amount of time (in hours) that goes by until he notices that he
catches his first fish (there is always a fish on his line by the end of the day,
because he fishes in a very densely populated pond). So X has density
3
fX (x) = (x2 + 8), for 0 ≤ x ≤ 5,
245
and fX (x) = 0 otherwise.
Exercise 28.8. Let X have density
2
fX (x) = x, for 0 ≤ x ≤ 5,
25
and fX (x) = 0 otherwise.

Exercise 28.9. Consider a pair of random variables X, Y with constant joint

density on the triangle with vertices at (0, 0), (3, 0), and (0, 3). Find the ex-
pected value E(X). (Notice that, by symmetry, E(Y ) is just the same!)
Exercise 28.10. Consider a pair of random variables X, Y with constant joint
density on the quadrilateral with vertices (0, 0), (2, 0), (2, 6), (0, 12).
 28.4. Exercises 363

a. Find the expected value E(X).

b. Find the expected value E(Y ).

Exercise 28.11. Let X, Y have joint density fX,Y (x, y) = 14e−2x−7y for x > 0
and y > 0; and fX,Y (x, y) = 0 otherwise.

a. Find the expected value E(X).

b. Find the expected value E(Y ).

Exercise 28.12. Let X, Y have joint density fX,Y (x, y) = 18e−2x−7y for 0 <
y < x; and fX,Y (x, y) = 0 otherwise.

a. Find the expected value E(X).

b. Find the expected value E(Y ).

Exercise 28.13. Suppose X, Y have joint density

(
1
(3 − x)(2 − y) if 0 ≤ x ≤ 3 and 0 ≤ y ≤ 2,
fX,Y (x, y) = 9
0 otherwise.

a. Find the expected value E(X).

b. Find the expected value E(Y ).

Exercise 28.14. Suppose X, Y have joint density

(
4
(x)(1 − x)(3 − y) if 0 ≤ x ≤ 1 and 0 ≤ y ≤ 3,
fX,Y (x, y) = 3
0 otherwise.

a. Find the expected value E(X).

b. Find the expected value E(Y ).

Exercise 28.15. Let X have density

 4 for 0 ≤ x ≤ 1,

3

fX (x) = 14 for 3 ≤ x ≤ 4,
0 otherwise.



Find E(X).

Exercise 28.16. Let X have density

 8 for 0 ≤ x ≤ 1,

7

fX (x) = 18 for 7 ≤ x ≤ 8,
0 otherwise.



Find E(X).
364 Chapter 28. Expected Values of Continuous Random Variables

Exercise 28.17. Consider the joint density of X and Y from Example 26.2:
1
fX,Y (x, y) = (1 − x2 )(3 − y) for −1 ≤ x ≤ 1 and −1 ≤ y ≤ 1,
8
and fX,Y (x, y) = 0 otherwise.
a. Find E(X).
b. Find E(Y ).
Exercise 28.18. Consider the joint density of X and Y from Example 26.4:
3
fX,Y (x, y) = xy for 0 ≤ x and 0 ≤ y and x + y ≤ 2,
2
and fX,Y (x, y) = 0 otherwise. Find E(Y ).
Exercise 28.19. As in Example 27.2, a bird lands in a grassy region described
as follows: 0 ≤ x, and 0 ≤ y, and x + y ≤ 10. Let X and Y be the coordinates
of the bird’s landing. Assume that X and Y have the joint density
fX,Y (x, y) = 1/50 for 0 ≤ x and 0 ≤ y and x + y ≤ 10,
and fX,Y (x, y) = 0 otherwise. Find E(Y ). (Hint: The answer is not “ E(Y ) =
5.”)

28.4.2 Extensions
Exercise 28.20. Consider two points that are independently placed on a line
of length 10, at locations X and Y . Thus the joint density of X and Y is
fX,Y (x, y) = 1/100 for 0 ≤ x ≤ 10 and 0 ≤ y ≤ 10,
and fX,Y (x, y) = 0 otherwise.
a. First, try to show that, if Z denotes the distance between X and Y (so
0 ≤ Z ≤ 10), then the cumulative distribution function FZ (z) of Z is
1 1 2
FZ (z) = z − z for 0 ≤ z ≤ 10
5 100
and FZ (z) = 0 for z ≤ 0 and FZ (z) = 1 for z ≥ 10. (Hint: Use the complement;
try to find P (Z > z), i.e., the probability that the distance between X and Y
is more than z. A picture should help.)
b. Regardless of whether you can accomplish the part above, just find fZ (z),
by differentiating FZ (z).
c. Use the density fZ (z) to find E(Z), i.e., the expected distance between
X and Y .
Exercise 28.21. Let X and Y have joint density
3 2
fX,Y (x, y) = (x + y), for 0 ≤ x ≤ 2 and 0 ≤ y ≤ 4,
80
and fX,Y (x, y) = 0 otherwise. Find E(X).
 28.4. Exercises 365

28.4.3 Advanced
Exercise 28.22. Three children are skating around the perimeter of a circle in
an ice rink of radius 50 feet. Assume that their locations around the perimeter
of the circle are independent, continuous random variables, each with constant
density. Their mother comes to the door of the ice rink (located at a fixed
position on perimeter of the circle). When she appears, what is the expected
distance around the perimeter from her to the closest of the three children?
Chapter 29

Variance of Continuous Random

Variables

If you do not expect the unexpected, you will not find it; for it is hard to be
sought out, and difficult.
—Heraclitus

To what extent are the students’ grades in a course near the average? How can
we quantify the spread of randomly thrown darts from the center of the board?

29.1 Variance of a Continuous Random Variable

In the previous chapter, we extensively studied the fact that the expected value
E(X) of a continuous random variable X with density fX (x) is
Z ∞
E(X) = xfX (x) dx.
−∞

The variance of a continuous random variable measures the spread of the random
variable’s distribution around the mean (just as the variance does with a discrete
random variable as well). We first point out two different random variables that
have the same expected value but different variances.

Example 29.1. For our first example, we define two random variables that
each have expected value 10/3. The random variable Y is from Exercise 28.19,
and the random variable X will be defined below. Since these two random
variables have the same expected value, we will use their variances to compare
them.

366
 29.1. Variance of a Continuous Random Variable 367

In Exercise 28.19, a bird lands in a grassy region described as follows: x ≥ 0,

and y ≥ 0, and x + y ≤ 10. Let X and Y be the coordinates of the bird’s
landing. Assume that X and Y have the joint density

fX,Y (x, y) = 1/50 for 0 ≤ x and 0 ≤ y and x + y ≤ 10,

and fX,Y (x, y) = 0 otherwise. We already found fY (y), for 0 ≤ y ≤ 10:

Z 10−y
fY (y) = 1/50 dx = (10 − y)/50
0

and fY (y) = 0 otherwise, and then we computed the expected value of Y :

Z 10
E(Y ) = (y)(10 − y)/50 dy = 10/3.
0

Now we compare to a different random variable that also has expected value
10/3: Suppose that X is the waiting time until the next bus arrives, given in
minutes. If X has density

3 −3x/10
fX (x) = e for x > 0,
10
and fX (x) = 0 otherwise, then the density of X is exponentially decreasing,
with λ = 3/10, so we can conclude, by Theorem 28.7 that E(X) = 1/λ = 10/3.
Now a question immediately arises: Both random variables, X and Y , have
expected value 10/3; can we say something more about the distributions of
X and Y to better understand how X and Y behave? With discrete random
variables, we sometimes looked at the masses; for continuous random variables,
we can look at the densities. The densities of Y and X are given, respectively,
on the left and right sides of Figure 29.1.

fX (x)
0.3

fY (y)
0.2 0.2

0.1 0.1
y x
2 4 6 8 10 2 4 6 8 10

Figure 29.1: Left: The density fY (y) = (10−y)/50 of Y . Right: The density
3 −3x/10
fX (x) = 10 e of X.
 368 Chapter 29. Variance of Continuous Random Variables

It is still difficult to quantify which distribution in Example 29.1 is more

“spread out” from the expected value. To quantify this—as we quantified such
a spread with discrete variables—we use the concept of the variance.

Definition 29.2. Variance

For a random variable X with expected value µX = E(X), the variance of X
is
Var(X) = E((X − µX )2 ).

We notice that

Var(X) = E((X − µX )2 ) = E(X 2 − 2µX X + µ2X ) = E(X 2 ) − 2µX E(X) + µ2X ,

but µX and E(X) are equal, so this yields:

Remark 29.3. Variance

For X with expected value µX = E(X), the variance of X is always

Var(X) = E(X 2 ) − (E(X))2 .

The spread of a random variable is sometimes quantified in terms of the

square root of the variance instead of the variance itself:

Definition 29.4. Standard deviation

For a random variable X with variance Var(X), the standard deviation of X
is p
σX = Var(X).

The definition of To find the variance, we need to know how to compute the expected value
variance is the of the square of X and also how to compute the expected value of the square
same for continuous of Y .
and discrete random
variables.
Definition 29.5. Expected value of the square of a continuous random
variable
If X is a continuous random variable X with density fX (x), then the expected
value of X 2 is Z ∞
2
E(X ) = x2 fX (x) dx.
−∞
 29.1. Variance of a Continuous Random Variable 369

Example 29.1 (continued) In Example 29.1, we introduced a random variable

Y such that
Z ∞
2
E(Y ) = y 2 fY (y) dy
−∞
Z 10
= y 2 (10 − y)/50 dy
0
Z 10
= (10y 2 − y 3 )/50 dy
0
10
= (10y 3/3 − y 4/4)/50 y=0
= (10000/3 − 10000/4)/50
= 50/3
and thus
Var(Y ) = E(Y 2 ) − (E(Y ))2 = 50/3 − (10/3)2 = 50/9 = 5.555556.
So the standard deviation of Y is
p
σY = 50/9 = 2.357.
We have to work a little harder to get the expected value of X 2 :
Z ∞
2
E(X ) = x2 fX (x) dx
−∞
Z ∞  
3
= (x2 ) e−3x/10 dx
0 10
and we use integration by parts, with u = x2 and dv = 10
3 −3x/10
e dx, and thus
du = 2x dx and v = −e −3x/10 , so
Z ∞
∞
−3x/10
2 2
E(X ) = (x )(−e ) − (2x)(−e−3x/10 ) dx
x=0 0
∞
The first part, (x2 )(−e−3x/10 ) x=0
, evaluates to 0. We can manipulate the
second part by factoring the “2” out, and by multiplying and dividing by 3/10,
so that we have
Z ∞     
2 2 3 −3x/10 2 20 10 200
E(X ) = (x) e dx = E(X) = = .
3/10 0 10 3/10 3 3 9
and finally we obtain
Var(X) = E(X 2 ) − (E(X))2 = 200/9 − (10/3)2 = 100/9 = 11.111111,
and p
σX = 100/9 = 3.333.
Since the standard deviation σX of X is greater than the standard deviation
σY of Y , then X is more “spread out” around its expected value, as compared
to how much Y is spread around its own expected value.
370 Chapter 29. Variance of Continuous Random Variables

Example 29.6. If X is a continuous random variable with density

(
1
(4x + 1) if 2 ≤ x ≤ 4,
fX (x) = 26
0 otherwise,

find the expected value of X 2 , and also the variance and standard deviation of
X.
The expected value of X 2 is
Z ∞
E(X 2 ) = x2 fX (x) dx,
−∞

but fX (x) = 0 for x’s outside the range 2 ≤ x ≤ 4. Thus, we have

Z 4
2 1
E(X ) = (x2 ) (4x + 1) dx
2 26
Z 4
1
= (4x3 + x2 ) dx
2 26
4
1 4 4 x3

= x +
26 4 3 x=2
43 23
   
1
= 44 + − 24 +
26 3 3
= 388/39
= 9.9487
From Example 28.2, we know E(X) = 121/39. Thus
Var(X) = E(X 2 ) − (E(X))2
= 388/39 − (121/39)2
= 491/1521
= 0.3228.
The standard deviation of X is
p 1√
σX = 491/1521 = 491 = 0.5682.
39

29.2 Expected Values of Functions of One Random

Variable
To compute the variance of a continuous random variable X, we need to cal-
culate E(X 2 ), the expected value of the square of X. Sometimes we need the
expected value of other functions of a random variables. Such expected values
are calculated in an analogous way:
 29.2. Expected Values of Functions of One Random Variable 371

Definition 29.7. Expected value of a function of a continuous random

variable
If X is a continuous random variable X with density fX (x), and if g is an
arbitrary function, then the expected value of g(X) is
Z ∞
E(g(X)) = g(x)fX (x) dx.
−∞

Example 29.8. Returning to the above Example, in which X is a continuous

random variable with density
(
1
(4x + 1) if 2 ≤ x ≤ 4,
fX (x) = 26
0 otherwise,
find the expected value of 1/X.
The expected value of 1/X is
Z ∞
E(1/X) = (1/x)fX (x) dx,
−∞

but fX (x) = 0 for x’s outside the range 2 ≤ x ≤ 4. Thus, we have

Z 4
1
E(1/X) = (1/x) (4x + 1) dx
2 26
Z 4  
1 1
= 4+ dx
2 26 x
4
1
= (4x + ln x)
26 x=2
1
= (((4)(4) + ln 4) − ((4)(2) + ln 2))
26
4 1
= + ln 2
13 26
= 0.3344
In particular, we emphasize that E(1/X) 6= 1/E(X). In this case, referring back
to Example 28.2, we have 1/E(X) = 39/121 = 0.3223, which is not equal to
E(1/X) = 0.3344.

Example 29.9. When Joyce reaches in her refrigerator for a drink from a 2-
liter bottle of soda, her roommates have usually already helped themselves to
some of the soda. So the amount of soda that remains is a random variable X.
If X has density fX (x) = 1/2 for 0 ≤ x ≤ 2, and fX (x) = 0 otherwise, calculate
the cost of the portion that has already been drunk by the roommates. Suppose
that the cost of soda is $1.38 for 2 liters, i.e., $0.79 per liter.
372 Chapter 29. Variance of Continuous Random Variables

The quantity that has been drunk is 2 − X liters, and its cost is (0.79)(2 − X).
So the desired expected value is
Z 2
E((0.79)(2 − X)) = (0.79)(2 − x)(1/2) dx
0
Z 2
= 0.395 (2 − x) dx
0
2
= 0.395 (2x − x2/2) x=0
= (0.395)(2)
= 0.79

Theorem 29.10. Expected value of a linear function of a continuous

random variable
If X is a continuous random variable X with density fX (x), and if a and b are
any constants, then the expected value of aX + b is

E(aX + b) = aE(X) + b.
To see that the theorem is true, just write
Z ∞
E(aX + b) = (ax + b)fX (x) dx
−∞
Z ∞ Z ∞
=a xfX (x) dx + b fX (x) dx
−∞ −∞
= aE(X) + b.

In Section 28.2, we learned that if the density of a random variable is con-

stant on the interval [a, b], then the expected value is (a + b)/2. In this case,
the density of X is the constant 1/2 on the interval [0, 2], so E(X) = 1. Thus,
using the boxed result above, we can decompose the soda example above, and
we again see that the expected cost of soda already drunk by the roommates is
E((0.79)(2 − X)) = E(1.58 − 0.79X) = 1.58 − 0.79E(X) = 0.79.

29.3 Expected Values of Functions of Two Random

Variables

Example 29.11. In Exercise 28.17 and Example 26.2, we considered:

1
fX,Y (x, y) = (1 − x2 )(3 − y) for −1 ≤ x ≤ 1 and −1 ≤ y ≤ 1,
8
and fX,Y (x, y) = 0 otherwise. Then we asked to find E(X) and to find E(Y ).
Now, revisiting this question, and considering the random variable X + Y ,
we ask what is the expected value of X + Y , i.e., what is E(X + Y )?
 29.3. Expected Values of Functions of Two Random Variables 373

One might guess (correctly, by the way!) that E(X + Y ) is just equal to
E(X) + E(Y ), and then use the answers from Exercise 28.17 to conclude that

E(X + Y ) = E(X) + E(Y ) = 0 − 1/9 = −1/9.

(Another response would be to let Z = X +Y , and then observe −2 ≤ Z ≤ 2,

and then find FZ (z), and then differentiate to find fZ (z), and finally to integrate
zfZ (z) over the range −2 ≤ z ≤ 2. This method of solution would be quite
complicated, so we do not use it.)
Fortunately, there is a straightforward method of computing the answer
(without relying on Exercise 28.17) that will prove even more useful. We just
integrate x+y over all possible x’s and y’s (i.e., using a double integral), weight-
ing each possible value x + y by the joint density, i.e., by fX,Y (x, y). Using this
method, we get:
Z ∞Z ∞
E(X + Y ) = (x + y)fX,Y (x, y) dy dx
−∞ −∞
Z1 Z 1
1
= (x + y) (1 − x2 )(3 − y) dy dx
−1 −1 8

One nice feature of this method is that it prevents us from having to take
the intermediate step to compute the density of either of the variables. The
downside is that it makes the integration a little more complicated, but we
handle the whole computation in one fell swoop! We get the following:
Z 1Z 1
1
E(X + Y ) = (x + y) (1 − x2 )(3 − y) dy dx
−1 −1 8
Z 1 Z 1
1
= (1 − x2 ) (3x − xy + 3y − y 2 ) dy dx
−1 8 −1
Z 1
1 h
1
i
= (1 − x2 ) (3xy − xy 2/2 + 3y 2/2 − y 3/3) y=−1 dx
−1 8
Z 1
1
= (1 − x2 )(6x − 2/3) dx
−1 8
Z 1
1
= (6x − 6x3 − 2/3 + 2x2/3) dx
−1 8
1 1
= (6x2/2 − 6x4/4 − 2x/3 + 2x3/9) x=−1
8
= −1/9

The ability to compute E(X + Y ) in one calculation is very nice. . . , but even
nicer is the fact that this method works much more generally. For any function
of X and Y , we can calculate the expected value of the function of X and Y
using this method.
374 Chapter 29. Variance of Continuous Random Variables

Definition 29.12. Expected values of functions of continuous random

variables
If X and Y are continuous random variables with joint density fX,Y (x, y), and
if g is any function of two variables, then
Z ∞Z ∞
E(g(X, Y )) = g(x, y)fX,Y (x, y) dy dx
−∞ −∞

Now, by using the new method for expected values, and letting g(x, y) = x, we
have a very succinct way to calculate the expected value of X directly, using a
joint density:

Remark 29.13. Expected values of a continuous random variable,

using a joint density
If X and Y are continuous random variables with joint density fX,Y (x, y),
then the expected value of X is
Z ∞Z ∞
E(X) = xfX,Y (x, y) dy dx
−∞ −∞

Generalizing the example above, we can always get the expected value of
the sum of two random variables, by just letting g(x, y) = x + y.

Remark 29.14. Expected value of the sum of two continuous random

variables
If X and Y are continuous random variables with joint density fX,Y (x, y),
then the expected value of X + Y is
Z ∞Z ∞
E(X + Y ) = (x + y)fX,Y (x, y) dy dx
−∞ −∞
Z ∞Z ∞ Z ∞Z ∞
= xfX,Y (x, y) dy dx + yfX,Y (x, y) dy dx
−∞ −∞ −∞ −∞
= E(X) + E(Y )

Example 29.15. In Example 25.9, a person throws a dart at a square dart

board. Let X and Y denote, respectively, the x- and y-coordinates (in feet)
of the location where the dart lands; the middle of the dart board is at (0, 0).
Suppose that the dart board is two feet wide and two feet tall, so that the dart
only lands on the dart board if −1 ≤ X ≤ 1, −1 ≤ Y ≤ 1. Also suppose
that the person always hits the dart board, and moreover, X and Y have joint
density:
9
fX,Y (x, y) = (1 − x2 )(1 − y 2 ) for −1 ≤ x ≤ 1, −1 ≤ y ≤ 1,
16
and fX,Y (x, y) = 0 otherwise.
 29.3. Expected Values of Functions of Two Random Variables 375

Back in Example 25.9, we already computed E(X) = 0, but we stopped in

the middle of that problem to compute fX (x), which was somewhat unnecessary.
Instead, with our new method, we can just compute directly, without stopping
in the middle.

We compute E(X) directly:

Z ∞ Z ∞
E(X) = xfX,Y (x, y) dy dx
−∞ −∞
Z 1Z 1  
9
= (x) (1 − x2 )(1 − y 2 ) dy dx
−1 −1 16
Z 1 Z 1
9 3
= (x − x ) dx (1 − y 2 ) dy
16 −1 −1
9 1 1
= (x2/2 − x4/4) x=−1 (y − y 3/3) y=−1
16
9
= (0)(4/3)
16
=0

Example 29.16. As a crowning achievement, we recompute Exercise 28.20:

Consider two points that are independently placed on a line of length 10, at
locations X and Y . Thus the joint density of X and Y is

fX,Y (x, y) = 1/100 for 0 ≤ x ≤ 10 and 0 ≤ y ≤ 10,

and fX,Y (x, y) = 0 otherwise.

Let Z = |X − Y | be the absolute value of the distance between X and Y .
Find the expected distance E(Z) = E(|X − Y |) between X and Y .

We compute

E(Z) = E(|X − Y |)
Z ∞Z ∞
= |x − y|fX,Y (x, y) dy dx
−∞ −∞
Z 10 Z 10
1
= |x − y| dy dx
0 0 100

The only thing we need to do is split the integral into two parts, according to
whether x ≥ y, in which case |x − y| = x − y, or whether x ≤ y, in which case
376 Chapter 29. Variance of Continuous Random Variables

|x − y| = y − x. So we continue

Z 10 Z x Z 10 Z 10
1 1
E(Z) = (x − y) dy dx + (y − x) dy dx
0 0 100 0 x 100
Z 10 x Z 10 10
2 1 2 1
= (xy − y /2) dx + (y /2 − xy) dx
0 100 y=0 0 100 y=x
Z 10 Z 10
2 1 2 1
= (x − x /2) dx + ((102/2 − 10x) − (x2/2 − x2 )) dx
0 100 0 100
Z 10
1
= (x2 − 10x + 50) dx
0 100
10
1
= (x3/3 − 10x2/2 + 50x)
100 x=0
1
= (1000/3 − 1000/2 + 500)
100
= 10/3

Notice that we did not need to compute FZ (z) or fZ (z) en route to comput-
ing E(Z) with this method. So there are some advantages to just computing
expected values using the powerful formula
Z ∞ Z ∞
E(g(X, Y )) = g(x, y)fX,Y (x, y) dy dx.
−∞ −∞

In the case of this example, we were using g(X, Y ) = |X − Y |.

29.4 More Friendly Facts about Continuous Random

Variables

(Several of these facts have analogous discrete versions; see Section 12.3.)
We already showed that E(aX + b) = aE(X) + b for any random variable
X and any constants a and b. More is true:

Theorem 29.17. Expected value of a (linear) sum of two continuous

random variables
If X and Y are continuous random variables with joint density fX,Y (x, y), and
if a and b are any constants, then the expected value of aX + bY is

E(aX + bY ) = aE(X) + bE(Y ).

 29.4. More Friendly Facts about Continuous Random Variables 377

To see that E(aX + bY ) = aE(X) + bE(Y ), just write

Z ∞Z ∞
E(aX + bY ) = (ax + by)fX,Y (x, y) dy dx
−∞ −∞
Z ∞Z ∞ Z ∞Z ∞
=a xfX,Y (x, y) dy dx + b yfX,Y (x, y) dy dx
−∞ −∞ −∞ −∞
= aE(X) + bE(Y )

Applying the above fact over and over shows that:

Corollary 29.18. Expected value of a (linear) sum of two or more

continuous random variables
If X1 , X2 , . . . , Xn are continuous random variables and if a1 , a2 , . . . , an are any
constants, then the expected value of a1 X1 + a2 X2 + · · · + an Xn is

E(a1 X1 + a2 X2 + · · · + an Xn ) = a1 E(X1 ) + a2 E(X2 ) + · · · + an E(Xn ).

Caution: The X and Y must be independent in order to use the

formulas in the box below!!!

Theorem 29.19. The expected value of the product of the functions

of two independent random variables equals the product of the ex-
pected values of the functions of the two random variables
If X and Y are independent random variables, and g and h are any two
functions, then
E(g(X)h(Y )) = E(g(X))E(h(Y )).

To see this, we compute

Z ∞Z ∞
E(g(X)h(Y )) = g(x)h(y)fX,Y (x, y) dy dx
−∞ −∞
Z ∞Z ∞
= g(x)h(y)fX (x)fY (y) dy dx since X, Y are indep.
−∞ −∞
Z ∞ Z ∞
= g(x)fX (x) dx h(y)fY (y) dy
−∞ −∞
= E(g(X))E(h(Y ))

Using identity functions for f and g, i.e., f (X) = X and g(Y ) = Y , the following
nice fact follows immediately:

Corollary 29.20. The expected value of the product of two indepen-

dent random variables equals the product of the expected values of
the two random variables
If X and Y are independent random variables, then

E(XY ) = E(X)E(Y ).
378 Chapter 29. Variance of Continuous Random Variables

Again, caution: The X and Y must be independent, to utilize these

results!!
The argument for the following is given as Exercise 29.32:

Theorem 29.21. If X1 , . . . , Xn are independent random variables, and

a1 , . . . , an are constants, then

Var(a1 X1 + · · · + an Xn ) = a21 Var(X1 ) + · · · + a2n Var(Xn ).

When n = 2, a1 = a, a2 = 1, X1 = X, and X2 = b (a constant), then

Var(X2 ) = E(X22 ) − (E(X2 ))2 = E(b2 ) − (E(b))2 = b2 − b2 = 0, so this gives us
immediately:

Corollary 29.22. If X is any random variable, and if a and b are any con-
stants, then
Var(aX + b) = a2 Var(X).

If we use ai = 1 for all i, then Theorem 29.21 yields:

Corollary 29.23. The variance of the sum of independent random

variables equals the sum of the variances
If X1 , . . . , Xn are independent random variables, then

Var(X1 + · · · + Xn ) = Var(X1 ) + · · · + Var(Xn ).

(We emphasize that the X1 , . . . , Xn are n independent values, not just n

copies of the same value.) For instance, if the random variables represent n
independent waiting times (e.g., at traffic lights), or heights of people, or other
kinds of independent measurements, we get the variance of the sum by adding
the variances of the individual random variables.

29.5 Exercises
29.5.1 Practice
For the densities in Exercises 29.1 to 29.8, find the variance of the
given random variable.

Exercise 29.1. A child has lost her ring somewhere along a 20-foot line in a
narrow strip of grass. Let X denote the location (lengthwise) along the grass
where it was dropped, so

1
fX (x) = , for 0 ≤ x ≤ 20,
20
and fX (x) = 0 otherwise.
 29.5. Exercises 379

Exercise 29.2. Let X have density

2
fX (x) = x, for 0 ≤ x ≤ 5,
25
and fX (x) = 0 otherwise.

Exercise 29.3. When meeting a friend at a coffee house, Donald sees that his
friend has not yet arrived, so he assumes that his friend’s arrival time, X, in
minutes, will have density
1
fX (x) = , for 0 ≤ x ≤ 5,
5
and fX (x) = 0 otherwise.

Exercise 29.4. The waiting time X, in minutes, until the next green car passes
has density
1
fX (x) = e−x/2 , for 0 < x,
2
and fX (x) = 0 otherwise.

Exercise 29.5. Darius always arrives at Wiley dining hall 50 minutes before
it closes. He estimates that the time X necessary to stand in line, in minutes,
has density
x
fX (x) = , for 0 ≤ x ≤ 10,
50
and fX (x) = 0 otherwise.

Exercise 29.6. Let X have density

1
fX (x) = , for 2 ≤ x ≤ 6,
4
and fX (x) = 0 otherwise.

Exercise 29.7. Upon arriving at a restaurant and finding that no table is

available, Diana and Markus have a waiting time X (in hours) with density

fX (x) = 2e−2x , for 0 < x,

and fX (x) = 0 otherwise.

Exercise 29.8. Let X be the time, in hours, that Christine spends talking on
the telephone. Then X has density

fX (x) = 3e−3x , for 0 < x,

and fX (x) = 0 otherwise.

380 Chapter 29. Variance of Continuous Random Variables

Exercise 29.9. Consider a pair of random variables X, Y with constant joint

density on the triangle with vertices at (0, 0), (3, 0), and (0, 3).
a. Find the expected value of the sum of X and Y , i.e., find E(X + Y ).
b. Find the variance of X, i.e., find Var(X).
Exercise 29.10. Consider a pair of random variables X, Y with constant joint
density on the quadrilateral with vertices (0, 0), (2, 0), (2, 6), (0, 12).
a. Find the variance of X, i.e., find Var X.
b. Find the variance of Y , i.e., find Var Y .
Exercise 29.11. Let X, Y have joint density fX,Y (x, y) = 14e−2x−7y for x > 0
and y > 0; and fX,Y (x, y) = 0 otherwise. Find the variance of the sum of X
and Y , i.e., find Var(X + Y ).
Exercise 29.12. Let X, Y have joint density fX,Y (x, y) = 18e−2x−7y for 0 <
y < x; and fX,Y (x, y) = 0 otherwise. Find the variance of Y .
Exercise 29.13. Suppose X, Y have joint density
(
1
(3 − x)(2 − y) if 0 ≤ x ≤ 3 and 0 ≤ y ≤ 2,
fX,Y (x, y) = 9
0 otherwise.

Find the expected value of X 2 + Y 3 , i.e., find E(X 2 + Y 3 ).

Exercise 29.14. Let X have density
1
fX (x) = , for 2 ≤ x ≤ 7,
5
and fX (x) = 0 otherwise.
a. Find the expected value of X 2 , i.e., E(X 2 ).
b. Find the expected value of 1/X 2 , i.e., E(1/X 2 ).
c. Find the variance Var(X).
d. Find the standard deviation σX .
Exercise 29.15. Two convicts are running away from prison. One man, John,
runs east X miles, and another man, Jeff, runs north Y miles. John cannot
run more than 4 miles, and Jeff cannot run more than 6 miles. At a random
point in time, their locations are spotted by a helicopter. Assume that the joint
density of their location is
1
fX,Y (x, y) = , for 0 ≤ x ≤ 6, and 0 ≤ y ≤ 4,
24
and fX,Y (x, y) = 0 otherwise. Let g(x, y) = x + y be the total distance that the
two men have run so far.
 29.5. Exercises 381

a. Find the expected value of the sum of these two distances.

b. Find the variance of the sum of these two distances.

Exercise 29.16. The distance X, in yards, that a small person can throw a
50-pound weight, has density

fX (x) = −0.0375x2 + 0.075x + 0.3, for 0 ≤ x ≤ 4,

and fX (x) = 0 otherwise. Find the variance of X.

For the densities in Exercises 29.17 to 29.22, and the given function
g(x, y) in each exercise, find E(g(X, Y )).

Exercise 29.17. Let X, Y have joint density

1
fX,Y (x, y) = , for 0 ≤ x ≤ 6, 0 ≤ y ≤ 6,
36
and fX,Y (x, y) = 0 otherwise. Let g(x, y) = 2x + y.

Exercise 29.18. Let X, Y have joint density

fX,Y (x, y) = 1, for 0 ≤ x ≤ 1, 0 ≤ y ≤ 1,

and fX,Y (x, y) = 0 otherwise. Let g(x, y) = x2 + y 2 .

Exercise 29.19. Let X, Y have joint density

1
fX,Y (x, y) = , for 0 ≤ x ≤ 4, 0 ≤ y ≤ 9,
36
and fX,Y (x, y) = 0 otherwise. Let g(x, y) = xy.

Exercise 29.20. Let X, Y have joint density

fX,Y (x, y) = 1, for 0 ≤ x ≤ 1, 0 ≤ y ≤ 1,

and fX,Y (x, y) = 0 otherwise. Let g(x, y) = x + y.

Exercise 29.21. Let X, Y have joint density

1
fX,Y (x, y) = , for 8 ≤ x ≤ 10, and 0 ≤ y ≤ 3,
6
and fX,Y (x, y) = 0 otherwise. Let g(x, y) = x2 y 3 .

Exercise 29.22. Let X, Y have joint density

xy 2
fX,Y (x, y) = , for 0 ≤ x ≤ 4, 0 ≤ y ≤ 3,
72
x−y
and fX,Y (x, y) = 0 otherwise. Let g(x, y) = 2 .
382 Chapter 29. Variance of Continuous Random Variables

29.5.2 Extensions
Exercise 29.23. Geoffrey does not like to be low on gas, so he randomly stops
to fill up his tank. He has a 14-gallon tank, and the current price of gas is $3.75
per gallon. Whenever he stops to buy gas, he always buys a candy bar for $1.30.
If X is the amount of gas (in gallons) in his tank when he stops for a purchase,
then fX (x) = 1/14 for 0 ≤ x ≤ 14, and fX (x) = 0 otherwise. He always fills
the tank, so he will always buy 14 − X gallons. Find the expected amount of
money Geoffrey spends on a purchase of gas and a candy bar.
Exercise 29.24. Let X have density
1
fX (x) = e−x/3 , for 0 < x,
3
and fX (x) = 0 otherwise.
a. Find the expected value of X 2 , i.e., E(X 2 ).
b. Find the variance Var(X).
c. Find the standard deviation σX .
Exercise 29.25. Let X and Y correspond to the horizontal and vertical co-
ordinates in the triangle with corners at (2, 0), (0, 2), and the origin. Let
15
fX,Y (x, y) = 28 (xy 2 + y) for (x, y) inside the triangle, and fX,Y (x, y) = 0
otherwise. Find E(XY ).
Exercise 29.26. Let X and Y correspond to the horizontal and vertical coor-
dinates in the rectangle with corners at (15, 0), (15, 10), (0, 10), and the origin.
Let fX,Y (x, y) = 150
1
for (x, y) inside the rectangle, and fX,Y (x, y) = 0 other-
wise. Find E(XY ).2

Exercise 29.27. Let X have density

2
fX (x) = ,
75
for x in the quadrilateral with vertices at (0, 0), (10, 0), (5, 5), (0, 5), and
fX (x) = 0 otherwise.
a. Find E(X).
b. Find Var(X).
Exercise 29.28. Let X have density
fX (x) = 25xe−5x , for 0 < x,
and fX (x) = 0 otherwise. Find Var(X).
Exercise 29.29. Let X have density
2 ln x
fX (x) = , for 1 ≤ x ≤ e,
x
and fX (x) = 0 otherwise. Find Var(X).
 29.5. Exercises 383

29.5.3 Advanced
Exercise 29.30. Every day, a student calls his mother and then (afterwards)
calls his girlfriend. Let X be the time (in hours) until he calls his mother, and
let Y be the time (in hours) until he calls his girlfriend. Since he always calls
his mother first, then X < Y . So let the joint density of the time be

fX,Y (x, y) = 10e−3x−2y for 0 < x < y,

and fX,Y (x, y) = 0 otherwise.

Let Z = |Y − X| be time in between the beginning of the two calls.

a. Find E(Z).
b. Find Var(Z).

Exercise 29.31. Let X have density

1
fX (x) = e−x/3 , for 0 < x,
3
and fX (x) = 0 otherwise.

a. Find the expected value of X 3 , i.e., E(X 3 ).

b. Find a general form, which handles all positive integers n, for E(X n ),
often called the nth moment of X.

Exercise 29.32. Show that, if X1 , . . . , Xn are independent random variables,

and a1 , . . . , an are constants, then

Var(a1 X1 + · · · + an Xn ) = a21 Var(X1 ) + · · · + a2n Var(Xn ).

Exercise 29.33. Suppose that X and Y have joint probability density function
fX,Y (x, y) = 16e−4x−4y for x > 0 and y > 0, and fX,Y (x, y) = 0 otherwise. Find
E(X + Y ).

Exercise 29.34. If the joint density of X and Y is 27 8

xy on the interior of the
triangle with corners at (0, 0), (3, 0), (0, 3) (and the joint density is 0 elsewhere),
find E(X).
Chapter 30

Review of Continuous Random

Variables

30.1 Summary of Continuous Random Variables

What can you do with a density?

• Remember that it is important to check that you have a valid density first.
• Find probabilities, e.g., P (1.7 ≤ X ≤ 3.2).
• Graph the density fX (x).
• Calculate expected value (average), variance, and standard deviation of
X and of functions of X.
• Calculate the CDF by integrating the density.

What can you do with a CDF?

• Find cumulative probabilities, e.g., FX (5.77) = P (X ≤ 5.77).

• Graph the CDF FX (x).
• Find the median and other percentiles.
• Calculate the density by taking the derivative of the CDF.

The formulas in this chapter were for continuous random variables only.
Earlier in the text, we discussed discrete random variable formulas which are
similar, but they did not use calculus.
How do you know if your density is valid?

• The density must account for every possible outcome. It must describe
everything that could happen from −∞ to ∞, although the density is
sometimes zero on large portions of the real line.

384
 30.1. Summary of Continuous Random Variables 385

• If you integrate the density from −∞ to ∞, the result is 1. (The area

under the density is 1.)
Z ∞
fX (x) dx = 1
−∞

How do you find the cumulative distribution function (CDF)?

Z a
FX (a) = P (X ≤ a) = fX (x) dx
−∞

How do you find the median?

Look at the CDF graph. The first x-value where the CDF is equal to 0.5 is
your median.
How do you find other percentiles?
To find the pth percentile, just find the x-value where FX (x) = P (X ≤
x) = p. This works for the median: the median is the value of x for which
FX (x) = P (X ≤ x) = 1/2. It works for other percentiles too. For instance, the
95th percentile is the x-value for which FX (x) = P (X ≤ x) = 0.95. (You find
the first x-value where the CDF is equal to 0.95.)
How do you get the mean/(weighted) average/expected value?
Z ∞
E(X) = xfX (x) dx
−∞
Z ∞
E(g(X)) = g(x)fX (x) dx
−∞
How do you get the variance and standard deviation?

Var(X) = E(X 2 ) − (E(X))2

p
σX = Var(X)
What do you do if you have a linear combination of X? Here a and b
are constants.

E(aX + b) = aE(X) + b
Var(aX + b) = a2 Var(X)
p p
σaX+b = a2 Var(X) = |a| Var(X)

What do you do if you have a sum of random variables and need

an expectation? (This works regardless of whether the variables are
independent or dependent.)
n
X  Xn
E Xi = E(Xi )
i=1 i=1
386 Chapter 30. Review of Continuous Random Variables

What if all of these random variables have the same distribution?

(Again, this works regardless of whether the variables are indepen-
dent or dependent.)
X n 
E Xi = nE(X1 )
i=1

What do you do if you need the variance of n independent random

variables?
n n
!
X X
Var Xi = Var(X)
i=1 i=1
v
u n
uX
σPni=1 Xi =t Var(X)
i=1

What if all of these random variables are independent and also have
the same distribution?
n
!
X
Var Xi = n Var(X)
i=1
p
σPni=1 Xi = n Var(X)

30.2 Exercises
Exercise 30.1. Using the following probability density function
(
2 if 0 ≤ x ≤ 1/2,
fX (x) =
0 otherwise,

a. Find the cumulative distribution function FX (x).

b. Find P (0.25 ≤ X ≤ 0.45).
c. Find P (X > 0.45).
d. Find P (X = 0.25).
e. Find the mean E(X).
f. Find the standard deviation σX .
g. Graph the density fX (x).
h. Graph the CDF FX (x). Identify the median on the graph of the CDF.

Exercise 30.2. Using the following probability density function,

(
1 −x/8
e if x > 0,
fX (x) = 8
0 otherwise,
 30.2. Exercises 387

a. Find the cumulative distribution function FX (x).

b. Find P (X > 4).
c. Find P (−2 ≤ X ≤ 12).
d. Find P (X < 240).
e. Find the mean E(X).
f. Find the standard deviation σX .
g. Graph the density fX (x).
h. Graph the CDF FX (x). Identify the median on the graph of the CDF.

Exercise 30.3. Using the following probability density function,

(
1
2 if x ≥ 1,
fX (x) = x
0 otherwise,

a. Find the cumulative distribution function FX (x).

b. Find P (X > 3).
c. Find P (−0.5 ≤ X < 5).
d. Find P (X < 4.2).
e. Find the mean E(X).
f. Graph the density fX (x).
g. Graph the CDF FX (x). Identify the 25th, 50th (median), and 75th
percentiles on the graph of the CDF.

Exercise 30.4. Using the following CDF,

if x < 2,

0

FX (x) = x−2
2 if 2 ≤ x ≤ 4,
if 4 < x,

1


find the corresponding density fX (x).

Exercise 30.5. What is the constant k that makes the following function a
valid density? (
k(e−x + e−4x ) if 0 ≤ x,
fX (x) =
0 otherwise,

Exercise 30.6. What is the constant k that makes the following function a
valid density? (
kx2 (1 − x)7 if 0 ≤ x ≤ 1,
fX (x) =
0 otherwise,
388 Chapter 30. Review of Continuous Random Variables

Exercise 30.7. Which of the following can be true? If an answer is false, state
why it is false.
a. A CDF FX (x) can have the value 4.3.
b. A density fX (x) can have the value 4.3.
c. A mass pX (x) can have the value 4.3.
Exercise 30.8. Which of the following can be true? If an answer is false, state
why it is false.
a. A CDF FX (x) can be 1 for two or more values of x.
b. A density fX (x) can be 1 for two or more values of x.
c. A mass pX (x) can be 1 for two or more values of x.
Exercise 30.9. Which of the following can be true? If an answer is false, state
why it is false.
a. The (indefinite) integral of FX (x) is fX (x).
b. The (indefinite) integral of fX (x) is FX (x).
c. The area under the curve of fX (x) between −∞ and x is FX (x).
d. The area under the curve of FX (x) between −∞ and x is fX (x).
e. The derivative of FX (x) is fX (x).
f. The derivative of fX (x) is FX (x).
Exercise 30.10. Which of the following can be true? If an answer is false,
state why it is false.
a. The area under the FX (x) curve from −∞ to ∞ is 1.
b. The area under the fX (x) curve from −∞ to ∞ is 1.
Exercise 30.11. Which of the following can be true? If an answer is false,
state why it is false.
a. The CDF FX (x) can be negative.
b. The density fX (x) can be negative.
c. The mass pX (x) can be negative.
Exercise 30.12. Which of the following can be true? If an answer is false,
state why it is false.
a. The graph of fX (x) can have jumps (i.e., can be discontinuous).
b. The graph of FX (x) for discrete random variables can have jumps.
c. The graph of FX (x) for continuous random variables can have jumps.
d. The graph of pX (x) can have jumps.
Part VI

Named Continuous Random

Variables

For discrete distributions, you learned the rules for general distributions and
then about some of the most widely used types of random variables, to make
your life easier in scenarios that match specific situations (Binomial, Geomet-
ric, etc.). For continuous distributions, you have already learned the rules for
general distributions during the previous part of the book. Now in the chapters
that follow, we will show you some formulas for scenarios matching very com-
mon types of situations, and the named continuous random variables that are
appropriate in these situations. For instance, we dedicate several chapters to
the Normal distribution, perhaps the most important continuous distribution
for probability and statistics. At the end of this part of the book, we will review
some strategies for comparing and contrasting all of these distributions, to help
you pick the correct distribution for your situation.
There are many other continuous distributions than the five that we cover
here, but these are the most common ones.
By the end of this part of the book, you should be able to:

1. Distinguish between Continuous Uniform, Exponential, Gamma, Beta,

and Normal distributions when reading a story.

2. Identify the variable and the parameters in a story, and state in English
what the variable and its parameters mean.

3. Use the formulas for the density, CDF, expected value, and variance to
answer questions and find probabilities.

4. Utilize joint densities to study problems about two or more random vari-
ables at the same time.

5. Use all discrete and continuous named distributions where appropriate.

389
390 Part VI. Named Continuous Random Variables

Math skills you will need: ex , derivatives, integrals (in particular, integration
by parts), and the Gamma function.
Additional resources: Computer programs (such as Excel, Maple, Mathe-
matica, Matlab, Minitab, R, SPSS, etc.) and calculators may be used to assist
in the calculations. You will also need the standard Normal table that appears
in Chapter 35 and also appears in the back of the book.
Chapter 31

Continuous Uniform Random

Variables

We make chance a matter of absolute calculation. We subject the unlooked for

and unimagined, to the mathematical formulæ of the schools.
—“The Mystery of Marie Rogêt” by Edgar Allan Poe (Snowden’s Ladies’
Companion, 1842 and 1843)

If you want to board a bus at a random time, and the bus circles campus once
every 30 minutes, how long do you expect to wait until the bus arrives? What
is the probability that the bus will come within the next 5 minutes? What is
the variance of the time until the next bus comes? What is the probability that
you just missed the bus by 1 minute or less?

31.1 Introduction
For the next few chapters, we focus on some very specific continuous random
variables. The most simple type of continuous random variable is called a
Continuous Uniform random variable, often just called a Uniform. We say that
X is Uniform anytime that the density of X is constant on some interval. In
such a case, the density of a Uniform random variable must be equal to 1 divided
by the length of the interval. The reason is that, when the density (which is
constant) is integrated over the length (or area, or volume, in the case of two
or three dimensions) of the region, the integral needs to evaluate to 1.
One of the nicest properties of Uniform random variables is that, since the
density is constant, the integrals for finding various probabilities are especially
easy, because we can essentially avoid having to perform an integration. We
can just multiply the constant (i.e., the density) by the length over which we

391
392 Chapter 31. Continuous Uniform Random Variables

are “integrating,” and then we obtain the desired probability, without actually
having to use any complicated integrating techniques.

The name “Continuous Uniform” (often abbreviated simply as “Uniform”)

is reminiscent of the Discrete Uniform random variables from Chapter 20. A
Discrete Uniform random variable has a mass that is constant on a collection of
points, just as a continuous random variable has a density that is constant on
some interval. (Similarly, a pair of jointly distributed Uniform random variables
has a density that is constant on some two-dimensional region.)

Continuous Uniform random variables

The common thread: Constant density. If a random variable has a
density that is constant on a region, the variable is Uniform. This means
that, for instance, the random variable is equally likely to be located in any
regions that have the same size. If the location X of a hidden object is is
Uniform on [0, 10], then X is equally likely to be found in [2.2, 3.4], or [0, 1.2],
or [8.74, 9.94], etc., etc., because all of these regions have length 1.2.
Things to look for: Constant density on an interval (or a pair of random
variables with constant joint density on a two-dimensional region).
The variable:
X = an exact position or arrival time
(for example: your exact arrival time, in minutes, if we know that you will
arrive sometime between 1:00 and 1:15 pm). There are no general limits to
what values X can take, but a specific range will be defined for each problem.
The parameters: a and b, the endpoints of the interval [a, b] or (a, b) where
the density of X is nonzero. Sometimes a pair of Continuous Uniform random
variables is defined on a two-dimensional region.
Density: (
1
b−a , a ≤ x ≤ b,
fX (x) =
0 otherwise

CDF: 
0
 x < a,
x−a
FX (x) = , a ≤ x ≤ b,
 b−a
1 b<x


Expected value formula:

E(X) = (a + b)/2

Variance formula:
Var(X) = (b − a)2/12
 31.2. Examples 393

The notation for a Uniform random variable on the interval (a, b) looks like:
X ∼ C.Uniform(a, b).
Where does the cumulative distribution function formula come from?
If we integrate the probability density function formula from −∞ to c, where
a ≤ c ≤ b, we get
Z c Z c
1 c−a
FX (c) = fX (x) dx = dx = .
−∞ a b − a b−a
Thus FX (x) = x−a
b−a for a ≤ x ≤ b.
However, it is also useful to think about the cumulative distribution function
as the area under the curve for x in the range [a, c].
Where does the expected value formula come from?
The expected value E(X) for the Uniform random variable is just the mid-
point of the range where the density fX (x) is not 0. (This is not true for other
types of continuous random variables.) Since the Continuous Uniform distribu-
tion is symmetric, the median (center) and the mean (expected value) are the
same.

31.2 Examples

Example 31.1. If X is a Continuous Uniform random variable on the interval

[a, b], then the expected value of X is E(X) = (a + b)/2, as shown in Theo-
rem 28.6. The expected value of X 2 is:
Z b
2 1
E(X ) = x2 dx
a b−a
b
x3 1
=
3 b−a x=a
b3 − a3 1
=
3 b−a
(b − a)(a2 + ab + b2 ) 1
=
3 b−a
2
a + ab + b 2
=
3
So the variance of X is:
a2 + ab + b2 a + b 2 4a2 + 4ab + 4b2 3a2 + 6ab + 3b2
 
− = −
3 2 12 12
2
a − 2ab + b2
=
12
(b − a)2
=
12
394 Chapter 31. Continuous Uniform Random Variables

Example 31.2. Sasha is a clerk in a convenience store. She needs to go to the

bathroom and leaves her counter unattended at 1:02 PM. When she returns at
1:08 PM, there is an impatient customer waiting at the counter.

a. Why is this a Continuous Uniform distribution situation? What are the

parameters? What is X?
The customer was equally likely to have arrived anytime in the continuous
range between 1:02 and 1:08. The parameters are a = 2 and b = 8. The random
variable X is the customer’s actual arrival time.
b. Show the graph of the probability density function for the impatient
customer’s arrival time.
The formula for the density is
(
1/6 if 2 ≤ x ≤ 8,
fX (x) =
0 otherwise

See the left side of Figure 31.1.

1 FX (x)

0.5
1/3 fX (x)
1/6
x x
2 4 6 8 10 2 4 6 8 10

Figure 31.1: Left: The density fX (x) = 1/6. Right: The CDF FX (x) = 6 ,
x−2

of Uniform random variable on [2, 8].

c. What is the expected time the customer arrived?

2+8
E(X) = = 5, which is 1:05 PM.
2

d. What is the standard deviation for the time the customer arrived?
The variance is
(8 − 2)2
Var(X) = = 3,
12
 31.2. Examples 395

so the standard deviation is

√
σX = 3 = 1.7321 minutes.

e. What is the probability the customer arrived within the last 2 minutes
(i.e., between 1:06 PM and 1:08 PM)?
We are looking for P (6 ≤ X ≤ 8). You could do this problem three different
ways.
Method #1. For the Continuous Uniform distribution X, the probability of
X being in some range is equal to the length of the range divided by the entire
length of the interval where X is defined. This is true in general, for Continuous
random variables, because the density we are integrating (to obtain the proba-
length of [6, 8]
bilities) is a constant function. Thus P (6 ≤ X ≤ 8) = = 1/3.
length of [2, 8]
(Note: If some of the desired interval falls outside the range where X is
defined, that part of the range must be ignored. For example, consider P (6 ≤
X ≤ 10). We know that X is between 2 and 8, so the portion from 8 to 10 can
be ignored. So, P (6 ≤ X ≤ 10) = P (6 ≤ X ≤ 8) = 2/6 = 1/3.)
Method #2. If you use the probability density function, you could integrate
Z 8
1
P (6 ≤ X ≤ 8) = = 2/6 = 1/3.
6 6

Method #3. If you know the cumulative distribution function, you could
just use

P (6 ≤ X ≤ 8) = P (X ≤ 8) − P (X ≤ 6) = FX (8) − FX (6).

See the right side of Figure 31.1 for a plot of the cumulative distribution func-
tion.
f. What is the cumulative distribution function? Also show the graph.
The CDF is 
0
 x ≤ 2,
x−2
FX (x) = 2 ≤ x ≤ 8,
 8−2
1 8 ≤ x,


The CDF is shown on the right in Figure 31.1.

g. Given that the impatient customer did not arrive in the first 2 minutes,
what is the probability that he arrived in the last 2 minutes of this interval?
This is a conditional probability. When working with Uniform random vari-
ables, if a condition removes some of the interval, all of the remaining condi-
tional probabilities are still Uniform on the remaining interval. E.g., in this
example, we have the condition X ≥ 4. So the portion [2, 4] of the original
396 Chapter 31. Continuous Uniform Random Variables

interval is removed. What remains is the rest of the interval, i.e., [4, 8]. Thus,
the conditional density is constant on the interval [4, 8]. So

length of [6, 8]
P (X > 6 | X > 4) = = 2/4 = 1/2.
length of [4, 8]

Remark 31.3. Conditional probabilities and Continuous Uniform

Random Variable
Consider X which is Uniform on [a, b]. If we are given that X > c, for some
c in the interval [a, b], then all of the resulting conditional probabilities are
Uniform on [c, b].
Similarly, if we are given X < c (for c in [a, b]), then all of the resulting
conditional probabilities are Uniform on [a, c].

This calculation can also be written as

P (X > 6 and X > 4) P (X > 6) 1 − FX (6)
P (X > 6 | X > 4) = = = .
P (X > 4) P (X > 4) 1 − FX (4)

Therefore, P (X > 6 | X > 4) == 1−(6−2)/6

1−(4−2)/6 = 1/2. As another way to compute
this probability, we can consider a new variable Y that is Uniform on [4, 8]. The
density of Y is (
1/4 4 ≤ y ≤ 8
fY (y) =
0 otherwise
This density is shown in Figure 31.2. The desired probability is P (Y > 6) =
2/4 = 1/2.

0.4 fX (x)

0.2
x
2 4 6 8 10

Figure 31.2: Density fY (y) = 1/4 of Uniform random variable Y on [4, 8].

Example 31.4. If a burglary happens in an apartment, the costs to the renter

from the theft, T , can be modeled by a Continuous Uniform random variable
ranging from no charge up to $4000.
 31.2. Examples 397

a. What is the probability distribution function for the cost of damage?

(
1
if 0 ≤ x ≤ 4000,
fT (x) = 4000
0 otherwise

b. What is the expected cost of damage?

0 + 4000
E(T ) = = $2000.
2

c. What is the standard deviation in the cost of damage?

(4000 − 0)2
Var(T ) = = 1,333,333,
12
p
σT = 1,333,333 = $1154.70.
d. An insurance company issues a renter’s policy with a deductible of $800.
This means that if there is a claim the insurance company will reimburse the
renter all damage costs between $800 and $4000, but the renter is responsible
for the first $800 in costs.
What is the expected out-of-pocket cost C to the renter if damage occurs?
We observe that T = C + M , where M is the amount that the insurance
company pays. The out-of-pocket cost C is just a function of T :
(
T if 0 ≤ T ≤ 800,
C=
800 if 800 < T ≤ 4000

Since C is just a function of T , then we can compute E(C) by integrating the

density of T times the cost:
Z 800 Z 4000
1 1
E(C) = x dx + 800 dx
0 4000 800 4000
800
x2 1 1
= + (3200)800
2 4000 x=0 4000
= 80 + 640
= $720.00

e. How much should the insurance company expect to pay the renter if damage
occurs?
Remember that the insurance company only pays if the damage is above
$800. Again, use M to denote the amount that the insurance company pays.
Then M is just a function of T :
(
0 if 0 ≤ T ≤ 800,
M=
T − 800 if 800 ≤ T ≤ 4000
398 Chapter 31. Continuous Uniform Random Variables

As before, since M is just a function of T , then we can compute E(M ) by

integrating the density of T times the amount M that the insurance company
pays:
Z 800 Z 4000
1 1
E(M ) = 0 dx + (x − 800) dx
0 4000 800 4000
 2  4000
x 1
= − 800x
2 4000 x=800
= $1280.00

Notice that the amount the insurance company expects to pay,

E(M ) = $1280.00,

added to the amount the renter expects to pay

E(C) = $720.00,

is equal to the expected total amount of damage

E(T ) = $2000.00

Now we return to the method of using the lengths of intervals, as discussed

in Example 31.2.
If X is Uniformly distributed on the interval [a, b], then

1
fX (x) = for a ≤ x ≤ b,
b−a
and fX (x) = 0 otherwise. So we can use the lengths of intervals, instead of
having to integrate.

Remark 31.5. Using lengths of intervals to calculate probabilities,

with Continuous Uniform random variables
If C is a subinterval of [a, b], then

length of C length of C
Z
1
P (X ∈ C) = dx = = .
C b−a b−a length of [a, b]

Similar concepts work in higher dimensions. When X and Y have a joint

density that is constant, we say that X and Y have a jointly Uniform distribu-
tion on a two-dimensional region. The joint density must be 1 divided by the
area of the region, so that the integral over the region is 1.
 31.2. Examples 399

Example 31.6. Let X and Y denote the latitude and longitude of a person
who is thought to be lost in a certain part of a forest. If we assume that the
densities of X and Y are each constant, say, 2 ≤ X ≤ 8.2 and 3.01 ≤ Y ≤ 6.3,
then
1
fX (x) = for 2 ≤ x ≤ 8.2,
8.2 − 2
and fX (x) = 0 otherwise, and
1
fY (y) = for 3.01 ≤ y ≤ 6.3,
6.3 − 3.01
and fY (y) = 0 otherwise. If, moreover, we assume that X and Y are indepen-
dent, then the joint density of X and Y is constant too:
fX,Y (x, y) = fX (x)fY (y)
  
1 1
= for 2 ≤ x ≤ 8.2 and 3.01 ≤ y ≤ 6.3,
8.2 − 2 6.3 − 3.01
and fX,Y (x, y) = 0 otherwise.

More generally, if X is Uniform on [a, b] and Y is Uniform on [c, d], and if

X and Y are independent, then
  
1 1
fX,Y (x, y) = for a ≤ x ≤ b and c ≤ y ≤ d,
b−a d−c
and fX,Y (x, y) = 0 otherwise.
We can also have X and Y Uniform on non-rectangular regions too.

Example 31.7. For instance, if X and Y are Uniform on a circular dartboard

of radius 2, then
fX,Y (x, y) = 1/(4π) for x, y on the dartboard,
and fX,Y (x, y) = 0 otherwise.
The reason that fX,Y (x, y) = 1/(4π) on the dartboard is that—in general—the
density of a Uniform random variable defined on a region is equal to 1 divided
by the length, area, volume, etc., of the region. That way, when integrating
the density (which is constant) over the length, area, volume, of the region,
the integral will evaluate to 1. In this case, the circular-shaped dartboard has
area 4π, so the density must be fX,Y (x, y) = 1/(4π) on the circle.
If, for instance, D represents the distance from the location of the dart to
the center of the board, then
area of circle of radius a πa2
FD (a) = P (D ≤ a) = = = a2/4,
area of circle of radius 2 π22
for 0 ≤ D ≤ 2. We can use such a cumulative distribution function to find out,
for instance, about the density of D, or the expected value of D, etc.
400 Chapter 31. Continuous Uniform Random Variables

Example 31.8. When a joint density is constant, then the conditional densities
(e.g., if the value of one random variable is given) are uniform too. For instance,
in Example 31.7, if X = 0 is given, then the conditional density of Y is uniform
on the interval [−2, 2], i.e.,

fY |X (y | 0) = 1/4 for −2 ≤ y ≤ 2.

As another example with the dartboard, if√Y =√1 is given, then the conditional
density of X is uniform on the interval [− 3, 3], i.e.,
1 √ √
fX|Y (x | 1) = √ for − 3 ≤ y ≤ 3.
2 3
For some other examples of this phenomenon, refer to Example 27.2 (where
fX|Y (x | 2) = 1/8 for 0 ≤ x ≤ 8) and Example 27.3 (where fX|Y (x | 2) = 2√
1
5
√ √
for − 5 ≤ x ≤ 5).

Remark 31.9. Constant joint densities and conditional distributions

If X, Y have a constant joint density (i.e., a joint Uniform distribution), then
the conditional densities are constant (i.e., Uniform) too.

Example 31.10. Now consider two students who each choose—independently,

and Uniformly—numbers between 0 and 10.

Let X the be first student’s choice, and let Y be the second student’s choice.
Since X and Y are independent, we can multiply to get the joint density of X
and Y :
fX,Y (x, y) = 1/100 for 0 ≤ x ≤ 10, 0 ≤ y ≤ 10,
and fX,Y (x, y) = 0 otherwise.
Suppose that we want to find the expected value of the minimum of the two
numbers.
Method #1: To find the expected value, we can integrate
Z 10 Z 10
1
E(min(X, Y )) = min(x, y) dy dx
0 0 100
It is difficult to know whether x or y is the minimum when we integrate over
all x’s and y’s at once, so we suggest breaking the integral into two pieces,
depending on whether Y or X is actually the minimum. In Figure 31.3, we
show these two regions.
 31.2. Examples 401

y
10
min(X, Y ) = X
9
i.e., X is the minimum
8
7
6
5
4
3
2 min(X, Y ) = Y
1 i.e., Y is the minimum
x
1 2 3 4 5 6 7 8 9 10

Figure 31.3: The regions where min(X, Y ) = X, and where min(X, Y ) = Y .

Now we integrate min(X, Y )fX,Y (x, y) over the entire square, shown in Fig-
ure 31.3. When we are in the lower triangle, where min(X, Y ) = Y , then

min(X, Y )fX,Y (x, y) = (y)(1/100).

When we are in the upper triangle, where min(X, Y ) = X, then

min(X, Y )fX,Y (x, y) = (x)(1/100).

We integrate over the range where Y is the min and then over the range where
X is the min:
Z 10 Z x Z 10 Z 10
1 1
E(min(X, Y )) = min(x, y) dy dx + min(x, y) dy dx
0 0 100 0 x 100
and we get
Z 10 Z x Z 10 Z 10
1 1
E(min(X, Y )) = y dy dx + x dy dx
0 0 100 0 x 100
If we wanted to, we could switch the bounds of integration on the second inte-
gral, so that we integrate over x on the inside and over y on the outside. The
reasons will be clear in a brief moment:
Z 10 Z x Z 10 Z y
1 1
E(min(X, Y )) = y dy dx + x dx dy
0 0 100 0 0 100
Finally, we see that we have essentially just written the same integral twice,
just swapping letters the second time, so we can save ourselves some time by
402 Chapter 31. Continuous Uniform Random Variables

just writing, more simply:

Z 10 Z x
1
E(min(X, Y )) = 2 y dy dx
0 0 100

Finally, we solve the integral

10 x
y2 1
Z
E(min(X, Y )) = 2 dx
0 2 100 y=0
10
x2 1
Z
=2 dx
0 2 100
10
x3 1
=2
6 100 x=0
= 10/3

Method #2: Another possible method is to always write U as the smaller of the
two values X and Y , and write V as the larger of the two values X and Y . In
other words,
U = min(X, Y ),

and
V = max(X, Y ).

Then U and V must have constant density on the triangle where 0 ≤ U ≤ V ≤

10, since this joint density is induced by the constant joint density for X and
Y . So the density must be 1/50 because the density must integrate to 1 over
the whole region (which has area 50). In other words,

fU,V (u, v) = 1/50 for 0 ≤ u ≤ v ≤ 10,

and fU,V (u, v) = 0 otherwise. This triangle is shown in Figure 31.4. So the

v
10
8
6
4
2
u
2 4 6 8 10

Figure 31.4: The region where 0 ≤ U ≤ V ≤ 10.

 31.3. Linear Scaling of a Uniform Random Variable 403

integral for the expected value becomes:

E(min(X, Y )) = E(U )
Z 10 Z v  
1
= (u) du dv
0 0 50
Z 10  2    v
u 1
= dv
0 2 50 u=0
Z 10  2   
v 1
= dv
0 2 50
 3    10
v 1
=
6 50 v=0
= 10/3,

which agrees with our result from the first method.

31.3 Linear Scaling of a Uniform Random Variable

Theorem 31.11. Linear scaling of a Uniform random variable
If X is a Uniform random variable on the interval [s, t], and if a, b are constants,
then aX + b is a Uniform random variable too.
If a > 0, then aX + b is Uniform on the interval [as + b, at + b].
If a < 0, then the interval is reversed, i.e., aX + b is Uniform on the interval
[at + b, as + b].

To see that this is true, we show the result for a > 0. Since s < X < t, then
as + b < aX + b < at + b. So if we write Y = aX + b, we see that fY (y) = 0 for
y outside the interval [as + b, at + b]. For y inside the interval [as + b, at + b],
we compute
y−b
−s
 
y−b a
FY (y) = P (Y ≤ y) = P (aX + b ≤ y) = P X≤ = ,
a t−s

and thus
d 1/a 1
fY (y) = FY (y) = = .
dy t−s a(t − s)
So the density of Y is constant on the interval [as + b, at + b] and is 0 otherwise.
Thus Y is a Uniform random variable on the interval [as + b, at + b].
404 Chapter 31. Continuous Uniform Random Variables

Example 31.12. With this information in hand, we return to Exercise 29.23

and calculate much more than just the expected value of the purchase price:
Geoffrey does not like to be low on gas, so he randomly stops to fill up
his tank. He has a 14-gallon tank, and the current price of gas is $2.75 per
gallon. Whenever he stops to buy gas, he also buy a candy bar for $1.30. If X
is the amount of gas (in gallons) in his tank when he stops for a purchase, then
fX (x) = 1/14 for 0 ≤ x ≤ 14, and fX (x) = 0 otherwise. He always fills the
tank, so he will always buy 14 − X gallons.

The price of the purchase is

Y = (2.75)(14 − X) + 1.30 = 39.80 − 2.75X.

Since X is Uniform on [0, 14], it follows that Y is Uniform on the interval

[39.80 − (2.75)(14), 39.80 − (2.75)(0)] = [1.30, 39.80].

Thus, we confirm the fact that the expected value of the purchase price is

39.80 + 1.30
E(Y ) = = 20.55.
2
We know much more now; for instance, we know that the variance of Y is

(39.80 − 1.30)2
Var(Y ) = = 123.5208333.
12
and we know that the density of Y is

1 1
fY (y) = = for 1.30 ≤ y ≤ 39.80,
39.80 − 1.30 38.50

and fY (y) = 0 otherwise.

31.4 Exercises
31.4.1 Practice
Exercise 31.1. Rope checks. In a certain manufacturing process, an auto-
mated quality control computer checks 10 yards of rope at a time. If no defects
are detected in that 10-yard section, that portion of the rope is passed on. How-
ever, if there is a defect detected, a person will have to check the rope over more
carefully to determine where (measured from the left side in yards) the defect
is. If exactly 1 defect is detected in a rope section, we would like to find the
probabilities for its location.
 31.4. Exercises 405

a. Why is this a Continuous Uniform problem?

b. What does X represent in this scenario?
c. What are the parameters in this scenario?
d. What is the expected value for the location of the defect?
e. What is the standard deviation?
f. What is the probability density function? Write it in function notation,
and also graph it.
g. What is the cumulative distribution function? Write it in function nota-
tion, and also graph it.
h. Find P (X > 8).
i. Find P (2.3 ≤ X ≤ 5.2)
j. Find P (X < 2|X < 5)

Exercise 31.2. Harry Potter and the Missed Exit. You are driving down
an interstate when you suddenly realize you have missed your exit because
you were busy listening to an exciting chapter of a Harry Potter book on CD.
According to the CD case, the chapter lasts for 14 minutes. Assume you are
driving 60 miles per hour (in other words, 1 mile per minute). You could have
passed your exit anytime over the past 14 minutes. You want to know about
the probabilities for where it was exactly (how far back in miles) as you turn
around to head back to find it.

a. Why is this a Continuous Uniform problem?

b. What does X represent in this scenario?
c. What are the parameters in this scenario?
d. What is the expected time it will take for you to find the missed exit?
e. What is the standard deviation?
f. What is the probability density function? Write it in function notation,
and also graph it.
g. What is the cumulative distribution function? Write it in function nota-
tion, and also graph it.
h. Find the probability that the exit was passed within a mile of the half-way
point.
i. Find the probability that the exit was passed in the first minute of the
chapter.
j. You are pretty sure that you were paying attention to the road until the
last 4 minutes of the chapter when Voldemort tried to attack Harry, so within
406 Chapter 31. Continuous Uniform Random Variables

that 4-minute interval, what is the probability that you missed the exit within
the last 30 seconds?
k. How many total miles do you expect to have to drive out of your way to
get back to the missed exit (miles past the exit ×2)?
l. Also find the standard deviation in the total miles out of your way that
you had to drive.
m. If you get reimbursed by your company 43 cents a mile for your trip plus
the $3.20 you spent on a grande mocha at Starbucks (conveniently located at
the turn-around spot) to wake yourself up as you get turned around, how much
is the expected amount of money this detour will cost the company?
n. What is the standard deviation in this amount of money that you get
reimbursed in part m?

Exercise 31.3. Network latency. The network latency per request from your
computer to a server is Uniform between 30 ms and 150 ms.

a. What is the probability that a particular request is 60 ms or less?

b. If your computer makes 320 requests to the server, how many are expected
to have a latency that is 60 ms or less?

Exercise 31.4. A fly is flying around. You believe that there is a fly some-
where less than 6 feet away from you. If you believe that he is located Uniformly
in a circle of radius 6 feet away from you, what is the probability that he is more
than 2 feet away from you?

Exercise 31.5. Bird on a wire. A bird lands at a location that is Uniformly

distributed along an electrical wire of length 150 feet. The wire is stretched
tightly between two poles. What is the probability that the bird is 20 feet or
less from one or the other of the poles?

Exercise 31.6. Touchdown. A quarterback needs to throw the ball quickly,

and in his haste, the location it lands is Uniformly distributed within a 30 × 120
sq. ft. area in the endzone. The receiver can only catch the ball the ball lands
within 7 feet of him, i.e., within a circle of radius 7 feet, completely contained
in the endzone. What is the probability that the receiver catches the ball?
(Assume that the receiver is positioned so that a circle of 7 feet around him is
completely contained within the endzone.)

Exercise 31.7. Target art. A wall in a room is 108 inches tall and 132 inches
wide. There is a painting on the wall that is 18 inches by 24 inches. If a
tennis ball is accidentally flung at the wall, and the location where it lands is
Uniformly distributed on the wall, what is the probability that the tennis ball
hits the painting?

Exercise 31.8. Solar cell. A solar cell measures 10 inches by 10 inches, and it
has 4 non-overlapping regions that generate electricity. Each of these regions is
 31.4. Exercises 407

4 inches by 4 inches in size (there is a border between the electricity-generating

regions). What is the probability that a photon that hits the solar cell at a
location which is Uniformly distributed on the solar cell will actually hit one of
these electricity-generating regions?

31.4.2 Extensions
Exercise 31.9. Cab ride. Suppose that the length X of a randomly selected
passenger’s trip in a cab is Uniformly distributed between 5 and 30 miles. The
charge induced for such a trip, in dollars is Y = 2.50X + 3.00.

a. How much should a randomly selected customer expect to pay?

b. What is the standard deviation of the amount that a randomly selected
customer pays?

Exercise 31.10. There’s a fly in the house! Consider the two-dimensional

house in Figure 31.5. Suppose that a fly is found somewhere in the house, with
a location that is Uniformly distributed.

a. Find the conditional density of Y given that X = 2.

b. Find the conditional probability that the fly is upstairs, i.e., Y > 10,
given that X = 2, i.e., find P (Y > 10 | X = 2).

y
20

15

10

x
5 10

Figure 31.5: Picture of the house.

Exercise 31.11. There’s still a fly in the house! Consider the two-
dimensional house in Figure 31.5. Suppose that a fly is still found somewhere
in the house, with joint density

fX,Y (x, y) = 1/150 for x, y in the house,

and fX,Y (x, y) = 0 otherwise. (We no longer assume X = 2.)

408 Chapter 31. Continuous Uniform Random Variables

a. Find E(X).
b. Find Var(X).
c. Find E(Y ).
d. Find Var(Y ).
Exercise 31.12. Long jump. Suppose that a particular long jumper assumes
that each of his jumps are Uniformly distributed between 6.5 and 7.2 meters.
He is happy whenever he jumps 7 meters or more. If he makes 10 such jumps,
what is the probability that he is happy with exactly 4 of his jumps?
Exercise 31.13. Student arrivals. Five students will arrive to the classroom
during the next 2 minutes. Assume that their arrival times are independent,
and each arrival time is Uniformly distributed between 0 and 2 minutes. What
is the probability that exactly two of the students arrive during the next 30
seconds?
Exercise 31.14. Manufactured cubes. A machine manufactures cubes with
a side length which varies Uniformly over the interval [0.2, 0.3] in millimeters.
For the following problems, make sure you use the correct units. (Assume the
sides of the base and the height are all the same.)
a. What is the expected side length?
b. What is the standard deviation of the side length?
c. What is the expected area of one of the square bases?
d. What is the standard deviation of one of the square bases?
e. What is the expected volume of one of the cubes?
f. What is the standard deviation of the volume of one of the cubes?
In the rest of the problem, assume that the cost to make the cubes is 12
cents per cubic millimeter and 6 cents for the general cost (labor, electricity,
etc.) per cube.
f. What is the expected cost for making 1 cube?
g. What is the variance in the cost for making 1 cube?
h. What is the expected cost for making 10 cubes?
i. What is the variance in the cost for making 10 cubes?
Exercise 31.15. Dartboard. Kelly throws a dart at a circular dartboard of
radius 3 feet. Let X and Y denote the location where the dart lands. Assume
that −3 ≤ X ≤ 3 and −3 ≤ Y ≤ 3 and X 2 + Y 2 ≤ 9, i.e., the dart lands on
the dartboard. Moreover, assume that the dart’s location is Uniform on the
dartboard, i.e.,

fX,Y (x, y) = 1/(9π) if x, y are on the dartboard, i.e., x2 + y 2 ≤ 9,

 31.4. Exercises 409

and fX,Y = 0 otherwise.

√
Let D = X 2 + Y 2 be the distance from the dart to the center of the
dartboard. Find E(D).
Exercise 31.16. Bugs on a wire. An automatic camera is taking pictures of
insects that land on a wire. The distance between each insect and the camera (at
the time of the picture) is random, with constant density on the interval [1, 6],
measured in feet. (Assume that the locations of the insects are independent.)
A “very good” picture is taken when an insect is less than 3 feet away. What is
the probability that none of the camera’s 3 attempted pictures are “very good”?
Exercise 31.17. More bugs on a wire. For the scenario in Exercise 31.16,
if X, Y, Z are the distances of the camera from the insect for his first, second,
and third pictures, respectively, then find the expected value of the smallest of
these three distances, i.e., find E(min(X, Y, Z)).
Exercise 31.18. Laundry cycles. You are doing two loads of laundry: one
was just put in the washer, the other was simultaneously put in the dryer.
You know it takes the dryer between 25 and 40 minutes to completely dry
your clothes, and the washer takes between 27 and 37 minutes to complete its
cycle. Those distributions are independent and Uniform on their respective time
intervals. What is the probability the dryer is already done if the washer just
finished?
Exercise 31.19. Pencil darts. In a game of chance, a circle of radius 3 inches
is drawn on a piece of paper that is 8.5 × 11 sq. in. While blindfolded, the
student tries to place her pencil tip inside the circle. She wins $3 if she is
successful, or loses $1 if unsuccessful. (If she misses the paper altogether, she
can try again until she hits the paper.) Assume that when she hits the paper,
the location of her pencil tip is Uniformly distributed on the paper’s surface.
What are her expected winnings/loses in this game?
Exercise 31.20. Clay pots. Emmanuel makes pots out of clay. The clay
costs $5.00 per pound. Each pot is made from a random amount X of clay
(in pounds), Uniformly distributed between 1.9 and 2.7 pounds. There is also
a charge to bake the pot in a kiln (a kiln is a big oven for making clay get
very hard). The charge to use the kiln to harden a pot is $0.40. What is the
probability that he spends more than $12.00 altogether (cost of clay plus cost
of using the kiln) to make a pot of clay?
Exercise 31.21. Lost toy. A child is playing in a sandbox and loses a very
small toy. If the sand box is a square the measures 5 feet by 5 feet, what is the
probability that the toy is actually located in a right triangle at the southwest
corner of the sandbox, with sides measuring 1.5 feet by 2.5 feet?
Exercise 31.22. Active child. Suppose that a piece of chalk is broken into
two pieces, with the breaking point Uniformly distributed along the length of
the chalk, which is 8 cm. What is the probability that, as a result, one or the
other of the pieces is shorter than 1 cm?
410 Chapter 31. Continuous Uniform Random Variables

Exercise 31.23. Mole rat. The location of a mole rat is Uniform inside a
circular enclosure that has diameter 40 feet. What is the probability that the
mole rat is within 2 feet from the edge of the enclosure?

Exercise 31.24. Broken chalk. Assume a mother’s child has a random lo-
cation that is Uniformly distributed across the 80 foot by 120 foot playground
shown in Figure 31.6 below.

a. What is the probability that the child is in the grass?

b. On the swings?
c. On the slides?
y
80
70 grass

50
playscope
slides swings
30
picnic
20
tables bench
10
x
10 40 55 75 100 120

Figure 31.6: A playground.

Exercise 31.25. Crayon factory. In a crayon factory, wax is rolled into

cylinders, each of which are exactly 3.6 inches long, but the radius (in inches) is
a Uniform random variable X on the interval [0.15, 0.17]. Find the probability
that the volume of a crayon exceeds 0.30 cubic inches. Hint: The volume of a
cylinder is πr2 ` where r is the radius and ` is the length.

Exercise 31.26. Let X, Y, Z be independent and uniformly distributed on the

interval [0, 10]. Find the probability that Y is the middle value, i.e., find P (X <
Y < Z or Z < Y < X).

Exercise 31.27. Suppose X, Y have constant joint density on the triangle with
vertices at (0, 0), (3, 0), and (0, 3). Find E(X).

31.4.3 Advanced
Exercise 31.28. A fly near the wall. There is a fly randomly located Uni-
formly in a room that is 10 feet high, 14 feet long, and 13 feet wide. What is
the probability that the fly is within 1 foot of the walls, ceiling, or floor?
 31.4. Exercises 411

Exercise 31.29. Let X1 , X2 , X3 be independent continuous random variables,

each Uniformly distributed in the interval [0, 10]. Let Y denote the middle of
the three values. Find the cumulative distribution function FY (a) = P (Y ≤ a)
of the random variable Y .

Exercise 31.30. If X has a Continuous Uniform distribution on the interval

(1, 10), find E(ln X).
Chapter 32

Exponential Random Variables

Waiting is painful. Forgetting is painful. But not knowing which to do is the

worst kind of suffering.
—By the River Piedra I Sat Down and Wept by Paulo Coelho (Harper, 1994)

What is the expected time until your best friend sends you a text message?
When a mother is waiting for her three children to call her, what is the proba-
bility that the first call will arrive within the next 5 minutes?

32.1 Introduction
Exponential random variables are often waiting times for the next event to
happen. For instance, the time until the telephone rings, or a guest arrives, or
black car passes, etc., are all modeled by Exponential random variables. We
saw examples of Exponential random variables, for example, in Example 25.5
and Example 28.8. Exponential random variables are always positive. (It would
not make sense, for instance, for a waiting time to be negative.) Exponential
random variables, as we will see, also have the memoryless property, as did
Geometric random variables. In many ways, Exponential random variables are
a continuous version of Geometric random variables. Besides being memoryless,
the density of an Exponential random variable decreases exponentially in x, just
as the mass of a Geometric random variable decreased exponentially.
The notation for an Exponential random variable looks like:

X ∼ Exponential(λ).

412
 32.2. Average and Variance 413

Exponential random variables

The common thread: Time until the 1st success occurs. For example,
the time until the 1st customer enters a shop, or the telephone rings, or one’s
girlfriend/boyfriend arrives at the door, or until the 1st black car drives by.
Things to look for: The waiting time until the first event occurs, or gaps
between events.
The variable:

X = time until the next event occurs, X ≥ 0

The parameters:

λ = the average rate, e.g., number of arrivals per time period.

When the time between arrivals is exponential, the number of arrivals in a

fixed time interval is Poisson with the mean λ.
Density: (
λe−λx , x > 0,
fX (x) =
0 otherwise

CDF: (
1 − e−λx , x > 0,
FX (x) =
0 otherwise
Expected value formula:
E(X) = 1/λ
Variance formula:
Var(X) = 1/λ2

32.2 Average and Variance

We use integration by parts to show why Exponential random variables have
expected value 1/λ. We use x = u and dx = du, as well as dv = λe−λx and
v = −e−λx , to get
Z ∞ Z ∞ ∞
Z ∞
E(X) = xfX (x) dx = xλe−λx dx = (−xe−λx ) − −e−λx dx.
−∞ 0 0 0

We observe that −xe−λx = 0 for x = 0, and −xe−λx → 0 as x → ∞. Thus

∞ ∞
e−λx
Z
−λx
E(X) = e dx = − = 1/λ.
0 λ 0
414 Chapter 32. Exponential Random Variables

The second moment is calculated similarly. We again use integration by parts,

with x2 = u and 2x dx = du, as well as dv = λe−λx and v = −e−λx , to get
Z ∞
E(X 2 ) = x2 fX (x) dx
Z−∞∞
= (x2 )(λ)e−λx dx
0
Z ∞
∞
2 −λx
= (−x e ) − −2xe−λx dx.
0 0

Since −x2Re−λx = 0 for x = 0, and since −x2 e−λx → 0 as x → ∞, then

∞
E(X 2 ) = 0 2xe−λx dx. We can multiply and divide by 2 and λ, to get what
remains into the form of the first moment, i.e.,
Z ∞
2 ∞
Z
2 2
E(X 2 ) = 2xe−λx dx = xλe−λx dx = E(X) = 2 .
0 λ 0 λ λ

So the variance is:

2 1
− 2 = 1/λ2 .
Var(X) =
2
λ λ
Some examples of the densities and CDFs for Exponential random variables
with various values of λ are given in Figure 32.1. These examples show that no

fX (x)
2
fX (x)
0.6 1
fX (x)
0.4 1
0.5
0.2
x x x
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

FX (x)
FX (x) FX (x)
1 1 1

0.5 0.5 0.5

x x x
2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Figure 32.1: Upper row: The density fX (x) of an Exponential random

variable with parameter λ equal to 1/2, 1, or 2 (respectively). Lower row: The
analogous CDF FX (x) for λ equal to 1/2, 1, or 2 (respectively).
 32.2. Average and Variance 415

matter what the λ parameter is, the density starts at λ when x = 0 and then
quickly moves closer to 0 as x → ∞. The CDF starts at 0 but quickly climbs
close to 1 as x → ∞. The density and CDF curves are steeper for larger values
of λ.

Example 32.1. The time between fatal car accidents on a stretch of desolate
highway between two cities was found to follow an Exponential distribution
with a mean of one accident every 44 days. If an accident has occurred today
(the count starts over), the sheriff’s office is interested in when the next accident
may occur.

a. What does X represent in this story? What values can X take?

The random variable X is the time until the next fatal accident. Since you
can’t have a negative waiting time, then X > 0.
b. Why is this an example of the Exponential distribution?
We know the rate these accidents occur, and we are waiting for the next
accident (one event). We are measuring our wait in days (continuous), not in
specific trials (discrete).
c. What is the parameter for this distribution?
The parameter is λ, which is the number of accidents/time, i.e., 1 acci-
dent/44 days or 0.02273 accidents/day.
d. How long should the sheriff’s office expect to wait for the next accident
to occur?

1 1
E(X) = = = 44 days/accident.
λ 1 accident/44 days
e. What is the standard deviation in the time the sheriff’s office will wait
for the next accident to occur?

1
Var(X) = = 1936 days2
(0.02273)2
√
σX = 1936 = 44 days
f. Find the probability that the next accident occurs within the next 31
days.

P (X ≤ 31) = FX (31) = 1 − e−(0.02273)(31) = 0.5057.

g. Today is August 1st. What is the probability that no accident will have
occurred by August 9th?
416 Chapter 32. Exponential Random Variables

This means that we will have to wait more than 8 days for the next accident
to occur. So the probability is

P (X > 8) = 1 − FX (8) = 1 − (1 − e−(0.02273)(8) ) = 1 − (1 − 0.8337) = 0.8337

h. You need an estimate of how long you would wait so that there is a
95% chance that the next accident would happen before that day. What is that
cut-off length of time?
We want the time a such that

P (X ≤ a) = 0.95,

or equivalently
FX (a) = 0.95 = 1 − e−0.02273x .
Thus e−0.02273x = 0.05. Taking the natural logarithm of both sides, we have

−0.02273x = −2.996,

so x = 131.7964. In other words, we are 95% certain that the next accident will
happen within the next 131.7964 days.
We saw an example of Exponential random variables in Example 26.7. The
“lifetime” of a music player might not seem like a “waiting time,” but it can be
interpreted that way, i.e., the time we wait until the music player dies.
We saw another example of waiting times near the start of section 29.1,
where X is the waiting time until the next bus arrives, given in minutes. That
waiting time was exponential with λ = 3/10. So now we can immediately
conclude that E(X) = 1/λ = 10/3.
We also saw waiting times in Exercise 26.6, about the response time for a
police car to an emergency call.

Example 32.2. Let X be the time (in minutes) that Maxine waits for a traffic
light to turn green, and let Y be the time (in minutes, at a different intersection)
that Daniella waits for a traffic light to turn green. Suppose that X and Y have
joint density

fX,Y (x, y) = 15e−3x−5y , for x > 0 and y > 0,

and fX,Y (x, y) = 0 otherwise.

In this case, we notice that the joint density can be factored, so there are some
constants c and d so that

fX (x) = ce−3x for x > 0,

 32.3. Properties of Exponential Random Variables 417

and fX (x) = 0 otherwise, and

fY (y) = de−5y for y > 0,

and fY (y) = 0 otherwise. Since each of these densities must integrate to 1, we

see that c = 3 and d = 5, and X and Y are independent Exponential random
variables, with parameters 3 and 5, respectively.

32.3 Properties of Exponential Random Variables

32.3.1 Complement of the CDF

If X is an Exponential random variable with parameter λ, then the CDF of X

is
FX (x) = 1 − e−λx for x > 0,

and FX (x) = 0 otherwise. Thus, for a > 0, we see that P (X > a) has an
especially nice form:

P (X > a) = 1 − P (X ≤ a) = 1 − (1 − e−λa ) = e−λa .

32.3.2 Memoryless Property of Exponential Random Variables

Just like the Geometric distribution, the Exponential distribution has the mem-
oryless property. The memoryless property helps streamline our ability to cal-
culate conditional distributions for Exponential random variables. (Recall from
Chapter 16 that the Geometric distribution also has a memoryless property.)
For example, if X is an Exponential random variable, then P (X > 5 | X >
2) = P (X > 3). Intuitively, if we have already waited at least two minutes for
an event to occur, then the conditional probability that we wait at least five
minutes altogether is equal to the probability that, starting from right now, we
will wait at least three additional minutes until the event occurs. One impor-
tant difference is that, since the Exponential distribution is continuous, then
P (X ≥ 5) and P (X > 5) are exactly the same value, because P (X = 5) = 0.
As another example, if we know that a waiting time X is bigger than 7, and
we want the probability that X is bigger than 11, then we only need to know
that—after 7 units of time have passed—we have to wait at least 4 additional
units of time. So, for example, it would make sense if

P (X > 11 | X > 7) = P (X > 4).

418 Chapter 32. Exponential Random Variables

We compute

P (X > 11 and X > 7)

P (X > 11 | X > 7) =
P (X > 7)
P (X > 11)
=
P (X > 7)
e−11λ
= −7λ
e
= e−4λ
= P (X > 4)

This phenomenon is true in general. If a and b are positive constants, and we

are given X > a, then the probability that X > a + b is just equal to the
(unconditional) probability that X > b:

P (X > a + b and X > a)

P (X > a + b | X > a) =
P (X > a)
P (X > a + b)
=
P (X > a)
e−(a+b)λ
=
e−aλ
−bλ
=e
= P (X > b)

This is called the “memoryless property,” which we also saw was true for Geo-
metric random variables. If we already know that X is bigger than a, then the
conditional probability (given X is bigger than a) that X is bigger than a + b
is just equal to the unconditional probability that X is bigger than b. This
is true with waiting times in life too, in many cases. For instance, if we have
already waited 10 minutes for the phone to ring, then the probability that we
wait a total of 12 minutes or less is just equal to the unconditional probability
of waiting 2 minutes or less from the outset—the previous 10 minutes do not
affect the future waiting time. (We are assuming that we do not know when
the phone will ring, i.e., that there is not a planned calling time in advance. So
this example works best, for example, if we imagine that we are answering a
telephone in an office, for a telethon, etc., when the timing of the next phone
call is variable and is not planned ahead of time.)

Theorem 32.3. Memoryless property of an Exponential Random

Variable
If X is an Exponential random variable with parameter λ, then for any positive
constants a and b,

P (X > a + b | X > a) = P (X > b).

 32.3. Properties of Exponential Random Variables 419

Example 32.4. (continued from Example 32.1) The time between acci-
dents for all fatal car accidents on a stretch of highway between two cities was
found to follow an Exponential distribution with a mean of one accident every
44 days. If an accident has occurred today (the count starts over), the sheriff’s
office is interested in when the next accident may occur.

Today is August 1st. Given that no accident has occurred by August 25th,
what is the probability that no accident will have occurred by September 2nd?
No accident occurring by August 25th means that there is no accident in
the next 24 days, so our wait time will be more than 24 days, so X > 24.
No accident occurring by September 2nd means that there is no accident in
the next 32 days so our wait time will be more than 32 days.
So the conditional probability is

P (X > 32 and X > 24)

P (X > 32 | X > 24) =
P (X > 24)
P (X > 32)
=
P (X > 24)
1 − FX (32)
=
1 − FX (24)
1 − (1 − e−(0.02273)(32) )
=
1 − (1 − e−(0.02273)(24) )
e−(0.02273)(32)
=
e−(0.02273)(24)
= e−(0.02273)(8)
= 0.8337

That was a lot of math to go through. However, thanks to the memoryless

property, the result is the same as if you rename the start time as August 25th
instead of August 1st.

P (X > 32 | X > 24) = P (X > 8) by the memoryless property,

so by the calculation from Example 32.1g, we have

P (X > 32 | X > 24) = P (X > 8) = 1 − FX (8) = 0.8337.

32.3.3 Minimum of Independent Exponential Random Variables

We already saw, in Exercise 25.17, that the maximum of two independent Ex-
ponential random variables does not have any type of form that we have seen
420 Chapter 32. Exponential Random Variables

before. . . but the minimum of two independent Exponential random variables is

an Exponential random variable too! In other words, if X and Y are indepen-
dent Exponential random variables with parameters λ1 , λ2 , respectively, and we
define Z = min(X, Y ), then Z is Exponential random variable as well. To see
this, we notice that Z must be positive since X and Y are both positive; thus
FZ (z) = 0 for z ≤ 0. Now consider z > 0. We see that

FZ (z) = P (Z ≤ z)
= 1 − P (Z > z)
= 1 − P (X > z and Y > z)
= 1 − P (X > z)P (Y > z)
= 1 − e−λ1 z e−λ2 z
= 1 − e−(λ1 +λ2 )z

So Z is an Exponential random variable with parameter λ1 + λ2 . Applying this

rule over and over, we get the following result:

Theorem 32.5. Minimum of a collection of Exponential Random

Variables
If X1 , X2 , . . . , Xn are independent Exponential random variables with param-
eters λ1 , λ2 , . . . , λn , respectively, and we define

Z = min(X1 , X2 , . . . , Xn ),

then Z is also an Exponential random variable, with parameter λ1 + · · · + λn .

Example 32.6. Now consider two random variables, X and Y , with joint
distribution
fX,Y (x, y) = 36e−5x−4y for 0 < x < y,
and fX,Y (x, y) = 0 otherwise. Then X has an exponential distribution but Y
does not.
We compute, for y > 0, the density of Y :
Z y
36
fY (y) = 36e−5x−4y dx = (e−4y − e−9y ).
0 5
The density of Y tells us that Y is not exponentially distributed. On the other
hand, for x > 0, when we compute the density of X,
Z ∞
fX (x) = 36e−5x−4y dy = 9e−9x ,
x

so X is an Exponential random variable with parameter λ = 9. Finally, since

0 < X < Y , we know that X and Y are dependent.
 32.3. Properties of Exponential Random Variables 421

The expected value of Y is

Z ∞  
36 −4y 13
E(Y ) = (y) (e − e−9y ) dy = .
0 5 36
Since X is Exponential with parameter 9, we do not need to evaluate an integral
to find the expected value. We know that the expected value is the inverse of
the parameter, i.e., E(X) = 1/9.

32.3.4 Poisson Process

There is an important link between the Poisson distribution (Chapter 18) and
the Exponential distribution in this chapter. There is a name for the process of
combining the Poisson, Exponential, and Gamma (next chapter) distributions.
(A sequence of events that have independent Exponential waiting times between
consecutive events is called a Poisson process.) Here we discuss how these
distributions are related.
Consider a process in which events occur, such as the times when cars pass
by an observer, when the telephone rings, when a text message arrives, etc. If
the times between consecutive occurrences are independent Exponential random
variables, each with parameter λ, then such a process is called a Poisson process.
This concept is fundamental in courses on Stochastic Processes, but we mention
here a basic property of such processes: The number of occurrences in a fixed
time interval of length t is a Poisson random variable with parameter λt.

Example 32.7. Suppose that the times between the arrival of consecutive
emails are independent Exponential random variables, each of which has an
average of 1/2 a minute, i.e., the parameter is λ = 2. So we expect 2 emails
to arrive per minute. The expected number of emails to arrive in a fixed 10
minute time interval (say, between 1:35 PM and 1:45 PM) is a Poisson random
variable with mean (2)(10) = 20.

Example 32.8. A traffic engineer has studied a lonesome desert highway, and
she believes that the average time between consecutive cars at an observation
point is 12 minutes, and these waiting times are independent Exponential ran-
dom variables. Thus, on average, 5 cars pass the observation point per hour.
Moreover, the number of cars that pass the observation point per hour is Pois-
son with mean 5. The number of cars that pass during a two-hour period is
Poisson with mean 10. The number of cars that pass during a 30-minute pe-
riod is Poisson with mean 2.5 (even though the number of cars that pass is an
integer-valued random variable, the parameter λ is allowed to be a non-integer).
422 Chapter 32. Exponential Random Variables

The probability that exactly 4 cars pass between 1:10 PM and 2:10 PM is
e−5 54
= 0.1755.
4!
The probability that exactly 13 cars pass between 1:35 PM and 3:35 PM is
e−10 1013
= 0.0729.
13!
The probability that exactly 2 cars pass between 4:30 PM and 5:00 PM is
e−2.5 2.52
= 0.2565.
2!

32.3.5 Moments of an Exponential Random Variable (Optional)

We refer to E(X k ), for a fixed integer k, as the kth moment of X. For instance,
for an Exponential random variable, the kth moment is:
Z ∞
E(X k ) = (xk )(λe−λx ) dx
0
= − (xk )(e−λx )

− (kxk−1 )(e−λx )/λ

− ((k)(k − 1)xk−2 )(e−λx )/λ2
− ((k)(k − 1)(k − 2)xk−3 )(e−λx )/λ3
∞
−λx k
− · · · − k!(e )/λ
x=0
= k!/λk .
Theorem 32.9. Moments of an Exponential Random Variable
If X is an Exponential random variable with parameter λ, then the kth mo-
ment of X is
E(X k ) = k!/λk .
In particular, the expected value of X is

E(X) = 1/λ,

and the second moment of X is

E(X 2 ) = 2/λ2 ,

so the variance of X is
 2
2 1
Var(X) = 2 − = 1/λ2 .
λ λ
 32.4. Exercises 423

32.4 Exercises
32.4.1 Practice
Exercise 32.1. Egg laying. Chickens at Rolling Meadows Farm lay an average
of 18 eggs per day. The farmer has rigged a fancy monitoring device to the
nesting boxes so that he can monitor exactly when the hens lay their eggs.
Assume that no 2 eggs will be laid at exactly the same time and that the eggs
(and chickens) are independent from each other. The farmer wants to know
how long he will have to wait (in minutes) for the next egg to be laid if he starts
monitoring at the first rooster crow in the morning.

a. Why is this an Exponential problem?

b. What does X represent in this scenario?
c. What is the parameter in units matching this specific question?
d. What is the expected length of time (in minutes) the farmer will have to
wait for the first egg to be laid?
e. What is the standard deviation?
f. What is the probability density function for the wait time (in minutes)
for the first egg? Write it in function form and also graph it.
g. What is the cumulative distribution function for the wait time (in min-
utes) for the first egg? Write it in function form and also graph it.
h. What is the probability the farmer will have to wait longer than it takes
him to make his cup of coffee (the first 10 minutes after the rooster crows)?
i. What is the probability the first egg will be laid while he is milking the
cow (between 30 minutes and an hour after the rooster crows)?
j. Given that the first egg did not come while he was making his coffee,
what is the probability that the first egg will come while he is milking the cow?
k. The farmer is out doing chores and can’t check the fancy egg monitor. If
the farmer wants to wait to go to the henhouse to feed the hens until he is 90%
sure the first egg has been laid, how long should he wait?

Exercise 32.2. Hurricanes. On the average, hurricane of category 4 or

stronger (on the Stafford/Simpson scale) strikes the United States once every
6 years. A hurricane of this strength has winds of at least 131 miles per hour
and can cause extreme damage. (http://www.aoml.noaa.gov/hrd/Landsea/
deadly/index.html) An insurance agency is considering whether it might want
to stop insuring oceanfront homes and wants to assess the risk involved. The
president of the company wants to know how long (in years) before the next
hurricane that is category 4 or stronger.

a. Why is this an Exponential scenario?

424 Chapter 32. Exponential Random Variables

b. What does X represent in this scenario?

c. What is the parameter?
d. What is the expected length of time (in years) between now and when
the next hurricane that is category 4 or stronger?
e. What is the variance in this length of time?
f. What is the probability density function for the length of time before the
next hurricane that is category 4 or stronger? Write your answer in function
form and show a graph.
g. What is the CDF for the length of time before the next hurricane that is
category 4 or stronger? Write your answer in function form and show a graph.
h. What is the probability that there will not be any hurricane that is
category 4 or stronger, during the next 3 years?
i. What is the probability that there will be a hurricane that is category 4
or stronger, during the period that is between 5 to 10 years from now?
j. Given that there are no hurricanes that are category 4 or stronger during
the next 3 years, what is the probability that there will not be any during the
next 10 years?
k. How long a waiting time do we need, if we want to be 75% sure that
there is a hurricane that is category 4 or stronger during the waiting time?

Exercise 32.3. Waiting by the phone. Bob is waiting for his girlfriend Alice
to call. His waiting time X is Exponentially distributed, with expected waiting
time E(X) = 0.20 hours, i.e., 12 minutes.

a. What is the probability that he must wait more than 12 minutes for her
to call?
b. What is the probability that he must wait more than 12 minutes (alto-
gether) for her to call, given that he has already waited 3 minutes?
c. What is the probability that he must wait at most 10 minutes (altogether)
for her to call, given that he has already waited 3 minutes?
d. Given that he waited less than 10 minutes, what is the probability that
he waited less than 3 minutes?
e. Find the unique time “a” in hours such that P (X ≤ a) = 1/2 and
P (X > a) = 1/2. This is the median waiting time; also see Exercise 32.13.

Exercise 32.4. Waiting for a bus. A student waits for a bus. Let X be
the number of hours that the student waits. Assume that the waiting time is
Exponential with average 20 minutes.

a. What is the probability that the student waits more than 30 minutes?
 32.4. Exercises 425

b. What is the probability that the student waits more than 45 minutes
(total), given that she has already waited for 20 minutes?
c. Given that someone waits less than 45 minutes, what is the probability
that they waited less than 20 minutes?
d. What is the standard deviation of the student’s waiting time?

Exercise 32.5. Waiting for a ride. The waiting time for rides at an amuse-
ment park has an Exponential distribution with average waiting time of 1/2 an
hour. Find the time “t” such that 80% of the people have waiting time t or less.

Exercise 32.6. Happy birthday. It is your birthday and you are waiting for
someone to write a “Happy Birthday” message on your Facebook wall. Your
waiting time is approximately Exponential with average waiting time of 10
minutes between such postings; assume that the times of the postings are inde-
pendent.

a. What is the probability that the next posting takes 15 minutes or longer
to appear?
b. What is the standard deviation of the time in between consecutive Happy
Birthday messages?
c. Suppose that the most recent posting was done at 1:40 PM, and it is now
1:45 PM (i.e., no postings have been made during the last five minutes). What
is the expected time for the next message to appear?

Exercise 32.7. Falling asleep. Lily estimates that her time to fall asleep
each night is approximately Exponential, with an average time of 30 minutes
until she falls asleep.

a. What is the probability that it takes her less than 10 minutes to fall
asleep?
b. What is the probability that it takes her more than 1 hour to fall asleep?
c. If she has already laid awake for 1 hour, what is the probability that it
will take her more than 90 minutes (altogether) to fall asleep?

32.4.2 Extensions
Exercise 32.8. Pizza delivery. Suppose that the times until Hector, Ivan,
and Jacob’s pizza arrives are independent exponential random variables, each
with average of 20 minutes. Find the probability that none of the waiting times
exceed 20 minutes, i.e., find P (max(X, Y, Z) ≤ 20).

Exercise 32.9. Flight delays. Suppose that, when an airplane waits on the
runway, the company must pay each customer a fee if the waiting time exceeds
3 hours. Suppose that an airplane with 72 passengers waits an exponential
amount of time on the runway, with average 1.5 hours. If the waiting time X,
426 Chapter 32. Exponential Random Variables

in hours, is bigger than 3, then the company pays each customer (100)(X − 3)
dollars (otherwise, the company pays nothing). What is the amount that the
company expects to pay for the 72 customers on the airplane altogether? (Of
course their waiting times are all the same.)

Exercise 32.10. Let X be uniform on [0, 10]. Let Y be exponential with

E(Y ) = 5. Find P (X < Y ).

Exercise 32.11. Let X be exponential with expected value 3. Let Y be another

random variable that depends on X as follows: if X > 5, then Y = X − 5;
otherwise, Y = 0.

a. Find the expected value of Y .

b. Find the variance of Y .

Exercise 32.12. Vending machine. The time between consecutive uses of a

vending machine is Exponential with average 15 minutes.

a. Given that the machine has not been used in the previous 5 minutes,
what is the probability that the machine will not be used during the next 10
minutes?
b. How many purchases are expected within the next hour?
c. What is the distribution of the number of purchases to be made within
the next hour?
d. Suppose that a person pays 75 cents for each beverage in the vending
machine. The supplier pays 40 cents per beverage in the machine, and thus
makes a profit of 35 cents per beverage. What is the expected profit made
during a given 8 hour period?

Exercise 32.13. Median. Suppose Y is an Exponential random variable with

parameter λ > 0. What is the unique value “a” such that P (Y ≤ a) = 1/2 and
P (Y > a) = 1/2? (This value “a” is called the “median.”)

Exercise 32.14. If Y is an Exponential random variable with parameter λ > 0,

the expected value of Y is E(Y ) = 1/λ. What is the probability that Y is bigger
than its expected value, i.e., what is P (Y > λ1 )?

Exercise 32.15. Still waiting by the phone. Dan, Dominic, and Doug
are waiting together in the living room for their girlfriends, Sally, Shellie, and
Susanne, to call. Their waiting times (in hours) are independent Exponential
random variables, with parameters 2.1, 3.7, and 5.5, respectively. What is the
probability that the phone will ring (i.e., the first call will arrive) within the
next 30 minutes (i.e., within the next 1/2 an hour)?

Exercise 32.16. Air traffic control. Air traffic control stations often have
insufficient numbers of air traffic controllers, sometimes just one person on duty.
In a recent study, a lone air traffic controller is managing an airstrip in which
 32.4. Exercises 427

the expected time between arrivals of airplanes is 15 minutes. Assume that

the times between consecutive arrivals are independent Exponential random
variables. If he falls asleep for a period of 5 minutes, what is the probability
that one or more airplanes arrive during this period (but cannot land because
the air traffic controller is not awake to guide them)?

32.4.3 Advanced
Exercise 32.17. Consider an Exponential random variable X with parameter
λ > 0. Let Y = bXc, which means we get Y by rounding X down to the
nearest integer (in particular, Y itself is a discrete random variable, because Y
is always an integer). For example, if X = 7.2, then Y = 7. If X = 12.9999,
then Y = 12. If X = 5.01, then Y = 5, etc.
So the mass of Y is exactly

pY (y) = P (Y = y) = P (y ≤ X < y + 1).

a. Find an expression for the mass of Y . (Your expression will have λ in it;
i.e., just integrate to find the value of P (y ≤ X < y + 1), and then simplify.)
b. Do you recognize the mass of Y ? (Yes, you should!) What type of
random variable is Y ? What are the parameters of Y ? (Hint: Y is one of the
types of named discrete random variables.)

Exercise 32.18. Let X be an Exponential random variable with E(X) = 1/λ.

Define Y = dXe, i.e., Y is the least integer that is greater than or equal to Y .
For instance, if X = 5.3, then Y = 6. If X = 4.99, then Y = 5. If X = 5.00001
then Y = 6. If X = 5.99999 then Y = 6. What is the distribution of Y ?

Exercise 32.19. A store manager is very impatient at the start of the day. Let
X be the time (in minutes) until his first customer arrives. The workers at the
store decided to find a way to measure his impatience level as a function of X:

g(X) = aX 2 + bX + c,

where a, b, c are fixed constants.

a. Calculate the manager’s expected impatience level.

b. Find the variance of the manager’s impatience level.

Exercise 32.20. Suppose that the time in between customers at a store are in-
dependent Exponential random variables, with an average of 2 minutes between
consecutive customers. Let X be the time until the 3rd customer arrives.

a. Find E(X).
b. Find Var(X).
c. Find the density of X.
Chapter 33

Gamma Random Variables

I have noticed that people who are late are often so much jollier than the people
who have to wait for them.
—E. V. (Edward Verrall) Lucas

An employee wonders how long it will take until the fourth customer arrives at
the store. How is this related to the length of time until the first arrival? Is
the time to the fourth arrival always four times as long as the time to the first
arrival? Why or why not?

33.1 Introduction
A Negative Binomial random variable is the sum of r independent Geometric
random variables, i.e., the number of (discrete) trials until the rth success oc-
curs. The Gamma random variable has a similar motivation, but in a continuous
setting. A Gamma random variable is the sum of r independent Exponential
random variables. If an Exponential random variable is viewed as the waiting
time until the first event occurs, then a Gamma random variable is the wait-
ing time until the rth event occurs. Thus, if X1 , X2 , . . . , Xr are independent
Exponential random variables that each have parameter λ, and if we define

X = X1 + X2 + · · · + Xr ,

then we say that X is a Gamma random variable with parameters λ and r.

The density of a Gamma random variable X with parameters λ and r is
λr r−1 −λx
fX (x) = x e for x > 0,
Γ(r)

and fX (x) = 0 otherwise.

428
 33.1. Introduction 429

Gamma random variables

The common thread: Time until the rth event occurs. For example,
the time until the 10th customer enters a shop, or the phone rings for the 3rd
time, or the time until the 10th black car drives by.
Things to look for: Repeated independent trials until rth event; same pa-
rameter for each waiting time.
The variable:

X = time until the rth event occurs, X ≥ 0

The parameters:

r = total number of arrivals/events that you are waiting for

λ = the average rate, e.g., number of arrivals per time period

The λ for the Gamma distribution is the same as the one for the Exponential
distribution. The r parameter is often called the “shape” parameter, and λ is
the “scale” parameter.
Density:
λr r−1 −λx
(
Γ(r) x e , x > 0,
fX (x) =
0 otherwise
where Γ(r) = (r − 1)! since r is a positive integer.
When r = 1, a Gamma random variable is just an Exponential random vari-
able.
CDF:
(λx)j
(
1 − e−λx r−1
P
FX (x) = j=0 j! , x > 0,
0 otherwise
Expected value formula:
E(X) = r/λ
Variance formula:
Var(X) = r/λ2

The notation for the Gamma looks like:

X ∼ Gamma(r, λ).

Since a Gamma random variable is the sum of r Exponential random variables,

then the expected value of a Gamma random variable is just equal to r times
the expected value of an Exponential random variable:

1 1 r
E(X) = E(X1 + · · · + Xn ) = E(X1 ) + · · · + E(Xr ) = + ··· + = .
λ λ λ
430 Chapter 33. Gamma Random Variables

Since the Xj ’s are independent, the variance of a Gamma random variable is

just equal to r times the variance of an Exponential random variable:

1 1 r
Var(X) = Var(X1 + · · · + Xn ) = Var(X1 ) + · · · + Var(Xn ) = 2
+· · ·+ 2 = 2 .
λ λ λ
Some examples of densities for Gamma random variables with various values
of r and λ are given in Figures 33.1, 33.2, and 33.3.

fX (x) fX (x) fX (x)

0.5 1 4

0.25 0.5 2

x x x
5 10 5 10 5 10

Figure 33.1: The density fX (x) of a Gamma random variable with r = 1

and (left) λ = 1/2, (middle) λ = 1, and (right) λ = 4. Notice that the scale of
the y-axis changes is changing as the values of λ change.

fX (x) fX (x) fX (x)

0.2 0.4 1.6

0.1 0.2 0.8

5 10 x 5 10 x 5 10 x

Figure 33.2: The density fX (x) of a Gamma random variable with r = 2

and (left) λ = 1/2, (middle) λ = 1, and (right) λ = 4.

fX (x) fX (x) fX (x)

0.1 0.2 0.8

0.05 0.1 0.4

5 10 x 5 10 x 5 10 x

Figure 33.3: The density fX (x) of a Gamma random variable with r = 5

and (left) λ = 1/2, (middle) λ = 1, and (right) λ = 4.
 33.2. Examples 431

33.2 Examples

Example 33.1. Customers arrive at a shoe store at an average rate of 1 every

5 minutes. The inter-arrival times from customer to customer are independent,
and each inter-arrival time is Exponentially distributed. Since the clerks work
on commission, arguments over which customers belong to which clerk can get
very heated. Bill has made a deal with his coworker Shirley that he will wait on
the first 3 customers, she will wait on the next 3 customers after that, and then
he takes over again. Shirley wants to grab a cup of coffee while she is waiting
for her turn, but she wants to estimate how much time she will have for her
break. In particular, Shirley wants to make sure that she is ready when the 4th
customer arrives.

a. What does X represent in this story? What values can X take?

The random variable X is the time Shirley waits for the 4th customer to
arrive. In particular, X is nonnegative. (Theoretically all 4 customers could
arrive near the very beginning of Bill’s shift, or the 4th customer might not
arrive for days.)
b. Why is this an example of the Gamma distribution?
Correct answer: The random variable X has a Gamma distribution because
we know the average customer arrival rate, and we are measuring the waiting
time until the 4th customer.
Other answers you might have considered (and why they are wrong):
If we were measuring the waiting time for just the 1st customer, this would
be an Exponential random variable.
If we were counting up how many customers arrive in the next 30 minutes,
this would be a Poisson random variable.
If we were counting the number of customers Shirley had to help until she
had her first customer who purchased red stiletto high heels (and if we know the
probability that customers in general want to purchase red stiletto high heels),
then this would be a Geometric random variable.
If we wanted to count the number of customers until we found the 4th
customer who purchased red stiletto high heels, then this would be a Negative
Binomial random variable.
c. What are the parameters for this distribution?
The parameter λ is the average waiting time between customer arrivals,
e.g., 2 customers every 10 minutes (or 1 customer every 5 minutes if we want
to simplify).
The parameter r is the number of customers Shirley is waiting for. She is
interested in the arrival time of the 4th customer, so r = 4.
432 Chapter 33. Gamma Random Variables

d. How long should Shirley expect to have for her coffee break before it is
her turn to wait on the customers?
r 4 customers
E(X) = = = 20 minutes.
λ 1 customer / 5 minutes
e. What is the standard deviation in the amount of time Shirley will have
for her coffee break?
r 4
Var(X) = 2
= = 100 minutes2
λ (1/5)2
√
σX = 100 = 10 minutes
f. What is the density of the time until the 4th customer of the day arrives?
(1/5)4 4−1 −(1/5)x
(
fX (x) = Γ(4) x e , x > 0,
0 otherwise
which, since Γ(4) = 3! = 6, simplifies to
(
1
x3 e−(1/5)x , x > 0,
fX (x) = 3750
0 otherwise

g. What is the probability that she will have at least 30 minutes before she
has her first customer (the fourth customer of the day)?
The probability is

P (X > 30) = 1 − P (X ≤ 30)

= 1 − FX (30)
4−1
!
X ((1/5)(30))j
=1− 1 − e−(1/5)(30)
j!
j=0
3
X 6j
= e−6
j!
j=0

= e−6 (1 + 6 + 18 + 36)
= 61e−6
= 0.1512

Example 33.2. David works at a customer call center. He talks to customers

on the telephone. The length (in hours) of each conversation is Exponential
with average 1/3, and the lengths of calls are independent. As soon as one
conversation is finished, he hangs up the phone, and immediately picks up the
 33.2. Examples 433

phone again to start another call (i.e., there are no gaps in between the calls).
Thus, if he conducts r phone calls in a row, the total amount of time he spends
on the telephone is X1 + · · · + Xr , where the Xj ’s are independent, and each
Xj has the same density.

We see that the total time it takes David to make r calls is exactly a Gamma
random variable X = X1 + · · · + Xr with parameters λ = 3 and r.
The expected time that it takes him to make r calls is
E(X) = r/3.
The variance of the time that it takes him to make r calls is
Var(X) = r/9,
and thus the standard deviation is r/9.
p
Here we are using
integration by parts.
We can also compute the probability that he completes two calls within the
first 1 hour, by using the density of X given above. When r = 2 and λ = 3,
then the density of X becomes
fX (x) = x2−1 32 e−3x /(2 − 1)! = 9xe−3x for x > 0,
and fX (x) = 0 otherwise. Completing two calls in the first hour means X < 1;
thus
Z 1
P (X < 1) = 9xe−3x dx
0
 1
= 9xe−3x /(−3) − 9e−3x /(9)
x=0
−3 −3


= −3e − e − (−1)
= 1 − 4e−3
= 0.8009

Example 33.3. Consider 300 students who are waiting for service at the regis-
trar’s office. Assume that their waiting times are independent exponential ran-
dom variables, and each waiting time (in minutes) has density fX (x) = 2e−2x
for x > 0, and fX (x) = 0 otherwise. Find the probability that the 300 students
collectively (i.e., altogether) spend between 145 and 152 hours waiting for their
appointments.

This is the same probability as 145 ≤ Y ≤ 152 where Y is a Gamma random

variable Y with parameters r = 300 and λ = 2. So the probability is
Z 152 300
2
P (145 ≤ Y ≤ 152) = x299 e−2x dx = 0.3122,
145 299!
but this requires a high powered calculator or symbolic computing environment
(such as Maple or Mathematica) to calculate.
434 Chapter 33. Gamma Random Variables

33.3 Exercises
33.3.1 Practice
Exercise 33.1. Egg laying. (See Exercise 32.1.) Chickens at Rolling Mead-
ows Farm lay an average of 18 eggs per day. The farmer has rigged a fancy
monitoring device to the nesting boxes so that he can monitor exactly when
the hens lay their eggs. Assume that no 2 eggs will be laid at exactly the same
time and that the eggs (and chickens) are independent from each other. The
farmer wants to know how long he will have to wait (in minutes) for the next
half-dozen (6) eggs to be laid so that he can bake a chocolate cake if he starts
monitoring at the first rooster crow in the morning.

a. Why is this a Gamma problem? What makes this a Gamma situation

instead of an Exponential situation?
b. What does X represent in this scenario?
c. What are the parameters in units matching this specific question?
d. What is the expected length of time (in minutes) the farmer will have to
wait for the half-dozen eggs to be laid?
e. What is the standard deviation?
f. What is the probability density function for the wait time (in minutes)
for the half-dozen eggs? Write it in function form and also graph it.
g. What is the cumulative distribution function for the wait time (in min-
utes) for the half-dozen eggs? Write it in function form and also graph it.
h. Find the probability the farmer will have to wait for his half-dozen eggs
longer than it takes him to do his morning chores and run his errands in town
(6 hours after the rooster crows).
i. Find the probability that the sixth egg will be laid while he is eating
dinner (between 12 and 13 hours after the rooster crows)?

Exercise 33.2. Hurricanes. (See Exercise 32.2.) On the average, a category 4

(on the Stafford/Simpson scale) or stronger hurricane strikes the United States
once every 6 years. A hurricane of this strength has winds of at least 131
miles per hour and can cause extreme damage. (http://www.aoml.noaa.gov/
hrd/Landsea/deadly/index.html) An insurance agency is considering whether
they might want to stop insuring oceanfront homes and wants to assess the risk
involved. The president of the company wants to know how long (in years)
before the next 3 hurricanes that are category 4 or stronger.

a. Why is this a Gamma scenario?

b. What does X represent in this scenario?
c. What are the parameters?
 33.3. Exercises 435

d. What is the expected length of time (in years) between now and when
the third hurricane that is category 4 or stronger will come?
e. What is the variance in this length of time?
f. What is the probability density function for the length of time before the
third hurricane that is category 4 or stronger will come? Write your answer in
function form and show a graph.
g. What is the CDF for the length of time before the third hurricane that is
category 4 or stronger? Write your answer in function form and show a graph.
h. What is the probability that the third hurricane that is category 4 or
stronger will arrive in the next 10 years?

Exercise 33.3. Flight delays. The time (in minutes) until a person’s flight
departs at an airport has density

1 −x/45
fX (x) = e , for x > 0,
45
and fX (x) = 0 otherwise.

a. What is the expected total waiting time of seven passengers on seven

different flights?
b. What is the standard deviation of the total waiting time for seven pas-
sengers on seven different flights?

Exercise 33.4. Waiting for a ride. The waiting time for rides at an amuse-
ment park has an Exponential distribution with average waiting time of 1/2 an
hour (assume that the waiting times are independent).

a. If a person rides 5 rides, what is the expected amount of time that the
person spends waiting in line?
b. If a person rides 5 rides, what is the standard deviation of the time that
the person spends waiting in line?
c. Find the probability that the person spends more than 1 hour altogether
while waiting for two rides (i.e., that their actual waiting time is longer than
the expected waiting time).

Exercise 33.5. Waiting for a bus. A student waits for a bus 10 times during
a week. Let X be the number of hours that the student waits. Assume that the
waiting times are independent Exponential random variables, each with average
30 minutes.

a. What is his expected time spent waiting for the 10 buses altogether?
b. What is the standard deviation of his waiting time for the 10 buses
altogether?
436 Chapter 33. Gamma Random Variables

Exercise 33.6. Homework. A student estimates that the time needed to solve
each homework problem is Exponentially distributed and is independent of all
the other homework problems. Each problem takes, on average, 15 minutes to
solve.

a. What is the expected time spent on a 6-question homework assignment?

b. What is the variance of the time spent on a 6-question homework assign-
ment?

Exercise 33.7. Waiting at the store. The time that each customer spends
in a grocery store line is Exponential with average waiting time 3 minutes. If I
am the 7th customer in line, how long do I expect to wait until all 7 of us have
had our groceries processed?

Exercise 33.8. Taking an exam. Thirteen students take an exam. Each

student spends an Exponential amount of time on problem #2, with average of
10 minutes per student. What is the density of the sum of the time that the 13
students spend on problem #2 altogether?

Exercise 33.9. Happy birthday. It is your birthday and you are waiting for
someone to write a “Happy Birthday” message on your Facebook wall. Your
waiting time is approximately Exponentially distributed with average waiting
time of 10 minutes between such postings; assume that the times of the postings
are independent.

What is the probability that you have to wait 24 minutes or less until you
get your second Happy Birthday message?

Exercise 33.10. Pinball. While listening to “Pinball Wizard,” you decide to

conduct an experiment in which you play pinball over and over again. Each
game takes an Exponential amount of time to finish, with expected value of 3
minutes.

a. What is the density of the time that you spend, if you play two games in
a row?
b. What is the probability that you will be finished with two games in a
total of five minutes or less?

Exercise 33.11. Meteor shower. An astronomer watches a meteor shower.

He believes that the density of the time in between each meteor is

fX (x) = 10e−10x , for x > 0;

otherwise, fX (x) = 0. He has also estimates that there are 500 meteors in this
shower. How many minutes does he expect to be watching this meteor shower?

Exercise 33.12. Starting a band. Five friends decide to start a band. They
each start to practice their own instruments until they can perform their first
 33.3. Exercises 437

song. Unfortunately, they practice independently of each other, so their waiting

times until they are ready to perform are independent Exponential random
variables, each with an average of 3 days. How much time do they expect to
spend practicing altogether (i.e., the sum of their practice times) until they are
all ready to get together and rehearse?

33.3.2 Extensions
Exercise 33.13. Lots of homework. Kelsey gets back to her apartment and
immediately starts all her homework assigned that day for her three classes. The
time to finish each is an Exponential random variable, with average 30 minutes;
these three times are independent. What is the probability that it takes her
more than 90 minutes (altogether) to complete the three assignments?

Exercise 33.14. Passing cars. A student procrastinates, watching cars pass

his house. After 10 black cars have passed, he will (finally) start his homework.
He notices that the time between consecutive cars that pass is Exponentially
distributed, and the times are independent, each with expected time of 1/5 of
a minute. He also assumes that each car has a 30% chance of being black, and
the colors of the cars are independent. How long should he expect to wait until
the 10th black car arrives?

Exercise 33.15. Waiting for a therapist. The waiting time to see a thera-
pist is Exponential with average of 10 minutes during each visit. What is the
probability that, during 3 separate visits to the therapist, a patient spends a
total of 40 minutes or more waiting?

33.3.3 Advanced
Exercise 33.16. If X is a Gamma random variable with r = 2, show that X
does not have the memoryless property.

Exercise 33.17. If X is a Gamma random variable with r = 2, find the prob-

ability that X exceeds E(X).
Chapter 34

Beta Random Variables

Life shrinks or expands in proportion to one’s courage.

—The Diary of Anais Nin, Volume 3 by Anaïs Nin (Swallow Press, 1971)

What is the percent of students who pass the SOA/CAS P/1 actuarial exam?

34.1 Introduction
The Beta distribution deals with percents, proportions, or fractions. The nota-
tion for the Beta distribution looks like:

X ∼ Beta(α, β)

When α and β are both 1, a Beta random variable is simply a Uniform random
variable.
Unfortunately, in general, there is no nice shortcut for the Beta cumulative
distribution function. Since the power of x depends on the parameters α and β,
the density (and, therefore, its integral) depends essentially on those parameters.
Thus, a concise form for the CDF is not available.
The density has a familiar shape for many pairs of parameters. We give
some examples. The plots are only drawn for the values 0 ≤ x ≤ 1 because
the density equals 0 elsewhere. We give four density and CDF example plots in
Figures 34.1, 34.2, 34.3, and 34.4.

438
 34.1. Introduction 439

fX (x) FX (x)
1 1

0.5 0.5
x x
0.5 1 0.5 1

Figure 34.1: When α = 1 and β = 1, (left) the density fX (x) = 1, and

(right) the CDF FX (x) = x of a Beta(1, 1) random variable.

2 fX (x) 1 FX (x)

1 0.5
x x
0.5 1 0.5 1

Figure 34.2: When α = 2 and β = 1, (left) the density fX (x) = 2x, and
(right) the CDF FX (x) = x2 of a Beta(2, 1) random variable.

2 fX (x) 1 FX (x)

1 0.5
x x
0.5 1 0.5 1

Figure 34.3: When α = 1 and β = 2, (left) the density fX (x) = 2(1 − x),
and (right) the CDF FX (x) = x(2 − x) of a Beta(1, 2) random variable.

1 FX (x)
1.5 fX (x)
1 0.5
0.5
x x
0.5 1 0.5 1

Figure 34.4: When α = 2 and β = 2, (left) the density fX (x) = 6x(1 − x),
and (right) the CDF FX (x) = x2 (3 − 2x) of a Beta(2, 2) random variable.
440 Chapter 34. Beta Random Variables

Beta random variables

The common thread: A two-variable distribution that describes the
percentage of some quantity.
A Beta random variable is, for instance, the percent of a job which will be
completed, fraction of resources to be used, proportion of area covered, etc.
Things to look for: Two parameters are needed to give the density’s shape.
The variable:

X = the proportion, fraction, or percent of a quantity of interest.

The parameters: The parameters, α and β, will either be given or can be

found by modeling the density based upon data from previous observations.
Density:
( Γ(α+β) α−1
Γ(α)Γ(β) x (1 − x)β−1 , for 0 ≤ x ≤ 1,
fX (x) =
0 otherwise

When α or β is a positive integer, it is helpful to remember that Γ(n) = (n−1)!.

Expected value formula:
α
E(X) =
α+β
Variance formula:
αβ
Var(X) =
(α + β)2 (α + β + 1)

34.2 Examples

Example 34.1. A soda company distributor wants to figure out how long he
should wait to deliver more sodas to a particular convenience store. The store’s
stock of sodas in a month can be modeled by a Beta distribution with α = 2
and β = 5, where X is the proportion of sodas that will be purchased.

a. What is the probability density function for X?

(
30x(1 − x)4 , for 0 ≤ x ≤ 1,
fX (x) =
0 otherwise

This density is depicted in Figure 34.5.

b. What is the cumulative distribution function for X?
 34.2. Examples 441

3 fX (x)

2.5
1.2
FX (x)
2 1

0.8
1.5
0.6
1
0.4
0.5
0.2
x x
0.2 0.4 0.6 0.8 1 1.2 0.2 0.4 0.6 0.8 1 1.2

Figure 34.5: When α = 2 and β = 5, a Beta(2, 5) random variable has

density fX (x) = 30x(1 − x)4 (left) and CDF FX (x) = 1 − 6(1 − x)5 + 5(1 − x)6
(right).

The CDF FX (x) is 0 for x < 0 and is 1 for x > 1. The interesting portion
of FX (x) is for 0 ≤ x ≤ 1. In this region, it is profitable to use u-substitution,
because the (1 − x) portion of the density has a higher power (namely, 4) than
the x portion of the density (namely, 1). So we calculate, for 0 ≤ a ≤ 1,
Z a
FX (a) = 30x(1 − x)4 dx.
0

Using u = 1 − x and du = −dx, we obtain

Z 1−a
FX (a) = −30(1 − u)u4 du
1
Z 1
= 30(u4 − u5 ) du
1−a
  1
1 5 1 6
= 30 u − u
5 6 u=1−a
= 1 − 6(1 − a) + 5(1 − a)6 .
5

This CDF is depicted in Figure 34.5.

c. What is the expected proportion of sodas that will be purchased?

2
E(X) = = 2/7 = 0.2857.
2+5
d. What is the variance and standard deviation of the proportion of sodas
442 Chapter 34. Beta Random Variables

that will be purchased?

(2)(5)
Var(X) = = 5/196 = 0.0255,
(2 + 5)2 (2 + 5 + 1)
p
σX = 5/196 = 0.1597.

e. What is the probability that more than 3/4 of the sodas the distributor
delivered will be purchased?

P (X > 3/4) = 1 − FX (3/4)

= 1 − 1 − 6(1 − 3/4)5 + 5(1 − 3/4)6

= 6(1/4)5 − 5(1/4)6
= 19/4096
= 0.0046

f. What is the probability that between 1/2 and 3/4 of the sodas the dis-
tributor delivered will be purchased?
The probability is P (1/2 ≤ X ≤ 3/4) = FX (3/4) − FX (1/2). Using the
results from part e, we know FX (3/4) = 1 − P (X > 3/4) = 1 − 19/4096 =
4077/4096. Also, we have

P (X ≤ 1/2) = FX (1/2) = 1 − 6(1 − 1/2)5 + 5(1 − 1/2)6 = 57/64,

so it follows that

P (1/2 ≤ X ≤ 3/4) = P (X ≤ 3/4) − P (X ≤ 1/2)

= 4077/4096 − 57/64
= 0.1047

g. The distributor was already told that more than half of the sodas delivered
earlier this month have been purchased, so what is the conditional probability
that less than 3/4 of the sodas will be purchased?

P (1/2 < X < 3/4)

P (X < 3/4 | X > 1/2) =
P (X > 1/2)
429/4096
=
1 − FX (1/2)
429/4096
=
1 − 57/64
= 429/448
= 0.9576
 34.3. Exercises 443

34.3 Exercises
34.3.1 Practice

Exercise 34.1. Qualifying exam. The proportion of people who pass a pro-
fessional qualifying exam on the first try has a Beta distribution with α = 3
and β = 4.

a. What is the expected proportion of people who will pass on the first try
at the next exam?
b. What is the standard deviation in the proportion of people who will pass
on the first try at the next exam?

Exercise 34.2. Shelf space. A grocery store chain is trying to decide how
much shelf space to devote to organic produce. If they don’t stock enough, their
more upscale customers will shop at the competing grocery store. If they stock
too much, then much of the expensive organic produce will have to be thrown
out when it expires. The percentage of organic produce which is purchased
during the week after delivery can be modeled with a Beta distribution with
α = 4 and β = 3.

a. What is the expected proportion of organic produce is purchased??

b. What is the standard deviation of the proportion of organic produce that
is purchased?

34.3.2 Extensions

Exercise 34.3. Horse coloration. The proportion of horses with a particular

coloration is modeled by the following function:
(
k(−2x3 + x4 + x2 ) if 0 ≤ x ≤ 1
fX (x) =
0 otherwise

a. Why is this a Beta distribution?

b. What are the parameters?
c. What is the value of k?
d. What is the expected value?

Exercise 34.4. Email checking. The percent of time on a workday that

employees at a company spend checking email can be modeled by the following
function: (
k(1 − 3x + 3x2 − x3 ) if 0 ≤ x ≤ 1
fX (x) =
0 otherwise
444 Chapter 34. Beta Random Variables

a. Why is this a Beta distribution?

b. What are the parameters?
c. What is the value of k?
d. What is the expected value?
Chapter 35

Normal Random Variables

Sir,—It has been wittily remarked that there are three kinds of falsehood: the
first is a ‘fib,’ the second is a downright lie, and the third and most aggravated
is statistics.
—“Letter to the Editor” of The National Observer, written by T. Mackay,
dated June 8, 1891, published June 13, 1891

What is the probability that a randomly chosen student is more than 6 feet tall?
What is the probability the randomly selected student will be between 5 and 6
feet tall? What is the cutoff length that separates the tallest 5% of people from
the rest of the population?

fX (x)
0.4

0.2
x
−4 −2 2 4

Figure 35.1: The density of a standard Normal random variable.

35.1 Introduction
We encounter Normal random variables in many situations. The shape of the
density of a standard Normal random variable, seen in Figure 35.1, is the famil-
iar bell curve that is so often associated with randomly distributed quantities

445
 446 Chapter 35. Normal Random Variables

and measurements. Many things that are clustered near their expected value
The density of a are Normally distributed. In fact, the bell curve looks like a bell because most
Normal random of the density is concentrated in a bell shape near the expected value. Also,
variable is often as with a bell, only a relatively small amount of the density is concentrated far
referred to as a bell from the center.
curve.
Weights and heights of many animals, plants, as well as manufactured prod-
ucts of all shapes and sizes, are often Normally distributed. The growth of
plants, the volumes of liquids in bottles, and many other biological, physical,
and financial random variables are Normally distributed.
Normal random variables
The common thread: Concentration (often of some kind of mea-
surement) near the expected value, so the distribution is like a bell
curve. For instance, the weight, height, lifespan, and intelligence scores of
living things are often Normally distributed. The actual volume of soda in a
can filled by a machine in a factory, the high temperatures for June 17th in
a given community, and many other variables in the biological, physical, and
financial world, are all Normally distributed.
Things to look for: Random quantities that are “close” to their expected
values a relatively large portion of the time.
The variable:

X = the actual height, volume, weight, score, etc.

The parameters:

µX = E(X) = the expected value

2
σX = the variance

Sometimes the standard deviation σX is given instead.

Density:
2 2
e−(x−µX ) /(2σ )
fX (x) =√ , −∞ < x < ∞
2πσ 2
Expected value formula:
E(X) = µX
Variance formula:
2
Var(X) = σX

Perhaps Normal random variables are the most prevalent kind of random
variables in everyday applications. The Normal distribution is often referred
to as the Gaussian distribution. (It is named for the 19th century mathemati-
cian Gauss; http://en.wikipedia.org/wiki/Carl_Friedrich_Gauss .) Since
 35.1. Introduction 447

we experience Normal random variables every day, why have we not practiced
working with the density of the Normal distribution yet? The reason is this:
The Normal distribution has a density that we can easily write down, but we
cannot easily integrate the density. In fact, we do not have any closed-formed
way to express probabilities that arise from the Normal distribution. Similarly,
we cannot even write down the cumulative distribution function of the Normal
distribution with a closed-formula.
For Normal random variables, with parameters µX and σX 2 , as we will see

in Section 35.2, the expected value is exactly E(X) = µX and the variance is
exactly Var(X) = σX 2 . As we will see in later chapters, the Normal distribution

plays a crucial role in describing the asymptotic behavior of the sum and of the
average of a large collection of random variables that are either independent
or loosely dependent. Many limiting theorems have been discovered that are
associated with the Normal distribution.
Another nice property of Normal random variables is that the sum of inde-
pendent Normal random variables is also a Normal random variable (this will
be established in the next chapter), and also if X is Normal, then aX + b is
Normal too (to be seen in Section 35.2). These properties of Normal random
variables make them very desirable to work with, even though we must consult
a chart to look up the probabilities associated with Normal random variables
(since we cannot integrate the density by hand).
The notation for a random variable X with Normal distribution is:
2
X ∼ N (µX , σX ).
Since there is not a closed formula for the CDF of the Normal distribution,
we are forced to read all of the probabilities for the Normal distribution from
a table. This is not because of any inadequacy in our mathematical ability
or understanding; nobody has such a closed formula. So mathematicians and
statisticians and practitioners do exactly the same thing—or one can use a
calculator (if it has a button for the Normal distribution), or a program on a
computer.
A Normal random variable with parameters 0 and 1 (respectively) is used
so frequently that we have a special name for it: a standard Normal random
variable. When working with Normal random variables, whenever we write Z,
we are referring to a standard Normal random variable.
Definition 35.1. Density of a Standard Normal Random Variable
We say Z is a standard Normal random variable if it has parameters 0 and 1
respectively. A standard Normal random variable Z has density
1 2
fZ (z) = √ e−z /2 for −∞ < z < ∞.
2π
One of the very nice things about Normal random variables—especially from
the perspective of using the Normal distribution—is that we do not need to
 448 Chapter 35. Normal Random Variables

The authors assume calculate integrals in order to get probabilities from the density. We are unable
that a significant to integrate the density in closed form. Instead, we provide a table that gives
portion of our (cumulative) probabilities of the form P (Z ≤ z), for z’s in the range 0.00 to
readers will be 3.09. The table is included in this chapter and also at the back of the book.
taking the Society of
Actuaries P exam/ It might appear that half of the values we need, P (Z ≤ z) for z < 0, are
Casualty Actuarial missing from the table, because we often need to calculate P (Z ≤ z) for negative
Society exam 1; only values of z too! Fortunately, the density of Z is symmetric around µZ = 0 (i.e.,
P (Z ≤ z) for z ≥ 0
about the origin), so we can calculate the missing probabilities by using the
are given.
complementary probabilities. We will practice this technique many times.
We also emphasize that only the probabilities for standard Normal random
variables are given in the table. Of course, for every possible combination
of µX and σX 2 , there is a unique Normal distribution. We cannot provide a

separate Normal table for every possible Normal distribution, so in Section 35.2
we will learn how to standardize or normalize whatever Normal curve we need,
by doing a transformation of any Normal random variable X to the standard
Normal random variable Z.
It is helpful to see how changing µX and σX will affect the shape of fX (x).
The shape is always a bell curve, with area 1 below the curve. The curves always
stretch from negative infinity to positive infinity, but there is never much area
once you get far beyond the middle of the curve. For smaller standard devia-
tions, the curve will be taller and skinnier. For larger standard deviations, the
curve will be shorter and wider. In Figure 35.2, we plot the density fX (x) when
2 = 1 is fixed but µ
σX X takes on values −2, 0, and 1, respectively. Changing
µX just shifts the density to the left or right. The curve with µX = −2 is on
the left; the curve with µX = 0 is in the middle; and the curve with µX = 1 is
on the right.

fX (x)
0.4

0.2
x
−4 −2 2 4

Figure 35.2: Three examples of the density of Normal random variables with
µX = −2 (loosely dashed), µX = 0 (solid line), and µX = 1 (densely dashed),
respectively. In each case, σX
2 = 1.

In Figure 35.3, we see that smaller values of σX give thin, tall densities,
because such densities are very tightly concentrated about µX . In contrast,
large values of σX give wide, short densities, because such densities are not very
concentrated around µX at all.
 35.2. Transforming Normal Random Variables 449

0.8 fX (x)
0.6
0.4
0.2
x
−8 −6 −4 −2 2 4 6 8

Figure 35.3: Three examples of the density of Normal random variables with
µX = 0 in each case, and σX = 0.5 (the tall, well concentrated one, drawn with
densely dashed lines), σX = 1 (the standard Normal, drawn with a solid line),
and σX = 3 (the short, very widely distributed one, drawn with loosely dashed
lines), respectively.

35.2 Transforming Normal Random Variables

If b, c are any fixed real numbers, and Z is standard Normal, then bZ + c is a
Normal random variable. To see this, first assume b > 0. We compute
2
(a−c)/b
e−z /2
Z
P (bZ + c ≤ a) = P (Z ≤ (a − c)/b) = √ dz
−∞ 2π
Now we substitute x = bz + c, and thus dx = b dz, to get
2 2)
a
e−(x−c) /(2b
Z
P (bZ + c ≤ a) = √ dx.
−∞ 2πb2
−(x−c)2/(2b2 )
Thus, bZ + c has density e √ 2 , so bZ + c is Normal with parameters
2πb
µbZ+c = c and σbZ+c
2 = b2 . Since Z is symmetric about 0, then −Z has the
same distribution as Z, i.e., both Z and −Z are standard Normal random
variables. Thus b(−Z) + c = −bZ + c also is Normal with the same parameters
as bZ +c. So every random variable of the form X = bZ +c is Normal (regardless
of whether b is positive or negative), with µX = c and σX
2 = b2 .

Theorem 35.2. If Z is a standard Normal random variable, and b, c are

any constants, then bZ + c is a Normal random variable with mean c and
variance c2 .

We claimed that if X has the density

2 2
e−(x−µ) /(2σ )
fX (x) = √ , −∞ < x < ∞,
2πσ 2
then E(X) = µ and Var(X) = σX 2 , but we did not prove it yet. We prove it

first for the standard Normal random variable Z, i.e., we now prove that Z has
450 Chapter 35. Normal Random Variables

expected value 0 and variance 1 (and thus standard deviation 1 too). To see
this, we note that
Z ∞  −z 2/2 
e
E(Z) = (z) √ dz, (35.1)
−∞ 2π
but Z  e−z 2/2  2
e−z /2
(z) √ dz = − √ .
2π 2π
Taking limits as z → −∞ and z → ∞, we verify that E(Z) = 0.
To see Var(Z) = 1, we write Var(Z) = E(Z 2 ) − (E(Z))2 ; since E(Z) = 0,
then Z ∞  e−z 2/2 
Var(Z) = E(Z 2 ) = (z 2 ) √ dz.
−∞ 2π
 −z2/2 
Using integration by parts, with u = z and du = dz, and dv = (z) e√2π dz
−z 2/2
and v = − e√2π , we obtain
Z ∞  e−z 2/2  2
e−z /2 ∞
Z ∞ 2
e−z /2
Var(Z) = (z 2 ) √ dz = −z √ − − √ dz.
−∞ 2π 2π z=−∞ −∞ 2π
−z 2/2
The first term is 0 because −z e√2π → 0 as z → −∞ and as z → ∞. The
R∞
second term is 1 because it is equal to −∞ fZ (z) dz, i.e., the integral of the
density of Z over all values of z. Thus Var(Z) = 1. (See Exercise 35.25 for one
additional note.)
More generally, consider a Normal random variable X with parameters µX
and σX 2 . By Theorem 35.2, σ Z + µ
X X is Normal too. A Normal random
variable’s distribution is completely and uniquely specified by the values of its
two parameters. Since X and σX Z + µX have the same expected value and
same variance and are both Normal random variables, then X and σX Z + µX
have the same distribution. Also, σX Z + µX has mean µX and variance σX 2 , so

it follows that the parameters of X are actually the mean and variance of X,
respectively. So we have proved:

Theorem 35.3. If X is any Normal random variable with parameters µX

and σX2 , then X has expected value µ
X and variance σX , and thus standard
2

deviation σX . In particular, a standard Normal random variable has expected

value µZ = 0 and variance σZ2 = 1, and thus standard deviation σZ = 1.

Within our proof, the following useful fact was also established (since the
density of a Normal random variable is completely determined by its two pa-
rameters).
Theorem 35.4. If Z is a standard Normal random variable, and if X is Nor-
mal with expected value µX and variance σX
2 , then X has the same distribution

as σX Z + µX .
 35.2. Transforming Normal Random Variables 451

Thus, every Normal random variable can be written in a standardized form,

also called a normalized form:

Corollary 35.5. Standardizing a Normal random variable

If X is any Normal random variable, with parameters µX and σX 2 , and if Z

is a standard Normal random variable, then X and σX Z + µX have the same

distribution. So X can be “scaled” to a standard Normal random variable, i.e.,
X − µX
is a standard Normal random variable.
σX

As another corollary of Theorem 35.4, suppose X is a Normal random vari-

able and r, s are any two constants. We know X and σX Z + µX have the same
distribution. Thus rX + s and r(σX Z + µX ) + s = rσX Z + (rµX + s) have the
same distribution too. So rX + s is a Normal random variable with standard
deviation rσX and expected value rµX + s.

Theorem 35.6. If X is a Normal random variable and r, s are any two con-
stants, then rX + s is a Normal random variable with standard deviation rσX
and expected value rµX + s.

As an analogy about scaling a Normal random variable: Think about mea-

suring a table in centimeters versus inches. The table is the same length, no
matter how we measure it. We just change the units of measurement. So, for
instance, if X is given in inches, then Y = 2.54X is the same measurement,
given in centimeters, because 1 inch equals 2.54 centimeters. This is compara-
ble to the situation above, with µ = 0 and σ = 2.54. See Figure 35.4 for another
example of scaling a Normal random variable.

0.15 fX (x) 0.45 fX (x)

0.1 0.3

0.05 0.15
x x
−5 5 10 −2 2

Figure 35.4: Scaling a Normal random variable essentially just changes the
labeling in both the x- and y-dimensions. Left: The density of a Normal random
variable with µX = 2 and σX = 3. Right: The density of a standard Normal
random variable, i.e., with µX = 0 and σX = 1. Notice that the shape stays
the same, but the center, spread, and height have changed.
452 Chapter 35. Normal Random Variables

Standard Normal Table

z . 0 . 1 . 2 . 3 . 4 . 5 . 6 . 7 . 8 . 9
0.0 0.5000 0.5040 0.5080 0.5120 0.5160 0.5199 0.5239 0.5279 0.5319 0.5359
0.1 0.5398 0.5438 0.5478 0.5517 0.5557 0.5596 0.5636 0.5675 0.5714 0.5753
0.2 0.5793 0.5832 0.5871 0.5910 0.5948 0.5987 0.6026 0.6064 0.6103 0.6141
0.3 0.6179 0.6217 0.6255 0.6293 0.6331 0.6368 0.6406 0.6443 0.6480 0.6517
0.4 0.6554 0.6591 0.6628 0.6664 0.6700 0.6736 0.6772 0.6808 0.6844 0.6879
0.5 0.6915 0.6950 0.6985 0.7019 0.7054 0.7088 0.7123 0.7157 0.7190 0.7224
0.6 0.7257 0.7291 0.7324 0.7357 0.7389 0.7422 0.7454 0.7486 0.7517 0.7549
0.7 0.7580 0.7611 0.7642 0.7673 0.7704 0.7734 0.7764 0.7794 0.7823 0.7852
0.8 0.7881 0.7910 0.7939 0.7967 0.7995 0.8023 0.8051 0.8078 0.8106 0.8133
0.9 0.8159 0.8186 0.8212 0.8238 0.8264 0.8289 0.8315 0.8340 0.8365 0.8389
1.0 0.8413 0.8438 0.8461 0.8485 0.8508 0.8531 0.8554 0.8577 0.8599 0.8621
1.1 0.8643 0.8665 0.8686 0.8708 0.8729 0.8749 0.8770 0.8790 0.8810 0.8830
1.2 0.8849 0.8869 0.8888 0.8907 0.8925 0.8944 0.8962 0.8980 0.8997 0.9015
1.3 0.9032 0.9049 0.9066 0.9082 0.9099 0.9115 0.9131 0.9147 0.9162 0.9177
1.4 0.9192 0.9207 0.9222 0.9236 0.9251 0.9265 0.9279 0.9292 0.9306 0.9319
1.5 0.9332 0.9345 0.9357 0.9370 0.9382 0.9394 0.9406 0.9418 0.9429 0.9441
1.6 0.9452 0.9463 0.9474 0.9484 0.9495 0.9505 0.9515 0.9525 0.9535 0.9545
1.7 0.9554 0.9564 0.9573 0.9582 0.9591 0.9599 0.9608 0.9616 0.9625 0.9633
1.8 0.9641 0.9649 0.9656 0.9664 0.9671 0.9678 0.9686 0.9693 0.9699 0.9706
1.9 0.9713 0.9719 0.9726 0.9732 0.9738 0.9744 0.9750 0.9756 0.9761 0.9767
2.0 0.9772 0.9778 0.9783 0.9788 0.9793 0.9798 0.9803 0.9808 0.9812 0.9817
2.1 0.9821 0.9826 0.9830 0.9834 0.9838 0.9842 0.9846 0.9850 0.9854 0.9857
2.2 0.9861 0.9864 0.9868 0.9871 0.9875 0.9878 0.9881 0.9884 0.9887 0.9890
2.3 0.9893 0.9896 0.9898 0.9901 0.9904 0.9906 0.9909 0.9911 0.9913 0.9916
2.4 0.9918 0.9920 0.9922 0.9925 0.9927 0.9929 0.9931 0.9932 0.9934 0.9936
2.5 0.9938 0.9940 0.9941 0.9943 0.9945 0.9946 0.9948 0.9949 0.9951 0.9952
2.6 0.9953 0.9955 0.9956 0.9957 0.9959 0.9960 0.9961 0.9962 0.9963 0.9964
2.7 0.9965 0.9966 0.9967 0.9968 0.9969 0.9970 0.9971 0.9972 0.9973 0.9974
2.8 0.9974 0.9975 0.9976 0.9977 0.9977 0.9978 0.9979 0.9979 0.9980 0.9981
2.9 0.9981 0.9982 0.9982 0.9983 0.9984 0.9984 0.9985 0.9985 0.9986 0.9986
3.0 0.9987 0.9987 0.9987 0.9988 0.9988 0.9989 0.9989 0.9989 0.9990 0.9990
3.1 0.9990 0.9991 0.9991 0.9991 0.9992 0.9992 0.9992 0.9992 0.9993 0.9993
3.2 0.9993 0.9993 0.9994 0.9994 0.9994 0.9994 0.9994 0.9995 0.9995 0.9995
3.3 0.9995 0.9995 0.9995 0.9996 0.9996 0.9996 0.9996 0.9996 0.9996 0.9997
3.4 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9998
3.5 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998
3.6 0.9998 0.9998 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999
For a standard Normal random variable Z, these are the values of the cu-
mulative distribution function FZ (z) = P (Z ≤ z), also corresponding to the
area under the density fZ to the left of z.

fZ (z)
0.4
For example, in this graph,
the area under the curve is
P (Z ≤ 0.75) = 0.7734.
z
−2 2
 35.2. Transforming Normal Random Variables 453

Now we practice using the standard Normal table. We have four types of
ways to use the table. In every case, we encourage you to quickly sketch a
Normal curve—accuracy is not required at all—because we just want you to see
whether the probability that results is bigger or smaller than 1/2. This will be
easy to tell when looking at the picture, because either more or less than 1/2 of
the area under the curve is shaded. Remember that the probability is just the
area under the curve of the density function.
In the next several examples, we will demonstrate how to use the standard
Normal table to calculate probabilities in several different situations.

Example 35.7. If z ≥ 0 and we want P (Z ≤ z), we just directly look up

P (Z ≤ z) in the table. For instance, in Figure 35.5 P (Z ≤ 1.24) = 0.8925 is
the area under the curve.

fZ (z)
0.4

z
−4 −2 2 4

Figure 35.5: The area under the curve is P (Z ≤ 1.24) = 0.8925.

Example 35.8. If z ≥ 0 and we want P (Z > z), we first look up the comple-
ment, P (Z ≤ z), in the table, and then we use P (Z > z) = 1 − P (Z ≤ z). For
instance, see Figure 35.6.

fZ (z) fZ (z)
0.4 0.4

0.2
z z
−4 −2 2 4 −2 2

Figure 35.6: Left: We want the probability P (Z > 0.88), but it is not in
the table. Right: We find P (Z ≤ 0.88) = 0.8106 in the table. So the desired
probability is P (Z > 0.88) = 1 − P (Z ≤ 0.88) = 1 − 0.8106 = 0.1894.
454 Chapter 35. Normal Random Variables

Example 35.9. If z ≤ 0 and we want P (Z ≥ z) (not on the table) we use

symmetry! We just use symmetry and then look up the value in the table. For
instance, both plots in Figure 35.7 have the same shaded area.

fZ (z) fZ (z)
0.4 0.4

z z
−4 −2 2 4 −4 −2 2 4

Figure 35.7: Both plots have the same shaded area. Left: The shaded area
is P (Z ≥ −1.10). Right: The shaded area is P (Z ≤ 1.10). Each one is 0.8643.

Example 35.10. If z ≤ 0 and we want P (Z ≤ z) (not on the table) we use

symmetry and the complement! For instance, the upper left and upper right
plots of Figure 35.8 (same shaded area) are mirror images. The complementary
probability is in the Normal table; it is displayed at the bottom of Figure 35.8.

fZ (z) fZ (z)
0.4 0.4

0.2
z z
−4 −2 2 4 −4 −2 2 4
fZ (z)
0.4

z
−4 −2 2 4

Figure 35.8: Upper left: The shaded area is P (Z ≤ −0.48). Upper right:
The shaded area is P (Z ≥ 0.48) (a mirror image). Bottom: The shaded area
is P (Z ≤ 0.48), which is 0.6844 (see the Normal table). So the upper left and
upper right ones are each P (Z ≥ 0.48) = 1−P (Z ≤ 0.48) = 1−0.6844 = 0.3156.
 35.2. Transforming Normal Random Variables 455

Finally, for the probability of Z occurring in some given range, we need two
separate probabilities. For P (a ≤ Z ≤ b) = P (Z ≤ b) − P (Z ≤ a), we compute
P (Z ≤ b) and P (Z ≤ a) separately.

Example 35.11. As an example of computing the probability of Z in some

range, consider the computation of
P (1.22 ≤ Z ≤ 1.92) = P (Z ≤ 1.92) − P (Z ≤ 1.22) = 0.9726 − 0.8888 = 0.0838.
The analogous computation is given in Figure 35.9.

fZ (z) fZ (z)
0.4 0.4

z z
−4 −2 2 4 −4 −2 2 4
fZ (z)
0.4

0.2
z
−4 −2 2 4

Figure 35.9: Step-by-step method for computing P (1.22 ≤ Z ≤ 1.92) =

P (Z ≤ 1.92) − P (Z ≤ 1.22) = 0.9726 − 0.8888 = 0.0838 in stages. Upper left:
The shaded area is P (Z ≤ 1.92) = 0.9726. Upper right: The shaded area is
P (Z ≤ 1.22) = 0.8888. Bottom: The shaded area is P (1.22 ≤ Z ≤ 1.92) =
P (Z ≤ 1.92) − P (Z ≤ 1.22) = 0.9726 − 0.8888 = 0.0838.

Example 35.12. Consider P (−1.51 ≤ Z ≤ −0.57), in which neither value

seems to be on the table. By symmetry, this is equal to P (0.57 ≤ Z ≤ 1.51).
We first turn the range of negative numbers into a range of positive numbers,
and then use the Normal table to get the relevant values.
P (−1.51 ≤ Z ≤ −0.57) = P (0.57 ≤ Z ≤ 1.51)
= P (Z ≤ 1.51) − P (Z ≤ 0.57)
= 0.9345 − 0.7157
= 0.2188.
456 Chapter 35. Normal Random Variables

The plots of the relevant Normal curves are given in Figures 35.10 and 35.11.

fZ (z) fZ (z)
0.4 0.4

0.2
z z
−4 −2 2 4 −4 −2 2 4

Figure 35.10: Left: The shaded area is P (−1.51 ≤ Z ≤ −0.57). Right: The
shaded area is P (0.57 ≤ Z ≤ 1.51) (mirror image).

fZ (z) fZ (z)
0.4 0.4

z z
−4 −2 2 4 −4 −2 2 4

Figure 35.11: The probability P (0.57 ≤ Z ≤ 1.51) = P (Z ≤ 1.51) −

P (Z ≤ 0.57) = 0.9345 − 0.7157 = 0.2188 is the same area under the curve as in
Figure 35.10.

Example 35.13. Consider P (−0.50 ≤ Z ≤ 1.50), in which only one value is

on the table. The value that is not on the table will have to be handled with
symmetry. We decompose the desired probability as:
P (−0.50 ≤ Z ≤ 1.50) = P (Z ≤ 1.50) − P (Z ≤ −0.50).
The latter probability, P (Z ≤ −0.50), is not in the Normal table. So we use a
mirror image and then a complementary probability:
P (Z ≤ −0.50) = P (Z ≥ 0.50)
= 1 − P (Z ≤ 0.50)
= 1 − 0.6915
So the entire calculation is:
P (−0.50 ≤ Z ≤ 1.50) = P (Z ≤ 1.50) − P (Z ≥ 0.50)
= P (Z ≤ 1.50) − (1 − P (Z ≤ 0.50))
= 0.9332 − (1 − 0.6915)
= 0.6247
 35.2. Transforming Normal Random Variables 457

The shaded area corresponding to the probability P (−0.50 ≤ Z ≤ 1.50) =

0.6247 is given in Figures 35.12.

fZ (z)
0.4

z
−4 −2 2 4

Figure 35.12: The area under the curve is P (−1.51 ≤ Z ≤ −0.57) =

P (0.57 ≤ Z ≤ 1.51) = P (Z ≤ 1.51) − P (Z ≤ 0.57) = 0.9345 − 0.7157 = 0.2188.

Example 35.14. It is helpful to know the probability that X is within one

or two or three standard deviations of its expected value, µX . The probability
that X is within one standard deviation of its µX is

P (µX − σX ≤ X ≤ µX + σX )
 
µX − σX − µX X − µX µ X + σ X − µX
=P ≤ ≤
σX σX σX
= P (−1 ≤ Z ≤ 1)
= P (Z ≤ 1) − P (Z ≤ −1)
= P (Z ≤ 1) − P (Z ≥ 1)
= P (Z ≤ 1) − (1 − P (Z ≤ 1))
= 0.8413 − (1 − 0.8413)
= 0.6826

The probability that X is within two standard deviations of µX is If X is between

µX − 2σX and
P (µX − 2σX ≤ X ≤ µX + 2σX ) µX + 2σX with
  probability 95%,
µX − 2σX − µX X − µX µX + 2σX − µX
=P ≤ ≤ how often is X
σX σX σX above µX + 2σX ?
= P (−2 ≤ Z ≤ 2)
= P (Z ≤ 2) − P (Z ≤ −2)
= P (Z ≤ 2) − P (Z ≥ 2)
= P (Z ≤ 2) − (1 − P (Z ≤ 2))
= 0.9772 − (1 − 0.9772)
= 0.9544
458 Chapter 35. Normal Random Variables

The probability that X is within three standard deviations of µX is

P (µX − 3σX ≤ X ≤ µX + 3σX )

 
µX − 3σX − µX X − µX µX + 3σX − µX
=P ≤ ≤
σX σX σX
= P (−3 ≤ Z ≤ 3)
= P (Z ≤ 3) − P (Z ≤ −3)
= P (Z ≤ 3) − P (Z ≥ 3)
= P (Z ≤ 3) − (1 − P (Z ≤ 3))
= 0.9987 − (1 − 0.9987)
= 0.9974

fZ (z) fZ (z) fZ (z)

0.4 0.4 0.4

z z z
−2 2 −2 2 −2 2

Figure 35.13: Left: The shaded area shows P (µX − σX ≤ X ≤ µX + σX ) =

P (−1 ≤ Z ≤ 1) = 0.6826. Middle: The shaded area shows P (µX − 2σX ≤
X ≤ µX + 2σX ) = P (−2 ≤ Z ≤ 2) = 0.9544. Right: The shaded area shows
P (µX − 3σX ≤ X ≤ µX + 3σX ) = P (−3 ≤ Z ≤ 3) = 0.9974.

Remark 35.15. For quick calculations—the Empirical Rule

People sometimes use the “68–95–99.7% rule” for Normal curves. This means
that if your distribution is Normal-shaped, then approximately:
68% of the distribution will fall between µX − σX and µX + σX . 95% of
the distribution will fall between µX − 2σX and µX + 2σX . 99.7% of the
distribution will fall between µX − 3σX and µX + 3σX .

Now we consider some examples that require us to convert Normal random

variables (which are not standard) into standard Normal random variables. We
will first convert the Normal random variable X to a standard Normal random
variable Z by using Z = X−µ σX . Then we will use the appropriate inequality
X

changes to make the form match the Normal table. Finally, we will look up the
probability from the table for the standard Normal distribution.
 35.2. Transforming Normal Random Variables 459

Example 35.16. Checking account balances at a bank are approximately Nor-

mally distributed with expected value $1325 and standard deviation $250. Bill
has a balance of $775. For each question below, we we give the related curve
for the standard Normal distribution with the relevant area shaded.

a. What is the probability that an account will have less money than Bill’s
account?
We let X ∼ N (µX = 1325, σX 2 = 62500) be the amount in the selected

account. We compute
 
X − µX 775 − 1325
P (X < 775) = P <
σX 250
= P (Z < −2.20)
= P (Z > 2.20)
= 1 − P (Z ≤ 2.20)
= 1 − 0.9861
= 0.0139.

See Figure 35.14 (left side) to visualize this probability.

fZ (z) fZ (z)
0.4 0.4

0.2 0.2
z z
−4 −2 2 4 −4 −2 2 4

Figure 35.14: Left: The shaded area is P (Z ≤ −2.20) = 0.0139. Right: The
shaded area is P (Z > 2.20) = 0.0139.

b. What is the probability that an account will have more than $1875?
Again let X ∼ N (1325, 62500). Then

 
X − µX 1875 − 1325
P (X > 1875) = P >
σX 250
= P (Z > 2.20)
= 1 − P (Z ≤ 2.20)
= 1 − 0.9861
= 0.0139.

See Figure 35.14 (right side) to visualize this as the shaded area under the
Normal curve.
460 Chapter 35. Normal Random Variables

c. What is the probability that an account will have exactly $1875?

Again let X ∼ N (1325, 62500). Then P (X = 1875) = 0, because the
probability that any continuous random variable equals a particular value is
always zero (this holds true in general, not just for Normal random variables).
d. What is the probability that an account will have less than $1325, which
is the expected value of the money in an account?
Since the density of Normal random variables are always symmetric about
the expected value, we do not need a calculator for this value. We know that
a Normal random variable is less than its expected value exactly half the time,
so the desired probability is 1/2. If we want to calculate this quantity formally,
we can again let X ∼ N (1325, 62500), and then:

 
X − µX 1325 − 1325
P (X < 1325) = P <
σX 250
= P (Z < 0)
= 0.5000

See Figure 35.15 to visualize the probability.

fZ (z)
0.4

z
−4 −2 2 4

Figure 35.15: The probability P (Z < 0) = 0.5000 is the area under the
curve.

e. What is the probability that an account will have less than $10?
Again let X ∼ N (1325, 62500). Then

 
X − µX 10 − 1325
P (X < 10) = P <
σX 250
= P (Z < −5.26)
=0

The reason that P (Z < −5.26) ≈ 0 is that Z is very well concentrated about
the mean; the probability P (Z < z) is practically 0 for relatively small values
of z, e.g., for z < −3.
 35.3. “Backward” Normal Problems 461

f. What is the probability that an account will have between $1075 and
$1825?
Again let X ∼ N (1325, 250). Then

 
1075 − 1325 X − µX 1825 − 1325
P (1075 < X < 1825) = P < <
250 σX 250
= P (−1 < Z < 2)
= P (Z < 2) − P (Z < −1)

The probability P (−1 < Z < 2) is given in Figure 35.16.

fZ (z)
0.4

z
−4 −2 2 4

Figure 35.16: The probability P (−1 < Z < 2) = 0.8185 is the area under
the curve.

We can look up P (Z < 2) in the table directly, i.e., P (Z < 2) = 0.9772.

On the other hand, we must manipulate P (Z < −1) learning the methods we
learning earlier:

P (Z < −1) = P (Z > 1) = 1 − P (Z < 1) = 1 − 0.8413 = 0.1587.

So we get
P (1075 < X < 1825) = 0.9772 − 0.1587 = 0.8185.
So an account has between $1075 and $1825 with probability 0.8185.

35.3 “Backward” Normal Problems

In the previous problems, we knew the x-value and we wanted to find the
probability P (X ≤ x) or P (X ≥ x).
What if you know the probability or percentile (for example, “top quarter”
or “bottom tenth” or “middle 50%”), but you don’t know the cut-off x-value that
will give you this probability? We will call this situation a “backward” Normal This is like working
problem because you solve for the x-value in a procedure that is backward from from the inside to
what you did in the previous types of problems in this chapter. the outside.
 462 Chapter 35. Normal Random Variables

1. Start with P (Z ≤ z) as the given probability. Work backward from the

probability in Normal table to get the desired corresponding z value.

2. Use the complement if necessary, e.g., P (Z > z) = 0.2 is the same as

P (Z < z) = 1 − 0.2 = 0.8.
If you have a two-sided probability, use

P (−z < Z < z) = P (Z < z) − P (Z < −z)

= P (Z < z) − P (Z > z)
= P (Z < z) − (1 − P (Z < z))
= 2P (Z < z) − 1 (35.2)

We rearrange 3. Convert the z to x by converting with x = µX + zσX .

Z = (X − µX )/σX
to solve for X.

Example 35.17. Consider the checking account scenario in Example 35.16,

in which the balances are approximately Normally distributed with a mean of
$1325 and a standard deviation of $250.

a. Find the range for the top 15% of balances.

We begin with P (Z > z) = 0.15, but again this is not in the table for the
Normal distribution, so we must use the complement. We have

0.15 = P (Z > z) = 1 − P (Z ≤ z),

as shown on the left side of Figure 35.17. Therefore, we also have

P (Z ≤ z) = 1 − 0.15 = 0.85,

as shown on the right side of Figure 35.17. Looking in the chart, this means
z = 1.04. We still need to convert from Z, the number of standard deviations

fZ (z) fZ (z)
0.4 0.4

0.2
z z
−4 −2 2 4 −4 −2 2 4

Figure 35.17: Left: The shaded area is P (Z > z) = 0.15. Right: The
shaded area is P (Z ≤ z) = 0.85.
 35.3. “Backward” Normal Problems 463

away from the mean our account balance is, to actual account balances.

x0 = µX + zσX
x0 = 1325 + (1.04)(250)
x0 = 1585

Now we can put this back into our inequality expression:

0.15 = P (Z > 1.04) = P (X > 1585).

b. Find the range for the bottom 20% of balances.

We start with P (Z ≤ z) = 0.20, but none of the probabilities on the table
for the Normal distribution (at the end of this chapter) are smaller than 0.5, so
we must use the complement. The technique is the same as in Example 35.10.
We describe this in a series of three plots, in Figure 35.18.

fZ (z) fZ (z)
0.4 0.4

0.2
z z
−4 −2 2 4 −4 −2 2 4
fZ (z)
0.4

z
−4 −2 2 4

Figure 35.18: Upper left: The shaded area is P (Z ≤ z) = 0.20. Upper

right: The shaded area is P (Z ≥ −z) = 0.20 (a mirror image). Bottom: The
shaded area is P (Z ≤ −z) = 0.80, so −z = 0.84 (see the Normal table).
So the upper left area is P (Z ≤ −0.84) = 0.20, and the upper right area is
P (Z ≥ 0.84) = 0.20.

We know that z will be negative because the desired probability P (Z ≤ z) = When we use the
0.20 is less than 0.5 (i.e., is not found on the chart for the standard Normal Normal table to
table); see Figure 35.18 (left side). The probability is the same if we look at work backward, we
the mirror image, but now our cut-off value is −z, i.e., P (Z ≥ −z) = 0.20; go from the inside
(probabilities) to
see Figure 35.18 (middle). Finally, we need to take the complement of this
outside (z-values).
probability so that we can look up the value on the table, i.e.,

0.20 = P (Z ≤ z) = P (Z ≥ −z) = 1 − P (Z ≤ −z).

See Figure 35.18 (right side).

464 Chapter 35. Normal Random Variables

Thus
P (Z ≤ −z) = 1 − 0.20 = 0.80
So the Normal table gives us −z = 0.84, i.e., z = −0.84. Now we use this
information about the standard Normal random variable Z to convert the cutoff
from z’s into x’s. We can relate the cutoffs x and z such that P (Z ≤ z) = 0.20 =
P (X ≤ x) as follows: x = µX + zσX , so we get

x = 1325 + (−0.84)(250) = 1115.

Thus, 20% of the balances are less than $1115, i.e.,

P (Z ≤ −0.84) = 0.20 = P (X ≤ 1115).

Example 35.18. Between what two central values do 40% of the balances fall?

We have

0.40 = P (−z ≤ Z ≤ z)
= 2P (Z < z) − 1 by (35.2)

So 2P (Z < z) = 1.40, and thus P (Z < z) = 0.70. So the table tells us z = 0.52,
or perhaps z = 0.53. We could compromise and use z = 0.525. Thus, the upper
balance is
X = µX + σX z = 1325 + (250)(0.525) = 1456.25,
and the lower balance is

X = µX + σX z = 1325 − (250)(0.525) = 1193.75.

35.4 Summary: “Forward” Vs. “Backward” Normal

Forward:
You know x and you are looking for a probability of percentage, e.g., find
P (X ≥ 2). Convert X to Z at beginning, using Z = (X − µX )/σX . Use the
Normal table from the outside (z-values) to the inside (probabilities).
Backward:
You know the probability or percentage, and you want x, e.g., such that
P (X ≥ x) = 0.25. Use the Normal table from the inside (probabilities) to the
outside (z-values). Convert Z to X at the end, using X = µX + σX Z.
 35.5. Exercises 465

35.5 Exercises
35.5.1 Practice
Exercise 35.1. Use the Normal table to find the following probabilities starting
from Z. Also sketch a standard Normal curve, and shade the region correspond-
ing to the given probability.

a. P (Z < 1.47)
b. P (Z > 1.47)
c. P (Z < −1.47)
d. P (Z > −1.47)
e. P (Z = −1.47)

Exercise 35.2. Use the Normal table to find the following probabilities starting
from Z. Also sketch a standard Normal curve, and shade the region correspond-
ing to the given probability.

a. P (Z ≤ 0.19)
b. P (Z ≤ 1.90)
c. P (Z ≥ 9.10)
d. P (0.19 < Z < 1.90)
e. P (Z ≥ −1.90)
f. P (Z = −1.90)

Exercise 35.3. Assume that X has a Normal distribution with a mean of 2

and a standard deviation of 3. Use the Normal table to find the following
probabilities starting from X. Also sketch a standard Normal curve, and shade
the region corresponding to the given probability.

a. P (X < 1.62)
b. P (X > −8.49)
c. P (−4 < X < 1)

Exercise 35.4. Assume that X has a Normal distribution with a mean of −1.32
and a standard deviation of 0.34. Use the Normal table to find the following
probabilities starting from X. Also sketch a standard Normal curve, and shade
the region corresponding to the given probability.

a. P (X > −2)
b. P (X < 2.56)
c. P (1.47 < X < 4.12)
466 Chapter 35. Normal Random Variables

Exercise 35.5. Use the standard Normal table to find the following cut-off
values for Z. Also sketch a standard Normal curve, and shade the region corre-
sponding to the given probability.

a. P (Z < z) = 0.95
b. P (Z > z) = 0.15
c. P (−z < Z < z) = 0.65

Exercise 35.6. Use the standard Normal table to find the following cut-off
values for Z. Also sketch a standard Normal curve, and shade the region corre-
sponding to the given probability.

a. P (Z < z) = 0.5
b. P (Z > z) = 0.87
c. P (−z < Z < z) = 0.25

Exercise 35.7. Assume that X has a Normal distribution with a mean of 2 and
a standard deviation of 3. Use the standard Normal table to find the following
cut-off values for X. Also sketch a standard Normal curve, and shade the region
corresponding to the given probability.

a. P (X < x) = 0.78
b. P (X > x) = 0.21
c. What are the two central values such that 90% of the X values are in the
range between these two numbers?

Exercise 35.8. Assume that X has a Normal distribution with a mean of −1.32
and a standard deviation of 0.34. Use the standard Normal table to find the
following cut-off values for X. Also sketch a standard Normal curve, and shade
the region corresponding to the given probability.

a. What is the cut-off for the top 5% of X values?

b. What is the cut-off for the bottom 10% of X values?
c. What are the boundaries for the middle 50% of X values?

Exercise 35.9. Find the value of a so that, if Z is a standard Normal random

variable, then
P (a ≤ Z ≤ 0.54) = 0.3898.

Exercise 35.10. Exam scores. The students in my class have Exam 1 scores
which are Normally distributed with a mean of 75 and a standard deviation of
9. If a student is selected at random,

a. What is the probability the student will have a score of more than 90 (an
A)?
 35.5. Exercises 467

b. What is the probability the student will have a score of less than 60 (an
F)?
c. What is the probability a student will have a score between 80 and 89
(the B range)?
d. What is the range of scores for the middle 50% of the student scores?
e. What is the range for the central 99.7% of the scores?
f. What is the lower cut-off for the top 3/4 of the students?
Exercise 35.11. Sugary candy. The quantity of sugar X (measured in grams)
in a randomly selected piece of candy is Normally distributed, with expected
value E(X) = µX = 22 and variance Var(X) = σX 2 = 8. Find the probability

that a randomly selected piece of candy has less than 20 grams of sugar.
Exercise 35.12. Annual precipitation. Assume that the annual precipi-
tation in a student’s hometown is Normally distributed, with expected value
µX = 36.3 inches and variance σX 2 = 8.41. A rare species of frog lives in the

town. This rare species of frog is known to reproduce during the year only if
the annual precipitation is between 35 and 39 inches. What is the probability
that the species of frog is able to reproduce this year?
Exercise 35.13. Getting to class. The distance a student lives (in miles)
from their probability classroom is approximately Normally distributed with a
mean of 3 miles and a standard deviation of 1.2 miles.
a. How far away do the closest 10% of students live?
b. What is the probability that a student will live too close to get a parking
permit (less than 1 mile away)?
c. What is the probability that a student will live further away than 5 miles
or less than 1 mile away?
Exercise 35.14. Movie length. Children’s movies run an average of 98 min-
utes with a standard deviation of 10 minutes. You check out a movie, selected
at random without reading the running time on the box, from the library to
entertain your kids so you can study for your probability test. Assume that
your kids will be occupied for the entire length of the movie.
a. What is the probability that your kids will be occupied for at least the 2
hours you would like to study?
b. What is range for the bottom quartile (lowest 25%) of time they will be
occupied?
c. What are the limits for the central 95% of times your kids will be occu-
pied?
Exercise 35.15. Weighing beagles. Let X be the weight (in pounds) of a
beagle. Then X is Normally distributed with µX = 17.2 pounds and σX
2 = 3.2.

Find the probability that a beagle weighs 20 pounds or less.

468 Chapter 35. Normal Random Variables

Exercise 35.16. Weighing elephants. A certain type of elephant has aver-

age weight 11,000 pounds, with standard deviation of 1,000 pounds. Find the
probability that such an elephant’s weight exceeds 13,000 pounds.

Exercise 35.17. Can of soda. The quantity of liquid in a can of soda is

Normally distributed with average 11.92 ounces and standard deviation 0.05
ounces. Find the probability that the can of soda contains 12 ounces of liquid
or more.

Exercise 35.18. Blonde hairs. Suppose that a randomly chosen blonde-

haired woman has 140,000 hairs on their head, with standard deviation 20,000.
Find the probability that such a person has fewer than 150,000 hairs on her
head.

Exercise 35.19. Sunflower heights. A full-grown sunflower stands 12 feet

tall, on average, with standard deviation of 1 foot. Find the probability that a
sunflower is between 11 to 13 feet tall.

Exercise 35.20. Weighing pumpkins. In a certain field, the weight of pump-

kins is 14 pounds, with standard deviation of 3.7 pounds. What is the proba-
bility of finding a pumpkin that weighs 15 pounds or more?

Exercise 35.21. Physics exams. In a large physics course, students have

average score of 82 percent, with standard deviation of 5 percent. Find the
probability that a randomly chosen student’s score exceeds 90 percent.

Exercise 35.22. Drag race. In a quarter-mile drag race, the average time of
completion is 13.2 seconds, with standard deviation of 0.11 seconds. Find the
probability that a car completes the race in 13 seconds or less.

Exercise 35.23. Weighing cereal. Chocolate-coated Sugar Fun Blast cereal

is filled into boxes by weight, with the weight approximately Normally dis-
tributed with an average of 16 ounces and a standard deviation of 0.2 ounces.

a. What is the probability that the box you buy (chosen at random) will
weigh less than 15.5 ounces?
b. What is the probability that the box you buy will weigh between 16 and
16.25 ounces?
c. What is the probability that the box you buy will weigh more than 17
ounces?
d. What is the upper cut-off for the 90th percentile (bottom 90%) of weights?
e. What is the range for the middle 60% of weights?
f. What is the range for the upper 2.5%?
 35.5. Exercises 469

35.5.2 Extensions
Exercise 35.24. Female heights. Assume that the height of an American
female is Normal with expected value µ = 64 and standard deviation σ = 2.5.

a. What is the probability that an American female’s height is 66 inches or

taller?
b. The heights of 10 American females are measured (in inches). Let Y be
the number of the 10 females whose height is 66 inches or taller. Find P (Y = 7).
What is the distribution of Y ?

35.5.3 Advanced
Exercise 35.25. We have not actually verified here that the density of a Normal
random variable integrates to 1, because the usual argument to do this requires
a perhaps surprising conversion to polar coordinates and a double integral; see,
for instance, Section
R ∞5.4 of Ross [5] or Section 5.3 of Pitman [4]. Construct such
an argument that −∞ fZ (z) dz = 1.
Chapter 36

Sums of Independent Normal

Random Variables

Probability is a mathematical discipline whose aims are akin to those, for ex-
ample, of geometry of analytical mechanics. In each field we must carefully
distinguish three aspects of the theory:
(a) the formal logical content,
(b) the intuitive background,
(c) the applications.
The character, and the charm, of the whole structure cannot be appreciated
without considering all three aspects in their proper relation.
—An Introduction to Probability Theory and its Applications, Volume 1, 3rd
edition by William Feller (Wiley, 1971)

What is the total amount of food eaten by all of the students in a dining hall on
a given day? In a given month? In a given year? How can we scale our results
to compare the average, variance, and standard deviation of these food totals?

36.1 Sums of Independent Normal Random Variables

We already demonstrated in the last chapter that the Normal random variable
is really useful in practice because many random variables encountered in real
life are Normal random variables. Many quantities associated with physical life
are Normally distributed (as we mentioned, lengths, volumes, widths, heights,
etc., related to all kinds of animals and plants and other physical beings, are
often Normally distributed). On these grounds alone, Normal random variables
are beautiful and important enough to study in a course of probability.

470
 36.1. Sums of Independent Normal Random Variables 471

We emphasize that Normal random variables are also extremely important

for another reason! Informally stated:
When many independent random variables (not necessarily Nor-
mal) are added together, the sum of the random variables is approx-
imately Normal. We will begin to formalize this idea in Chapter 37. Also,
the average of many independent random variables is approximately Normal
too. The study of these “limit laws” is a major area of probability theory, in
which people investigate the limiting properties of random variables. One of
the most important parts of the theory of limit laws is devoted to the study
of Normal random variables and their usefulness in approximating the sum of
random variables and the averages of random variables. In Chapter 37, we will
state some of the most commonly used limit laws and discuss their applications.
Some of the results are useful and pleasantly surprising! Before studying limit
laws, however, we discuss sums of independent Normal random variables.
In the previous chapter, Theorem 35.6 gave us the remarkable fact:
If X is a Normal random variable and r, s are any two constants, then rX + s
is a Normal random variable with standard deviation rσX and expected value
rµX + s.

The following theorem, to be proved in Section 36.2, is also very useful:

Theorem 36.1. The sum of independent Normal random variables is
a Normal random variable too (First Version)
If X1 , X2 , . . . , Xn are independent Normal random variables, with expected
values µ1 , µ2 , . . . , µn , respectively, and with variances σ12 , σ22 , . . . , σn2 , respec-
tively, then

X1 + · · · + Xn is also a Normal random variable

with expected value µ1 + · · · + µn and variance σ12 + · · · + σn2 .

The amazing part of the fact above is that if X1 , . . . , Xn are independent

Normal random variables, then X1 + · · · + Xn is a Normal random variable too.
The expected value must be
E(X1 + · · · + Xn ) = E(X1 ) + · · · + E(Xn ) = µ1 + · · · + µn ;
(this is not surprising at all; this works for all sums of all random variables, with
or without the independent assumption and with or without the assumption
that the random variables are Normal). The variance must be
Var(X1 + · · · + Xn ) = Var(X1 ) + · · · + Var(Xn ) = σ12 + · · · + σn2 ;
(this is not surprising either; this works for all sums of independent random
variables, with or without the assumption that the random variables are Nor-
mal). Once again, we emphasize that the delightful part of the boxed statement
above is that the sum of independent Normals is Normal too.
472 Chapter 36. Sums of Independent Normal Random Variables

In the case in which the Normal random variables are not only independent,
but also all have the same expected values and the same variances, the boxed
formula above has the following nice form:

Corollary 36.2. The sum of independent Normal random variables

is a Normal random variable too (Identically Distributed Version)
If X1 , X2 , . . . , Xn are independent Normal random variables, which each have
expected value µ and variance σ 2 , then

X1 + · · · + Xn is also a Normal random variable

with expected value nµ and variance nσ 2 .

These two nice, boxed results have analogous versions, if we subtract the
expected value from the sum of the random variables and divide by the variance.
In the case of the first version, we get:

Corollary 36.3. The sum of independent Normal random variables,

properly scaled, is a standard Normal random variable (First Ver-
sion)
If X1 , X2 , . . . , Xn are independent Normal random variables, with expected
values µ1 , µ2 , . . . , µn , respectively, and with variances σ12 , σ22 , . . . , σn2 , respec-
tively, then

X1 + · · · + Xn − (µ1 + · · · + µn )
p is a standard Normal random variable.
σ12 + · · · + σn2

In the case of the second version, we get:

Corollary 36.4. The sum of independent Normal random variables,

properly scaled, is a standard Normal random variable (Alternate
Version)
If X1 , X2 , . . . , Xn are independent Normal random variables, which each have
expected value µ and variance σ 2 , then
X1 + · · · + Xn − nµ
√ is a standard Normal random variable.
nσ 2

Example 36.5. In Exercise 35.11, we encountered a type of candy such that the
quantity of sugar X (measured in grams) in a randomly selected piece of candy
is Normally distributed, with expected value E(X) = µX = 22 and variance
Var(X) = σX 2 = 8.
 36.1. Sums of Independent Normal Random Variables 473

Suppose that Hector eats ten pieces of this candy on Halloween night. What
is the probability that he consumed more than 200 grams of sugar?
If X1 , . . . , X10 denote the amount of sugar in the first, second, . . . , tenth
pieces of candy, respectively, then the total amount of sugar that he ate is
X1 + · · · + X10 . We know that the amount of sugar in the jth piece of candy is a
Normal random variable Xj ∼ N (µ = 22, σ 2 = 8), so the total amount of sugar
is also a Normal random variable with expected value 10µ = (10)(22) = 220
and variance 10σ 2 = (10)(8) = 80. So

X1 + · · · + X10 − 220
Z= √
80
is a standard Normal random variable. With this in mind, we compute the
probability that Hector consumed more than 200 grams of sugar. The desired
probability is:

 
X1 + · · · + X10 − 220 200 − 220
P (X1 + · · · + X10 > 200) = P √ > √
80 80
= P (Z > −2.24)
= P (Z < 2.24)
= 0.9875

So there is a 98.75% chance that Hector ate more than 200 grams of sugar
on Halloween night.

Example 36.6. Consider a species of ant whose body weight (in milligrams)
is Normally distributed with mean µ = 5 and variance σ 2 = 1.3.
Suppose that an ant colony contains 100,000 of these ants. What is the
probability that the ant colony weighs more than 500,500 milligrams?

Let X1 , . . . , X100,000 denote the weights of the 100,000 ants. So the total
weight of the colony is X1 + · · · + X100,000 . The weight of each ant is Normally
distributed, with Xj ∼ N (µ = 5, σ 2 = 1.3) for each j. Thus the total weight
of the colony is also a Normal random variable with expected value 100,000µ =
(100,000)(5) = 500,000 and variance 100,000σ 2 = (100,000)(1.3) = 130,000.
We use this information to shift and scale the sum of the weights. The sum of
the weights, after this adjustment, is

X1 + · · · + X100,000 − 500,000
Z= √ ,
130,000

a standard Normal random variable.

474 Chapter 36. Sums of Independent Normal Random Variables

Now we calculate the probability that the colony weighs more than 500,500
milligrams:

 100,000  P100,000
Xj − 500,000
 
X j=1 500,500 − 500,000
P Xj > 500,500 = P √ > √
j=1
130,000 130,000
= P (Z > 1.39)
= 1 − P (Z ≤ 1.39)
= 1 − 0.9177
= 0.0823

Thus, there is an 8.23% chance that the weight of the ant colony exceeds 500,500
milligrams.

Example 36.7. In planning for a new bookstore, a bookseller estimates that

the thickness of each book to be sold is Normally distributed, with expected
value µ = 1.125 inches and standard deviation σ = 0.25 inches.
The bookseller is thinking about the thickness of a book of the month to be on
a feature display at the front of her store. (If the book is too thick, and the
shelf is too thin, the book will fall off the shelf.) She wants to be 95 percent (or
more) sure that the book will fit onto her feature display. How thick must her
feature display be, to guarantee that this month’s book of the month will fit on
it? (The feature display must be as deep—or deeper—than the thickness of the
book itself.)
If X is the thickness of the book of the month, then
X − 1.125
Z=
0.25
is a standard Normal random variable. Let a be the required thickness of the
feature display so that P (X ≤ a) ≥ 0.95. Then we have

0.95 ≤ P (X ≤ a)
 
X − 1.125 a − 1.125
=P ≤
0.25 0.25
 
a − 1.125
=P Z≤
0.25

Looking at the Normal table, we must have a−1.125

0.25 ≥ 1.65, in order for the
probability above to exceed 95%. So we get
a − 1.125
≥ 1.65,
0.25
 36.1. Sums of Independent Normal Random Variables 475

which simplifies to
a ≥ (0.25)(1.65) + 1.125 = 1.5375.
So the feature display must be at least 1.5375 inches deep, in order to accom-
modate the book of the month at least 95% of the time.

Example 36.8. In the scenario from the previous example, within the store
itself, the bookseller wants to be able to place n books on a regular shelf that is
72 inches long, and still be 95 percent (or more) sure that all of these n books
will fit. What is the largest value of n that she can plan to place on each shelf?

Let X1 , X2 , . . . denote the thickness of the books. The total thickness of n

books is X1 + · · · + Xn . We are given that Xj ∼ N (µ = 1.125, σ 2 = 0.0625),
for each j. Thus the total width of the n books on the shelf is also a Normal
random variable with expected value nµ = 1.125n and variance nσ 2 = 0.0625n.
So
X1 + · · · + Xn − 1.125n
Z= √
0.0625n
is a standard Normal random variable. The bookseller wants P (X1 +· · ·+Xn ≤
72) ≥ 0.95. So we calculate:
0.95 ≤ P (X1 + · · · + Xn ≤ 72)
 
X1 + · · · + Xn − 1.125n 72 − 1.125n
=P √ ≤ √
0.0625n 0.0625n
 
72 − 1.125n
=P Z≤ √
0.0625n

Thus, looking at the chart of the Normal distribution, we see that we must have
72 − 1.125n
1.645 ≤ √ ,
0.0625n
or equivalently (rearranging the terms),
√
0 ≥ 1.125n + 1.645 0.0625n − 72.
To find the value of n where equality is reached (i.e., where 0 is obtained)√ in
√
the previous equation, we can write: x = n and a = 1.125, b = 1.645 0.0625,
and c = −72. Then we just need √to remember that the quadratic equation
b2 −4ac √
0 = ax2 +bx+c has solution x = −b± 2a . In this case, we know x (i.e., n) is
√
positive, so we only need the larger of the two solutions. So we get n = 7.8193
and n = 61.1416. Thus, the maximum value of n that the bookseller can use,
and still maintain the desired conditions, is n = 61. In other words, if the
bookseller puts 61 books onto a shelf, she can still be at least 95% confident
that they will all fit onto the 72 inch shelf.
476 Chapter 36. Sums of Independent Normal Random Variables

If X, Y are independent Normal random variables, then −1X = −X is

Normal, and if we add Y , Theorem 36.1 implies that Y − X is Normal too.
Corollary 36.9. The difference of independent Normal random vari-
ables is a Normal random variable too
If X, Y are independent Normal random variables, then Y − X is a Normal
random variable too. We have E(Y − X) = E(Y ) − E(X), and Var(Y − X) =
Var(Y ) + Var(−X) = Var(Y ) + Var(X).

Example 36.10. As in Exercise 35.24, assume that the height (in inches) of an
American female is Normal with expected value µ1 = 64 and standard deviation
σ1 = 2.5. Also assume that the height of an American male is Normal with
expected value µ2 = 69 and standard deviation σ2 = 3.0. Let X denote the
female’s height and Y denote the male’s height. What is the probability that a
randomly selected male is taller than a randomly selected female?
The expected value of Y − X is
E(Y − X) = E(Y ) − E(X) = 69 − 64 = 5.
The variance of Y − X is
Var(Y − X) = Var(Y ) + Var(X) = 3.02 + 2.52 = 15.25.
√
The standard deviation of Y − X is Var(Y − X) = 15.25 = 3.91.
p

The probability that a randomly selected male is taller than a randomly

selected female is
P (Y > X) = P (Y − X > 0)
 
Y −X −5 0−5
=P >
3.91 3.91
= P (Z > −1.28)
= P (Z < 1.28)
= 0.8997

36.2 Why a Sum of Independent Normals Is Normal

(Optional)
Near the start of this Chapter, we claimed that: If X1 , X2 , . . . , Xn are indepen-
dent Normal random variables, with expected values µ1 , µ2 , . . . , µn , respectively,
and with variances σ12 , σ22 , . . . , σn2 , respectively, then
X1 + · · · + Xn is also a Normal random variable
 36.2. Why a Sum of Independent Normals Is Normal (Optional) 477

with expected value µ1 + · · · + µn and variance σ12 + · · · + σn2 .

Here we present all of the details to support that claim.
To see that X1 + · · · + Xn is a Normal random variable, we will show that
X1 + X2 is Normal. This implies that

(X1 + X2 ) + X3 is Normal,

which implies that

(X1 + X2 + X3 ) + X4 is Normal
etc., and by similar reasoning, the sum of n independent Normals is seen to be
Normal. To see that X1 + X2 is Normal, we note that X1 and X2 have the same Note: if σ1 = 0, do
distribution as σ1 Z1 +µ1 and σ2 Z2 +µ2 , where Z1 , Z2 are independent standard not divide by 0;
Normal random variables. So it suffices to show that σ1 Z1 + σ2 Z2 + µ1 + µ2 is instead,
Normal. In fact, using the rule “X is Normal implies αX + β is Normal too,” σ1 Z1 +σ2 Z2 +µ1 +µ2
would simplify to
with α = σ1 and β = µ1 + µ2 , it is enough to show the simpler statement that
σ2 Z2 + µ1 + µ2 and
Z1 + σσ12 Z2 is Normal. For simplicity, we write c = σ2 /σ1 , and we prove that is thus already
Z1 + cZ2 is Normal with expected value 0 and variance 1 + c2 . We work with known to be
the CDF of Z1 + cZ2 to accomplish this. For any real-valued a, we show that Normal, so we’d be
the CDF of Z1 + cZ2 is done.
Z a
t2
 
1
FZ1 +cZ2 (a) = exp − dt.
2(1 + c2 )
p
−∞ 2π(1 + c2 )

The rest of the argument is dedicated to show that FZ1 +cZ2 (a) has this form.
We compute

FZ1 +cZ2 (a) = P (Z1 + cZ2 ≤ a)

Z ∞ Z (a−z1 )/c  2  2
1 z 1 z
= √ exp − 1 √ exp − 2 dz2 dz1 .
−∞ −∞ 2π 2 2π 2

We want to reverse the order of integration, so we must get rid of the z1 ’s in the
range of z2 . Thus, we use the transformation t = z1 + cz2 (i.e., z2 = (t − z1 )/c),
so dt = c dz2 , and in particular, when z2 = (a − z1 )/c, we have t = a. This
yields
Z ∞Z a  2
(t − z1 )2 1
 
1 z1 1
FZ1 +cZ2 (a) = √ exp − √ exp − dt dz1 .
−∞ −∞ 2π 2 2π 2c2 c

Now we expand the (t − z1 )2 and extract the terms that have t’s but no z1 ’s.
We also switch the order of integration. We get
Z ∞Z a  2  2
t − 2tz1 + z12 1

1 z1 1
FZ1 +cZ2 (a) = √ exp − √ exp − dt dz1
−∞ −∞ 2π 2 2π 2c2 c
Z a  2 Z ∞  2
2tz1 − z12

1 −t 1 1 z1
= √ exp √ exp − + dz1 dt.
−∞ 2π 2c2 c −∞ 2π 2 2c2
478 Chapter 36. Sums of Independent Normal Random Variables

We simplify the z1 expression:

z12 2tz1 − z12 c2 z12 − 2tz1 + z12

− + = −
2 2c2 2c2
(1 + c )z12 − 2tz1
2
=−
2c2
2

(1 + c ) 2 2tz1
=− z1 − .
2c2 1 + c2

Then we “complete the square” on the z1 expression:

z12 2tz1 − z12 (1 + c2 ) t2 (1 + c2 ) t2

 
2 2tz1
− + = − z 1 − + + ,
2 2c2 2c2 1 + c2 (1 + c2 )2 2c2 (1 + c2 )2

which simplifies to
2
z 2 2tz1 − z12 (1 + c2 ) t2

t
− 1 + = − z1 − + .
2 2c2 2c2 1 + c2 2c2 (1 + c2 )

Now we substitute back into the expression for the CDF to get
Z a  2
1 −t 1
FZ1 +cZ2 (a) = √ exp 2
−∞ 2π 2c c
 2
 2 
Z ∞ exp − (1+c2 ) z1 − t 2 + 2 t2 2
2c 1+c 2c (1+c )
× √ dz1 dt.
−∞ 2π
Combining the exponential expressions with t’s, using

−t2 t2 t2
+ = − ,
2c2 2c2 (1 + c2 ) 2(1 + c2 )

the CDF expression simplifies to

Z a
t2
 
1 1
FZ1 +cZ2 (a) = √ exp − 2
−∞ 2π 2(1 + c ) c
Z ∞ 2 !
(1 + c2 )

1 t
× √ exp − z1 − dz1 dt.
−∞ 2π 2c2 1 + c2

The expression for the z1 terms almost looks like the integral for the density of
a Normal random variable. To see this more clearly, we write µt = 1+c t
2 and
c2
σ2 = 1+c2
, so that we have
a
t2
Z  
1 1
FZ1 +cZ2 (a) = √ exp − 2
−∞ 2π 2(1 + c ) c
Z ∞  
1 1
× √ exp − 2 (z1 − µt )2 dz1 dt.
−∞ 2π 2σ
 36.3. Exercises 479
√
1+c2
Then we multiply and divide by a factor of c = 1
σ = √1 , to get
σ2
a
t2
Z  
1 1 c
FZ1 +cZ2 (a) = √ exp − 2)
√
−∞ 2π 2(1 + c c 1 + c2
Z ∞  
1 1 2
× √ exp − 2 (z1 − µt ) dz1 dt.
−∞ 2πσ 2 2σ
When evaluating the inner integral, the
 value of2 t (and thus of µt ) is fixed,
−µt )
so that the inner integrand, √ 2 exp − (z12σ
1
2 is exactly the density of a
2πσ
Normal random variable with expected value µt and variance σ 2 . So the inner
integral evaluates to 1, because we are integrating over all z1 . Thus, the entire
expression simplifies to
Z a
t2
 
1
FZ1 +cZ2 (a) = exp − dt.
2(1 + c2 )
p
−∞ 2π(1 + c2 )
So, the CDF of Z1 + cZ2 shows Z1 + cZ2 is Normal with E(Z1 + cZ2 ) = 0 and
Var(Z1 + cZ2 ) = 1 + c2 .

36.3 Exercises
36.3.1 Practice
Exercise 36.1. Haircuts. The time that it takes a random person to get a
haircut is Normally distributed, with an average of 23.8 minutes and a standard
deviation of 5 minutes. Assume that different people have independent times
of getting their hair cut. Find the probability that, if there are four customers
in a row (with no gaps in between), they will all be finished getting their hair
cut in 1.5 hours (altogether) or less.
Exercise 36.2. Annual precipitation. As in Exercise 35.12, assume that
the annual precipitation in a student’s hometown is Normally distributed, with
expected value µ = 36.3 inches and variance σ 2 = 8.41. Also assume that the
amount of precipitation is independent in distinct years.
Let X1 , . . . , X10 denote the precipitation in the 10 distinct years of a decade.
Find the probability that the total rainfall during the decade exceeds 380 inches.
Exercise 36.3. Printer pages. A printer can produce, on average, 30 pages
per minute, i.e., one page every 2 seconds. Each page’s printing time has stan-
dard deviation of 0.3 seconds. If the pages run times are independent, find the
probability that the total print time for a 30 page job is 62 seconds or less.
Exercise 36.4. Counting calories. Let X denote the number of calories
that a person eats in a single day. Suppose that X is Normally distributed with
µX = 2000 and σX 2 = 10,000. If the person’s eating habits are independent

from day to day, find the probability that the person eats 735,000 calories or
more during a 365-day year.
480 Chapter 36. Sums of Independent Normal Random Variables

Exercise 36.5. Customers’ purchases. At a local store, there are 14 cus-

tomers during a given evening. Each customer has an expected purchase of
$4.90, with a standard deviation of $1.50. If the customers’ purchases are as-
sumed to be independent and approximately Normal, what is the approximate
probability that the revenue exceeds $74 on a given evening?
Exercise 36.6. Weighing books. At the library, a student checks out 10
books. She estimates that each book has a Normally distributed weight, with
mean 12 ounces and standard deviation of 3 ounces. What is the probability
that the total weight of the books exceeds 125 ounces altogether?
Exercise 36.7. Candy sticks. A student has 23 candy sticks in a bag, with
lengths that are Normally distributed. Each stick is, on average 1.8 cm long,
with standard deviation 0.5 cm. What is the probability that the total length
of the candy is less than 40 cm?
Exercise 36.8. Waiting for a bus. Dylan waits at the bus stop an average of
6 minutes per day, with a standard deviation of 1.5 minutes. He believes that
his waiting times are approximately Normally distributed.
a. If he waits for the bus on 20 mornings, what is the probability that he
spends more than 115 minutes at the bus stop altogether in the morning?
b. If he takes into account the evening waiting times too, and he believes that
the mornings and evenings both have about the same kind of distributions (e.g.,
independent, Normally distributed, etc.), approximate the probability that he
spends at least 230 minutes at the bus stop altogether in the mornings and the
evenings during the month.
Exercise 36.9. Weighing apples. According to http://www.fowlerfarms.
com/apple_a_day.htm, a certain kind of apple weighs 150 grams, on average.
Suppose that the standard deviation of this type of apple is 20 grams. When
picking 66 such apples, what is the probability that they weigh (altogether)
9966 grams or more?
Exercise 36.10. Waiting for girls. Preparing for their dates on a Friday
night, it takes each girl (approximately) a Normally distributed amount of time
to prepare, with an average of 40 minutes (per girl) to get ready, and a standard
deviation of 15 minutes. The 50 men who are waiting for them in the lobby are
mostly science majors, and they start to wonder: What is the probability that
the average waiting time exceeds 36 minutes (where the average is taken over
the 50 preparation times of the women)?

36.3.2 Extensions
Exercise 36.11. Tree heights. After planting 20 evergreen tree saplings on
his farm, Don wonders about their heights. If the heights are independent and
Normally distributed, with average height 3 feet, and standard deviation of 1.2
feet, what is the probability that none of the 20 trees’ heights exceeds 5 feet?
 36.3. Exercises 481

Exercise 36.12. Distances between cars. On a busy interstate highway,

there are 21 cars in a particular lane. Let X1 , . . . , X20 denote the 20 distances
between these 21 cars (i.e., X1 is the distance between the first and second cars;
X2 is the distance between the second and third cars; etc.). At a particular
moment, the Xj ’s are judged to be approximately Normal, with an average of
500 feet between consecutive cars and standard deviation of 75 feet. Find the
probability that the row of 21 cars is less than two miles long (each mile contains
5280 feet) if:

a. The length of each car is assumed to be negligible, i.e., we do not take

into account the lengths of the 21 cars themselves.
b. The length of each car is assumed to be fixed, i.e., 13.5 feet long. Thus,
we want the probability of

X1 + · · · + X20 + (13.5)(21) < (5280)(2).

c. The lengths of the cars are assumed to be independent and Normally

distributed too, each with average length 13.5 feet and standard deviation 1
foot. In this case, if Y1 , . . . , Y21 are the length of the cars Thus, we want the
probability of

X1 + · · · + X20 + Y1 + · · · + Y21 < (5280)(2).

Exercise 36.13. Long jumps. An athletic director is recording long-jump

scores for a group of students. The expected value of each of their long jumps
is 7 meters, and the standard deviation is 0.2 meters. If 20 such jumps are
independent and approximately Normally distributed, find the probability that
the average of these 20 jumps is between 6.95 and 7.05.

Exercise 36.14. Baking cakes. Roberto and Sally are busy making cakes for
a bake sale. They need to make 35 cakes for the sale. The cooking times of the
cakes are independent and Normally distributed. Sally makes 19 of the cakes
and Roberto makes 16 of them. Sally’s have average cook time of 45 minutes
each, with standard deviation of 3 minutes. Roberto’s cook, on average, 42
minutes each, with standard deviation of 4 minutes. What is the probability
that the average of their 35 baking times is actually between 43 to 44 minutes
altogether?

Exercise 36.15. Losing weight. A new low-carb diet author claims that
people following his plan will lose an average of 3 pounds with a standard
deviation of 1.4 pounds each week. An author of a low-fat diet claims that
people following her plan will lose an average of 2 pounds with a standard
deviation of 1.8 pounds each week. What is the probability that a low-carb
dieter will lose at least a pound more than a low-fat dieter in a week?
482 Chapter 36. Sums of Independent Normal Random Variables

36.3.3 Advanced
Exercise 36.16. Consider a group of n independent, identically distributed
Normal-loss contracts, X1 , . . . , Xn , each with average µ and standard deviation
σ. A company plans to keep an amount of funds R on reserve so that X1 +
· · · + Xn ≤ R with probability 95%, i.e., so that the company is 95% sure that
all of its losses can be covered.

a. If n = 25, find the needed reserve R. What is the value of R/n, i.e., the
reserve needed per contract?
b. Same question, with n = 100.
c. Same question, with n = 10,000.
d. Same question, for general n. (Notice that, for a bigger aggregation of
contracts, the amount of reserve, per contract, gets smaller; in other words, R/n
is a decreasing function of n, which gets closer to the average loss per contract.)
Chapter 37

Central Limit Theorem

If people do not believe that mathematics is simple, it is only because they do

not realize how complicated life is.
—John von Neumann (remark at the first national meeting of the Associa-
tion for Computing Machinery, 1947)

A basketball player plans to practice shooting hoops until she gets at least 10
successful baskets. She’s meeting up with some friends after practice, so she
wants to plan how long that will take, given that she usually makes about 60%
of the baskets she shoots. If she wants to be 95% sure of her planning, how
many baskets should she plan on attempting, to be sure that she has at least
10 successful shots?

37.1 Introduction
Gerolamo Cardano (1501–1576) studied games of chance using methods that
were among the earliest predecessors to the modern discipline of probability.
Five hundred years after Cardano, many others have contributed to the study of
probability as it related to the limiting properties of randomness; these scholars
include: Bernoulli (actually several Bernoullis), Laplace, Poisson, Chebyshev,
Markov, Borel, Cantelli, Kolmogorov, and Khinchin. Probability is a major area
of study that transcends mathematics and has applications in every discipline
that involves randomness or uncertainty or chance. Within the study of prob-
ability theory, the subtopic of limiting properties receives a significant amount
of attention. Many articles and books have been devoted to the study of lim-
iting properties of random phenomena. We will only present some of the most
commonly referenced laws about limits in probability theory. This is just the
first stone off a large mountain of known results about the limiting properties
about random variables.

483
484 Chapter 37. Central Limit Theorem

We do not prove the Weak or Strong Law of Large Numbers or the Central
Limit Theorem. More advanced techniques (for instance, characteristic func-
tions) are needed. See, for instance, Billingsley [1] or Durrett [2] for details.

37.2 Laws of Large Numbers

Theorem 37.1. The Weak Law Of Large Numbers
Consider a sequence of independent random variables X1 , X2 , X3 , . . . that each
have finite expected value µ. Also consider any positive number  > 0. The
weak law of large numbers states that the average of the first n of the Xj ’s
will not be too far from µ. More specifically, the probability that the average
of X1 , . . . , Xn is more than  away from µ will converge to 0, as n → ∞. In
other words,  
X1 + · · · + Xn
lim P − µ <  = 1,
n→∞ n
or equivalently,
 
X1 + · · · + Xn
lim P − µ ≥  = 0,
n→∞ n

So if we fix a small  (for instance,  = 1/1000), then the probability of the

average X1 +···+X
n
n
of the Xj ’s being more than  away from µ will converge to
0 as n → ∞.

Theorem 37.2. The Strong Law Of Large Numbers

Consider a sequence of independent random variables X1 , X2 , X3 , . . . that each
have finite expected value µ. The strong law of large numbers states that the
average of the first n of the Xj ’s will converge as n → ∞ to µ with probability
1. In other words, the probability that the average of X1 , . . . , Xn converges to
µ as n → ∞ is 100%. In other words,
 
X1 + · · · + Xn
P lim = µ = 1.
n→∞ n

(Note: It is possible for a sequence of random variables to converge weakly

(i.e., the average converges in probability) without the sequence of random vari-
ables converging strongly. The opposite is impossible, since strong convergence
implies weak convergence. The Strong Law of Large Numbers is a stronger
statement than the Weak Law of Large Numbers.)

37.3 Central Limit Theorem

The name of the next theorem is a bit of a misnomer, because there are many
versions of central limit theorems. The following is one of the most simple and
widely used:
 37.4. CLT for Sums of Continuous Random Variables 485

Theorem 37.3. The Central Limit Theorem

Consider a sequence of independent random variables X1 , X2 , X3 , . . . that each
have finite expected value µ and finite variance σ 2 . Let Z ∼ N (0, 1) be a
standard Normal random variable. The central limit theorem states that the
probability of the average of X1 ,√. . . , Xn , properly scaled (i.e., with subtrac-
tion of nµ and then division by nσ 2 ), being less than “a,” will converge to
the cumulative distribution function of a standard Normal random variable,
evaluated at a. In other words,
  Z a
X1 + · · · + Xn − nµ 1 2
lim P √ ≤a = √ e−z /2 dz,
n→∞ nσ 2 −∞ 2π
or equivalently,
 
X1 + · · · + Xn − nµ
lim P √ ≤ a = FZ (a).
n→∞ nσ 2

The Central Limit Theorem, often affectionately called the CLT, is perhaps
surprising because we do not need the random variables X1 , X2 , X3 , . . . to be
Normal. This theorem basically says that sums of n independent random vari-
ables (of any type) are distributed similarly to a Normal random variable when
n is large. (There is no minimum n necessary before the CLT applies, but the
CLT is more effective, the larger n is.) This is a truly powerful concept that
is used throughout the sciences and beyond. We give some applications in the
next several examples.

37.4 CLT for Sums of Continuous Random Variables

Example 37.4. Consider the volumes of soda remaining in 100 cans of soda
that are nearly empty. Let X1 , . . . , X100 , denote the volumes (in ounces) of
cans one through one hundred, respectively. Suppose that the volumes Xj are
independent, and that each Xj is Uniformly distributed between 0 and 2.

Find the probability that the 100 cans of soda contain less than 90 ounces
of soda total.
The expected value of Xj is E(Xj ) = (0 + 2)/2 = 1. The variance of Xj is
Var(Xj ) = (2 − 0)2/12 = 1/3. Thus, the expected value of X1 + · · · + X100 is

E(X1 + · · · + X100 ) = (100)(1),

and the variance of X1 + · · · + X100 (since the Xj ’s are independent) is

Var(X1 + · · · + X100 ) = (100)(1/3).

486 Chapter 37. Central Limit Theorem

The content of the Central Limit Theorem, applied to this scenario, is that
X1 + · · · + X100 − (100)(1)
p
(100)(1/3)

is approximately Normally distributed, i.e.,

!
X1 + · · · + X100 − (100)(1)
P p ≤ a ≈ FZ (a).
(100)(1/3)

Suppose we want to find the probability that the total amount is no more than
90 ounces. We compute
!
X1 + · · · + X100 − (100)(1) 90 − (100)(1)
P (X1 + · · · + X100 ≤ 90) = P p ≤ p
(100)(1/3) (100)(1/3)
≈ P (Z ≤ −1.73)
= P (Z ≥ 1.73)
= 1 − P (Z ≤ 1.73)
= 1 − 0.9582
= 0.0418

In other words, the probability is approximately 4.18% that the 100 cans of
soda contain a total of less than 90 ounces of soda.
Example 37.4 (continued) Now we find the probability that the total volume
of soda remaining is between 97 and 103 ounces.
We compute

P (97 ≤ X1 + · · · + X100 ≤ 103)

!
97 − (100)(1) X1 + · · · + X100 − (100)(1) 103 − (100)(1)
=P p ≤ p ≤ p
(100)(1/3) (100)(1/3) (100)(1/3)
≈ P (−0.52 ≤ Z ≤ 0.52)
= P (Z ≤ 0.52) − P (Z ≤ −0.52)
= P (Z ≤ 0.52) − P (Z ≥ 0.52)
= P (Z ≤ 0.52) − 1 + P (Z ≤ 0.52)
= 2P (Z ≤ 0.52) − 1
= (2)(0.6985) − 1
= 0.3970

So the probability is approximately 39.70% that the 100 cans of soda contain
between 97 and 103 ounces of soda.
 37.4. CLT for Sums of Continuous Random Variables 487

Example 37.5. We return to the setup in Exercise 33.2: David works at a

customer call center. He talks to customers on the telephone. The length (in
hours) of each conversation has density fX (x) = 3e−3x for x > 0 and fX (x) = 0
otherwise. The lengths of calls are independent. As soon as one conversation
is finished, he hangs up the phone, and immediately picks up the phone again
to handle another call (i.e., there are no gaps in between the calls). Thus, if
he conducts n phone calls in a row, the total amount of time he spends on the
telephone is X1 + · · · + Xn , where the Xj ’s are independent, and each Xj has
the density above.

What is the probability that David can handle 135 calls within a 40 hour
period? In other words, what is P (X1 + · · · + X135 ≤ 40)?
If we only use the fact that X1 + · · · + X135 is a Gamma random variable,
we would need to compute 135 nested integrals. Yuck! Computing 135 nested
integrals would definitely be unreasonable, but such a computation would be
necessary to get the exact probability.
Fortunately, we can use the Central Limit Theorem to get an excellent ap-
proximation to the answer, and we will not need to solve any integrals! Since
E(Xj ) = 1/3 for each j, so
E(X1 + · · · + X135 ) = (135)(1/3).
Since the Xj ’s are independent, we can add the variances to obtain
Var(X1 + · · · + X135 ) = (135)(1/9).
Thus, we compute P (X1 + · · · + X135 ≤ 40) as follows:

!
X1 + · · · + X135 − (135)(1/3) 40 − (135)(1/3)
P p ≤ p ≈ P (Z ≤ −1.29)
(135)(1/9) (135)(1/9)
= P (Z ≥ 1.29)
= 1 − P (Z ≤ 1.29)
= 1 − 0.9015
= 0.0985
Thus, we conclude that the probability is approximately 38.59% that David can
handle 135 calls within 40 hours.

Example 37.6. As in Example 37.5, consider the lengths of calls handled by

David in a call center. The calls are independent Exponential random variables,
and each call lasts, on average, 1/3 of an hour. On a particular day, David
records the lengths of 24 consecutive calls. What is the probability that the
average of these 24 calls exceeds 1/4 of an hour?
488 Chapter 37. Central Limit Theorem

Let X1 , . . . , X24 denote the lengths of the calls. The average length of the 24
calls is (X1 + · · · + X24 )/24. So the average exceeds 1/4 of an hour if
 
X1 + · · · + X24 1
P > .
24 4
Multiplying by 24 on both sides of the inequality, we have, equivalently,

P (X1 + · · · + X24 > 6) .

Now we normalize the sum of the random variables, to get

!
X1 + · · · + X24 − (24)(1/3) 6 − (24)(1/3)
P p > p ≈ P (Z > −1.22)
(24)(1/9) (24)(1/9)
= P (Z < 1.22)
= 0.8888

37.5 CLT for Sums of Discrete Random Variables

When we apply the Central Limit Theorem to sums of discrete random vari-
ables, the method of setup and solution is almost completely the same. The
only change is a concept called continuity correction. We only use continuity
correction when working with sums of discrete random variables that are integer
valued.
To motivate the need for continuity correction, consider a sum of random
variables X1 , X2 , . . . , Xn that are integer-valued. Then X1 + · · · + Xn is also
integer valued. In other words, X1 + · · · + Xn cannot take on fractional values
that are not integers. So, for instance, if we are trying to compute

P (X1 + · · · + Xn ≤ 40),

this is exactly the same as

P (X1 + · · · + Xn < 41).

Which number should we use in our computation, 40 or 41??? The answer will
not be greatly affected either way, but the rule of continuity correction tells us to
use the number halfway in between the two candidate values. So we must first
determine which two numbers are candidates (one value will correspond to a less-
than-or-equal or a greater-than-or-equal, and the other value will correspond to
a strictly-less-than or a strictly-greater-than). Afterwards, we take the average
of the two numbers, in this case, 40.5. So we would compute
 
X1 + · · · + Xn − nµ 40.5 − nµ
P √ ≤ √ .
nσ 2 nσ 2
 37.5. CLT for Sums of Discrete Random Variables 489

Remark 37.7. Caution: Continuity correction is only used for appli-

cations of the Central Limit Theorem to sums of discrete random
variables. We do not need continuity correction when applying the
Central Limit Theorem to sums of continuous random variables.

Example 37.8. One thousand students participate in a survey to see how many
breakfasts that they each eat, within a two week period. Let X1 , X2 , . . . , X1000
denote the numbers of breakfasts that student 1, 2, . . . , 1000 eats, respectively.
Since Xj is the number of breakfasts consumed by the jth student, then 0 ≤
Xj ≤ 14 for each j. Assume that Xj is Binomial, with n = 14 and p = 0.6, and
assume that the students behave independently.

Find the probability that the students eat strictly more than 8350 breakfasts
altogether during the two week period.
The expected value of Xj is E(Xj ) = np = (14)(0.6) = 8.4. The variance of
Xj is Var(Xj ) = np(1 − p) = (14)(0.6)(0.4) = 3.36. Thus, the expected value
of X1 + · · · + X1000 is

E(X1 + · · · + X1000 ) = (1000)(8.4) = 8400,

and the variance of X1 + · · · + X1000 (since the Xj ’s are independent) is

Var(X1 + · · · + X1000 ) = (1000)(3.36).

The content of the Central Limit Theorem, applied to this scenario, is that

X1 + · · · + X1000 − (1000)(8.4)
p
(1000)(3.36)

is approximately Normally distributed, i.e.,

!
X1 + · · · + X1000 − (1000)(8.4)
P p ≤a ≈ FZ (a).
(1000)(3.36)

The only way that the continuity correction affects our calculation is in deciding
whether to compute
P (X1 + · · · + X1000 > 8350)

or
P (X1 + · · · + X1000 ≥ 8351).

These are the same probabilities, but they will generate slightly different results
in our approximation below. So we pick the average value of 8350 and 8351
490 Chapter 37. Central Limit Theorem

(which is 8350.5), and we proceed as follows: So we compute


X1 + · · · + X1000 − (1000)(8.4)
P (X1 + · · · + X1000 > 8350.5) = P p
(1000)(3.36)

8350.5 − (1000)(8.4)
> p
(1000)(3.36)
≈ P (Z > −0.85)
= P (Z < 0.85)
= 0.8023

In other words, the probability is approximately 80.23% that the students eat
strictly more than 8350 breakfasts during the two week period.
Example 37.8 (continued) Suppose that we now want to find the probability
that between 8380 and 8435 breakfasts (inclusive) are served, i.e., find

P (8380 ≤ X1 + · · · + X1000 ≤ 8435).

We could equivalently compute

P (8379 < X1 + · · · + X1000 < 8436),

because that would be the exact same probability. So we settle for the average
of each endpoint. In other words, we compute

P (8379.5 < X1 + · · · + X1000 < 8435.5).

The derivation is the following:

P (8379.5 ≤ X1 + · · · + X1000 ≤ 8435.5)

!
8379.5 − 8400 X1 + · · · + X1000 − 8400 8435.5 − 8400
=P p ≤ p ≤p
(1000)(3.36) (1000)(3.36) (1000)(3.36)
≈ P (−0.35 ≤ Z ≤ 0.61)
= P (Z ≤ 0.61) − P (Z ≤ −0.35)
= P (Z ≤ 0.61) − P (Z ≥ 0.35)
= P (Z ≤ 0.61) − 1 + P (Z ≤ 0.35)
= 0.7291 − 1 + 0.6368
= 0.3659

So the probability is approximately 36.59% that the students eat between 8380
and 8435 breakfasts (inclusive) during the two week period.
 37.5. CLT for Sums of Discrete Random Variables 491

Example 37.9. We modify the scenario from Example 17.3 of Chapter 17

as follows: Suppose that 220 basketball players are under consideration for a
position on a basketball team. Suppose that each player makes 80% of her free
throws successfully. Each player attempts to make a basket as many times as
necessary, until scoring the first basket. Let Xj denote the number of attempts
that the jth player needs. So X1 + · · · + X220 is the total number of attempts
that all 220 players need (collectively) until they have each made one basket
successfully. Also let the Xj ’s be independent, i.e., the number of attempts by
one player does not affect the number of attempts from any of the other players.

What is the probability that strictly between 260 and 280 attempts are
needed? In other words, what is

P (260 < X1 + · · · + X220 < 280)?

Equivalently, what is

P (261 ≤ X1 + · · · + X220 ≤ 279)?

Since we are approximating an integer-valued sum of random variables, we again

need to use continuity correction. So we compute

P (260.5 ≤ X1 + · · · + X220 ≤ 279.5).

We know from Example 17.3 that each player has expected number of shots
E(Xj ) = 1/0.80 = 1.25 and variance Var(Xj ) = 0.20/0.802 = 0.3125.
We compute:

P (260.5 ≤ X1 + · · · + X220 ≤ 279.5)

P220 !
260.5 − (220)(1.25) i=1 Xi − (220)(1.25) 279.5 − (220)(1.25)
=P p ≤ p ≤ p
(220)(0.3125) (220)(0.3125) (220)(0.3125)
≈ P (−1.75 ≤ Z ≤ 0.54)
= P (Z ≤ 0.54) − P (Z ≤ −1.75)
= P (Z ≤ 0.54) − P (Z ≥ 1.75)
= P (Z ≤ 0.54) − 1 + P (Z ≤ 1.75)
= 0.7054 − 1 + 0.9599
= 0.6653

So the probability is approximately 66.53% that (strictly) between 260 and 280
attempts are needed altogether.
 492 Chapter 37. Central Limit Theorem

37.6 Approximations of Binomial Random Variables

The Central Limit Theorem states that the sum of a large number of indepen-
dent random variables behaves approximately like a Normal random variable.
The CLT can also be applied to one Binomial(n, p) random variable (not a sum
of Binomial random variables) to show that a Binomial has some similarities
to a Normal random variable if np(1 − p) is significantly large. In practice,
np(1 − p) ≥ 10 is sometimes used a rule of thumb for this application of the
CLT. Notice that, if np(1−p) is large, then n is large too, because n ≥ np(1−p)
always. There is no definitive, comprehensive rule that states “n is large enough”
for all such approximations.
Consider such a Binomial(n, p) random variable X, with np(1 − p) large.
The main idea is that we can view X as the sum of a large number of in-
dependent Bernoulli random variables X1 , X2 , . . . , Xn , each with parameter p.
For instance, if X is a Binomial random variable with parameters n = 300
and p = 1/4 then np(1 − p) = 56.25, so np(1 − p) is sufficiently large that
X will roughly behave like a Normal random variable. In such a case, we let
X1 , . . . , X300 be 300 independent Bernoulli random variables, each with param-
eter p = 1/4, and then both X and X1 + · · · + X300 have the same mass:
pX (j) = P (X = j)
 
300
= (1/4)j (3/4)300−j
j
= P (X1 + · · · + X300 = j)
= pX1 +···+X300 (j).
Since Binomial As another example, if X is a Binomial random variable with parameters n =
random variables 1000 and p = 9/10 then np(1 − p) = 90, so np(1 − p) is large enough to make X
are discrete, approximately Normal. We let X1 , . . . , X1000 be independent Bernoulli random
continuity correction variables, each with parameter p = 9/10, and then
is needed when  
doing a Normal 1000
approximation to a pX (j) = (9/10)j (1/10)1000−j = pX1 +···+X1000 (j).
j
Binomial random
variable. So, in other words, if X is a Binomial random variable with parameters n
and p such that np(1 − p) is large, then X can be treated as the sum of n
independent Bernoulli random variables, each with parameter p, and thus, by
the Central Limit Theorem, X is approximately a Normal random variable, with
E(X) = np and Var(X) = np(1 − p).
Another way to see that Binomial random variables with parameters n and
p (such that np(1 − p) is sufficiently large) are approximately Normal random
variables is to graph the mass of such random variables. The mass of a Binomial
random variable looks like the density of a Normal random variable, i.e., like a
bell curve. In contrast to this idea, the mass of a Geometric random variable
does not look like a bell curve at all, and a Geometric random variable is not
approximated by a Normal random variable.
 37.6. Approximations of Binomial Random Variables 493

Note that the random variables are only approximately Normal, not com-
pletely Normal. For instance, the mass of a Binomial random variable is still
discrete (although it looks like a bell curve) and the mass pX (j) is only defined
for integers 0 ≤ j ≤ n. See Figures 37.1 and 37.2. We will still use continuity
correction with such Binomial random variables, since they are discrete.

Remark 37.10. Approximating a Binomial random variable by a

Normal random variable
If X is a Binomial random variable with parameters n and p such that np(1−p)
is large (e.g., np(1 − p) ≥ 10), then X behaves approximately like a Normal
random variable with expected value np and variance np(1 − p) and standard
deviation np(1 − p). Thus
p

X − µX X − np
Z≈ =p
σX np(1 − p)

i.e., we approximately have a standard Normal random variable, when we

standardize the X. In other words,
!
X − np
P p ≤ a ≈ P (Z ≤ a) = FZ (a).
np(1 − p)

Such approximations should also include continuity correction, which makes

a slight adjustment to the value of a, as discussed earlier in this chapter.

pX (x)
0.06

0.05

0.04

0.03

0.02

0.01
x
40 50 60 70 80 90 100 110

Figure 37.1: The mass of a Binomial random variable with n = 300 and
p = 1/4, near the expected value, np = 75.
494 Chapter 37. Central Limit Theorem

pX (x)
0.05
0.04
0.03
0.02
0.01
x
850 860 870 880 890 900 910 920 930 940 950

Figure 37.2: The mass of a Binomial random variable with n = 1000 and
p = 9/10, near the expected value, np = 900.

Example 37.11. Consider the independent single-baby births, in which the

probability of a boy or girl in each birth is assumed to be equally likely. Let X
be the number of girls.

The number X of girls is Binomial with expected value np = (1000)(1/2) =

500 and variance np(1−p) = (1000)(1/2)(1/2) = 250. Thus, X is approximately
Normal with expected value 500 and variance 250.
The probability that strictly more than 470 of the babies are girls is exactly
P (X > 470) = P (X = 471) + P (X = 472) + · · · + P (X = 1000)
   
1000 471 529 1000
= (1/2) (1/2) + (1/2)472 (1/2)528
471 472
 
1000
+ ··· + (1/2)1000 (1/2)0
1000
but this is very unwieldy to calculate, and many calculators could not han-
dle such a computation anyway! So we use a Normal approximation to the
Binomial, with continuity correction, noting that
P (X > 470) = P (X ≥ 471),
so we use P (X > 470.5), and we get
P (X > 470) = P (X > 470.5)
 
X − 500 470.5 − 500
=P √ > √
250 250
≈ P (Z > −1.87)
= P (Z < 1.87)
= 0.9693
 37.6. Approximations of Binomial Random Variables 495

Example 37.11 (continued) Now, what if we want to approximate a specific

probability? For instance, what if we want to know the probability that exactly
500 of the 1000 babies are girls?
We know that
 
1000
P (X = 500) = (1/2)500 (1/2)500 ,
500

but again, this is very difficult for many calculators to handle. So we reformulate
this as
P (500 ≤ X ≤ 500),
or equivalently
P (499 < X < 501),
and with continuity correction, this becomes

P (499.5 ≤ X ≤ 500.5),

which will give a pretty good approximation. We calculate:

 
499.5 − 500 X − 500 500.5 − 500
P (499.5 ≤ X ≤ 500.5) = P √ ≤ √ ≤ √
250 250 250
≈ P (−0.03 ≤ Z ≤ 0.03)
= P (Z ≤ 0.03) − P (Z ≤ −0.03)
= P (Z ≤ 0.03) − P (Z ≥ 0.03)
= P (Z ≤ 0.03) − 1 + P (Z ≤ 0.03)
= 0.5120 − 1 + 0.5120
= 0.0240

By the way, the exact answer, using a computer, is 0.02522501818 . . ., so

our approximation is very comparable to the actual value.

Example 37.12. At a certain local restaurant, students are known to prefer

Japanese pan noodles 40% of the time (it is a very popular and tasty dish!).
Consider 2000 randomly chosen students; let X denote the number of students
that order Japanese pan noodles.

We know that X is Binomial with n = 2000 and p = 0.40, so E(X) =

np = (2000)(0.40) = 800 and Var(X)
p= np(1 − p) = √(2000)(0.40)(0.60) = 480,
so the standard deviation of X is np(1 − p) = 480 = 21.9089. So X is
approximately Normal with expected value 800 and variance 480.
 496 Chapter 37. Central Limit Theorem

The probability that at most 840 of the students eat Japanese pan noodles
is P (X ≤ 840), i.e., P (X < 841), so we use P (X ≤ 840.5) for the continuity
correction, and we calculate
 
X − 800 840.5 − 800
P (X ≤ 840.5) = P √ ≤ √
480 480
≈ P (Z ≤ 1.85)
= 0.9678

37.7 Approximations of Poisson Random Variables

A Poisson random variable with large parameter λ will be distributed like a
Since Poisson Normal random variable. One way to see this is to graph the mass of such a
random variables Poisson random variable, which also looks like a bell curve; see Figure 37.3.
are discrete,
continuity correction pX (x)
is needed when
0.018
doing a Normal
approximation to a 0.016
Poisson random
variable.
0.014
0.012
0.010
0.008
0.006
0.004
0.002
x
400 420 440 460 480 500 520 540 560 580 600

Figure 37.3: The mass of a Poisson random variable with λ = 500, near the
expected value, λ = 500.

Of course, Poisson random variables are discrete, and moreover the mass
pX (j) is defined only for the nonnegative integers 0, 1, 2, . . .. Thus, we must use
continuity correction with approximations of Poisson random variables too.
Having the parameter λ ≥ 10 is one possible rule of thumb for getting rela-
tively good approximations of a Normal random variable to a Poisson random
variable. We can view X as the sum of a large number of Poisson random
variables. As an example, if X is a Poisson random variable with parameter
λ = 500, we can let X1 , . . . , X500 be a collection of independent Poisson random
 37.7. Approximations of Poisson Random Variables 497

variables, each with parameter 1, so both X and X1 + · · · + X500 have the same
mass:

pX (j) = e−500 500j /j! = pX1 +···+X500 (j).

This comes from the fact that the sum of independent Poisson random variables
is also a Poisson random variable, and that the parameter of the sum is equal
500
z }| {
to the sum of the parameters. In this case, 1 + 1 + · · · + 1 = 500. So we
have just decomposed a Poisson random variable X with parameter 500 into a
sum of 500 independent Poisson random variables X1 , . . . , X500 that each have
parameter 1. Thus, by the Central Limit Theorem, X is approximately a Normal
random variable. We note that X √ has expected value λ = 500 and variance 500,
and therefore standard deviation 500.

We can even decompose the very same Poisson random variable more finely.
For instance, if X is Poisson with parameter λ = 500, we can let X1 , . . . , X3000
be a collection of independent Poisson random variables, each with parameter
1/6 (we use 1/6 here, so that the mean is 500 = (3000)(1/6)), then both X and
X1 + · · · + X3000 have the same mass:

pX (j) = e−500 500j /j! = pX1 +···+X3000 (j).

3000
z }| {
This works because 1/6 + 1/6 + · · · + 1/6 = 500.

Methods like this allow us to even decompose Poisson random variables with
parameters that are not integers! For instance, if X is Poisson with parameter
372.9, we can let X1 , . . . , X1000 be a collection of independent Poisson random
variables, each with parameter 0.3729, and then both X and X1 , . . . , X1000 have
the same mass:

pX (j) = e−372.9 372.9j /j! = pX1 +···+X1000 (j).

1000
z }| {
This works because 0.3729 + 0.3729 + · · · + 0.3729 = 372.9.
498 Chapter 37. Central Limit Theorem

Remark 37.13. Approximating a Poisson random variable using a

Normal random variable
If X is a Poisson random variable with parameter λ sufficiently large, e.g.,
λ ≥ 10, then X behaves approximately like a Normal√random variable with
expected value λ, variance λ, and standard deviation λ. Thus
X − µX X −λ
= √
σX λ
behaves approximately like a standard Normal random variable Z. In other
words,  
X −λ
P √ ≤ a ≈ P (Z ≤ a) = FZ (a).
λ
As with Binomial random variables, these approximations of Poisson random
variables should also include continuity correction, resulting in a slight adjust-
ment to the value of a, as discussed earlier.

Example 37.14. A certain grocery store is believed to have a Poisson number

of customers each day, with parameter λ = 1200.

What is the probability that the number of customers on a certain day is

strictly between 1150 and 1250?
We let X denote the number of customers. So E(X) = λ = 1200 and
Var(X) = λ = 1200 too. We want to compute

P (1150 < X < 1250),

or equivalently,
P (1151 ≤ X ≤ 1249),
so with continuity correction, we use

P (1150.5 ≤ X ≤ 1249.5).

Now we compute
 
1150.5 − 1200 X − 1200 1249.5 − 1200
P (1150.5 ≤ X ≤ 1249.5) = P √ ≤ √ ≤ √
1200 1200 1200
≈ P (−1.43 ≤ Z ≤ 1.43)
= P (Z ≤ 1.43) − P (Z ≤ −1.43)
= P (Z ≤ 1.43) − P (Z ≥ 1.43)
= P (Z ≤ 1.43) − 1 + P (Z ≤ 1.43)
= 0.9236 − 1 + 0.9236
= 0.8472
 37.7. Approximations of Poisson Random Variables 499

For comparison, the actual answer, using a computer, is

P (1150 < X < 1250) = P (X = 1151) + P (X = 1152) + · · · + P (X = 1249)

1249
X e−1200 1200j
=
j!
j=1151

= 0.8470129469 . . .

We could not perform such a calculation very easily by hand, because it has
nearly 100 terms that would need to be calculated, and each term includes the
use of very large numbers, e.g., 1200j and j! for 1151 ≤ j ≤ 1249, and also very
small numbers, e.g., e−1200 for 1151 ≤ j ≤ 1249.

Example 37.15. A publisher estimates that a 500 page book has a Poisson
number of errors with average λ = 100 errors/book. Estimate the probability
that there are strictly more than 80 errors in such a book.

If we let X denote the number of errors in a newly published 500 page book,
the probability that there are strictly more than 80 errors is P (X > 80), which
is equivalent to P (X ≥ 81).
Why can we not calculate this manually, without Normal approximation?
If we try a manual calculation, we first realize that there are infinitely many
terms:
e−100 10081 e−100 10082 e−100 10083 e−100 10084
P (X ≥ 81) = + + + + ··· ,
81! 82! 83! 84!
which is hopeless. So we rearrange and use the complement:

P (X ≥ 81) = 1 − P (X ≤ 80)
e−100 1000 e−100 1001 e−100 1002 e−100 10080
= + + + ··· + ,
0! 1! 2! 80!
but this is still hopeless, because there are 81 terms in the summation. So the
value of a Normal approximation—which can be evaluated in one fell swoop—
should now seem apparent.
Also remember that we use P (X ≥ 80.5), for the purposes of continuity
correction. This gives us
 
80.5 − 100 X − 100
P (80.5 ≤ X) = P √ ≤ √
100 100
≈ P (−1.95 ≤ Z)
= P (1.95 ≥ Z)
= P (Z ≤ 1.95)
= 0.9744
500 Chapter 37. Central Limit Theorem

Example 37.16. (Continuation of Example 37.15)

If we buy 10 different books from this publisher, each containing 500 pages,
the number of our books that have strictly more than 80 errors each is Binomial
with n = 10 and p = 0.9744. The probability that exactly nine of the books
have 80 or more errors each (and the remaining book has 79 or fewer errors) is
   
10 9 1 10
p (1 − p) = (0.9744)9 (0.0256)1 = 0.2027.
9 9

37.8 Exercises
37.8.1 Practice
Exercise 37.1. Concert tickets. At a certain university, each student who
tries to purchase concert tickets for an upcoming show is successful at connecting
to the concert ticket website with probability 0.85. If unsuccessful in logging
on, he or she tries again a few minutes later, over and over, until finally getting
tickets. The concert is large (tens of thousands of seats), so assume that such
attempts are independent. Within a group of 300 students, find the probability
that strictly more than 360 attempts are necessary for these 300 students to
successfully get tickets.
Exercise 37.2. Blind auction bids. At an auction, exactly 282 people place
requests for an item. The bids are placed “blindly,” which means that they are
placed independently, without knowledge of the actions of any other bidders.
Assume that each bid (measured in dollars) is a Continuous Uniform random
variable on the interval [10.50, 19.30]. Find the probability that the sum of all
the bids exceeds $4150.
Exercise 37.3. Baseball statistics. A certain baseball player has, on average,
0.7 RBI’s (runs batted in) per game, with standard deviation of 0.2. What is
the approximate probability that the player has at most 110 RBI’s during a
given season, which contains 162 games?
Exercise 37.4. Tootsie Pops. Back in 1970, a television commercial for Toot-
sie Pops (a candy) asked how many licks would be needed to get to the Tootsie
Roll Center of a Tootsie Pop. This question has become somewhat famous and
well-known among children. High school, undergraduate, and Ph.D. students
have all performed experiments to uncover the answer (some have even built
licking machines to test this question). Suppose that an average 364 licks are
required, with a standard deviation of 40 licks. If a group of 50 children test
their licking abilities, what is the approximate probability that, among the 50
children, the average number of licks per student is 380 or more?
 37.8. Exercises 501

Exercise 37.5. Strawberry milkshakes. The number of strawberry milk-

shakes that are sold at a local diner in a given day has an average of 97.5
per day and standard deviation of 7.8. In a given 30-day month, what is the
approximate probability that they sell more than 3000 milkshakes?
Exercise 37.6. Spelling. A student is learning to spell some difficult words.
He estimates that it takes him 6 attempts to learn each word, with a variance of
3.5 per word. Find the approximate probability that he will need 125 or more
attempts to learn 20 words.
Exercise 37.7. Water bottles. Water is sold in bottles that are advertised
to contain 1-liter each. Barbara is an inspector for the company and notices
that the amount of water in each bottle, unfortunately, has an average of only
0.99 liters, and surprisingly high standard deviation of 0.03 liters. Estimate the
probability that a 12-pack of water will actually contain 12 liters or more of
water.
Exercise 37.8. Memory chip. Rafael is pursuing a major in computer sci-
ence. He notices that a memory chip containing 212 = 4096 bits is full of data
that seems to have been generated, bit-by-bit, at random, with 0’s and 1’s
equally likely, and the bits are stored independently. If the each bit is equally
likely to be a 0 or 1, estimate the probability that there are actually 4120 or
more 1’s stored in the memory chip?
Exercise 37.9. Compost worms. Josephine estimates that the worms in her
composter can eat, on average, 3.05 pounds of waste per week, with standard
deviation of 0.3 pounds. She tests this conjecture with her worms over a 50-
week period. Estimate the probability that the average over the 50-week period
exceeds 3 pounds per week.
Exercise 37.10. Shower length. Each morning a student takes a shower that
lasts 15 minutes, with a standard deviation of 4 minutes. Find an estimate of
the probability that the student spends between 11 and 12 hours in the shower
during a 45-day period.
Exercise 37.11. Student enrollment. The registrar estimates that each
student enrolls for 15.3 credits on average, with a standard deviation of 1.2
credits per student. Estimate the probability that, on a campus with 4000
enrolled students, the total number of enrolled credit hours is strictly between
61,100 and 61,300.
Exercise 37.12. Toy geyser. A student builds a toy geyser for her engineering
class that has height (in inches) of density
1
fX (x) = e−x/8 for x > 0,
8
and fX (x) = 0 otherwise. If she makes 20 such geysers, what is the rough
probability that the average geyser height, among these 20 geysers, is 7 inches
or more?
502 Chapter 37. Central Limit Theorem

Exercise 37.13. Die rolls. A group of students in a probability course is

skeptical about Geometric random variables, so one day, they all bring a die to
class. Each of the 40 students rolls her/his die repeatedly. Each student stops
upon their first 5 that appears, independent of the other students. What is the
estimated probability that it takes the students a total of 250 or more die rolls
to accomplish their experiment?
Exercise 37.14. Phone calls. Suppose that each call a student makes to
his girlfriend has an average length of 12 minutes, and the standard deviation
of the length of each call is also 3 minutes. Estimate the probability that he
completely uses up a 300-minute calling card while talking to his girlfriend
during 24 conversations.
Exercise 37.15. Journal pages. Suppose that the number of pages in an issue
of a journal is Uniformly distributed between 256 and 384. During 40 years of
publication, the journal publishes a total of 160 issues. Give an estimate for
the probability that between 51,000 and 52,000 pages (inclusive) were used for
these 160 issues.
Exercise 37.16. Car sales. A car dealership expects to sell 6 cars per day,
with standard deviation of 2.1 cars per day. During a 100-day period, estimate
the probability that they sell strictly more than 625 cars.
Exercise 37.17. Rainfall. If the amount of rainfall in a given region is as-
sumed to be Uniform on the interval [0.2, 4.0] (in inches) each month, what is
the approximate probability that there are 53 or more inches of rain during a
24-month period?
Exercise 37.18. Class attendance. In a math class with 200 students, sup-
pose that the students’ decisions to attend the class are independent, and each
student attends with probability 93%. On a given day, find the approximate
probability that strictly fewer than 179 students attend.
Exercise 37.19. Student workers. Consider a group of students whose are
assigned to work a random number of hours. Their hours per student, per week,
are modeled by a Binomial random variable with n = 20 and p = 0.8. (Each
hour assigned to work will count as a “success” in the Binomial model.) If there
are 100 students in the fraternity, and the numbers of hours spent working are
independent, find an estimate for the probability that they work between 1580
and 1620 hours altogether during a given week.
Exercise 37.20. Weighing burgers. A restaurant claims that each of their
burgers has a 1/4 pound of meat after it is prepared. Of course this is not exact.
The students in a local high school measured some of these burgers for a science
project, and they concluded that the expected weight of such a burger is 0.251
pounds, but the standard deviation is 0.02 pounds. Afterwards, the 30 students
celebrate by each ordering 1 burger. What is the probability that, among the
30 students, the average weight of a burger is less than the advertised weight of
1/4 pound?
 37.8. Exercises 503

Exercise 37.21. Customer arrivals. Consider customers who arrive at a

checkout counter with an average rate of 8 per hour following a Poisson distri-
bution. Find the probability that strictly more than 70 customers arrive during
the eight-hour shift.

Exercise 37.22. Free throws. A particular basketball player successfully

scores about 75% of his free throw attempts. Approximate the probability that
he makes strictly more than 145 of 200 attempts.

Exercise 37.23. Jackpot. If a certain type of slot machine has only a proba-
bility of 0.0001 of yielding a jackpot on each game at a certain casino in Vegas,
and the patrons play those slots 250,000 times during a given month, estimate
the probability that the casino will have 30 or more jackpot rewards to payout
during the month.

Exercise 37.24. Ice cream patrons. An ice cream shop estimates that its
number of patrons per day is Poisson with mean 19. What is the estimated
probability that they have 20 or more customers on a given day?

Exercise 37.25. Failing processors. In the early testing stages of processor

manufacturing, 40% of the processors will fail in some way. If 500 processors
are manufactured at this stage, what is a rough estimate for the probability
that strictly less than 180 of the processors will fail in some way?

Exercise 37.26. Rotten berries. When buying a pack of strawberries at the

farmer’s market, there is a 0.10 chance that there will be a rotten berry in the
pack. If a restaurant buys 300 packs of strawberries for their desserts, what
is the approximate probability that fewer than 25 packs will contain a rotten
berry?

Exercise 37.27. Airport security. If 6% of all passengers are screened with

two rounds of security at the airport, and Southwest has 8 flights with 180
passengers each, what is the approximate probability that 80 or more of these
Southwest passengers will receive this extra level of screening?

Exercise 37.28. Microwave popcorn. A bag of microwave popcorn comes

with 200 kernels of corn. When microwaved for exactly 3 minutes, an individual
kernel pops 90% of the time. What is the approximate probability that there
are strictly less than 10 unpopped kernels in such a bag?

37.8.2 Extensions
Exercise 37.29. Burritos. Twenty-five students want burritos, but the dining
hall only has one burrito maker. Each burrito takes an average of 72.5 seconds
to cook, with standard deviation 3.2 seconds.

a. Estimate the probability that all twenty-five students can cook their
burritos in half an hour or less, if we ignore the time in between the consecutive
students.
504 Chapter 37. Central Limit Theorem

b. Estimate the probability that all twenty-five students can cook their
burritos in 2300 seconds or less, if we assume that there is an exact 20-second
delay in between each pair of consecutive students. (Hint: There are 24 such
delays between 25 students.)
c. Estimate the probability that all twenty-five students can cook their
burritos in 2300 seconds or less, if we assume that there is a Normally distributed
delay in between each pair of consecutive students, with average delay of 20
seconds between consecutive pairs of students, and standard deviation of 4
seconds.

Exercise 37.30. Online vs. traditional lecture. At a certain university, a

large course has 1500 registered students; the course can be viewed online if a
student does not feel like walking to the lecture hall. Each student decides each
day (independently of the other students and independent of her/his own prior
behavior) whether to attend the class in the lecture hall or just view the online
version. The probability a student attends class in the lecture hall is only 0.30,
so the registrar has quite a headache.

a. What is the expected number of students to attend class in the lecture

hall on a particular day?
b. If the registrar assigns a lecture hall with 500 seats for the class, what
is the probability that the lecture hall overflows (i.e., strictly more than 500
attend) on a particular day?
c. To save money, the registrar wants to use an even smaller lecture hall next
semester, but the professor insists on having enough seats so the probability is
95% (or more) of no overflow on a given day. How many seats does the professor
need in her lecture hall?

Exercise 37.31. Playing the lottery. A person plays a certain lottery game
10,000 times during his life. The chance of winning is 1/5000 during each
attempt, and the attempts are independent. Approximate the probability that
he wins 2 or more times during his life.
Chapter 38

Review of Named Continuous

Random Variables

38.1 Summing-up: How To Tell Random Variables

Apart

Your first step should be to identify whether X is discrete (how many?) or con-
tinuous (how long or how much?). Once you have decided that X is continuous,
think about what values of X would be allowed.
The actual waiting time until your 1st customer is an Exponential random
variable. The actual waiting time until your 3rd customer is a Gamma random
variable. The customer could arrive at a time that is Uniform between 8:00
and 8:15. The percentage of customers who buy ice cream could be a Beta
random variable. The number of customers during a long period of time is
approximately Normally distributed.
The following are some Pairs of Random Variables that are Related:
Waiting until the 1st success:
• Geometric: X is the # of people you have to ask until you get your first
“yes” answer (discrete)

• Exponential: X is the time you have to wait until you get your first
“yes” answer (continuous)

Waiting until the 4th success:

• Negative Binomial: X is the # of people you have to ask until you get
your 4th “yes” answer (discrete)

• Gamma: X is the time you have to wait until you get your 4th “yes”
answer (continuous)

505
 Summary of Named Continuous Random Variables
Name Uniform Exponential Gamma Beta Normal
2 2
λe−λx (λx)r−1 Γ(α+β)xα−1 (1−x)β−1 e−(x−µ) /(2σ )
Density 1/(b − a) λe−λx Γ(r) Γ(α)Γ(β)
√
2πσ 2
fX (x)
Domain a≤x≤b x≥0 x≥0 0≤x≤1 −∞ < x < ∞
(λx)j
CDF 1 − e−λx ; 1 − e−λx r−1 convert to std. Normal
P
(x − a)/(b − a) j=0 j! ,
FX (x) if r ∈ N Z = (X − µX )/σX
E(X) (a + b)/2 1/λ r/λ α/(α + β) µ
αβ
Var(X) (b − a)2/12 1/λ2 r/λ2 (α+β)2 (α+β+1)
σ2
Param- a, b are λ is average λ is average usually some µ = exp. value;
eters endpoints # of successes # of successes given σ = st. dev.
506

per time unit per time unit constraints

wait time ’til wait time ’til fraction, practical random
What location in 1st event rth event percentage, variable for
X is an interval (like continuous (like continuous proportion, quantities clustered
Geometric) Neg. Binom.) around avg. value
When spread wait time wait time Bayesian Central Limit Thm.
used evenly until 1st event until rth event statistics and applications

Note: Γ(r) = (r − 1)! if r ∈ N.

 38.2. Exercises 507

Remember, Exponential distributions have the memoryless property. This

means that, for instance,
P (X > 20 | X > 8) = P (X > 12).
In practice, the question would read, given that nobody has arrived in the first
8 minutes, what is the probability nobody will arrive in the next 12 minutes?
The first 8 minutes don’t matter to us now. For Exponential, you can start
from NOW. (Geometric distributions also have the memoryless property.)

38.2 Exercises
For Exercises 38.1 through 38.12, state which continuous or discrete distribution
would be most appropriate and why you think so. The choices are:

Bernoulli Hypergeometric Exponential

Binomial Poisson Gamma
Geometric Discrete Uniform Beta
Negative Binomial Continuous Uniform Normal
Exercise 38.1. Let X be the diameter of a chocolate chip cookie selected at
random if you know that the average diameter is 6 inches with a standard
deviation of 0.4 inches and if you know that the distribution of cookies has a
bell-shaped curve.
Exercise 38.2. Let X be the percent of flour which will be used up by the end
of the day at the bakery.
Exercise 38.3. Let X indicate whether the next customer will want to buy a
chocolate chip cookie.
Exercise 38.4. Let X be the length of time that the baker will have to wait
until a customer wants to buy a chocolate chip cookie.
Exercise 38.5. Let X be the number of customers who will come into the store
until a customer wants to buy a chocolate chip cookie.
Exercise 38.6. Let X be the number of customers, out of the next 20 who
come in, who will want to buy a chocolate chip cookie.
Exercise 38.7. Let X be the number of customers who will want to buy a
chocolate chip cookie in the next hour if, according to bakery records, an average
of 3 customers per hour want to buy a chocolate chip cookie.
Exercise 38.8. Let X be the exact arrival time of the first customer if we know
that exactly one customer arrived between 8:03 and 8:12 AM.
Exercise 38.9. Let X be the length of time that the baker will have to wait
to bring out a new batch of dozen chocolate chip cookies if he just put out the
first dozen and if on average 3 customers per hour want to buy a chocolate chip
cookie.
508 Chapter 38. Review of Named Continuous Random Variables

Exercise 38.10. Let X be the position of the cookie on the tray (numbered 1
to 12) that the cookie you will be served has if all cookies are equally convenient
to the waitress.

Exercise 38.11. Let X be the number of chocolate chip cookies which are
brought to you if you ask the waitress to randomly select 4 cookies from a
selection of 5 chocolate chip, 3 oatmeal, and 7 peanut butter.

Exercise 38.12. Let X be the number of cookies you will eat that until you
find the 5th one that has more than 7 chocolate chips in it.

Exercise 38.13. Dining hall dinners. A total of 40,000 students on a uni-

versity campus independently choose whether to go to the dining hall for dinner
each day. Each student has dinner there, on a given day, with probability 0.84.

a. What is the distribution of the total number of visits by the students to

the dining hall during a 7-day week?
b. Estimate the probability that 235,000 or fewer students go to the dining
hall during a given week.

Exercise 38.14. Survey. Five hundred people participate in a 12-week sur-

vey to see how many times that they go to a religious service. Each person’s
attendance has a Binomial distribution with n = 12 and p = 0.7. Assume that
the people behave independently with regard to attendance.

a. What is the actual distribution of the number of religious services at-

tended altogether during the 12-week period?
b. Estimate the probability that the 500 people attended between 4170 and
4230 religious services altogether during the 12-week period.

Exercise 38.15. Battery comparison. It is known that the lifetime of Save-

Bucks batteries follows an Exponential distribution with a mean of 40 hours.
The lifetime of LittleMoola batteries follows an Exponential distribution with
a mean of 30 hours.

a. What is the probability a SaveBucks battery will last more than 48 hours?
b. What is the probability a LittleMoola battery will last more than 48
hours?
c. Given that a SaveBucks battery has lasted more than 48 hours, what is
the probability it will last more than 96 hours? What property applies to this
situation?
d. The batteries look identical, so if you have a battery you selected at
random which has lasted more than 48 hours, what is the probability it is a
SaveBucks battery? Which rule do you use to answer this question?
 38.2. Exercises 509

Exercise 38.16. Power outage. The power has gone out in your house, and
you are using a small flashlight to read your probability book. You have three
batteries available to you, all of the same brand, but you can’t tell which brand
in the dark. The flashlight uses only one battery at a time. Use the information
about the two brands of batteries from Exercise 38.15 to answer the questions
below.
a. If the 3 batteries are from SaveBucks, how long do you expect to have
light from your flashlight? What is the standard deviation? What distribution
are you using, and what are the parameters? What important assumption do
you have to make about battery lives?
b. If the 3 batteries are from LittleMoola, how long do you expect to have
light from your flashlight? What is the standard deviation? What distribution
are you using and what are the parameters?
c. What is the probability you will need more than 3 batteries if you use the
flashlight every evening for a week because you forgot to pay your electricity
bill? (Assume there are 12 hours of darkness each night and 7 nights in the
week.) What distribution are you using, and what are the parameters?
Exercise 38.17. Distribution of batteries sold. The SaveBucks battery
company wants to wait until 60% of the batteries are sold at the corner conve-
nience store before they ship more. After years of study, they have determined
that the proportion of batteries which are sold after 6 weeks follows a Beta
distribution with α = 2 and β = 3.
a. What is the probability that more than 60% of the batteries will be sold
at the end of the 6-week period?
b. What is the expected proportion of batteries which will be sold at the
end of the 6-week period?
c. What is the standard deviation in the proportion of batteries which will
be sold at the end of the 6-week period?
Exercise 38.18. Where is my engagement ring? While walking from the
car into your dormitory, you dropped your engagement ring somewhere in the
snow. The path is straight and 30 feet long. You are distraught because the
density of its location seems to be constant along this 30-foot route.
a. What is the probability that the ring is within 12 feet of your car?
b. What is the probability that the ring is between 9 to 11 feet away from
the car?
c. What is the expected distance and the standard deviation for the distance
from the car where the ring fell?
Exercise 38.19. Lifeguard. A lifeguard has to jump in to save a swimmer in
trouble at the lake at Possum Park an average of 2 times a week, according to
a Poisson distribution.
510 Chapter 38. Review of Named Continuous Random Variables

a. What is the probability that he will save more than 25 swimmers in the
10-week season he works?
b. If you know he saved exactly 1 swimmer on his 8-hour shift today, what
is the chance he did it in the last hour that he worked?
c. If there are 5 lifeguards working independently, and they each have the
same average rate for saving swimmers, what is the probability that exactly
three of them will save more than 25 swimmers (each) in the 10-week season?
d. The lifeguards receive a small raise if they save 10 swimmers. How long
will the lifeguard expect to wait for his raise?
e. What is the probability our lifeguard have to wait at least 1 week after
the start of the summer until he saves his first swimmer?
f. How many independent lifeguards will we expect to have to ask until we
find our second lifeguard who had to wait at least 1 week after the start of the
summer until he/she saved his/her first swimmer?

Exercise 38.20. Disease diagnosis. A patient has been diagnosed with a

serious disease. The lifetimes of patients with this disease are approximately
Exponential with expected value 10 years, after their initial diagnosis.

a. What is the probability that a patient who is diagnosed today will still
be alive 15 years from now?
b. Given that a patient is still alive after 5 years, what is the probability
the patient will still be alive 15 years after the initial diagnosis?
c. If 5 people are diagnosed with this disease today, what is the probability
that all 5 of them will be alive 15 years from now?

Exercise 38.21. Nursing care. A nurse specializes in caring for patients

diagnosed with the disease described in Exercise 38.20.

a. The nurse cares for one patient at a time since the care is given in the
patient’s home, and constant monitoring of the patient is needed. She begins
caring for that patient as soon as the diagnosis is given, and she cares for the
patient until death. Then she is reassigned to a new patient. What is the
expected time until her 3rd patient dies?
b. The nurse wakes up and realizes that her patient’s “call” button is lit.
Unfortunately she doesn’t know how long it has been lit. She knows that it
wasn’t lit 10 minutes ago, when she fell asleep. So it could have been pressed
at anytime during the previous 10 minutes, and she is worried that her patient
has been waiting too long. What is the probability that the patient pressed the
call button in the first 30 seconds that the nurse was asleep?
c. Given that the call button was not pressed in the first 30 seconds, what
is the probability it was pressed during the last 2 minutes?
 38.2. Exercises 511

Exercise 38.22. Apple juice. The fraction of juice which can be squeezed
from an apple is a random variable with the following density:
(
k(1 − x)6 if 0 ≤ x ≤ 1
fX (x) =
0 elsewhere

a. Find the value of k that makes this a valid probability density function.
b. What is the expected fraction of juice which can be squeezed from an
apple?
c. What is the standard deviation in the fraction of juice which can be
squeezed from an apple?
d. What is the probability that an apple will have more than half of the
juice squeezed?
Exercise 38.23. Identify the distribution. If X has a mean 0.5 and a
variance 0.25, find the parameters of the distribution (if possible) if X is:
a. Binomial
b. Exponential
c. Normal
d. Uniform
e. Geometric
Exercise 38.24. Identify the distribution. If X has a mean 3 and a variance
2.5, find the parameters of the distribution (if possible) if X is:
a. Binomial
b. Exponential
c. Normal
d. Uniform
e. Geometric
Exercise 38.25. Cable repairs. The cable technician guarantees he’ll arrive
sometime between 8 AM and noon, but he can’t be any more specific than that.
a. What is the probability that he will come in between 9:30 and 10:45 AM?
b. You have a meeting at noon, and you’re hoping that the cable technician
will come before 11:30 AM so you’ll have time to get to the meeting. If the
cable technician has not come by 10 AM, what is the probability that he will
come before 11:30 AM?
c. What is the probability that the next 4 times you need the cable techni-
cian to come out, he will come between 9:30 AM and 10:45 AM on at least 3 of
those occasions?
512 Chapter 38. Review of Named Continuous Random Variables

Exercise 38.26. Cell phone usage. The amount of time you use on your cell
phone each month is approximately Normally distributed with a mean of 623
minutes and a standard deviation of 24 minutes.

a. What is the probability that you will talk more than 700 minutes (when
the extra charges apply) next month?
b. What range of talk-times represents the middle 50% of the times you will
talk?
c. How many monthly bills do you expect to receive until you get the 6th
one with extra charges (talk more than 700 minutes)?

Exercise 38.27. Student health center. The arrival times of students at

the health center at a university is known to have a Poisson distribution with
an average rate of 6 sick students arriving per hour.

a. What is the expected number of students who arrive in the next 30

minutes?
b. What is the expected time that will elapse between the 3rd and 4th
arrivals? What is the standard deviation?
c. What is the expected time that will elapse between the 4th and 7th
arrivals? What is the standard deviation?
d. What is the probability that there will be at least 15 minutes between
each of the next 3 arrivals?
e. What is the probability there will be fewer than 3 visits to the health
center during the next hour?
f. The health center is open from 9 AM to 5 PM. What is the probability
that there will be exactly one hour (9–10, 10–11, etc.) with strictly fewer than
3 visits?
g. Yesterday, a temporary secretary was working the admissions desk. He
kept records on how many students visited each hour, but not their exact times
of arrival. If we know exactly one student visited the health center during the
2–3 PM hour, what is the probability the student arrived within 5 minutes of
1:30?
h. The fraction of students who leave the health center each day with a pre-
scription for little blue Amoxycillin pills is modeled by the following probability
density function:
(
k(x2 − x3 ) if 0 ≤ x ≤ 1
fX (x) =
0 elsewhere
What is the expected fraction of students who will receive this prescription next
week? (Hint: It might help to solve for k first.)
 38.2. Exercises 513

Exercise 38.28. Accident insurance. An insurance company expects that

10% of its moderate-risk drivers will be involved in an accident during the first
31 days of the year.
a. What is the average number of days that you would expect to wait for a
moderate-risk driver, chosen at random, to be involved in an accident?
b. What portion of moderate-risk drivers are expected to be involved in an
accident during the next quarter (90 days)?
Exercise 38.29. More accident insurance. For low-risk drivers, the waiting
time for their first accident follows an Exponential distribution with a mean of
7 years. For high-risk drivers, the waiting time for their first accident follows
an Exponential distribution with a mean of 2 years. Assuming that low-risk
drivers and high-risk drivers are independent,
a. What is the probability that a low-risk driver and a high-risk driver will
both have their first accident within the next 3 years?
b. What is the probability that either a low-risk driver or a high-risk driver
will have an accident within the next 3 years?
c. Given that the low-risk driver will have an accident within the next 3
years, what is the probability that the low-risk driver will also have an accident
within the next 3 years afterwards (i.e., years 4 through 6)?
d. Give an example of how a low-risk driver and a high-risk driver would
NOT be independent. How reasonable is our assumption of independence? Why
is the assumption of independence so important for our calculations?
Exercise 38.30. Renter’s insurance. A renter’s insurance policy is written
to cover a loss, X, where X has a Uniform distribution with boundaries ranging
from no loss up to a maximum of $2000.
a. What is the expected loss?
b. What is the variance in the loss?
c. If a deductible is set at $500, what is the probability that the loss would
be above the deductible?
d. How much money will the insurance company expect to pay on a claim
if the deductible is set at $500?
e. How much money will the renter expect to pay on a claim if the deductible
is set at $500?
f. At what level must the deductible be set in order for the expected amount
of money the insurance company pays on a claim to be only 10% of what it would
be if there were no deductible?
Exercise 38.31. Stock investments. Three investments in high-tech stocks
are thought to have returns which are Exponential random variables with means
514 Chapter 38. Review of Named Continuous Random Variables

of 2, 3, and 4. Determine the probability that the maximum return from these
3 investments is more than 5. (Hint: Think about the opposite case.)

Exercise 38.32. Computer lifetimes. The lifetime of a computer component

purchased at World Wide Computers is Exponentially distributed with a mean
of 2 years.

a. What is the probability that the component fails in the first year?
b. What is the probability that the component fails in the second or third
year (between the 1-year mark and the 3-year mark)?
c. What is the probability that the component fails after the third year?

Exercise 38.33. Computer refunds. Use the information in Exercise 38.32

about the computer components. If the component fails during the first year
after purchase, the owner will refund the buyer’s purchase price. If the compo-
nent fails during the second or third year after purchase, he will refund 25% of
the buyer’s purchase price. Each component is purchased for $300.

a. How much does the owner expect to pay in refunds if a single component
is purchased?
b. How much does the owner expect to pay in refunds if 50 components are
purchased?

Exercise 38.34. Energy-saving bulb. The time an energy-saving light bulb

will last without failing has an Exponential distribution with a median of 3
years. Calculate the probability that a light bulb, selected at random, will
last at least 4 years. (Hint: Use the definition of median to help you get any
parameters you may need.)
Part VII

Additional Topics

In this part, we discuss some more advanced topics that not all courses will
have time to cover but are still important and interesting. Depending on the
interests of the students and teachers, some of these topics (like moment gener-
ating functions) may be introduced earlier in the course. These topics regularly
appear on the SOA/CAS P/1 exam.
By the end of this part of the book, you should be able to:

1. Calculate the variance of sums of random variables.

2. Calculate the covariance and correlation, and understand how these terms
define the relationship between two random variables.

3. Calculate conditional expectations.

4. Use the Markov and Chebyshev inequalities to calculate limiting proba-

bilities.

5. Calculate the PDF, CDF, joint PDF, and joint CDF for order statistics
for independent, identically distributed, continuous random variables.

6. Understand what moment generating functions are and how they can be
utilized.

7. Calculate various moments.

8. Transform random variables for calculations.

Math skills you will need: determinants, partial derivatives.

Additional resources: Calculators or a mathematical computation program
may be used to assist in the calculations.

515
Chapter 39

Variance of Sums; Covariance;

Correlation

William Feller was a probability theorist at Princeton University. One day he

and his wife wanted to move a large table from one room of their large house
to another, but, try as they might, they couldn’t get it though the door. They
pushed and pulled and tipped the table on its side and generally tried everything
they could, but it just wouldn’t go.
Eventually, Feller went back to his desk and worked out a mathematical
proof that the table would never be able to pass through the door.
While he was doing this, his wife got the table through the door.
—Professor Stewart’s Hoard of Mathematical Treasures: Another Drawer
from the Cabinet of Curiosities by Ian Stewart (Basic Books, 2009)

I believe there is a direct correlation between love and laughter.

—Yakov Smirnoff

This morning, your mom made two different types of cookies, peanut butter
and oatmeal raisin. She put them all in a big cookie jar in the kitchen. You
and your brother both randomly (and hungrily!) grab cookies out of the jar, but
your brother grabs first. Does the number of peanut butter cookies your brother
grabs have an impact on the number of peanut butter cookies you later grab?
What are the average and variance in the number of peanut butter cookies
selected by each of you?

39.1 Introduction
Many times during this course, we have used the fact that the expected value
of a sum of random variables is equal to the sum of the expected values of the

516
 39.2. Motivation for Covariance 517

random variables, i.e.,

E(X1 + · · · + Xn ) = E(X1 ) + · · · + E(Xn ).
This nice fact holds regardless of whether the Xj ’s are independent or dependent.
It works essentially because the sums or integrals used for the expected values
are both linear, e.g., if X and Y are continuous:
Z ∞Z ∞
E(X + Y ) = (x + y)fX,Y (x, y) dx dy
Z−∞∞ Z−∞
∞ Z ∞Z ∞
= xfX,Y (x, y) dx dy + yfX,Y (x, y) dx dy
−∞ −∞ −∞ −∞
= E(X) + E(Y ) (39.1)
Unfortunately, the analogous concept is more subtle for variances. If the Xj ’s
are independent, then the variance of the sum of random variables is equal to
the sum of the variances of the random variables. (Without the independence
of the Xj ’s, this rule falls apart!) The reason that things do not work quite as
nicely for variances is that Var(X) = E(X 2 ) − (E(X))2 does not have the same
linearity properties used above, to show E(X + Y ) = E(X) + E(Y ).

39.2 Motivation for Covariance

If we dive deeper into Var(X1 + · · · + Xn ), we recall from Section 12.2 that the
variance of a random variable Y is
Var(Y ) = E(Y 2 ) − (E(Y ))2 .
Pn
Using Y = i=1 Xi ,
this gives
n
X  n
 X n
2   X 2
Var Xi = E Xi − E Xi . (39.2)
i=1 i=1 i=1

For the left term in (39.2), we just multiply to expand ( ni=1 Xi )2 as follows:
P

X n 2 X n Xn
Xi = Xi Xj
i=1 i=1 j=1
n
XX n
= Xi Xj ,
i=1 j=1

but the expected value of the sum equals the sum of the expected values, so the
left term in (39.2) is
 Xn 2  X n Xn 
E Xi =E Xi Xj
i=1 i=1 j=1
n
XX n
= E(Xi Xj ).
i=1 j=1
518 Chapter 39. Variance of Sums; Covariance; Correlation

For thePright term in (39.2), we first convert the expected

Pn value of the sum,
n
i.e., E( i=1 Xi ), into the sum of expected values, i.e., i=1 E(Xi ), and then
multiply to expand, so
n
 X n
2  X 2
E Xi = E(Xi )
i=1 i=1
n
X n
X
= E(Xi ) E(Xj )
i=1 j=1
n X
X n
= E(Xi )E(Xj ).
i=1 j=1

Thus, equation (39.2) becomes

n
X  n X
X n n X
X n
Var Xi = E(Xi Xj ) − E(Xi )E(Xj ),
i=1 i=1 j=1 i=1 j=1

or equivalently, collecting analogous terms, we get

n
X  n X
X n
Var Xi = (E(Xi Xj ) − E(Xi )E(Xj )). (39.3)
i=1 i=1 j=1

A key point from all of these manipulations with the sums and the squaring is
that the terms E(Xi Xj ) − E(Xi )E(Xj ) play a fundamental role in the variance
of the sum of random variables.

39.3 Properties of the Covariance

Due to the primary importance of E(Xi Xj ) − E(Xi )E(Xj ) above, we introduce
the concept of the covariance of two random variables, which can be defined
in two equivalent ways:

Definition 39.1. Covariance of two random variables

The covariance of two random variables X and Y is defined as

Cov(X, Y ) = E((X − E(X))(Y − E(Y ))).

Another form of the covariance is very commonly used, as we briefly explain:

E((X − E(X))(Y − E(Y ))) = E (XY − Y E(X) − XE(Y ) + E(X)E(Y ))

= E(XY ) − E (Y E(X))
− E (XE(Y )) + E (E(X)E(Y ))

We note that E(X) is constant, so it can be pulled out of the term E (Y E(X)) =
E(Y )E(X). Similarly, E(Y ) is a constant, so it can be pulled out of the term
 39.3. Properties of the Covariance 519

E (XE(Y )) = E(X)E(Y ). Finally, E(X) and E(Y ) are both constants, and thus
E (E(X)E(Y )) = E(X)E(Y ). So we obtain

E((X − E(X))(Y − E(Y ))) = E(XY ) − E(Y )E(X) − E(X)E(Y ) + E(X)E(Y ),

which simplifies to just

E((X − E(X))(Y − E(Y ))) = E(XY ) − E(X)E(Y ).

Therefore, we have the following form, which we often utilize:

Theorem 39.2. Covariance of two random variables The covariance of

two random variables X and Y can be expressed as

Cov(X, Y ) = E(XY ) − E(X)E(Y ).

The covariance gives some information about how far two random variables
are spread from their expected values. Just like with the expected value or the
variance, the covariance is only one number, so it can only provide a relatively
small amount of information. Also, it is possible that many different pairs of
X’s and Y ’s could have exactly the same covariance. So the covariance does not
tell the whole story about the relationship between two variables, but it does
help. Very informally—just to build intuition—when a covariance between X
and Y is positive, then the larger X turns out to be, the larger Y will tend to
be, and vice versa. If the covariance is negative, then the larger that one of the
random variables tends to be, the smaller the other will tend to be.
As an informal example, suppose X and Y are indicator random variables
for the events that Alicia and Brent (respectively) get a peanut butter cookie
from the same cookie jar (which contains several types of cookies). When Alicia
gets a peanut butter cookie, then X = 1, and there is one less peanut butter
cookie remaining for Brent, so Y becomes less likely to be 1 than it would be
otherwise. Thus, X and Y have negative covariance. We elaborate on this
specific example in Example 39.9.
Using the notation in equation (39.3), we conclude that the variance of the
sum of random variables is equal to the sum of covariances of all pairs of the
random variables:

Theorem 39.3. Variance of the sum of random variables (version 1)

If X1 , . . . , Xn are random variables (that are not necessarily independent),
then
n X
X n
Var(X1 + · · · + Xn ) = Cov(Xi , Xj ).
i=1 j=1

Using n = 1, we see that the covariance of a random variable with itself is

exactly equal to the variance of the random variable:

Cov(X, X) = E((X − E(X))2 ) = Var(X),

520 Chapter 39. Variance of Sums; Covariance; Correlation

which can also be recognized by writing

Cov(X, X) = E(XX) − E(X)E(X)

= E(X 2 ) − (E(X))2
= Var(X).

Corollary 39.4. Covariance of a random variable with itself

The covariance of X with itself is equal to the variance of X:

Cov(X, X) = Var(X).

Isolating all n of the terms of this flavor from Theorem 39.3, we get another
way to express the variance of the sum of random variables:

Corollary 39.5. Variance of the sum of random variables (version 2)

If X1 , . . . , Xn are random variables (that are not necessarily independent),
then
n
X n X
X
Var(X1 + · · · + Xn ) = Var(Xi ) + Cov(Xi , Xj ).
i=1 i=1 j6=i

We also note that the covariance is symmetric in terms of the two variables,
i.e.,

Cov(X, Y ) = E(XY ) − E(X)E(Y )

= E(Y X) − E(Y )E(X)
= Cov(Y, X).

Theorem 39.6. Covariance is symmetric

The covariance of X and Y equals the covariance of Y and X:

Cov(X, Y ) = Cov(Y, X).

With this symmetry in mind, we see that each term Cov(Xi , Xj ) in Corol-
lary 39.5 will also appear as a Cov(Xj , Xi ) term. So the following formulation
makes calculations shorter to perform, since we are combining all of these du-
plicated pairs:

Corollary 39.7. Variance of the sum of random variables (version 3)

If X1 , . . . , Xn are random variables (that are not necessarily independent),
then
n
X XX
Var(X1 + · · · + Xn ) = Var(Xi ) + 2 Cov(Xi , Xj ).
i=1 i<j
 39.4. Examples of Covariance 521

We have already observed that for any random variables X and Y and
functions g and h, if X and Y are independent, then:

E(g(X)h(Y )) = E(g(X))E(h(Y )).

(Recall: This was proved for discrete random variables in Section 12.2 and for
continuous random variables in Section 29.4. The motivation for this is that
independence of X and Y allows the joint mass pX,Y (x, y) to be factored into
pX (x)pY (y), or in the continuous case, the joint density fX,Y (x, y) to be factored
into fX (x)fY (y). So the double sum or double integral for E(g(X)h(Y )) can be
recomputed as the product of two sums or two integrals, i.e., E(g(X))E(h(Y )).)
As a result, using simply g(X) = X and h(Y ) = Y , we see that if X and Y
are independent, then the covariance of X and Y is 0, i.e.,

Cov(X, Y ) = E(XY ) − E(X)E(Y ) = 0.

The covariance may or may not be 0 when X and Y are dependent; in fact, the
covariance of dependent random variances is usually not 0. (For the interested
reader, however, we give Examples 39.14 and 39.15, which show dependent
random variables that nonetheless have 0 covariance.)

Theorem 39.8. Covariance of independent random variables is 0

The covariance of two independent random variables X and Y is 0:

Cov(X, Y ) = 0.

This nice little fact, used in the context of Corollary 39.5 or Corollary 39.7,
reinforces our understanding that, for independent random variables X1 , . . . ,
Xn , the variance of the sum of the Xj ’s equals the sum of the variances of the
Xj ’s:
Var(X1 + · · · + Xn ) = Var(X1 ) + · · · + Var(Xn ).
We put two more facts about covariance into Section 39.5: (1) the covariance
of two sums equals two sums of covariances, and (2) constants can be factored
out of covariances. (This is usually referred to as linearity.) First, however, we
study some examples about covariance.

39.4 Examples of Covariance

Example 39.9. Consider an example in which Alicia grabs a cookie (without

replacement) from a cookie jar that has 13 cookies, 5 of which are peanut butter.
Let X = 1 if she gets a peanut butter cookie, and X = 0 otherwise, i.e., X is
an indicator for whether she gets a peanut butter cookie. After Alicia picks, let
Brent grab a cookie from the 12 that remain, and let Y = 1 if he gets a peanut
butter cookie, or Y = 0 otherwise. We find several properties of X and Y .
 522 Chapter 39. Variance of Sums; Covariance; Correlation

Find the mass of X and the mass of Y :

When we focus on just one of the random variables at a time, the other
random variable does not affect the situation. The mass of X is

pX (1) = 5/13, and pX (0) = 8/13.

The mass of Y is the same as the mass of X, although this might take a minute’s
thought to justify: Brent is equally likely to get any of the 13 cookies, so his
chances of getting a peanut butter cookie must be 5/13. To see this rigorously,
we can compute using the Bayes’ rule, i.e., by conditioning on the value of X,
i.e., by conditioning on whether Alicia received a peanut butter cookie:

P (Y = 1) = P (Y = 1 and X = 0) + P (Y = 1 and X = 1)
= P (Y = 1 | X = 0)P (X = 0) + P (Y = 1 | X = 1)P (X = 1)
= (5/12)(8/13) + (4/12)(5/13)
= 5/13

Remember that So Y has the same mass as X. In other words, Brent and Alicia are each equally
indicator random likely to get a peanut butter cookie. On the other hand, when we take Alicia’s
variables are just situation into account first, there is an effect on the ability of Brent to get a
Bernoulli random cookie. So X and Y are dependent. When Alicia does not get a peanut butter
variables, i.e., just 1
cookie, then Brent is more likely to get one, because more peanut butter cookies
or 0, depending on
whether some event remain, i.e.,
happens. P (Y = 1 | X = 0) = 5/12.
When she does get one, then he is less likely to get one, because there are fewer
peanut butter cookies remaining, i.e.,

P (Y = 1 | X = 1) = 4/12.

So
P (Y = 1 | X = 0) 6= P (Y = 1 | X = 1),
and these are both different from P (Y = 1) = 5/13. So Y is dependent on X.
Find the expected value of X + Y :
Even though X and Y are dependent, we can calculate the expected value
of X + Y , the total number of peanut butter cookies eaten by Alicia and Brent
altogether, by adding the expected values:

E(X + Y ) = E(X) + E(Y ) = 5/13 + 5/13 = 10/13.

(Note that E(X) = 5/13 since X is a Bernoulli random variable, and similarly,
E(Y ) = 5/13 since Y is also a Bernoulli.)
Find the variance of X + Y :
Since X and Y are dependent, we cannot just get the variance of X + Y by
adding the variances of X and Y separately. We do still know that Var(X) =
 39.4. Examples of Covariance 523

(5/13)(8/13) since X is Bernoulli, and similarly, Var(Y ) = (5/13)(8/13) since

Y is Bernoulli.
We have Cov(X, Y ) = E(XY )−E(X)E(Y ). We emphasize that the product
XY itself is a Bernoulli random variable because XY can only take on values
0 or 1. The only way to have XY = 1 is to have X = 1 and Y = 1 (otherwise,
X = 0 or Y = 0 so XY = 0). In other words,

E(XY ) = 1P (X = 1 and Y = 1)
+ 0P (X = 0 and Y = 1)
+ 0P (X = 1 and Y = 0)
+ 0P (X = 0 and Y = 0)

So we just have
E(XY ) = P (X = 1 and Y = 1).
To get X = 1 and Y = 1, we need X = 1, and then conditioned on X = 1, we The product of
need Y = 1 too. If both Alicia and Brent want peanut butter cookies, Alicia Bernoullis is a
needs to get one, and then (given that she got one), Brent will need one too. Bernoulli too.
So

E(XY ) = P (X = 1 and Y = 1)
= P (Y = 1 | X = 1)P (X = 1)
= (4/12)(5/13).

So
Cov(X, Y ) = (4/12)(5/13) − (5/13)(5/13).
In summary,

Var(X + Y ) = Var(X) + Var(Y ) + Cov(X, Y ) + Cov(Y, X)

= (2)(5/13)(8/13) + (2)((4/12)(5/13) − (5/13)(5/13))
= 220/507
= 0.4339

Example 39.10. Consider an eight-hour work day. Huiping and Ravi each
check their email just once per day. Let X and Y denote the time, respec-
tively, until Huiping and Ravi each check their emails during the day. Assume
that Huiping always checks email first, and that X and Y are Uniform on the
region where 0 ≤ X ≤ Y ≤ 8, so that the joint density is constant on the
triangle in Figure 39.1. Since the joint density is constant on a triangle of area
(8)(8)(1/2) = 32, then the joint density must be the reciprocal of the area, i.e.,

fX,Y (x, y) = 1/32 for 0 ≤ x ≤ y ≤ 8,

and fX,Y (x, y) = 0 otherwise.

524 Chapter 39. Variance of Sums; Covariance; Correlation

y
8
7
6
5
4
3
2
1
x
1 2 3 4 5 6 7 8

Figure 39.1: Region where X and Y have positive density.

Find the expected value and variance of Y − X, which is the time of the interval
in between Huiping and Ravi checking their email.
The expected time E(X) until Huiping checks email is obtained by integrat-
ing xfX,Y (x, y) over the triangle. We can use integration with respect to y, for
the outer integral; we can use integration over the x’s in the range 0 ≤ x leqy,
for the inner integral. This setup is shown in Figure 39.2.
y
8
7
6
5
4
3
2
1
x
1 2 3 4 5 6 7 8

Figure 39.2: Fixed value of y (here, for example y = 3.2), and x ranging
from 0 to y.

So the expected time until Huiping checks email is

Z 8Z y
E(X) = x/32 dx dy = 8/3.
0 0

The expected time until Ravi checks email is

Z 8Z y
E(Y ) = y/32 dx dy = 16/3.
0 0
 39.4. Examples of Covariance 525

Thus
E(Y − X) = E(Y ) − E(X) = 16/3 − 8/3 = 8/3.
We cannot simply add the variances of X and Y , since X and Y are dependent.
So we compute

Var(Y − X) = E((Y − X)2 ) − (E(Y − X))2

= E((Y − X)2 ) − (8/3)2 ,

and now we need to integrate (y − x)2 fX,Y (x, y) over the triangle, to get the
value of E((Y − X)2 ). Since fX,Y (x, y) = 1/32 on the triangle, we have
Z 8Z y
2
E((Y − X) ) = (y − x)2/32 dx dy
0 0
= 32/3.

So, in summary,

Var(Y − X) = 32/3 − (8/3)2

= 32/9.

As an alternative way to find the variance of Y −X, we could use the covariance:

Var(Y − X) = Cov(Y − X, Y − X)
= Cov(Y, Y ) + (−1) Cov(Y, X)
+ (−1) Cov(X, Y ) + (−1)(−1) Cov(X, X)
= Var(X) + Var(Y ) − 2 Cov(X, Y ).

The variances of X and Y are:

Z 8Z y
Var(X) = x2/32 dx dy − (8/3)2 = 32/3 − (8/3)2 = 32/9
0 0

and
Z 8Z y
Var(Y ) = y 2/32 dx dy − (16/3)2 = 32 − (16/3)2 = 32/9,
0 0

and the covariance is

Z 8Z y 
2 Cov(X, Y ) = 2 xy/32 dx dy − (8/3)(16/3)
0 0
= 2(16 − (8/3)(16/3))
= 32/9.

So we have verified that

Var(Y − X) = 32/9 + 32/9 − 32/9 = 32/9.

 526 Chapter 39. Variance of Sums; Covariance; Correlation

Example 39.11. Consider a selection of 10 songs on a playlist, of which exactly

7 are rock songs. If two such songs are chosen, and repetition is not allowed,
and all outcomes are equally likely, let X denote the number of songs that are
rock songs.

We let X1 and X2 be indicator random variables that denote whether the first
Remember that, for and second songs (respectively) are rock songs. Thus the expected number of
an indicator (i.e., a rock songs chosen is
Bernoulli random
variable) with E(X) = E(X1 + X2 ) = E(X1 ) + E(X2 ) = 7/10 + 7/10 = 14/10 = 7/5 = 1.4.
probability of
success p, and Now consider the variance of X. We have
probability of failure
q = 1 − p, the Var(X) = Var(X1 + X2 ) = Var(X1 ) + Var(X2 ) + 2 Cov(X1 , X2 ).
expected value is p
and the variance is Each of the Xj ’s is a Bernoulli random variable, so
pq.
Var(Xj ) = P (Xj = 1)P (Xj = 0)
= (probability song is rock)(probability song is not rock)
= (7/10)(3/10)
= 21/100.

Also
Cov(X1 , X2 ) = E(X1 X2 ) − E(X1 )E(X2 ).
We know E(X1 ) = 7/10 and E(X2 ) = 7/10. Also X1 and X2 are Bernoullis, so
X1 X2 is Bernoulli too. Thus the expected value of X1 X2 is just equal to the
probability that X1 X2 is equal to 1. So

E(X1 X2 ) = P (X1 X2 = 1)
= P (X1 = 1 and X2 = 1)
= P (X1 = 1)P (X2 = 1 | X1 = 1).

As before, P (X1 = 1) = 7/10. If X1 = 1, then the second song is equally likely

to be any of 9 remaining songs, of which 6 are rock songs. Thus

E(X1 X2 ) = P (X1 = 1)P (X2 = 1 | X1 = 1)

= (7/10)(6/9).

Putting this all together, we have

21 21  7  6   7  7 
Var(X) = + +2 −
100 100 10 9 10 10
= 28/75
= 0.37333.
 39.4. Examples of Covariance 527

Notice that the scenarios in Examples 39.9 and 39.11 are specific cases of the
Hypergeometric distribution. E.g., in Example 39.9, X + Y has a Hypergeo-
metric distribution, since we select n = 2 out of N = 13 items (all cookies);
exactly M = 5 (the peanut butter cookies) are desirable. Thus, as we learned
in the Hypergeometric chapter:
 
M 5 10
E(X + Y ) = n = (2) = ,
N 13 13
and (as we will see below in the more general case, in Example 39.12) we have:
   
M M N −n 5 5 13 − 2 220
Var(X + Y ) = n 1− = (2) 1− = .
N N N −1 13 13 13 − 1 507
In Example 39.11, the random variable X1 + X2 is Hypergeometric, because
there are N = 10 songs available altogether; n = 2 of the songs are chosen; and
M = 7 of the songs are desirable.

Example 39.12. Recall from Chapter 19 that a Hypergeometric random vari-

able is defined as follows: Consider N items altogether: M of the items are
“desirable” and the other N − M are “undesirable.” We pick a collection of n
items. Let X denote the number of “desirable” items that we get. Then X is a
Hypergeometric random variable with parameters N , M , and n. The mass of
X is
M N −M



j n−j
pX (j) = P (X = j) = N

.
n
In Chapter 19, we already discussed the fact that a Hypergeometric random
variable X can be decomposed as
X = X1 + X2 + · · · + Xn ,
where Xj = 1 if the jth item selected is desirable, and Xj = 0 otherwise. Thus
E(Xj ) = P (Xj = 1). There are N possible items that can be selected as the
jth item, and M of these items are desirable. So E(Xj ) = P (Xj = 1) = M/N .
Thus
E(X) = E(X1 + · · · + Xn )
= E(X1 ) + · · · + E(Xn )
= M/N + · · · + M/N
= nM/N.
In Remark 19.4 of Chapter 19, we computed the variance of a Hypergeometric
random variable. Although the computation is very similar, it is perhaps helpful
to recast this argument using the covariance. We have
n
X XX
Var(X) = Var(Xj ) + 2 Cov(Xi , Xj ).
j=1 i<j
528 Chapter 39. Variance of Sums; Covariance; Correlation

Since each Xj is a Bernoulli, then

Var(Xj ) = P (Xj = 1)P (Xj = 0)

 M  M
= 1− .
N N
For i < j, we have

Cov(Xi , Xj ) = E(Xi Xj ) − E(Xi )E(Xj ),

and we know E(Xi ) = M/N and E(Xj ) = M/N . Also, Xi and Xj are each 0
or 1, so Xi Xj is 0 or 1. In other words, Xi and Xj are each Bernoulli, so Xi Xj
is Bernoulli too. Thus

E(Xi Xj ) = P (Xi Xj = 1)
= P (Xi = 1 and Xj = 1)
= P (Xi = 1)P (Xj = 1 | Xi = 1).

As before, P (Xi = 1) = M/N . If Xi = 1, then the jth item is equally likely to

be any of N − 1 remaining items, of which M − 1 are desirable. Thus

E(Xi Xj ) = P (Xi = 1)P (Xj = 1 | Xi = 1)

 M  M − 1 
= .
N N −1
Putting this all together, we have
n  X X  M  M − 1   M  M 
X M  M
Var(X) = 1− +2 − .
N N N N −1 N N
j=1 i<j

The n terms of the first type are all alike; none depend on j. The n2 =

(n)(n − 1)/2 terms of the second type are also alike; none of them depend on i
or j. Thus
 M  M (n)(n − 1)  M  M − 1   M  M 
Var(X) = n 1− +2 − .
N N 2 N N −1 N N
After a bit of simplification using the common denominator N 2 (N − 1), this
yields the result that the variance of a Hypergeometric random variable is
 
nM (N − M )(N − n) M M N −n
Var(X) = =n 1− .
N 2 (N − 1) N N N −1

As motivation for the next two examples:

Remark 39.13. Comparing independence and covariance

If two random variables X and Y are independent, then Cov(X, Y ) = 0. The
converse is not necessarily true. For instance, it is possible to have two random
variables X and Y such that Cov(X, Y ) = 0 but X and Y are not independent.
 39.4. Examples of Covariance 529

Consider, for example, the following:

Example 39.14. Let X have mass pX (x) = P (X = x) = 1/5 for x =

−2, −1, 0, 1, 2. Define Y such that Y = 1 if X = 0, and Y = 0 otherwise.

In this example,

E(X) = −2(1/5) − 1(1/5) + 0(1/5) + 1(1/5) + 2(1/5) = 0.

Also XY = 0 always, since either X = 0, or otherwise Y = 0. Thus E(XY ) = 0.

So

Cov(X, Y ) = E(XY ) − E(X)E(Y )

= 0 − (0)E(Y )
= 0.

On the other hand, X and Y are not independent, because the value of Y is
completely dependent on the value of X.

The idea in the previous example can be generalized quite a bit:

Example 39.15. Let X be any random variable with expected value 0, and
let Y be defined such that

Y =0 whenever X 6= 0,

and let
Y 6= 0 otherwise, e.g., let Y = 13 when X = 0.
Then the distribution of Y again depends on the distribution of X, but also X
and Y again have covariance 0, since XY = 0 in all cases in this example, and
thus

Cov(X, Y ) = E(XY ) − E(X)E(Y )

= 0 − (0)E(Y )
= 0.
530 Chapter 39. Variance of Sums; Covariance; Correlation

39.5 Linearity of the Covariance

Theorem 39.16. Covariance of two sums of random variables equals
two sums of covariances
If X1 , . . . , Xn and Y1 , . . . , Ym are random variables (not necessarily indepen-
dent), and a1 , . . . , an and b1 , . . . , bm are constants, then
n X
X m
Cov(a1 X1 + · · · + an Xn , b1 Y1 + · · · + bm Ym ) = ai bj Cov(Xi , Yj ).
i=1 j=1

To see this, we compute (in a similar fashion to the argument for sums of
variances from Section 12.3),

Cov(a1 X1 + · · · + an Xn , b1 Y1 + · · · + bm Ym )
X n m
X  X n m
 X 
=E ai Xi bj Yj − E ai Xi E bj Yj
i=1 j=1 i=1 j=1
n X
X m  Xn m
X
=E ai bj Xi Yj − ai E(Xi ) bj E(Yj )
i=1 j=1 i=1 j=1
n
XX m n
XX m
= ai bj E(Xi Yj ) − ai bj E(Xi )E(Yj )
i=1 j=1 i=1 j=1
n X
X m
= ai bj (E(Xi Yj ) − E(Xi )E(Yj ))
i=1 j=1
Xn X m
= ai bj Cov(Xi , Yj )
i=1 j=1

This result is often used in a context without any constants (i.e., all ai = 1 and
bj = 1), because we often need to take the covariance of two sums of random
variables. This happens, for instance, when we want to take the covariance of
a sum of indicators with another sum of indicators. As another example, if n
and m are large, and the Xi ’s are independent from each other, and the Yj are
independent from each other, but the Xi ’s and Yj ’s have some dependencies,
then X1 + · · · + Xn is approximately Normal, and Y1 + · · · + Ym is approximately
Normal too, and the covariance tells us how the two sums are related.

Corollary 39.17. Covariance of two sums of random variables equals

two sums of covariances (version without constants)
If X1 , . . . , Xn and Y1 , . . . , Ym are random variables (not necessarily indepen-
dent), then
n X
X m
Cov(X1 + · · · + Xn , Y1 + · · · + Ym ) = Cov(Xi , Yj ).
i=1 j=1
 39.6. Correlation 531

When m = 1 and n = 1, and we write a1 = a and b1 = b, we get another

useful form:

Corollary 39.18. Constants can be factored out of covariances

If X and Y are random variables (not necessarily independent), and a and b
are coefficients, then

Cov(aX, bY ) = ab Cov(X, Y ).

For instance, if X is the charge for an oil change, and a = 1.07, then aX is the
charge for an oil change, with the 7% additional tax included. Similarly, if Y is
the number of containers of oil that are needed, and b is the price per container,
then bY is the price of the oil itself. Thus we can switch from Cov(aX, bY ) to
instead focusing on ab Cov(X, Y ). This could be helpful, especially if the charge
for an oil change is directly related to the number of containers of oil itself. The
other things, like the tax and the charge per container of oil, can be factored out
of the covariance equation, to simplify the situation and allow us to calculate
Cov(X, Y ) without the complications of tax or charge per container of oil.

39.6 Correlation

The correlation of two random variables X and Y is a one-number summary of

the relationship between X and Y . The correlation is always between −1 and
+1, as we will see below. A correlation near 1 indicates that Y basically increases
as X increases. A correlation near −1 indicates that Y tends to decrease as X
increases. A correlation near 0 indicates that Y is basically unaffected by the
increase or decrease of X. These are all imprecise statements, as they must be,
because the correlation is only a one-number way of stating how X and Y are
related. The relationship between X and Y might be quite complicated, and
one number (such as the correlation) cannot fully describe the relationship. So
a rough approximation to the behavior of the correction is all that we can hope
to quantify with such a one-number summary.

Example 39.19. If X is the number of hours that a student spends studying,

and if Y is the score a student earns on an exam, then Y will (roughly speaking)
increase as X increases. So the correlation will be a positive number and might
even be somewhat close to 1, which indicates that more studying is strongly
correlated to better performance on an exam. Of course, Y is capped (often at,
say, Y = 100 for a perfect score), and X may only be capped by the possible
studying time allowed before the exam takes place. So the correlation will
probably not be exactly 1, but might be high, say, 0.723 or 0.881.
532 Chapter 39. Variance of Sums; Covariance; Correlation

In the previous example, if X had been, for instance, time spent procrastinating,
then X and Y would have had a negative correlation.
The correlation of X and Y is defined as follows:
Definition 39.20. Correlation of two random variables
The correlation of two random variables X and Y , usually written as ρ, is
defined as
Cov(X, Y )
ρ(X, Y ) = p .
Var(X) Var(Y )
Note: The variance is only zero when a random variable is constant. So,
as long as X and Y are not constant, then the correlation between them is
well-defined.
To see that the correlation of two random variables X and Y is always
between −1 and +1, it is very helpful to write σX and σY as the standard 
deviations of X and Y , and then to study how Var σXX + σYY and Var σXX − σYY
behave. In fact, the correlation is like the covariance, but scaled by the sizes of
the variances of the random variables.
We note that variances are always positive, so
X Y 
0 ≤ Var +
σX σY
X Y  X Y  Y X
= Var + Var + Cov , + Cov ,
σX σY σX σY σY σX
but covariance is symmetric so the last two terms above
  are the same. Also, we
can factor out the standard deviations to get Var σXX = σ12 Var(X) = 1 and
   X

Var σY = σ2 Var(Y ) = 1. So we get 0 ≤ 2 + 2 Cov σX , σY , i.e.,

Y 1 X Y
Y
X Y 
−1 ≤ Cov , . (39.4)
σX σY
Similarly,
X Y 
0 ≤ Var −
σX σY
X Y  X Y  Y X
= Var + Var − Cov , − Cov ,
σX σY σX σY σY σX
 
but again we simplify to get 0 ≤ 2 − 2 Cov σXX , σYY , i.e.,
X Y 
1 ≥ Cov , . (39.5)
σX σY
Putting together (39.4) and (39.5) we get
X Y 
−1 ≤ Cov , ≤ 1,
σX σY
 39.6. Correlation 533
Cov(X,Y )
but σX σY = ρ(X, Y ), so in other words,

−1 ≤ ρ(X, Y ) ≤ 1.

Theorem 39.21. Correlation of two random variables is always be-

tween −1 and +1
The correlation ρ(X, Y ) of two random variables X and Y is always between
−1 and +1:
−1 ≤ ρ(X, Y ) ≤ +1.

Example 39.22. Let X be a Uniform random variable on the interval [0, 1],
and let Y = X 2 . Find the correlation between X and Y .

We see that
Z 1 Z 1
E(X) = (x)(1) dx = 1/2 and 2
E(X ) = (x2 )(1) dx = 1/3,
0 0

so
Var(X) = 1/3 − (1/2)2 = 1/12.
Also
Z 1
E(Y ) = E(X 2 ) = 1/3 and E(Y 2 ) = E(X 4 ) = (x4 )(1) dx = 1/5,
0

so
Var(Y ) = 1/5 − (1/3)2 = 4/45.
Also
Cov(X, Y ) = E(XY ) − E(X)E(Y ) = E(X 3 ) − E(X)E(Y ),
R1
and we know E(X) = 1/2 and E(Y ) = 1/3, and we see that E(X 3 ) = 0 x3 dx =
1/4, so
Cov(X, Y ) = 1/4 − (1/2)(1/3) = 1/12.
Thus X and Y are positively correlated, and the correlation between X and Y
is √
Cov(X, Y ) 1/12 15
ρ(X, Y ) = p p =p p = = 0.968.
Var(X) Var(Y ) 1/12 4/45 4

Example 39.23. Roll a die. Let X denote the value on the die. Let Y = 1 if
the die is 4, 5, or 6, and let Y = 0 otherwise. Find the correlation between X
and Y . Informally, when Y = 1, we know X must be larger (than when Y = 0).
So we anticipate X and Y being strongly positively correlated.
534 Chapter 39. Variance of Sums; Covariance; Correlation

We see that
6
X 6
X
E(X) = (j)(1/6) = 7/2, and 2
E(X ) = (j 2 )(1/6) = 91/6,
j=1 j=1

so
Var(X) = 91/6 − (7/2)2 = 35/12.
Also
E(Y ) = (3/6)(1) + (3/6)(0) = 1/2,
and
E(Y 2 ) = (3/6)(12 ) + (3/6)(02 ) = 1/2,
so
Var(Y ) = 1/2 − (1/2)2 = 1/4.
Also
Cov(X, Y ) = E(XY ) − E(X)E(Y ),
and we know that

E(XY ) = (1/6)(1)(0) + (1/6)(2)(0) + (1/6)(3)(0)

+ (1/6)(4)(1) + (1/6)(5)(1) + (1/6)(6)(1)
= 5/2.

So we have
Cov(X, Y ) = 5/2 − (7/2)(1/2) = 3/4.
Thus X and Y are positively correlated, and the correlation between X and Y
is
√
Cov(X, Y ) 3/4 3 105
ρ(X, Y ) = p p =p p = = 0.878.
Var(X) Var(Y ) 35/12 1/4 35

Example 39.24. Roll a die. Let X denote the value on the die. Let Y = 1 if
the die is even, i.e., is 2, 4, or 6, and let Y = 0 otherwise. Find the correlation
between X and Y . Informally, when Y = 1, we know X must be slightly larger
(than when Y = 0), because for instance X is 2 not 1, or X is 4 not 3, or X is
6 not 5. So we anticipate X and Y being a little bit positively correlated.

The calculation goes just the same as above, with

E(X) = 7/2,
Var(X) = 35/12,
E(Y ) = 1/2,
Var(Y ) = 1/4,
Cov(X, Y ) = E(XY ) − E(X)E(Y ),
 39.6. Correlation 535

but in this case

E(XY ) = (1/6)(1)(0) + (1/6)(2)(1) + (1/6)(3)(0)

+ (1/6)(4)(1) + (1/6)(5)(0) + (1/6)(6)(1)
= 2,

so
Cov(X, Y ) = 2 − (7/2)(1/2) = 1/4.
Thus X and Y are positively correlated, and the correlation between X and Y
is √
Cov(X, Y ) 1/4 105
ρ(X, Y ) = p p =p p = = 0.293.
Var(X) Var(Y ) 35/12 1/4 35
Thus, X and Y are positively correlated, but are not as strongly correlated as
in the previous example.

Example 39.25. Roll a die. Let X denote the value on the die. Let Y = 1 if
the die is odd, i.e., is 1, 3, or 5, and let Y = 0 otherwise. Find the correlation
between X and Y . Informally, when Y = 1, we know X must be slightly smaller
(than when Y = 0), because for instance X is 1 not 2, or X is 3 not 4, or X is
5 not 6. So we anticipate X and Y being a little bit negatively correlated. This
is completely symmetric to Example 39.24.

Same as above, but now

E(XY ) = (1/6)(1)(1) + (1/6)(2)(0) + (1/6)(3)(1)

+ (1/6)(4)(0) + (1/6)(5)(1) + (1/6)(6)(0)
= 3/2,

so we have
Cov(X, Y ) = 3/2 − (7/2)(1/2) = −1/4.
Thus, X and Y are negatively correlated, and the correlation between X
and Y is
√
Cov(X, Y ) −1/4 105
ρ(X, Y ) = p p =p p =− = −0.293.
Var(X) Var(Y ) 35/12 1/4 35

Example 39.26. Roll a die. Let X denote the value on the die. Let Y = 1
if the die is 1, 2, or 3, and let Y = 0 otherwise. Find the correlation between
X and Y . Informally, when Y = 1, we know X must be smaller (than when
Y = 0). So we anticipate X and Y being have a strong negative correlation.
This is completely symmetric to Example 39.23.
536 Chapter 39. Variance of Sums; Covariance; Correlation

Same as above, but now

E(XY ) = (1/6)(1)(1) + (1/6)(2)(1) + (1/6)(3)(1)

+ (1/6)(4)(0) + (1/6)(5)(0) + (1/6)(6)(0)
= 1,

so
Cov(X, Y ) = 1 − (7/2)(1/2) = −3/4.
Thus X and Y are negatively correlated, and the correlation between X and Y
is
√
Cov(X, Y ) −3/4 3 105
ρ(X, Y ) = p p =p p =− = −0.878.
Var(X) Var(Y ) 35/12 1/4 35

Example 39.27. Consider a random variable X that indicates if event A has

occurred, and another random variable Y that indicates if event B has occurred.

In this example, if the occurrence of B makes A more likely to occur (i.e.,

if P (A | B) > P (A)), then X and Y have positive correlation.
If the occurrence of B makes A less likely to occur (i.e., if P (A | B) < P (A)),
then X and Y have negative correlation.
If the occurrence of B does not affect the probability of A (i.e., if P (A | B) =
P (A)), then X and Y have zero correlation. In fact, X and Y are independent
in this last case.
To see all of this, we compute

E(X) = P (X = 1) = P (A),
Var(X) = P (X = 1)P (X = 0) = P (A)P (Ac ),
E(Y ) = P (Y = 1) = P (B),
Var(Y ) = P (Y = 1)P (Y = 0) = P (B)P (B c ),
E(XY ) = P (X = 1 and Y = 1) = P (A ∩ B),

so
Cov(A, B)
ρ(X, Y ) = p p
Var(X) Var(Y )
P (A ∩ B) − P (A)P (B)
=p p .
P (A)P (Ac ) P (B)P (B c )

The denominator is always positive. Thus, the sign of the correlation of X and
Y only depends on the sign of the numerator.
 39.7. Exercises 537

We see that the numerator is

P (A ∩ B) − P (A)P (B),

which can be rewritten as

P (A | B)P (B) − P (A)P (B) = (P (A | B) − P (A))P (B).

Thus, if P (A | B) > P (A), then P (A | B) − P (A) > 0, so the correlation of

X and Y is positive.
If P (A | B) < P (A), then P (A | B) − P (A) < 0, so the correlation of X
and Y is negative.
If P (A | B) = P (A), then P (A | B) − P (A) = 0, so the correlation of X
and Y is zero, and A and B are independent in this case, so X and Y are
independent too.

39.7 Exercises
39.7.1 Practice
Exercise 39.1. Pizza for lunch. Each day, Amy eats lunch at the cafeteria.
She chooses pizza as her main dish with probability 40%, and her behavior each
day is independent of all the other days. Let X denote the number of days she
chooses pizza in a 10-day period. Let Y = 10 − X denote the number of days
in which Amy does not eat pizza.

a. Are X and Y dependent or independent?

b. Find the covariance Cov(X, Y ) of X and Y .
c. Find the correlation ρ(X, Y ) of X and Y .

Exercise 39.2. Let X be Uniformly distributed on the interval [0, 10], and let
Y = 10 − X.

a. Find the covariance of X and Y .

b. Find the correlation of X and Y .

Exercise 39.3. Client meeting. An accountant must meet with one more
client before he can go home. The amount of time X (in minutes) that he meets
with the client is Uniformly distributed on the interval [40, 60]. The total length
of time Y (also in minutes) that he must remain in the office is 1.3X + 10.

a. Find the covariance of X and Y .

b. Find the correlation of X and Y .
538 Chapter 39. Variance of Sums; Covariance; Correlation

Exercise 39.4. Broken crayons. You are babysitting for two children, Abby
and Bill. They have a bucket of 30 crayons, 20 of which are unbroken, and
the other 10 are broken. They each choose a crayon, without replacement. Let
X = 1 if Abby gets an unbroken crayon, and X = 0 otherwise. Similarly, let
Y = 1 if Bill gets an unbroken crayon, and Y = 0 otherwise.

a. Find the covariance of X and Y .

b. Find the correlation of X and Y .

39.7.2 Extensions
Find (a) the covariance, and (b) the correlation of the random variables X and
Y defined in Exercises 39.5–39.15

Exercise 39.5. Let X be any random variable, and Y = a − X, where “a” is

a constant.

Exercise 39.6. Let X be Uniformly distributed on the interval [0, π], and let
Y = cos X.

Exercise 39.7. Sweet and sour. Henry and Sally each choose 1 candy from
a bag of treats, without replacement. There are 20 sweet candies and 3 sour
candies. Let X = 1 if Henry’s candy is sweet, or X = 0 otherwise. Let Y = 1
if Sally’s candy is sweet, or Y = 0 otherwise.

Exercise 39.8. Knitting sweaters. A grandmother is knitting wool sweaters

to donate to a non-profit organization. She knits pink and blue ones for girls
and boys, respectively. It takes 3 balls of yarn to knit one. She has promised to
donate 10 sweaters, but she has forgotten how many of the recipients are boys
and how many are girls. So the grandmother decides to roll a 6-sided die to
determine how many blue sweaters to knit. Let X be the number of the die
(and, thus, the number of blue sweaters that she knits), and let Y = 10 − X be
the number of pink sweaters that she knits.

Exercise 39.9. Let X and Y have a joint Uniform distribution on the triangle
with corners at (0, 2), (2, 0), and the origin.

Exercise 39.10. Let X and Y be Uniformly distributed on the diamond-shaped

region with corners located at (1, 0), (0, 1), (−1, 0), and (0, −1).

Exercise 39.11. Pizza pieces. Lloyd eats X pieces of pizza, where X is an

integer-valued random variable that is equally likely to be any of the values 1
through 8, inclusive. He takes the extra pizza home to his family, but if he eats
too much pizza, he buys an extra pizza to take home too. Let Y be the number
of pieces that he takes home to his family. Then, if 1 ≤ X ≤ 4, he does not
need to buy any more pizza, so Y = 8 − X is the leftover. On the other hand,
if 5 ≤ X ≤ 8, then Y = 16 − X, because he buys a second pizza in this case.
 39.7. Exercises 539

Exercise 39.12. Let X and Y be jointly distributed on the portion of the

√
Cartesian plane between the curves y = x and y = x2 . Let the joint density
8y
of X, Y be fX,Y (x, y) = 3x in this regions, and fX,Y (x, y) = 0 otherwise. (Notice
R 1 R √x 8y
that this is a density because 0 x2 3x dy dx = 0.)
Exercise 39.13. Roll two dice. Let X denote the maximum value that appears,
and let Y denote the minimum value that appears.
Exercise 39.14. Let X be Uniform on the interval [0, 1], and let Y = eX .
Exercise 39.15. Let X be an Exponentially distributed random variable with
expected value 1/2. Let Y = X 2 .

Exercise 39.16. Consider X and Y such that the joint density fX,Y (x, y) of
X and Y is Uniform on the square where 0 ≤ X, Y ≤ 1. In other words,

fX,Y (x, y) = 1 if 0 ≤ x ≤ 1 and 0 ≤ y ≤ 1,

and fX,Y (x, y) = 0 otherwise.

a. Are X and Y dependent or independent?
b. Find the covariance Cov(X 2 , X + Y ) of X 2 and X + Y .
c. Find the correlation Cov(X 2 , X + Y ) of X 2 and X + Y .
Exercise 39.17. Roll two 4-sided dice (not 6-sided dice). Let X be the mini-
mum value, and let Y be the maximum value. Find the covariance of X and Y .

39.7.3 Advanced
Exercise 39.18. Let X and Y correspond to the horizontal and vertical co-
ordinates in the triangle with corners at (2, 0), (0, 2), and the origin. Let
15
fX,Y (x, y) = 28 (xy 2 + y) for (x, y) inside the triangle, and fX,Y (x, y) = 0
otherwise.
a. Find the covariance of X and Y .
b. Find the correlation of X and Y .
Exercise 39.19. If X and Y have a constant joint density on the triangle where
0 ≤ y ≤ x ≤ 1, compute Cov(X, Y ).
Exercise 39.20. Draw 5 cards, without replacement, from a standard deck of
cards. Let X be the number of hearts selected. Find the variance of X.
Exercise 39.21. Suppose that Sasha picks an integer X at random between 1
and 6 (inclusive). Then Ravi picks a different integer Y at random, from the
5 integers that remain. Assume that all (6)(5) = 30 choices are equally likely.
Find the correlation ρ(X, Y ) of X and Y .
540 Chapter 39. Variance of Sums; Covariance; Correlation

Exercise 39.22. A total of 3n bears are in a bucket: n are red, n are yellow,
and n are blue. A child begins grabbing the bears at random, with all selections
equally likely. The bears are selected “without replacement,” i.e., she never puts
the bears back after she grabs them. Find the variance of the number of bears
she grabs until the first red bear appears.

Exercise 39.23. Roll 10 differently colored, 6-sided dice. Make a list of all
10
triples using the 10 values that appear. Let X be the total number


3 = 120
of such triples for which the three values in the triple agree. Find Var(X).
Chapter 40

Conditional Expectation

Statistically the probability of any one of us being here is so small that you would
think the mere fact of existence would keep us all in a contented dazzlement
of surprise. We are alive against the stupendous odds of genetics, infinitely
outnumbered by all the alternates who might, except for luck, be in our places.
—The Lives of a Cell: Notes of a Biology Watcher by Lewis Thomas
(Viking, 1974)

If a woman is waiting on her spouse, how does her arrival time affect the ex-
pected time until he arrives?

40.1 Introduction

The idea of conditional expectation of one random variable (say, X), given the
value of another random variable (say, Y = y), is to use the conditional mass or
the conditional density to calculate the expected value of one random variable
(X) when the value of the other is known (Y ). In other words, we know some
information ahead of time about one of our random variables that may affect
the mass or density—and, thus, the probabilities and expected value—of the
other random variable.

541
542 Chapter 40. Conditional Expectation

Definition 40.1. Conditional Expectation

In the continuous case, the conditional expectation of X, given Y = y, is
Z ∞
E(X | Y = y) = xfX|Y (x | y) dx.
−∞

In the discrete case, the conditional expectation of X, given Y = y, is

X
E(X | Y = y) = xpX|Y (x | y).
x

40.2 Examples
As an example, returning to the situation in Example 39.23, we have the fol-
lowing:

Example 40.2. Roll a die. Let X denote the value on the die. Let Y = 1 if
the die is 4, 5, or 6, and let Y = 0 otherwise.
Given Y = 1, what is the expected value of X?
When Y = 1 is known, then X is equally likely to be any of the values 4, 5,
or 6, and thus E(X | Y = 1) = 13 (4) + 31 (5) + 31 (6) = 5.
On the other hand, when Y = 0 is known, then X is equally likely to be
any of the values 1, 2, or 3, and thus E(X | Y = 0) = 13 (1) + 31 (2) + 31 (3) = 2.

As another example, returning to the situation in Example 39.24, we have the

following:

Example 40.3. Roll a die. Let X denote the value on the die. Let Y = 1 if
the die is even, i.e., is 2, 4, or 6, and let Y = 0 otherwise. Given Y = 1, what
is the expected value of X?

When Y = 1 is known, then X is equally likely to be any of the values 2, 4,

or 6, and thus E(X | Y = 1) = 13 (2) + 31 (4) + 31 (6) = 4.
On the other hand, when Y = 0 is known, then X is equally likely to be
any of the values 1, 3, or 5, and thus E(X | Y = 0) = 13 (1) + 31 (3) + 31 (5) = 3.

Now we turn our attention to conditional expectation for pairs of continuous

random variables. As an example:
 40.2. Examples 543

Example 40.4. Consider the scenario in Example 33.2. David works at a

customer call center. He talks to customers on the telephone. The length (in
hours) of each conversation has density fX (x) = 3e−3x for x > 0 and fX (x) = 0
otherwise. The lengths of calls are independent. As soon as one conversation is
finished, he hangs up the phone, and immediately picks up the phone again to
start another call (i.e., there are no gaps in between the calls).
Thus, if David conducts 2 phone calls in a row, the total amount of time he
spends on the telephone is X1 + X2 , where the Xj ’s are independent, and each
Xj has the Exponential density above.

We let X (also called X1 ) denote the time of the first call, so X has the
density above. We let Y (also called X1 + X2 ) denote the total time of call
one plus call two, combined, so as in Example 33.2, Y is a Gamma random
variable with n = 2 and λ = 3, so Y has density fY (y) = 9ye−3y for y > 0, and
fY (y) = 0 otherwise.
Given that the first two calls take 5/6 of an hour total, i.e., given that
Y = 5/6, find the expected length of the first call. We compute
Z ∞ Z ∞
fX,Y (x, 5/6)
E(X | Y = 5/6) = xfX|Y (x | 5/6) dx = x dx.
−∞ −∞ fY (5/6)
We see that
fY (5/6) = 9(5/6)e−(3)(5/6) = (15/2)e−5/2 .
Also
fX,Y (x, y) = fY |X (y | x)fX (x).
Once X is known, then Y > X. For any y > X, we have P (Y < y | X = x) =
P (Y − x < y − x | X = x) = P (X2 < y − x). Differentiating with respect to y
gives fY |X (y | x) = fX2 (y − x). (Less formally, this says that, once the length
of the first call is known, the remaining length until the end of both calls is
just equal to the length of the second call.) So fY |X (y | x) = 3e−(3)(y−x) for
y − x > 0, i.e., for y > x. Therefore, for 0 < x < 5/6, we have
fX,Y (x, 5/6) = 3e(−3)((5/6)−x) 3e−3x = 9e−5/2
Now we compute
Z ∞
fX,Y (x, 5/6)
E(X | Y = 5/6) = x dx
−∞ fY (5/6)
5/6
9e−5/2
Z
= x dx
0 (15/2)e−5/2
Z 5/6
= (x)(6/5) dx
0
= 5/12
544 Chapter 40. Conditional Expectation

So given that the two calls last 5/6 of an hour altogether, then the first call is
expected to last 5/12 of an hour.
This should make sense intuitively, because when only the total length of
two calls is known, we have no reason to believe that one call will be longer
than the other, so the expected length of the first call plus the expected length
of the second call must equal the total length of the two calls, i.e., 5/6. So the
conditional expected length of each call must be 5/12. (Each call is expected
to last half of the total time.)

Now we focus on computing the expected value of one random variable by

first conditioning, and then taking an expected value. Here, we want to com-
pute E(X). We first suppose that Y = y and we use integration to compute
E(X | Y = y). Then we integrate over all possible y’s, weighting each value by
fY (y), to get the expected value of X, regardless of what Y is. This is shown
below; at the end, no Y ’s or y’s remain:
Z ∞Z ∞
E(X) = xfX,Y (x, y) dx dy
−∞ −∞
Z ∞Z ∞
= xfX|Y (x | y)fY (y) dx dy
−∞ −∞
Z ∞Z ∞
= xfX|Y (x | y) dx fY (y) dy
−∞ −∞
Z ∞
= E(X | Y = y) fY (y) dy
−∞
In the last line above, note that only “y” remains; no “x” is still in the pic-
ture. Thus, we are basically computing the expected value of a function that
only depends on Y , i.e., E(X | Y = y). So the last line above is equal to
E( E(X | Y = y) ), where the outer integral is of course taken with regard to
Y , and the inner integral is with regard to X. This is sometimes written as
E( E(X | Y ) ). So the outer expected value refers to possible Y , while the inner
expected value refers to possible X. So we have
Z ∞
E(X) = E(X | Y = y) fY (y) dy
−∞
= E( E(X | Y = y) )
= E( E(X | Y ) ).
Similarly, if X and Y are discrete, we have
XX
E(X) = xpX,Y (x, y)
y x
XX
= xpX|Y (x | y)pY (y)
y x
X
= E(X | Y = y) pY (y)
y
 40.2. Examples 545

As before, in the last line, only “y” remains; no “x” is still in the picture.
So again we are computing the expected value of E(X | Y = y), which only
depends on Y . Thus, the last line is equal to E( E(X | Y = y) ), and again the
outer summation is taken with regard to Y , while the inner summation is taken
over all possible X. This is also sometimes written as E( E(X | Y ) ), where once
again the outer expected value is with regard to Y , and the inner expected value
is with regard to X. So we have
X
E(X) = E(X | Y = y) pY (y)
y

= E( E(X | Y = y) )
= E( E(X | Y ) ).

Example 40.5. If X and Y are continuous random variables that are indepen-
dent, then
E(X | Y = y) = E(X).
To see this, we just write
Z ∞
E(X | Y = y) = xfX|Y (x | y) dx,
−∞

but fX|Y (x | y) = fX (x) since X and Y are independent, so

Z ∞
E(X | Y = y) = xfX (x) dx = E(X).
−∞

Of course the arguments in the previous example and the following example
only work when X and Y are independent. This argument would fail if X and
Y are dependent.

Example 40.6. If X and Y are discrete random variables that are independent,
then
E(X | Y = y) = E(X).
To see this, we just write
X
E(X | Y = y) = xpX|Y (x | y),
x

but pX|Y (x | y) = pX (x) since X and Y are independent, so

X
E(X | Y = y) = xpX (x) = E(X).
x
546 Chapter 40. Conditional Expectation

Theorem 40.7. Conditional Expected Values of Independent Ran-

dom Variables If X and Y are independent random variables—either con-
tinuous or discrete—then

E(X | Y = y) = E(X).

Example 40.8. Suppose that Alice and Bob each eat one cookie per day,
on five consecutive days, so that they eat a total of ten cookies altogether.
They both love peanut butter cookies, but the cookie selection in each of their
dormitories is random. Let X1 , . . . , X5 be indicators for Alice’s five cookies, and
let X6 , . . . , X10 be indicators for Bob’s five cookies, where Xj = 1 if the jth
cookie is peanut butter, and Xj = 0 otherwise. Assume that the Xj ’s are all
independent Bernoulli random variables, each with parameter p = 0.40, because
there is always a fresh supply of cookies (in particular, Alice and Bob’s choices
each day do not affect each other).
Write X = X1 + · · · + X5 for the total number of peanut butter cookies that
Alice gets to eat over a five day period. Write Y = X1 + · · · + X10 for the total
number of peanut butter cookies that the couple gets to eat. Find the expected
number of cookies that Alice got to eat, given that the couple got to eat Y = 7
peanut butter cookies altogether, i.e., find E(X | Y = 7).
Given Y = 7, we claim that X has conditional distribution as a Hypergeo-
metric random variable with parameters M = 5 and N = 10 and n = 7. Using
what we already know about the Hypergeometric distribution, this yields
E(X | Y = 7) = nM/N = (7)(5)/10 = 3.5.
This makes sense because given Y = 7, there are 7 peanut butter cookies that
the couple gets, so the number of cookies that Alice gets is Hypergeometric,
i.e., it is 5 out of these 7 peanut butters and 3 non-peanut-butters. Each of
the 7 peanut butter cookies is equally likely to belong to Alice or Bob, so Alice
expects to get 3.5 peanut butter cookies, and Bob expects to get 3.5 peanut
butter cookies.
In case the reader is unconvinced by the claim that the conditional distri-
bution of X is Hypergeometric (given Y = 7), we can instead compute the
conditional mass P (X|Y = 7) before we can get to the conditional expected
value.
pX,Y (x, 7) pX (x)pY |X (7 | x)
pX|Y (x | 7) = =
pY (7) pY (7)
Since X is the number of peanut butter cookies Alice ate over five days, then
X is Binomial with parameters n = 5 and p = 0.40, so
 
5
pX (x) = (0.4)x (0.6)5−x .
x
 40.2. Examples 547

Since Y is the total number of peanut butter cookies eaten by the couple over
five days, then Y is Binomial with parameters n = 10 and p = 0.40, so
 
10
pY (7) = (0.4)7 (0.6)3 .
7
Now we need the conditional probability for PY |X (7|x), i.e., for the probability
that the couple has eaten 7 peanut butter cookies total over the five days, given
that Alice ate x of them. Basically, the piece we are missing is the number
of peanut butter cookies that Bob ate. The only way that they eat 7 peanut
butter cookies altogether—given that Alice ate x of them—is for Bob to eat
7 − x of them. The number of peanut butter cookies that Bob eats is Binomial
with n = 5 and p = 0.4. So
 
5
pY |X (7 | x) = P (Y = 7 | X = x) = (0.4)7−x (0.6)5−(7−x) .
7−x
Thus, putting all the pieces together for the probability pX|Y (x | 7),
5 x (0.6)5−x 5 (0.4)7−x (0.6)5−(7−x) 5



5

(0.4)
pX|Y (x | 7) = x 10

7−x = x 107−x
.
(0.4) 7 (0.6)3
7 7

This verifies the claim that X has a Hypergeometric distribution, given that
Y = 7.

Example 40.9. Suppose that each person who logs onto an online retailer’s
website on Black Friday is expected to spend $27.50. Exactly 120 people are
surveyed to see how much is spent by the people within this group. Each behaves
independently of the others, and each visits the online retailer with probability
0.90. How much money do we expect the group of 120 people to spend at the
online retailer?

We let Y denote the number of the people within the group of 120 who
decide to visit the online retailer. Thus Y is Binomial with parameters n = 120
and p = 0.90. Once Y = y is given, we know that exactly y people visit the
store, so the money spent by the group is X1 + · · · + Xy , so the expected money
spent by the group given Y = y is
y
z }| {
E(X1 + · · · + Xy | y) = 27.50 + · · · + 27.50 = (y)(27.50).

Thus, the expected money spent by the group, averaged over all possible values
of Y = y, is
Y
X 
E Xj = E((Y )(27.50)) = (120)(0.90)(27.50) = 2970.
j=1
548 Chapter 40. Conditional Expectation

This can also be written as

X Y 120 X
 X Y 
E Xj = E Xj | Y = y pY (y)
j=1 y=0 j=1

X y
120 X 
= E Xj pY (y)
y=0 j=1
120
X
= (27.50)(y)pY (y)
y=0
120
X
= 27.50 (y)pY (y)
y=0

= 27.50 E(Y )
= (27.50)(120)(0.9)
= 2970.

Example 40.10. Example 40.9 is a specific case of a more general phenomenon:

Consider a nonnegative, integer valued random variable Y , and a sequence
of identically distributed random variables X1 , X2 , X3 , . . . . Also assume Y
and all of the Xj ’s are independent. Now suppose that we want to find the
expected value of the first Y of the Xj ’s. In other words, we want to find
E(X1 + · · · + XY ). We emphasize that there are a random number of Xj ’s that
we are summing. Once we know that value of Y , the problem is easy:
E(X1 + · · · + XY | Y = y) = E(X1 + · · · + Xy | Y = y)
= E(X1 + · · · + Xy ) since Xj ’s & Y are indep.
= E(X1 ) + · · · + E(Xy )
= yE(X1 ) since Xj ’s are identically distributed.
Thus (using the outer E for Y and the inner E for X), we have
∞
X
E(E(X1 + · · · + XY | Y )) = E(X1 + · · · + XY | Y = y)P (Y = y)
y=0
∞
X
= yE(X1 )P (Y = y)
y=0
∞
X
= E(X1 ) yP (Y = y)
y=0

= E(X1 )E(Y ).
So the expected value of X1 + · · · + XY is just equal to the expected value of
one of the Xj ’s, multiplied by the expected value of Y , i.e., by the number of
random variables that we expect to add.
 40.2. Examples 549

Theorem 40.11. If Y is a nonnegative, integer valued random variable and if

X1 , X2 , X3 , . . . is a sequence of identically distributed random variables, which
are independent of each other and independent of Y , then (using the outer E
for Y and the inner E for X)

E(E(X1 + · · · + XY | Y )) = E(X1 )E(Y ).

Example 40.12. Suppose that the number of shoppers coming into a store
follows a Poisson distribution with an expected value of 10 per hour. Suppose
also that each shopper is—independent of all other shoppers—equally likely to
be a man or a woman.

Given that exactly 7 men are shopping, how many women do we expect are
shopping?
Let X and Y denote the number of men and women shoppers, respectively,
and let N = X + Y be the total number of shoppers, which is known to be a
Poisson random variable with mean 10.
We see that

pX,Y (x, y) = P (X = x and Y = y)

= P (X = x and Y = y | N = x + y)P (N = x + y)
(x + y)! e−10 10x+y
= (1/2)x (1/2)y
x!y! (x + y)!
 −5 x x
  −5
e (1/2)y 10y

e (1/2) 10
=
x! y!
 −5 x   −5 y 
e 5 e 5
=
x! y!

So we have factored the joint mass into “x stuff” and “y stuff,” each of which
is a mass, and thus X and Y must be independent. We see from each of their
masses that X and Y are independent Poissons, each with expected value 5.
So, X is Poisson with expected value 5 and Y is Poisson with expected
value 5. Also X and Y are independent. Thus, regardless of how many men
are shopping (e.g., regardless of the fact that exactly 7 men are shopping), the
number of female shoppers is unaffected by the value of X. Thus, even when
given that 7 men are shopping, the expected number of women shopping is
still 5.
550 Chapter 40. Conditional Expectation

The phenomenon in Example 40.12 can be generalized.

Theorem 40.13. Consider a set of N objects, where N is a Poisson random

variable with mean λ. Suppose that, independent of N , and independent of
the other objects, each object is of exactly 1 of j possible different types, as
follows: it is of type k with probability pk , where (of course) p1 + · · · + pj = 1.
Now let Xk denote the number of the N objects that are of type k, so that
X1 + · · · + Xj = N (since all of the objects have some type). Then, in
general, the random variables X1 , . . . , Xj ’s each turn out to be Poisson random
variables, which also turn out to be independent, and E(Xj ) = pj λ.

For comparison, in Example 40.12, there was a Poisson number N of shoppers,

each of which was one of j = 2 types, that is, a man (type 1) with probability
1/2 or a woman (type 2) with probability 1/2. In general, the pk probabilities
do not have to be the same.

40.3 Exercises
40.3.1 Practice
Exercise 40.1. Let X and Y have a joint uniform distribution on the triangle
with corners at (0, 2), (2, 0), and the origin. Find E(Y | X = 1/2).

Exercise 40.2. Errors on a page. Suppose that the number of errors per page
of a book has a Poisson distribution with parameter λ = 0.12. Also suppose
that the expected number of pages in a randomly selected new book is 400.
Find the mean number of errors in such a book.

Exercise 40.3. Arrival times. Consider a man and a woman who arrive at a
certain location; whoever arrives first will wait for the other to arrive. If X and
Y denote (respectively) the arrival times of the man and the woman after noon,
in minutes, assume that X and Y are independent and each Uniformly dis-
tributed on [0, 60]. (In other words, the man and woman arrive independently,
at Uniform times between noon and 1 PM.)
If the woman arrives at 12:35 PM, find the expected time spent waiting, i.e.,
find
E( |X − Y | | Y = 35 ).
Hint: Since we know that Y = 35, we may as well just substitute “35” for Y , so
that we only have X’s in our lives. So we just need to find

E( |X − 35| ).

Exercise 40.4. Female heights. As in Exercise 35.24 and Example 36.10, as-
sume that the height (in inches) of an American female is Normal with expected
value µ1 = 64 and standard deviation σ1 = 2.5. Also assume that the height
 40.3. Exercises 551

of an American male is Normal with expected value µ2 = 69 and standard

deviation σ2 = 3.0.
A man and a woman are chosen at random. The woman’s height is measured,
and she is found to be exactly 68.2 inches tall. How much taller do we expect
the man to be (as compared to this particular woman)?

Exercise 40.5. Annual precipitation. Let X be the amount of annual

snowfall in a region in North America. Assume that, given the value Y of
annual precipitation in the region, we know X is Normally distributed with
E(X | Y ) = 0.3Y and σX = 0.1Y . Given that the annual precipitation value Y
in a student’s hometown is 36.3 inches, what is the expected amount of snowfall?

Exercise 40.6. Mac and cheese. While Juanita waits for her Mac and Cheese
to boil, she works on her homework for 1/2 of the time. She has been waiting
for 3 minutes already (so she has already done 1.5 minutes of homework). She
doesn’t know how much more time is needed. She estimates that the remaining
time (in minutes) until the water boils is an Exponential random variable Y
with density fY (y) = 41 e−y/4 for y > 0, and fY (y) = 0 otherwise. Find the total
length of time that she expects to work on her homework while waiting for the
water to boil.

Exercise 40.7. Flowers. Sally and David each pick 10 flowers from the case
without paying attention to what type of flowers they are picking. There are a
large quantity of flowers available, 20% of which are roses. Let X be the number
of roses that Sally picks, and let Y be the number of roses that the couple picks
altogether. Find the number of roses that we expect Sally to pick if the total
number picked is Y = 12.

Exercise 40.8. Selling cookies. In a particular girl scout troop, each girl
sells an average of 30 boxes of cookies. Let Y be the number of girls in the
troop, and let X be the number of boxes of cookies sold. Find E(X | Y = y).

Exercise 40.9. Candy machine. There are 20 pieces of candy in a machine:

5 Hershey kisses, 6 Kit Kats, 4 Snickers, and 5 Paydays. The machine randomly
selects one of the remaining candies when someone makes a purchase, and the
candies are not replenished after the purchase (i.e., purchases are made without
replacement). If X Kit Kats are contained in a purchase of Y candies, find
E(X | Y = 4).

Exercise 40.10. Wake up times. Suppose that, on a random Sunday, your

roommate wakes up at some time Uniformly distributed between 12 noon and
3 PM. Suppose that you awake at 2 PM that day. If your roommate is not yet
awake when you get up at 2 PM, when do you expect him to wake up?

Exercise 40.11. Doctor’s appointments. A doctor has two consecutive ap-

pointments with patients. The duration of each appointment is Exponential
with expected value 20 minutes. The durations of the two appointments are
552 Chapter 40. Conditional Expectation

assumed to be independent. Given that the total duration of the two appoint-
ments turns out to be 38 minutes altogether, how long do we expect the first
appointment to be?

Exercise 40.12. First aid. A nurse has 10 minutes to administer first aid
to various soldiers in the middle of a battle. The time (in minutes) devoted to
the first soldier is Y , and the time (in minutes) devoted to the second soldier is
X. Assume that (X, Y ) is Uniformly distributed on the triangle where X ≥ 0,
Y ≥ 0, and X + Y ≤ 10. If we know that Y = 3, i.e., that she spends exactly 3
minutes with the first soldier, how much time do we expect that she will spend
with the second soldier?

Exercise 40.13. Let X1 and X2 be independent exponential random variables,

each with mean 1. Let Y = X1 + X2 . Find E(X1 | Y = 3).

40.3.2 Extensions
Exercise 40.14. Roll two 6-sided dice. Let X denote the minimum value that
appears, and let Y denote the maximum value that appears.

a. Find E(Y | X = 3).

b. Find E(X + Y | X = 3).

Exercise 40.15. A child rolls a pair of dice, one of which is blue and one of
which is red.

a. Given that the sum of the dice is 8, find the probability that the red die
shows the value 4.
b. Now the child rolls a pair of dice that look the same (i.e., which are not
painted). Given that the sum of the dice is 8, find the probability that both of
the dice simultaneous show the value 4.
c. Are your answers the same or different in the two parts above? Why?

Exercise 40.16. Two 6-sided dice are rolled. Let X be the minimum of the
two values, and let Y be the absolute value of the difference of the two values.
If D1 , D2 are the two values on the two dice, then

X = min(D1 , D2 ),

and
Y = |D1 − D2 |.
If Y = 4, what is the expected value of X?

Exercise 40.17. Guitar songs. Helen and Joe play guitar together every day
at lunchtime. The number of songs that they play on a given day has a Poisson
distribution, with an average of 5 songs per day. Regardless of how many songs
they will play that day, Helen and Joe always flip a coin at the start of each
 40.3. Exercises 553

song, to decide who will play the solo on that song. If we know that Joe plays
exactly 4 solos on a given day, then how many solos do we expect that Helen
will play on that same day?

Exercise 40.18. Black Jack. Let Y denote the value of the dealer’s card
that can be seen by all the players, in a game of Black Jack. Let X be a
Bernoulli random variable that indicates whether the dealer must stay (i.e., not
take another card). Given Y = 10, find the expected value of X. Hint: In Black
Jack, if the dealer has a total of 17 or greater in her hand, then she must stay;
if her total is 16 or less, she will draw. For the purposes of this question, the
Ace has value 11, and each face card (Jack, Queen, or King) has the value 10.

Exercise 40.19. Let X and Y denote, respectively, the x- and y-coordinates of

the location of a random chosen point that is Uniformly distributed in a region.
Given Y = 1, find the expected value of X when the region is:

a. a circle of radius 2, centered at the origin

b. a semicircle corresponding to the right-hand portion of the circle from
part a.

Exercise 40.20. Sandra rolls two 10-sided dice. Let Y be the sum of the two
dice, and let X be the value on the first die that she rolls. Given Y = 15, what
is the expected value of X?

Exercise 40.21. Date wait. Suppose that Harrison comes to pick up Rosita
for a date. Rosita will be ready at a time that is Uniformly distributed between
7 PM and 7:10 PM. If Harrison arrives at 7:07 PM, how long does he expect
to have to wait until Rosita is ready? (If she is already ready when he arrives,
then his waiting time is 0, since a waiting time cannot be negative.)

Exercise 40.22. A point is chosen Uniformly inside a circle of radius 2, centered

at the origin. If we are given that the point lands exactly on the x-axis, find
the expected distance of the point to the origin.

40.3.3 Advanced
Exercise 40.23. Let X and Y correspond to the horizontal and vertical co-
ordinates in the triangle with corners at (2, 0), (0, 2), and the origin. Let
15
fX,Y (x, y) = 28 (xy 2 + y) for (x, y) inside the triangle, and fX,Y (x, y) = 0
otherwise. Find E(X | Y = 1.5).

Exercise 40.24. Show that the generalization of Example 40.12, given in a box
at the end of the chapter, is true.
Chapter 41

Markov and Chebyshev

Inequalities

The worst form of inequality is to try to make unequal things equal.

—Aristotle

If we know the average speed of drivers on a highway, what kind of bounds can
we give on the probability that a randomly selected driver is speeding? How
about bounds on the probability that a randomly chosen driver is driving within
5 miles of the average speed?

41.1 Introduction
One subdiscipline within probability theory is the study of probability inequal-
ities. These inequalities are used to put bounds on how large or small a proba-
bility can be, under certain conditions. Some bounds are stronger than others.
Some inequalities have many conditions, and other inequalities are quite simple.
Many different types of inequalities are used in limiting situations, i.e., what
happens to a limit of a sequence of random variables and/or their associated
probabilities. As with many topics covered during this course, the study of
probability inequalities is very rich and deep. We could spend many chapters
on probability inequalities, but here we only scratch the surface.

41.2 Markov Inequality

The Markov inequality gives bounds on how large (the absolute value of) a
random variable can be.

554
 41.2. Markov Inequality 555

Theorem 41.1. Markov Inequality

If X is any random variable and a > 0, then

E( |X| )
P (|X| ≥ a) ≤ .
a

To see that the Markov inequality is true, we just consider whether |X| ≥ a.
More precisely, we let Y = 1 if |X| ≥ a, and Y = 0 otherwise. Then Y is an
indicator, so

P (|X| ≥ a) = P (Y = 1)
= E(Y ).

The novel step is that Y is always less than |X|/a:

• If Y = 1, then |X| ≥ a, so we have

Y = 1 ≤ |X|/a.

• If Y = 0, then
Y = 0 ≤ |X|/a
because |X| and a are both greater than or equal to 0.

Since Y ≤ |X|/a always, then

E(Y ) ≤ E(|X|/a) = E(|X|)/a.

Putting this together with the series of equations above, we get the Markov
inequality:
E( |X| )
P (|X| ≥ a) ≤ .
a
As a straightforward extension of the original Markov inequality, we notice
that, if X is always nonnegative (i.e., if X ≥ 0 always), then |X| = X. So, in
such a case, the Markov inequality works for not just |X| but also for X too.
So the second version of the Markov inequality gives bounds on how large a
nonnegative random variable can be.

Corollary 41.2. Markov Inequality (version 2)

If X is a nonnegative random variable and a > 0, then

E(X)
P (X ≥ a) ≤ .
a

Notice that the Markov inequality does not require us to know anything
about the distribution of the random variable X, except that we need to know
the expected value. We do not even need to know if X is discrete or continuous.
556 Chapter 41. Markov and Chebyshev Inequalities

Example 41.3. Consider a class of students for which the class average is 60%.

Find a bound for the probability that a randomly chosen student’s score is
72% or higher. We let X denote a randomly chosen student’s score. Using the
Markov inequality, we have

E(X) 60
P (X ≥ 72) ≤ = = 5/6.
72 72
Notice, in this example, that the distribution of X is not even given! All that
is required is that we know the expected value of X.

Example 41.4. The average salary of actuaries in a certain specialty is known

to be $83,000 per year.

If an actuary in that specialty is chosen at random, we can let X denote the

actuary’s salary. By the Markov inequality, the probability that the actuary
makes $90,000 or more can be bounded as follows:

E(X) 83,000
P (X ≥ 90,000) ≤ = = 83/90 = 0.92.
90,000 90,000

Example 41.5. On a certain highway, the speed limit is 55 miles per hour, but
most drivers are not driving so fast. The average speed on the highway is 49
miles per hour.

If X denotes a randomly chosen driver’s speed, then the probability that such
a person is driving faster than the speed limit is

E(X) 49
P (X ≥ 55) ≤ = = 0.89.
55 55

One important thing to emphasize is that, in none of these examples do we

get to actually calculate a probability directly. We only get upper bounds on
probabilities. The advantage of using the Markov inequality, however, is that
we do not need to know anything about the distribution of a random variable
X except for the expected value. Thus, the Markov inequality can be applied
in lots and lots of situations. We do not need to know much to use it, but on
the other hand, it tells us relatively little, as compared to a situation in which
 41.3. Chebyshev Inequality 557

we can build an inequality based on having more information about the random
variable’s actual distribution.
One unfortunate thing about the Markov inequality is that we do not nec-
essarily get sharp bounds (i.e., precise bounds) for P (|X| ≥ a). Another unfor-
tunate thing about the Markov inequality, however, is that it does not help us
to calculate values P (|X| ≥ a) when a is smaller than the expected value of |X|
(and thus E(|X|)/a is bigger than 1). If we try to apply the Markov inequality
when a > E(|X|), then the results that we get are not too interesting, because
we will have
E(|X|)
P (|X| ≥ a) ≤
a
but if E(|X|)
a > 1, this is not very informative, because we can automatically
write the tighter bound:
P (|X| ≥ a) ≤ 1,
which is true trivially, just because all probabilities of all events are bounded
above by 1. So we do not gain any additional information in such a case. We
give one such useless example.

Example 41.6. Return to the scenario of Example 41.3, in which a class of

students has a class average is 60%.

Find a bound for the probability that a randomly chosen student’s score is
55% or higher. We let X denote a randomly chosen student’s score. Using the
Markov inequality, we have

E(X) 60
P (X ≥ 55) ≤ = = 1.09.
55 55
This is completely uninteresting, because we could just as well have written

P (X ≥ 55) ≤ 1,

and we have still learned nothing new at all. So the Markov inequality is not
useful to us in such a situation.

41.3 Chebyshev Inequality

The Chebyshev inequality gives a different type of information about a random
variable. As with the Markov inequality, in order to apply the Chebyshev
inequality, we do not need to know the distribution of the random variable
being considered. We just need to know the expected value of the random
558 Chapter 41. Markov and Chebyshev Inequalities

variable, as well as the variance, which automatically tells us the standard

deviation too. Since we must provide two facts about the random variable here
(both the average value and the variance), then we should expect to get a more
informative bound, as compared to the Markov inequality (which only requires
use of the average value).
Theorem 41.7. Chebyshev Inequality
If X is any random variable and k is any positive number, then

Var(X)
P (|X − E(X)| ≥ k) ≤ .
k2
To see that the Chebyshev inequality is true, we just apply the Markov
inequality, using (X − E(X))2 as the random variable, and using a = k 2 . Then
the Markov inequality yields
E((X − E(X))2 )
P ((X − E(X))2 ≥ k 2 ) ≤ .
k2
On the left hand side, (X − E(X))2 ≥ k 2 if and only if |X − E(X)| ≥ k, so we
can rewrite the left hand side as P (|X − E(X)| ≥ k). On the right hand side,
E((X − E(X))2 ) = Var(X). So we get
Var(X)
P (|X − E(X)| ≥ k) ≤ ,
k2
which is exactly the statement of the Chebyshev inequality.
We can get a second version of the Chebyshev inequality if we just use
“k = aσX ” in the version above, where σX is the standard deviation of X. This
yields
Var(X)
P (|X − E(X)| ≥ aσX ) ≤ ,
(aσX )2
2 = 1, this simplifies to
for any a > 0. Since Var(X)/σX
1
P (|X − E(X)| ≥ aσX ) ≤ .
a2
The choice of letter that we use is arbitrary, so we write the second version of
the Chebyshev inequality using k’s instead of a’s:
Corollary 41.8. Chebyshev Inequality (version 2)
If X is any random variable and k is any positive number, then

P (|X − E(X)| ≥ kσX ) ≤ 1/k 2 .

If we read the second version of the Chebyshev inequality literally, it means

that a random variable is more than k standard deviations away from its
expected value at most 1/k 2 of the time. This is a way of quantifying the fact
that a random variable is “relatively close” to its expected value “most of the
time.” Chebyshev gives bounds that quantify both “how close” and “how much
of the time.”
 41.3. Chebyshev Inequality 559

We could also take complements on both sides of this equation, and rewrite
the equation
P (|X − E(X)| ≥ kσX ) ≤ 1/k 2 .
as the following:
1
1 − P (|X − E(X)| ≥ kσX ) ≥ 1 − ,
k2
or equivalently,
k2 − 1
P (|X − E(X)| ≤ kσX ) ≥ .
k2
This says that a random variable is within k standard deviations from its ex-
2
pected value at least k k−1
2 of the time.

Corollary 41.9. Chebyshev Inequality (version 3)

If X is any random variable and k is any positive number, then

k2 − 1
P (|X − E(X)| ≤ kσX ) ≥ .
k2

Example 41.10. Return to the scenario of Example 41.3, in which a class of

students has a class average is 60%. Also suppose it is known that the standard
deviation of the class’s scores is 10.

Find a bound for the probability that a randomly chosen student’s score is
between 45% and 75%. We let X denote a randomly chosen student’s score.
Using the Chebyshev inequality, since the standard deviation is σX = 10, we
use k = 1.5, and then we have

P (45 ≤ X ≤ 75) = P (|X − 60| ≤ 15)

= P (|X − 60| ≤ (1.5)(10))
(1.5)2 − 1
≥
(1.5)2
= 0.5556.

As with the Markov inequality, the Chebyshev inequality does not require that
we know that distribution of the scores at all. We just need to know the expected
value and the variance or the standard deviation. Also, as with the Markov
inequality, we do not get a precise answer, but rather, we only get some bounds
on the desired probability.
560 Chapter 41. Markov and Chebyshev Inequalities

Example 41.11. Return to the scenario of Example 41.4, in which the average
salary of actuaries in a certain specialty is known to be $83,000 per year. Also
assume that the standard deviation is $20,000 per year.

Let X be the salary of a randomly chosen actuary in the profession under

consideration. The probability that such an actuary makes more than $123,000
or less than $43,000 is

P (X ≥ 123,000 or X ≤ 43,000) = P (|X − 83,000| ≥ 40,000)

= P (|X − 83,000| ≥ (2)(20,000))
≤ 1/22
= 1/4.

Another way to express this is to say that the probability a randomly chosen
actuary’s salary is between $43,000 and $123,000 is

P (43,000 ≤ X ≤ 123,000) = P (|X − 83,000| ≤ 40,000)

= P (|X − 83,000| ≤ (2)(20,000))
22 − 1
≥
22
= 3/4.

We do not have sufficient information to tell the exact probability that a ran-
domly chosen actuary’s salary is between $43,000 and $123,000, but the bounds
given here are better than nothing.

Example 41.12. Return to the scenario of Example 41.5, in which, on a certain

highway, the speed limit is 55 miles per hour, but most drivers are not driving
so fast. The average speed on the highway is 49 miles per hour. Also assume
that the variance is 144.

If X denotes a randomly chosen driver’s speed, then the variance of X is 144,

so the standard deviation of X is 12. The probability that a randomly selected
driver is travelling between 29 and 69 miles per hour is

P (29 ≤ X ≤ 69) = P (|X − 49| ≤ 20)

= P (|X − 49| ≤ (20/12)(12))
(20/12)2 − 1
≥
(20/12)2
= 16/25
= 0.64
 41.4. Exercises 561

We already noted, after Example 41.10, that the Chebyshev inequality does
not require us to know the distribution of the random variable, but it does re-
quire us to know the expected value and to know the variance, or equivalently,
to know the standard deviation. We only get bounds on the probabilities from
the Chebyshev inequality. Also, we do not get any information about the frac-
tion of the time that a random variable is less than (or is more than) k standard
deviations of the expected value, for k < σX in version 1, or equivalently for
k < 1 in versions 2 or 3. The Chebyshev inequality is just not helpful in such
situations. In version 1, the right hand side will be Var(X)/k 2 > 1 in such a
case, which is not helpful. In version 2, the right hand side will be 1/k 2 > 1,
2
which is not helpful. In version 3, the right hand side will be k k−12 < 0, which
again is not helpful. So these extreme cases do not yield any useful information
from the Chebyshev inequality.

41.4 Exercises
41.4.1 Practice

Exercise 41.1. Waiting for a bus. While waiting for the bus on a snowy
morning, the expected waiting time (including unusual delays for snow!), is 12
minutes.

a. Use the Markov inequality to find an upper bound on the probability

that the bus takes 15 minutes or more to arrive.
b. In the same scenario as above, assume that the standard deviation of
the waiting time is 3. Use the Chebyshev inequality to find a bound on the
probability that the bus takes between 6 to 18 minutes to arrive.

Exercise 41.2. Students in class. The average number of students in a class

at a certain university is 31.

a. Use the Markov inequality to find an upper bound on the probability

that a class selected at random will have 40 or more students.
b. In the scenario above, now assume that the variance of the class size is
64. Use the Chebyshev inequality to find a bound on the probability that a
class selected at random will have between 17 and 45 people.

Exercise 41.3. Basketball shots. A basketball player has improved his scor-
ing ability. During a game, he can be expected to make 12 shots.

a. Give a bound on the probability that he makes at least 16 shots.

b. If the number of shots successfully scored by the same basketball player
is known to have a standard deviation of 2.5, find a bound on the probability
that he successfully makes between 7 and 17 shots.
562 Chapter 41. Markov and Chebyshev Inequalities

Exercise 41.4. Guitar players. The expected number of guitar players in a

classroom is 8. Given an upper bound on the probability that there are 11 or
more guitar players in a classroom.

Exercise 41.5. Chicken feathers. A certain type of chicken’s wing has 138
feathers on average, with standard deviation 5. Find a bound on the probability
that the chicken has between 120 and 156 feathers.

Exercise 41.6. Sneezing. Henry caught a cold recently and therefore he has
been sneezing a lot. His expected waiting time between sneezes is 45 seconds.
The standard deviation of the waiting time between his sneezes is 8 seconds.
Find a bound on the probability that the time between two consecutive sneezes
is between 30 and 60 seconds.

Exercise 41.7. Sleepy dog. The expected time for a student’s dog to fall
asleep is 12 minutes. Give an upper bound on the time that it takes the dog
more than 20 minutes to fall asleep.

Exercise 41.8. Music library. In a student’s music library on his mp3 player,
the expected length of a randomly chosen song is 3.2 minutes.

a. Find an upper bound on the probability that a randomly chosen song is

at least 5 minutes long.
b. If the standard deviation in the song lengths is known to be 0.8 minutes,
find a bound on the probability that the song is between 2.9 and 3.5 minutes
long.

Exercise 41.9. Snowy winter. The average amount of snow in a student’s

hometown during the winter months is 10 inches.

a. Find a bound on the probability that the snowfall in a given winter will
exceed 16 inches.
b. If the standard deviation of snowfall is 3.25 inches, find a bound on the
probability that there is between 6 and 14 inches of snow.

Exercise 41.10. Soccer goals. A talented soccer player is expected to score

21 goals in a given soccer season.

a. Find an upper bound on the probability that he scores 30 goals or more

during the season.
b. If he has a standard deviation of 6.5 goals during the season, give a bound
on the probability that he scores between 8 and 34 goals during the season.

Exercise 41.11. Flight time. The flight time of a plane flight from Denver
to New York City is has an average flight time of 3 hours and 16 minutes, and
standard deviation is 30 minutes.
 41.4. Exercises 563

a. Give a bound on the probability that such a flight takes 6 hours or longer.
b. Give a bound on the probability that the announced arrival time (3 hours,
16 minutes) and the actual arrival time differ by 1 hour or more.
Exercise 41.12. Cold weather. The average temperature in December is 27◦ .
a. Find a bound on the probability of the temperature falling outside the
range −50◦ to 50◦ .
b. If the standard deviation of the temperature is 7◦ , then what is the
probability that the temperature falls outside the range of 14◦ to 40◦ ?
Exercise 41.13. Final exam. Let X be the number of problems on the final
exam. The professor puts, on average, 30 problems on each exam. The standard
deviation of the number of problems he puts on a final exam is 2. Given a bound
for the likelihood that the final exam contains between 25 and 35 problems.
Exercise 41.14. Video download. A video can be downloaded from the
web and moved to a student’s mobile device in 12 minutes, on average. If the
standard deviation of the time required is 2 minutes, then give a bound on the
probability that 9–15 minutes are needed.
Exercise 41.15. Eating habits. In a study on eating habits, a particular
participant averages 750 cm3 of food during per meal.
a. It is extraordinarily rare for this participant to eat more than 1000 cm3
of food at once. Find a bound on the probability of such an event.
b. If the standard deviation of a meal size is 100 cm3 , the find a bound on
the event that the meal is either too much food, i.e., more than 1000 cm3 , or
an insufficient amount of food, namely, less than 500 cm3 .
Exercise 41.16. Long jumps. An athletic director is recording long-jump
scores for a group of students. The expected value of each of their long jumps
is 7 meters, and the standard deviation is 0.2 meters. If he chooses a student
at random, find a bound on the probability that the student’s long-jump is
between 6.7 and 7.3 meters.
Exercise 41.17. Sled runs. An energetic student can manage to get 9 sled
runs down a hill, on average, within a 30 minute time period.
a. Find an upper bound on the probability that the student achieves 12 or
more runs during a one hour period.
b. If the standard deviation of the number of runs that they can manage in
a 30 minute period is 2, then give a bound on the probability that the student
gets between 6 and 12 sled runs during a 30 minute period.
Exercise 41.18. String theory. Suppose that the expected number of stu-
dents who take String Theory is 23, with variance 169. Suppose that the as-
signed classroom can hold only 40 students. Also, 6 students is the minimum
564 Chapter 41. Markov and Chebyshev Inequalities

enrollment, i.e., if 5 or less people sign up, the class will be combined with
another. What is a bound on the probability that the class is held (i.e., has
a sufficient number of students) and can be taught in the intended classroom
(i.e., does not have too many students)?

Exercise 41.19. Fair coin. Trying to figure out if a coin is fair, a student
flips it one thousand times (!!!). Give an upper bound on the probability that
he gets 700 or more heads, assuming that the coin was actually a fair coin.

Exercise 41.20. Failing lights. On a strand of lights, the expected number

of lights that fail to work correctly is 20.

a. What is an upper bound on the probability that 30 or more lights fail to

work?
b. If the standard deviation is 5, what is a bound on the probability that
between 12 and 28 lights fail to work correctly?

Exercise 41.21. Buying gifts. This holiday season, assume that a person
will, on average, spend $673 on gifts, with a standard deviation of $79. Use
a Chebyshev inequality to find a bound on the probability that a randomly
selected person spends between $568 and $778.

Exercise 41.22. Accountant age. Assume that the average age of tax ac-
countants at a certain CPA firm is 44.

a. Find a bound for the probability that a randomly chosen employee of the
firm is 48 or older.
b. If the standard deviation of the age of a randomly chosen employee is 5
years, then find a bound for the probability that a randomly chosen employee’s
age is between 36 and 52.

41.4.2 Extensions
Exercise 41.23. With the same assumptions as in Exercise 41.2, three classes
are independently selected at random.

a. Let A denote the event that all three of the classes have 40 or more
students (i.e., 40 or more in each class). Find a bound on the probability of A.
Hint: Separate A into three independent events, find a bound on the probability
of each, and then think about how to appropriately combine your bounds.
b. In the scenario above, let B denote the event that all three classes
selected at random will have between 20 and 42 people (i.e., 20 to 42 people
in each class). Find a bound on the probability of B. Hint: Again, separate
B into three independent events, find a bound on each, and then recombine
appropriately.
Chapter 42

Order Statistics

Good order is the foundation of all great things.

—Edmund Burke
Order is the shape upon which beauty depends.
—Pearl S. Buck

When five students gather and compare their grades, what is the probability
that the highest grade exceeds 97%? What is the probability that at least one
student failed the exam? What is the probability that three or more students
earned a “B” grade or higher?

42.1 Introduction
The concept of order statistics is usually used with continuous random variables,
because the idea of determining the rank of the random variables (i.e., which is
smallest, which is largest, which is second-smallest, etc.) is key to the concept
of order statistics. Any “ties” between two random variables—i.e., any two
random variables that happen to equal exactly the same value—complicate the
issue. With continuous random variables, the probability of any ties is 0, so the
use of continuous random variables allows us to safely ignore the possibility of
ties. It is usually impossible to remove this complication with discrete random
variables, so we do not discuss order statistics of discrete random variables in
this text at all.
Usually we speak of the “order statistics” among independent, identically
distributed random variables. So if X1 , X2 , . . . , Xn are the random variables
under consideration, then the Xj ’s are independent, and all of their densities
fj (x) are the same function (or, equivalently, all of their cumulative distribution

565
566 Chapter 42. Order Statistics

functions Fj (x) are the same function). This is not always the case when work-
ing with order statistics (i.e., we could talk about order statistics of random
variables that are either not independent or not identically distributed), but
the independent, identically distributed situation is the most common scenario
for studying order statistics.
In this text, when speaking about order statistics, we always as-
sume that the random variables under study are independent, iden-
tically distributed, continuous random variables.

42.2 Examples

Example 42.1. Consider the waiting times X1 and X2 (in minutes) until
Samuel hears from two of his friends, Mary and Josephine, respectively. He
doesn’t know which one will call first. Both of the Xj ’s are Exponential with
expected value 5. The Xj ’s are also independent. Then the first order statistic
is X(1) = min{X1 , X2 }, i.e., the minimum of the two waiting times. The second
order statistic is X(2) = max{X1 , X2 }, i.e., the maximum of the two waiting
times.

In this case, X(1) is an Exponential random variable. To see this, for any
a > 0,

P (X(1) > a) = P (X1 > a and X2 > a)

= P (X1 > a)P (X2 > a)
= e−a/5 e−a/5
= e−2a/5

Thus X(1) is Exponential with E(X(1) ) = 5/2. We emphasize that the parameter
of X(1) is different that the parameters of X1 and X2 . With this in mind, we
can treat X(1) in just the same way that we would treat any other Exponential
random variable with expected value 5/2. For instance, we know immediately
that the variance of the time until he hears from his first friend is (5/2)2 =
25/4 = 6.25 minutes. Also, the probability he hears from the first friend within
the first four minutes is P (X(1) ) = 1 − e−2(4)/5 = 0.798.
On the other hand, we emphasize that X(2) is not an Exponential random
variable. For instance, the CDF of X(2) does not have the form of the CDF of
an Exponential random variable. For a > 0:

P (X(2) ≤ a) = P (X1 ≤ a and X2 ≤ a)

= P (X1 ≤ a)P (X2 ≤ a)
= (1 − e−a/5 )(1 − e−a/5 ).
 42.2. Examples 567

Incidentally, we can also find the densities of X(1) and X(2) . Since X(1) is
Exponential with expected value 5/2, then no calculation is necessary. We know
2
fX(1) (x1 ) = e−2x1 /5 for x1 > 0,
5
and fX(1) (x1 ) = 0 otherwise.
To find the density of X(2) , we just differentiate the CDF

FX(2) (x2 ) = P (X(2) ≤ x2 ) = (1 − e−x2 /5 )(1 − e−x2 /5 )

with respect to x2 , and we get

2
fX(2) (x2 ) = e−x2 /5 (1 − e−x2 /5 ) for x2 > 0,
5
and fX(2) (x2 ) = 0 otherwise.

Example 42.2. Consider the heights X1 , X2 , X3 of three randomly chosen peo-

ple, Alicia, Bernadette, and Charlotte, respectively. Suppose that the Xj ’s are
independent and identically distributed, e.g., if their heights are measured in
inches then perhaps the expected value of each height is 63.5 and the standard
deviation is 2.5.
In this example, the first order statistic, X(1) , is the minimum of the three
people’s heights. The second order statistic is X(2) , the height of the middle
person (who is neither shortest nor tallest). Finally, the third order statistic is
X(3) , the maximum of the three people’s heights. So, for example, if

X1 = 63.076
X2 = 62.849
X3 = 63.870

then the first, second, and third order statistics are, respectively:

X(1) = 62.849
X(2) = 63.076
X(3) = 63.870

Definition 42.3. Order Statistics

If we put X1 , X2 , . . . , Xn in order, so that X(1) is the smallest of these random
variables, and X(2) is the second-smallest of these random variables, and in
general, X(j) is the jth-smallest of these random variables, and thus X(n) is
the largest of these random variables, then we say that X(1) , X(2) , . . . , X(n) are
the 1st, 2nd, . . . , nth order statistics.
568 Chapter 42. Order Statistics

Example 42.4. For example, if a driver experiences the following ten waiting
times at ten red lights on a trip,

X1 = 29.186, X2 = 26.169, X3 = 3.422, X4 = 12.995, X5 = 3.127,

X6 = 1.360, X7 = 29.617, X8 = 10.420, X9 = 9.484, X10 = 18.837.

then we could put the waiting times in order, to get the order statistics.

The first order statistic is the minimum, written as X(1) = 1.360. The second
order statistic is the second-smallest value, written as X(2) = 3.127. The tenth
order statistic is the maximum value, written as X(10) = 29.617. So the 1st,
2nd, 3rd, . . . , 10th order statistics are, respectively:

X(1) = 1.360, X(2) = 3.127, X(3) = 3.422, X(4) = 9.484, X(5) = 10.420,
X(6) = 12.995, X(7) = 18.837, X(8) = 26.169, X(9) = 29.186, X(10) = 29.617.

Example 42.5. The joint density of the order statistics can only be nonzero
in the region where, for instance, X(1) < X(2) < X(3) . If we consider the joint
density fX(1) ,X(2) ,X(3) (2.7, 5.22, 3.9) of the first, second, and third order statistics
of three random variables X1 , X2 , X3 that are each defined in the interval [0, 10],
then we know that the joint density will be zero at this point, because we cannot
have the second order statistic X(2) = 5.22 and a smaller third order statistic
X(3) = 3.9. On the other hand, fX(1) ,X(2) ,X(3) (2.7, 3.9, 5.22) will be positive
because the order statistics are in ascending order!

42.3 Joint Density and Joint CDF of Order Statistics

Throughout this section, we assume that the random variables X1 , . . . , Xn under
consideration are independent and identically distributed.

Example 42.6. As in Example 42.1, consider the waiting times X1 and X2

(in minutes) until Samuel hears from two of his friends, Mary and Josephine,
respectively. He doesn’t know which one will call first. Both of the Xj ’s are
Exponential with expected value 5. The Xj ’s are also independent. Then the
first order statistic is X(1) = min{X1 , X2 }, i.e., the minimum of the two waiting
times. The second order statistic is X(2) = max{X1 , X2 }, i.e., the maximum of
the two waiting times.
 42.3. Joint Density and Joint CDF of Order Statistics 569

Consider any a, b with 0 < a < b. In order to calculate the joint density of
(X(1) , X(2) ) evaluated at (a, b), we first calculate the joint CDF:

FX(1) ,X(2) (a, b) = P (X(1) ≤ a and X(2) ≤ b)

To have X(1) ≤ a and X(2) ≤ b, we need either:

• Both X1 and X2 are smaller than a, or
• Only X2 < a, and a < X1 < b, or
• Only X1 < a, and a < X2 < b.
Thus

FX(1) ,X(2) (a, b) = P (X1 ≤ a and X2 ≤ a)

+ P (a < X1 ≤ b and X2 ≤ a)
+ P (X1 ≤ a and a < X2 ≤ b)
= (1 − e−a/5 )(1 − e−a/5 )
+ (e−a/5 − e−b/5 )(1 − e−a/5 )
+ (1 − e−a/5 )(e−a/5 − e−b/5 )
= (1 − e−a/5 )(1 − e−a/5 ) + 2(e−a/5 − e−b/5 )(1 − e−a/5 )

Thus, for 0 < x1 < x2 ,

FX(1) ,X(2) (x1 , x2 ) = (1 − e−x1 /5 )(1 − e−x1 /5 ) + 2(e−x1 /5 − e−x2 /5 )(1 − e−x1 /5 ).

Now we differentiate with respect to x1 and x2 to get the joint density of

(X(1) , X(2) ). It does not matter which order we choose to differentiate, so we
first differentiate with respect to x2 , so that the entire first term (which depends
only on x1 ) is removed. We get
 
∂ 1 −x2 /5
FX ,X (x1 , x2 ) = 2 e (1 − e−x1 /5 )
∂x2 (1) (2) 5
and then
  
∂ ∂ 1 −x2 /5 1 −x1 /5
fX(1) ,X(2) (x1 , x2 ) = FX ,X (x1 , x2 ) = 2 e e .
∂x1 ∂x2 (1) (2) 5 5
Thus, if we write
1
fX (x) = e−x/5 ,
5
which is the common distribution of both X1 and X2 , then we see that the joint
density of X(1) and X(2) is

fX(1) ,X(2) (x1 , x2 ) = 2fX (x1 )fX (x2 ) for 0 < x1 < x2 ,

and fX(1) ,X(2) (x1 , x2 ) = 0 otherwise.

570 Chapter 42. Order Statistics

Now we emphasize that this concept works much more generally.

Example 42.7. Consider n random variables X1 , . . . , Xn that are independent

and all have the same distribution.

The joint density of X1 , . . . , Xn can be factored into a product of the densities

of the Xj ’s, since the Xj ’s are independent.

fX1 ,X2 ,...,Xn (x1 , x2 , . . . , xn ) = fX1 (x1 )fX2 (x2 ) · · · fXn (xn ).

Of course, not only are the Xj ’s all independent, but they also all have the
same distribution, which we can just write as f (x). Thus, the joint density of
X1 , . . . , Xn becomes

fX1 ,X2 ,...,Xn (x1 , x2 , . . . , xn ) = f (x1 )f (x2 ) · · · f (xn ).

Thus, if we integrate all Xj ’s over all real numbers, we have

Z ∞Z ∞ Z ∞
··· f (x1 )f (x2 ) · · · f (xn ) dx1 dx2 · · · dxn = 1.
−∞ −∞ −∞

The joint density of the order statistics X(1) , . . . , X(n) can also be written as a
product, but we must insist now that the xj ’s are in ascending order. The main
point driving this idea is that there are n! equally likely ways that the underlying
X1 , . . . , Xn can be placed in order. So by multiplying by n!, the integral of the
joint density of the order statistics will be 1. Therefore, the joint density of
X(1) , . . . , X(n) is

fX(1) ,...,X(n) (x1 , . . . , xn ) = n!f (x1 ) · · · f (xn ) for x1 < · · · < xn ,

and fX(1) ,...,X(n) (x1 , . . . , xn ) = 0 otherwise.

Theorem 42.8. Joint Density of Order Statistics

If X1 , . . . , Xn are independent, continuous random variables that each have
density fX (x), then the order statistics X(1) , X(2) , . . . , X(n) have joint density

fX(1) ,...,X(n) (x1 , . . . , xn ) = n!f (x1 ) · · · f (xn ) for x1 < · · · < xn ,

and fX(1) ,...,X(n) (x1 , . . . , xn ) = 0 otherwise.
 42.3. Joint Density and Joint CDF of Order Statistics 571

Now we can easily recompute some of the previous problems from earlier chap-
ters.

Example 42.9. As in Exercise 31.29, let X1 , X2 , X3 be independent continuous

random variables, each Uniformly distributed in the interval [0, 10]. As an
example, X1 , X2 , X3 might be the arrival times of three people to a bus stop
between 8:50 AM and 9:00 AM, when they are not wearing wrist watches, so
they arrive Uniformly during a 10-minute interval. Let Y denote the middle of
the three values, i.e., the arrival time of the 2nd person.

In the language of order statistics, Y is always the second-smallest of X1 , X2 , X3 ,

so we have Y = X(2) . The joint density of X(1) , X(2) , X(3) is

fX(1) ,X(2) ,X(3) (x1 , x2 , x3 ) = 3!f (x1 )f (x2 )f (x3 ) for x1 < x2 < x3 ,
 1  1  1 
=6 for 0 ≤ x1 < x2 < x3 ≤ 10,
10 10 10
6
= for 0 ≤ x1 < x2 < x3 ≤ 10,
1000

and fX(1) ,X(2) ,X(3) (x1 , x2 , x3 ) = 0 otherwise.

Thus, the density of X(2) can be found by integrating X(1) and X(3) out of
the picture. If we integrate the joint density fX(1) ,X(2) ,X(3) (x1 , x2 , x3 ) over all
X(1) ’s that are less than X(2) , and integrate over all X(3) ’s that are greater than
X(2) , then we get the density fX(2) (x2 ) of the middle value, Y = X(2) . So the
density of X(2) is
Z 10 Z x2
6 6
fX(2) (x2 ) = dx1 dx3 = x2 (10 − x2 ) for 0 ≤ x2 ≤ 10,
x2 0 1000 1000

and fX(2) (x2 ) = 0 otherwise. We rewrite this result in the following way:
 1  x  10 − x 
2 2
fX(2) (x2 ) = 6 for 0 ≤ x2 ≤ 10,
10 10 10
and fX(2) (x2 ) = 0 otherwise. If we write fX (x) = 1/10 as the density of each of
the Xj ’s, and we write FX (x) = 10 x
as the cumulative distribution function of
each of the Xj , and 1 − FX (x) = 1 − 10 x
, then we notice that fX(2) (x2 ) has the
really nice form

fX(2) (x2 ) = 3!fX (x2 )FX (x2 )(1 − FX (x2 )) for 0 ≤ x2 ≤ 10,

and fX(2) (x2 ) = 0 otherwise. This might not look very general, but it is actually
a very special case of the following nice, general result:
572 Chapter 42. Order Statistics

Example 42.10. Consider independent continuous random variables X1 , X2 ,

. . . , Xn that each have density fX (x) and cumulative distribution function
FX (x). Then the density of the jth order statistic, fXj (xj ), is exactly
 
n
fX(j) (xj ) = fX (xj )(FX (xj ))j−1 (1 − FX (xj ))n−j ,
j − 1, 1, n − j
where  
n n!
= .
j − 1, 1, n − j (j − 1)!1!(n − j)!
n
In particular, j−1,1,n−j n!
is exactly the number of ways to pick


= (j−1)!1!(n−j)!
exactly 1 of the n variables X1 , . . . , Xn to be the jth order statistic (i.e., to be
the jth smallest of the collection X1 , . . . , Xn ), while also choosing j − 1 of the
variables to be smaller than Xj , and finally letting the other n − j variables be
larger than Xj .

Theorem 42.11. Density of a Particular Order Statistic

If X1 , X2 , . . . , Xn are independent, continuous random variables that each have
density fX (x), then the density of the jth order statistic, fXj (xj ), is exactly
 
n
fX(j) (xj ) = fX (xj )(FX (xj ))j−1 (1 − FX (xj ))n−j ,
j − 1, 1, n − j

where  
n n!
= .
j − 1, 1, n − j (j − 1)!1!(n − j)!
is exactly the number of ways to pick exactly 1 of the n variables X1 , . . . , Xn
to be the jth order statistic (i.e., to be the jth smallest of the collection
X1 , . . . , Xn ), while also choosing j − 1 of the variables to be smaller than Xj ,
and finally letting the other n − j variables be larger than Xj .
In the special case discussed in Example 42.9, we have
n=3 and j = 2,
as well as
x x
fX (x) = 1/10 and FX (x) = and 1 − FX (x) = 1 − .
10 10
Therefore, we are able to quickly verify the earlier result, from Example 42.9.
For 0 ≤ x2 ≤ 10,
 
3
fX(2) (x2 ) = fX (x2 )FX (x2 )(1 − FX (x2 ))
1, 1, 1
 1  x  10 − x 
2 2
=6 .
10 10 10
Otherwise, fX(2) (x2 ) = 0.
 42.3. Joint Density and Joint CDF of Order Statistics 573

Example 42.12. As in Example 42.1, consider the waiting times X1 and X2

(in minutes) until Samuel hears from two of his friends, Mary and Josephine,
respectively. He doesn’t know which one will call first. Both of the Xj ’s are
Exponential with expected value 5. The Xj ’s are also independent.

Using what we learned in Example 42.10, we can verify the calculation of the
densities of X(1) and X(2) from Example 42.1. We have
 
2 1 −x1 /5 2
fX(1) (x1 ) = e (1 − e−x1 /5 )0 (e−x1 /5 )1 = e−2x1 /5 .
0, 1, 1 5 5

Also
 
2 1 −x2 /5 2
fX(2) (x2 ) = e (1 − e−x2 /5 )1 (e−x2 /5 )0 = e−x2 /5 (1 − e−x2 /5 ).
1, 1, 0 5 5

This agrees with the calculations of the densities made in Example 42.1.

Example 42.13. Consider seven students whose grades are independent and
are each Uniform on the interval from 82 to 98. Let X1 , . . . , X7 denote their
grades. Rank the student’s scores in order, and denote the order statistics as,
respectively, X(1) , . . . , X(7) .

Since each Xj is Uniform on the interval from 82 to 98, then the density of
each Xj is 98−82
1
on the interval 82 ≤ x ≤ 98. Also, the cumulative distribution
function of each Xj is FX (x) = 98−82x−82
for 82 ≤ x ≤ 98. Thus, applying the
general model from Example 42.10, we see that the fifth-smallest student’s score,
X(5) , has the following density, for 82 ≤ x5 ≤ 98:
 
7 1  x5 − 82 5−1  x5 − 82 7−5
fX(5) (x5 ) = 1− .
4, 1, 2 98 − 82 98 − 82 98 − 82

Otherwise, fX(5) (x5 ) = 0.

This can be simplified to the following:
 1  x − 82 4  98 − x 2
5 5
fX(5) (x5 ) = 105 for 82 ≤ x5 ≤ 98,
16 16 16
and fX(5) (x5 ) = 0 otherwise.
574 Chapter 42. Order Statistics

Example 42.14. Now we generalize the formula for the density of the jth order
statistic of n independent random variables X1 , . . . , Xn that are each Uniform
on the interval from a to b. As usual, we rank the random variables in order,
and we denote the order statistics as, respectively, X(1) , . . . , X(n) .
Since each Xj is Uniform on the interval from a to b, then the density of each
Xj is b−a
1
on the interval a ≤ x ≤ b. Also, the cumulative distribution function
of each Xj is FX (x) = x−ab−a for a ≤ x ≤ b. Thus, using the model from
Example 42.10,

Theorem 42.15. The Order Statistics of Independent, Identically

Distributed Uniform Random Variables
If X1 , X2 , . . . , Xn are independent and Uniformly distributed on the interval
[a, b], then the density of the jth order statistic, i.e., the density of the jth
smallest of the n Uniform random variables, is
 
n 1  xj − a j−1  xj − a n−j
fX(j) (xj ) = 1− ,
j − 1, 1, n − j b−a b−a b−a

for a ≤ xj ≤ b, and fX(j) (xj ) = 0 otherwise.

In particular, we notice that the jth order statistic, in a collection of inde-

pendent random variables that have identical Uniform distributions, is not a
Uniform random variable itself.

Corollary 42.16. The Order Statistics of Independent, Identically

Distributed Uniform Random Variables
If X1 , X2 , . . . , Xn are independent and Uniformly distributed on the inter-
val [a, b], then none of the order statistics X(1) , X(2) , . . . , X(n) are Uniformly
distributed.

This observation makes sense intuitively as well. For instance, if we look at

the smallest random variable, i.e., the first order statistic, X(1) , from a collection
of 10 Uniform random variables distributed on the interval [0, 2], we know that
X(1) should have a density that is more heavily weighted toward the “0” end
of [0, 2] (i.e., toward the smaller end of the possibilities), and only very lightly
weighted on the “2” end of the interval [0, 2]. It would be relatively unlikely, for
instance, that the smallest of 10 random variables is at-or-above 1.72, because
that would mean that the other 9 random variables (which must be larger than
X(1) ) are forced to be larger than X(1) too, and, in particular, are forced to be
larger than 1.72. It would be much more likely for X(1) to be, say, between 0
and 0.3, i.e., somewhere closer to the lower end of the spectrum of possibilities,
because this puts less limitations on the ranges of possible values of the other
nine random variables.
 42.3. Joint Density and Joint CDF of Order Statistics 575

Example 42.17. Consider 10 people who are each waiting for an email to
arrive, and the waiting times are independent Exponential random variables,
each with average 30 minutes.

In this example, the waiting times can be referred to as X1 , . . . , Xn , and we

are given E(Xj ) = 30 for each j, so each Xj has parameter λ = 1/30. Thus the
density of each Xj is
1 −x/30
fXj (x) = e for x > 0,
30
and fXj (x) = 0 otherwise. Also, each Xj has cumulative distribution function

FXj (x) = 1 − e−x/30 for x > 0,

and FXj (x) = 0 otherwise. Thus, the density of the jth order statistic, i.e., the
density of the time of the jth email to arrive, is
 
10 1 −xj /30  j−1  10−j
fX(j) (xj ) = e 1 − e−xj /30 e−xj /30 .
j − 1, 1, 10 − j 30
Nothing in this equation is special about 10 emails arriving altogether, or about
the particular parameter λ = 1/30. We can generalize the previous example—
which also generalizes Examples 42.1 and 42.12—as follows:

Example 42.18. Consider n Exponential random variables (usually viewed as

waiting times) that are independent and have the same parameter, say, λ.

In this example, we refer to the Exponential random variables as X1 , . . . , Xn ,

and we are told that all of the Xj ’s have the same parameter λ. Then the density
of each Xj is
fXj (x) = λe−λx for x > 0,
and fXj (x) = 0 otherwise. Also, each Xj has cumulative distribution function

FXj (x) = 1 − e−λx for x > 0,

and FXj (x) = 0 otherwise. Thus,

Theorem 42.19. The Order Statistics of Independent, Identically

Distributed Exponential Random Variables
If X1 , X2 , . . . , Xn are independent and Exponentially distributed with parame-
ter λ (i.e., with expected value 1/λ), then the density of the jth order statistic,
i.e., the density of the jth smallest of the n Exponential random variables, is
 
n  j−1  n−j
fX(j) (xj ) = λe−λxj 1 − e−λxj e−λxj for xj > 0,
j − 1, 1, n − j

and fX(j) (xj ) = 0 otherwise.

576 Chapter 42. Order Statistics

In particular, we notice that the jth order statistic, in a collection of in-

dependent random variables that have identical Exponential distributions, is
not an Exponential random variable itself, with one exception: The first order
statistic—the minimum of the collection of independent Exponential random
variables—is an Exponentially distributed random variable.

Corollary 42.20. The Order Statistics of Independent, Identically

Distributed Exponential Random Variables
If X1 , X2 , . . . , Xn are independent and Exponentially distributed with ex-
pected value 1/λ, then the first order statistic X(1) is the only one of the
order statistics order statistics X(1) , X(2) , . . . , X(n) that is also Exponentially
distributed.
The density of the first order statistic is
 
n  n−1
fX(1) (x1 ) = λe−λx1 e−λx1 = nλe−nλx1 for x1 > 0,
0, 1, n − 1

and fX(1) (x1 ) = 0 otherwise. So X(1) is Exponentially distributed with pa-

rameter nλ, i.e., with expected value E(X(1) ) = nλ
1
.

Finally, we conclude with a few more examples.

Example 42.21. As in Example 22.5, consider the following: The lifetime of a

music player (before it permanently fails) is a random variable X with density
1
fX (x) = e−x/3 for x > 0,
3
and fX (x) = 0 otherwise.
If seven friends have these music players, and X1 , . . . , X7 denote the lifetimes
of the 7 devices, then the 5th order statistic, X(5) , is the time until the 5th one
out of 7 permanently fails. For x5 > 0, the 5th order statistic has density
 
7  5−1  7−5
fX(5) (x5 ) = (1/3)e−x5 /3 1 − e−x5 /3 e−x5 /3
4, 1, 2
 4  2
= (105)(1/3)e−x5 /3 1 − e−x5 /3 e−x5 /3 ;

otherwise, fX(5) (x5 ) = 0.

Example 42.22. As in Example 26.6, consider the following: Each time a

pitcher delivers a fastball, the speed is distributed between 90 and 100 miles
per hour, with density 1/10. Assume that the speeds of pitches are independent.
Suppose that the pitcher throws n such fastballs.
 42.3. Joint Density and Joint CDF of Order Statistics 577

By the results of Example 42.14, applied to Uniform random variables on

the interval [90, 100], we know that the density of the jth order statistic, i.e.,
the jth slowest of the fastballs is
n!  1  x − 90 j−1  xj − 90 n−j
j
fX(j) (xj ) = 1− ,
(j − 1)!1!(n − j)! 10 10 10
for 90 ≤ xj ≤ 100, and fX(j) (xj ) = 0 otherwise.
If the pitcher throws 5 fastballs, and we want to compute the probability
that 2 or more of them are below 93 miles per hour, it suffices to prove that the
2nd order statistic (i.e., the 2nd-slowest of the 5 fastballs) is less than 93 miles
per hour.
The density for j = 2, i.e., for the 2nd order statistic, i.e., for the 2nd-slowest
of the n = 5 fastballs, is, for 90 ≤ x2 ≤ 100,
5!  1  x − 90 2−1  x2 − 90 5−2
2
fX(2) (x2 ) = 1−
(2 − 1)!1!(5 − 2)! 10 10 10
 1  x − 90  x2 − 90  3
2
= 20 1− .
10 10 10
Otherwise, fX(j) (x2 ) = 0.
So the probability that the 2nd-slowest of the fastballs is less than 93 miles
per hour is
Z 93 Z 93  
1 x − 90  x − 90 3 23589
fX(2) (x) dx = 20 1− dx = = 0.47178.
90 90 10 10 10 50000

Example 42.23. As in Example 42.14, consider independent random variables

X1 , . . . , Xn that are each Uniform on the interval from a to b. Rank the random
variables in order, and denote the order statistics as, respectively, X(1) , . . . , X(n) .
We showed in Example 42.14 that the jth-smallest order statistic, X(j) , has the
following density:
 
n 1  xj − a j−1  xj − a n−j
fX(j) (xj ) = 1− ,
j − 1, 1, n − j b−a b−a b−a
for a ≤ xj ≤ b, and fX(j) (xj ) = 0 otherwise.
In particular, if a = 0 and b = 1, i.e., if X1 , . . . , Xn are each Uniform on
[0, 1], then the jth order statistic has density
n!
fX(j) (x) = xj−1 (1 − x)n−j for 0 ≤ x ≤ 1,
(j − 1)!1!(n − j)!
 
n j−1
=j x (1 − x)n−j for 0 ≤ x ≤ 1,
j
578 Chapter 42. Order Statistics

and fX(j) (x) = 0 otherwise. Thus, the expected value of the jth order statistic,
X(j) , is
Z 1
E(X(j) ) = xfX(j) (x) dx
0
Z 1  
n j
= j x (1 − x)n−j dx
0 j
 Z 1
n
=j xj (1 − x)n−j dx
j 0

Also
1
(n − j)!j!
Z
xj (1 − x)n−j dx =
0 (n + 1)!
(the inquisitive reader may want to consider why—the reasoning is connected
to the density of a Beta random variable), and thus we conclude

j
E(X(j) ) = .
n+1

Example 42.24. In general (i.e., when a and b are not necessarily 0 and 1),
then if X1 , . . . , Xn are independent and Uniform on [a, b], then we can normalize
that Xj ’s so that they are distributed on the interval [0, 1]. To do this, we define
Xj −a
Yj = b−a , and then each Yj is Uniform on [0, 1]. So the jth order statistic Y(j)
j
of Y1 , . . . , Yn has expected value E(Y(j) ) = n+1 , by the results of the previous
 
X(j) −a j
example. Equivalently, E b−a = n+1 , so the expected value of the jth
order statistic X(j) is
j(b − a)
E(X(j) ) = + a.
n+1

42.4 Exercises
42.4.1 Practice
Exercise 42.1. Dancers. In the middle of a complicated dance routine, five
dancers are located across the width of the stage. Let X1 , . . . , X5 denote their
positions across the stage. Assume that the Xj ’s are independent and that each
Xj is Uniformly distributed on [0, 100], where “0” denotes the far left side of the
stage, and “100” denotes the far right side of the stage, and the measurements
are given in feet.
 42.4. Exercises 579

a. Find the density of the person located closest to the left-hand side of the
stage, i.e., find fY (y) for Y = min(X1 , . . . , X5 ). (Hint: Y is just the 1st order
statistic, i.e., Y = X(1) .)
b. Find the probability that nobody is located within 20 feet of the left-hand
side of the stage.
c. Find the probability that at least two people are located within 30 feet
of the left-hand side of the stage. (Hint: You can use the 2nd order statistic for
this part.)
Exercise 42.2. Throwing darts. Alfredo, Barbara, and Cathy throw darts at
a dartboard of radius 9 inches. Let X1 , X2 , X3 denote the distance of Alfredo,
Barbara, and Cathy’s darts (respectively) from the center of the board. Thus,
X(1) (the first order statistic) is the minimum distance from the center to any
of the darts; X(3) is the distance to the farthest dart; and X(2) is the distance
to the middle of the three darts with respect to the origin.
a. Find the density of X(2) .
R9
b. Check that fX(2) (x2 ) is a density, i.e., check that 0 fX(2) (x2 ) dx2 = 1.
Exercise 42.3. Phone calls. Suppose that three male students are making
telephone calls to their girlfriends. Let X1 , X2 , X3 denote the times of their
phone calls. Suppose that the Xj ’s are independent Exponential random vari-
ables, each with expected value 20 minutes.
a. Find the density of X(1) , the first order statistic. (Easy check: X(1) is
the minimum of three independent Exponentials, so we remember that X(1) is
Exponential too.)
b. Find the probability that none of the calls are less than 12 minutes long,
i.e., they all exceed 12 minutes.
c. Find the probability that two or more of the calls are less than 23 minutes
each.
Exercise 42.4. Let X1 , X2 , X3 be three independent random variables that
are Uniformly distributed on the interval [0, 20]. Find the probability that the
minimum of the three random variables is between 12 and 15.
Exercise 42.5. Standing in a line. Five people stand along a street waiting
for a bus to arrive. The street is 9 yards away from a building.
√
The distance (in
x
yards) of each person from the street has density fX (x) = 18 , for 0 ≤ x ≤ 9.
These distances are independent. Find the density of the greatest of the five
distances from the street.
Exercise 42.6. Student presentations. In four sections of a course, running
(independently) in parallel, there are four students giving presentations that
are each Exponential in length, with expected value of 10 minutes each. How
time much do we expect to be needed until all four of the presentations are
completed?
580 Chapter 42. Order Statistics

Exercise 42.7. Soda cans. The weight of a can of soda is Uniformly dis-
tributed, between 238 and 242 grams. Find the density of:
a. the minimum of the three weights among three cans of soda;
b. the maximum of the three weights among three cans of soda;
c. the 2nd-largest of the three weights among three cans of soda.
Exercise 42.8. Pizza delivery. Suppose that the delivery time for a pizza
delivery is Uniformly distributed between 15 to 20 minutes. Now suppose that
4 people from a dormitory independently order pizza from 4 different drivers
(so the times for delivery are independent). What is the density of:
a. the minimum time until a pizza arrives?
b. the maximum time until a pizza arrives?
c. the time until the second pizza?
d. the time until the third pizza?
Exercise 42.9. Homework times. The time in which a student starts his
homework each night—during a seven day week—is Uniformly distributed be-
tween 6 PM and 8 PM. Let X and Y denote, respectively, the earliest and latest
times that he starts his homework that week.
a. Find the density of X.
b. Find the density of Y .
c. Find the expected value of X.
d. Find the expected value of Y .
Exercise 42.10. Marble drops. On a Rube Goldberg machine, three mar-
bles are dropped from a ledge, and they land at a location which is Uniformly
distributed between 0 to 12 inches away from a wall. What is the expected
location of the marble that lands farthest from the wall?
Exercise 42.11. Student arrivals. A class of 40 students is supposed to arrive
at 8:30 AM, but in practice they each arrive at a time Uniformly distributed
between 8:20 AM and 8:35 AM, and their arrival times are independent.
a. What is the expected arrival time of the last student to arrive?
b. What is the probability that 38 or more of the students have arrived by
8:30 AM?
Exercise 42.12. Shot put throws. Three athletes, competing in the shot
put event at a track and field competition, throw the ball (called the “shot”) as
far as they are able. The athletes have relatively similar strengths, so assume
that the distances they throw the ball are independently, Uniformly distributed
random variables on the interval [16.5, 19.5], measured in meters. What is the
probability that at least two of the three throws exceed 17 meters?
 42.4. Exercises 581

42.4.2 Extensions
Exercise 42.13. Violinist’s hands. When an extremely talented violin player
is soloing, the position of her hand along the violin’s neck seems to be Uniform
on a 9-inch interval (the total length is about 13 inches from nut to bridge; but
not all of the string is used). The violinist is photographed 6 times during a
performance. Assume that the location of her hand is independently placed in
each of the photos.

a. Find the jth order statistic of the location of her hand (1 ≤ j ≤ 6).
b. Find the probability that her hand is within 3 inches of the nut during
at least 2 of the 6 photographs.

Exercise 42.14. Let X1 , X2 , X3 , X4 be independent random variables, each

uniformly distributed on the interval [0, 10]. Let X(1) = min{X1 , X2 , X3 , X4 }
be the first order statistic. Find the probability that X(1) is smaller than 3, i.e.,
find P (X(1) ≤ 3).
 Chapter 43

Moment Generating Functions

A generating function is a clothesline on which we hang up a sequence of num-

bers for display.
—generatingfunctionology by Herbert S. Wilf (A. K. Peters, 2006)

How can we systematically calculate the moments of a random variable?

43.1 A Brief Introduction to Generating Functions

The study of generating functions is a topic in mathematics that has applications

and meanings in many topics, including discrete mathematics, complex analysis,
probability theory, asymptotics, algorithm analysis, etc. Generating functions
are also used in applied mathematics and in other disciplines (e.g., computer
science) as well.
Neil Sloane’s The whole idea of generating functions simply goes back to Maclaurin series,
On-Line a fundamental topic in calculus. Remember that a Maclaurin series is just a
Encyclopedia of Taylor series expansion of a function g(t) around the point t = 0. In other
Integer Sequences, words, a Maclaurin series is just a way to write a function g(t) as an infinite
http://oeis.org/,
series, in which the jth term is the jth derivative of g, evaluated at t = 0,
contains 250,000
integer sequences divided by j!.
∞
that might be of
X g (j) (0) j
g(t) = t .
interest. We can j!
j=0
also think about
non-integer The idea of Maclaurin series can perhaps appear more friendly upon first en-
sequences as well. counter if we consider them from Wilf’s point of view, as a “clothesline on which
(j)
we hang up a sequence of numbers for display.” If the numbers g j!(0) are of
some interest to us (e.g., because they form a sequence we want to study), then
we can view a Maclaurin series as a nice way to display these numbers. We just

582
 43.1. A Brief Introduction to Generating Functions 583

g (j) (0)
put j! as a coefficient of tj , and then we sum up all of the terms; anytime we
(j)
need g j!(0) , it will be sitting in front of tj , ready to be retrieved. The clothesline
looks like this:
g(0) 0 g 0 (0) 1 g 00 (0) 2 g 000 (0) 3 g (4) (0) 4 g (5) (0) 5 g (6) (0) 6
g(t) = t + t + t + t + t + t + t +· · ·
0! 1! 2! 3! 4! 5! 6!
At this point, one might wonder what to do if we do not have a function
in mind with interesting derivatives. Can we start from a sequence of numbers
and build such a function? Yes, we can start with a sequence of numbers
and construct—from the sequence—a new function g(t) that has interesting
coefficients in its Maclaurin series, so we want to display them on a clothesline.
One motivation for doing this is that a series representation is often a much
more compact way to store a sequence of integers, especially when no general,
succinct form of the numbers themselves are available. To build such functions
ourselves, we start with an interesting sequence of numbers a0 , a1 , a2 , . . ., and
we can just put them onto such a clothesline and add things up, and see what
function we get as a result. We can define g by writing
∞
X aj
g(t) = tj
j!
j=0
a2 2 a3 3 a4 4 a5 5 a6 6 a7 7
= a0 + a1 t + t + t + t + t + t + t + ···
2! 3! 4! 5! 6! 7!
Taking j derivatives, we get aj = g (j) (0) from this function g that we created!
The reason behind aj = g (j) (0) should be clear after taking a few derivatives
and trying it, but nonetheless, we put the reasoning in the Appendix to this
chapter; see Section 43.5.
Definition 43.1. Generating Function
If a0 , a1 , a2 , . . . is an interesting sequence of numbers from which we want to
build a generating function g(t), we write
∞
X aj
g(t) = tj
j!
j=0
a2 2 a3 3 a4 4 a5 5 a6 6 a7 7
= a0 + a1 t + t + t + t + t + t + t + ···
2! 3! 4! 5! 6! 7!
and then we get
aj = g (j) (0)
from this function g that we created!

There are several kinds of generating functions that are interesting to study,
e.g., ordinary generating functions, Exponential generating functions, multi-
variate generating functions, etc. (See, for instance, Analytic Combinatorics by
P. Flajolet and R. Sedgewick [3] for a comprehensive introduction to generating
functions.) In this chapter, we build moment generating functions.
584 Chapter 43. Moment Generating Functions

43.2 Moment Generating Functions

The moment generating function associated with random variable X, is
defined by defining aj as
aj = E(X j ),
the jth moment of a random variable. Thus, the moment generating function
g(t) or MX (t) associated with X should have the property that
∞ ∞
X aj X E(X j )
MX (t) = tj = tj .
j! j!
j=0 j=0

(We use g(t) and MX (t) interchangeably in the rest of the discussion, since
moment generating functions are the only kind of generating function we handle
in this book.) The equation above lets us wrap all moments E(X j ) of a random
variables into a compact form. They are divided by j! and then conveniently
stored as the coefficients of MX (t). Since the t has nothing to do with X, i.e.,
the t is a constant with regard to X, we can pull the t from each term into the
expected value, and we get
∞
X E((tX)j )
MX (t) = .
j!
j=0

The sum of expected values is equal to the expected value of the sum, so this
implies
∞
(tX)j
X 
MX (t) = E .
j!
j=0

Finally, we note that, regardless of the value of X, we always have

∞
X (tX)j
= etX ,
j!
j=0

so we conclude that the moment generating function associated with X is

MX (t) = E(etX ).

Definition 43.2. Moment generating function The moment generating

function MX (t) of a random variable X is equal to the expected value of etX :
∞ ∞
E(X j ) (tX)j
X X 
j
MX (t) = t =E = E(etX ).
j! j!
j=0 j=0

We know that, as discussed above, for all generating functions of this form,
aj = g (j) (0). In this case, aj = E(X j ) and MX (t) = E(etX ). Thus, if we com-
pute E(etX ), and we then take j derivatives and substitute in t = 0, we will get
 43.2. Moment Generating Functions 585

the jth moment of X. This is sometimes referred to as “moment pumping” be-

cause it just causes moment after moment of the random variable to be pumped
out. One new moment appears each time we take a derivative.
(We emphasize that t is just used as a variable (not a random variable),
which has nothing to do with X. It is, in a sense, a dummy variable or a place-
holder, that allows the scheme of the generating functions to work properly.)
The formula MX (t) = E(etX ) gives us the usual, straightforward way to
calculate the moment generating function of a random variable. We are just
calculating the expected value of the function etX , and we have lots and lots
of experience with calculating the expected value of a function of a random
variable. If X is discrete, then
X
MX (t) = E(etX ) = etx pX (x).
x

If X is continuous, then
Z ∞
MX (t) = E(etX ) = etx f (x) dx.
−∞

Definition 43.3. Moment generating function: discrete and continu-

ous cases
If X is discrete, the moment generating function of X is
X
E(etX ) = etx pX (x).
x

If X is continuous, the moment generating function of X is

Z ∞
tX
E(e ) = etx f (x) dx.
−∞

At this point, one might wonder what is so great about moment generating
function. For instance, if we want the 2nd moment of a random variable X, we
can just compute X
E(X 2 ) = x2 pX (x),
x

if X is discrete. Alternatively, if X is continuous, we have

Z ∞
E(X 2 ) = x2 f (x) dx.
−∞

The trouble, however, is that if we want another moment of X, (say,

for instance, the 3rd moment of X), we have to just turn around and
do the whole computation again (using, for instance, x3 instead of x2 ).
So the advantage of using moment generating functions is that, if we compute
586 Chapter 43. Moment Generating Functions

MX (t) = E(etX ) at the start, then we can get any moment of X by simply
taking derivatives of MX (t) and then evaluating at t = 0. To get the jth
moment of X, we just take j derivatives and then evaluate at t = 0. Taking
derivatives is generally much, much easier than taking sums or evaluating inte-
grals, so the method of moment generating functions has some advantages over
the direct calculation of moments. Based on our experience from calculus, we
are already good at taking derivatives, so the method of moment generating
functions should be appealing.

43.3 Moment Generating Function: Discrete Case

Example 43.4. Consider a Binomial random variable X with parameters n

and p. Then the mass of X is
 
n x
pX (x) = p (1 − p)n−x for x = 0, 1, 2, . . . , n,
x
and pX (x) = 0 otherwise. So the moment generating function of X is
MX (t) = E(etX )
n  
tx n
X
= e px (1 − p)n−x
x
x=0
n  
X n
= (et p)x (1 − p)n−x
x
x=0
= (e p + 1 − p)n ,
t

where the last line follows from the Binomial theorem, i.e., from
n  
X n j n−j
a b = (a + b)n .
j
j=0

So the moment generating function of a Binomial random variable X with

parameters n and p is
MX (t) = E(etX ) = (et p + 1 − p)n .
Thus, taking one derivative with respect to t, we see that the first moment of
X is
d t
E(X) = g 0 (0) = (e p + 1 − p)n
dt t=0
t n−1 t
= (n)(e p + 1 − p) (e p) t=0
n−1
= (n)(p + 1 − p) (p)
= np,
 43.3. Moment Generating Function: Discrete Case 587

and the second moment of X is

E(X 2 ) = g 00 (0)
d2 t
= (e p + 1 − p)n
dt2 t=0
d
= (n)(et p + 1 − p)n−1 (et p)
dt t=0
 
t n−2 t 2
= (n)(n − 1)(e p + 1 − p) (e p) + (n)(et p + 1 − p)n−1 (et p)
t=0
= (n)(n − 1)(p + 1 − p)n−2 (p)2 + (n)(p + 1 − p)n−1 (p)
= (n)(n − 1)(p)2 + np.

So the variance of X is

Var(X) = E(X 2 ) − (E(X))2

= (n)(n − 1)(p)2 + np − (np)2
= (n)(−1)(p)2 + np
= np(1 − p).

All of this agrees with what we computed earlier in the course about Binomial
random variables. This technique also enables us to readily calculate E(X j ) for
higher values of j.

Example 43.5. If X is a Poisson random variable with parameter λ, the mass

of X is
e−λ λx
pX (x) = for x = 0, 1, 2, . . .,
x!
and pX (x) = 0 otherwise. So the moment generating function of X is

MX (t) = E(etX )
∞
X e−λ λx
= etx
x!
x=0
∞
−λ
X (et λ)x
=e
x!
x=0
−λ (et λ)
=e e
((et −1)λ)
=e .

So the moment generating function of a Poisson random variable X with pa-

rameter λ is
t
MX (t) = E(etX ) = e((e −1)λ) .
588 Chapter 43. Moment Generating Functions

Thus, the first moment of X is

E(X) = g 0 (0)
d ((et −1)λ)
= e
dt t=0
((et −1)λ) t
=e eλ
t=0
= e(0λ) λ
= λ,

and the second moment of X is

E(X 2 ) = g 00 (0)
d2 ((et −1)λ)
= e
dt2 t=0
d ((et −1)λ) t
= e eλ
dt t=0
 t t

((e −1)λ) t 2
= e (e λ) + e((e −1)λ) et λ
t=0
 
= e(0λ) λ2 + e(0λ) λ
= λ2 + λ.

So the variance of X is

Var(X) = E(X 2 ) − (E(X))2

= λ2 + λ − (λ)2
= λ.

All of this agrees with what we computed earlier in the course about Poisson
random variables. This technique also enables us to readily calculate E(X j ) for
higher values of j.

43.4 Moment Generating Function: Continuous Case

Example 43.6. If X is a Uniform random variable on the interval [a, b], the
density of X is
1
fX (x) = for a ≤ x ≤ b,
b−a
 43.4. Moment Generating Function: Continuous Case 589

and fX (x) = 0 otherwise. So the moment generating function of X is

MX (t) = E(etX )
Z ∞
= etx fX (x) dx
−∞
Zb
1
= etx dx
a b−a
b
1 etx
=
b−a t x=a
etb − eta
= .
(b − a)(t)

So the moment generating function of a random variable X that is Uniform on

[a, b] is
etb − eta
MX (t) = E(etX ) = .
(b − a)(t)

We use the series expansion of etb and eta because the computation of the
derivative with quotient rule (especially for the second moment) becomes quite
intricate and much more tedious. The first moment of X is

E(X) = g 0 (0)
d etb − eta
=
dt (b − a)(t) t=0

X (tb)j ∞ ∞
1 d 1 1 X (ta)j 
= + − −
b − a dt t j!t t j!t
j=1 j=1
t=0
P∞ (tb)j P∞ (ta)j
since etb = j=0 j! and eta = j=0 j!
∞ ∞
1 d  X tj−1 bj X tj−1 aj 
= −
b − a dt j! j!
j=1 j=1
t=0
∞ ∞
1 X (j − 1)tj−2 bj X (j − 1)tj−2 aj 
= −
b−a j! j!
j=1 j=1
t=0
1  b2 a2 
= − only the j = 2 term remains when t = 0
b − a 2! 2!
(b + a)(b − a)
=
(b − a)(2)
a+b
= ,
2
590 Chapter 43. Moment Generating Functions

and the second moment of X is

E(X 2 ) = g 00 (0)
d2 etb − eta
=
dt2 (b − a)(t) t=0
∞ ∞
1 d  X (j − 1)tj−2 bj X (j − 1)tj−2 aj 
= −
b − a dt j! j!
j=1 j=1
t=0
∞ ∞
1  X (j − 1)(j − 2)tj−3 bj X (j − 1)(j − 2)tj−3 aj 
= −
b−a j! j!
j=1 j=1
t=0
1  (2)(1)b3 (2)(1)a3 
= − only the j = 3 term remains, as t = 0
b−a 3! 3!
b3 − a3
=
(b − a)(3)
(b − a)(a2 + ab + b2 )
=
(b − a)(3)
a + ab + b2
2
= .
3

So the variance of X is

Var(X) = E(X 2 ) − (E(X))2

a2 + ab + b2  a + b 2
= −
3 2
(b − a)2
= .
12

All of this agrees with what we computed earlier in the course about Uniform
random variables. This technique also enables us to readily calculate E(X j ) for
higher values of j.

Example 43.7. If X is a standard Normal random variable, i.e., with E(X) = 0

and Var(X) = 1, then (using exp for the Exponential function)

1
fX (x) = √ exp(−x2/2) for all x.
2π
 43.4. Moment Generating Function: Continuous Case 591

So the moment generating function of X is

MX (t) = E(etX )
Z ∞
= etx fX (x) dx
−∞
Z ∞
1
= etx √ exp(−x2/2) dx
−∞ 2π
Z ∞
1  x2 
= √ exp tx − dx.
−∞ 2π 2

By adding and subtracting t2 in the numerator, we see that tx − x2/2 can be

written as the difference of two squares:
x2 2tx − x2 t2 − (t2 − 2tx + x2 ) t2 − (x − t)2
tx − = = = .
2 2 2 2
Writing things in this way will help us to change variables in the computation
above. Returning to our computation,
Z ∞  t2 − (x − t)2  Z ∞  −(x − t)2 
1 t2/2 1
MX (t) = √ exp dx = e √ exp dx.
−∞ 2π 2 −∞ 2π 2
Now we substitute u = x − t and du = dx, so that the equation becomes
Z ∞  −u2 
2 1 2
MX (t) = et /2 √ exp du = et /2 ,
−∞ 2π 2
2
where the last line follows since √12π exp( −u
2 ) is exactly the density of a standard
Normal random variable, and we get “1” when we integrate it over all u’s.
So the moment generating function of a standard Normal random variable
X is
2
MX (t) = E(etX ) = et /2 .
We can easily check now that the first moment of X is 0 as follows:
d t2/2 2
E(X) = g 0 (0) = e = tet /2 = 0,
dt t=0 t=0

and the second moment of X is

d2 t2/2 d t2/2 2 2
E(X 2 ) = g 00 (0) = e = te = t2 et /2 + et /2 = 0 + 1 = 1.
dt2 t=0 dt t=0 t=0

So the variance of X is

Var(X) = E(X 2 ) − (E(X))2 = 1 − 02 = 1.

All of this agrees with what we already knew about X since X is a standard
Normal random variable. This technique also enables us to readily calculate
E(X j ) for higher values of j.
592 Chapter 43. Moment Generating Functions

Note: The 3rd and 4th moments of a random variable are called the skew-
ness and the kurtosis, respectively. The skewness measures how “skewed” the
distribution is, i.e., whether the density or mass is heavier on one side than the
other. The kurtosis measures whether the concentration of the density or mass
is tall and narrow, or short and wide.

Example 43.8. If Y is a Normal random variable with parameters E(Y ) =

µ and Var(Y ) = σ 2 , then we note that Y σ−µ is a standard Normal random
variable. Thus, writing X = Y σ−µ , and using the facts about standard Normal
random variables from the previous example, we see that the moment generating
function of Y σ−µ is
2
E(et(Y −µ)/σ ) = E(etX ) = et /2
but we also note that

E(et(Y −µ)/σ ) = E(e(tY /σ)−(tµ/σ) ) = e−tµ/σ E(etY /σ )

Putting these together, we have

2
e−tµ/σ E(etY /σ ) = et /2

or, multiplying on both sides by etµ/σ ,

 t2 tµ 
E(etY /σ ) = exp +
2 σ
Temporarily replacing t/σ by v and t by σv, this yields
 (σv)2 
E(evY ) = exp + vµ
2
and thus the moment generating function of a Normal random variable Y with
E(Y ) = µ and Var(Y ) = σ 2 is
 (σv)2 
MY (v) = E(evY ) = exp + vµ .
2
We can easily check now that the first moment of Y is µ as follows:

E(Y ) = g 0 (0)
d  (σt)2 
= exp + tµ
dt 2 t=0
 2(σt)(σ)   (σt)2 
= + µ exp + tµ
2 2 t=0
= µ,
 43.5. Appendix: Building a Generating Function 593

and the second moment of Y is

E(Y 2 ) = g 00 (0)
d2  (σt)2 
= exp + tµ
dt2 2 t=0
d 2(σt)(σ)
   (σt)2 
= + µ exp + tµ
dt 2 2 t=0
 2(σt)(σ) 2  (σt)2   (σt)2 
= + µ exp + tµ + σ 2 exp + tµ
2 2 2 t=0
2 2
=µ +σ .

So the variance of Y is

Var(Y ) = E(Y 2 ) − (E(Y ))2 = µ2 + σ 2 − (µ)2 = σ 2 .

All of this agrees with what we already knew about Y . This technique also
enables us to readily calculate E(Y j ) for higher values of j.

43.5 Appendix: Building a Generating Function

In this chapter, we claim that if we start with a sequence a0 , a1 , a2 , . . . of num-
bers that are interesting to us, and if we if define g by writing
∞
X aj
g(t) = tj
j!
j=0
a2 2 a3 3 a4 4 a5 5 a6 6 a7 7
= a0 + a1 t + t + t + t + t + t + t + ···
2! 3! 4! 5! 6! 7!

then we get aj = g (j) (0) from this function g that we created! Now we explain
the reasoning behind this phenomenon.
If we evaluate at t = 0, we get g(0) on the left hand side, and on the right
hand side, we get a0 ; all of the other terms on the right-hand-side are just 0;
thus
g(0) = a0 .
If we take one derivative with respect to t, we get
a2 a3 a4 a5 a6
g 0 (t) = a1 + 2t + 3t2 + 4t3 + 5t4 + 6t5 + · · ·
2! 3! 4! 5! 6!
and then substitute in t = 0, we conclude

g 0 (0) = a1 .
594 Chapter 43. Moment Generating Functions

If we take two derivatives with respect to t, we get

a2 a3 a4 a5 a6
g 00 (t) = (2)(1) + (3)(2)t + (4)(3)t2 + (5)(4)t3 + (6)(5)t4 + · · ·
2! 3! 4! 5! 6!
and then substitute in t = 0, we conclude

g 00 (0) = a2 .

If we take three derivatives with respect to t, we get

a3 a4 a5 a6
g 000 (t) = (3)(2)(1) + (4)(3)(2)t + (5)(4)(3)t2 + (6)(5)(4)t3 + · · ·
3! 4! 5! 6!
and then substitute in t = 0, we conclude

g 000 (0) = a3 .

More generally, if we take j derivatives with respect to t, we get

aj aj+1 aj+2
g (j) (t) = j! + (j + 1)j t + (j + 2)j t2 + · · ·
j! (j + 1)! (j + 2)!
and then substitute in t = 0, we conclude

g (j) (0) = aj .

Thus, the jth derivative of g(t), evaluated at t = 0, is exactly aj , as desired.

43.6 Exercises
Exercise 43.1. If X is a Geometric random variable with parameter p (in other
words,
P (X = x) = pX (x) = (1 − p)x−1 p for x = 1, 2, 3, . . .,
and pX (x) = 0 otherwise), then find the moment generating function of X.

Exercise 43.2. Use the moment generating function from Exercise 43.1 to
verify that, if X is Geometric with parameter p, then E(X) = 1/p.

Exercise 43.3. Now use the moment generating function from Exercise 43.1
to verify that, if X is Geometric with parameter p, then E(X 2 ) = (2 − p)/p2 .

Exercise 43.4. Verify that if X is Geometric with parameter p, then Var(X) =

q/p2 .

Exercise 43.5. If X is a Negative Binomial random variable with parameters

p and r, then find the moment generating function of X. (Hint: Use the result
from Exercise 43.1 as an aid.)

Exercise 43.6. Use the moment generating function from Exercise 43.5 to
verify that, if X is Negative Binomial with parameters p, r, then E(X) = r/p.
 43.6. Exercises 595

Exercise 43.7. Now use the moment generating function from Exercise 43.5
to verify that, if X is Negative Binomial with parameters p, r, then E(X 2 ) =
r(r + 1 − p)/p2 .

Exercise 43.8. Verify that if X is Negative Binomial with parameters p, r,

then Var(X) = rq/p2 .

Exercise 43.9. If X is an Exponential random variable with parameter λ, then

find the moment generating function of X. (Assume, for this problem, that we
are using t in the range where t < λ.)

Exercise 43.10. Use the moment generating function from Exercise 43.9 to
verify that, if X is Exponential with parameter λ, then E(X) = 1/λ.

Exercise 43.11. Now use the moment generating function from Exercise 43.9
to verify that, if X is Exponential with parameter λ, then E(X 2 ) = 2/λ2 .

Exercise 43.12. Verify that if X is Exponential with mean 1/λ, then Var(X) =
1/λ2 .

Exercise 43.13. If X is a Gamma random variable with parameters λ and r,

then find the moment generating function of X. (Hint: Use the result from
Exercise 43.9 as an aid. Again, for this problem, that we are using t in the
range where t < λ.)

Exercise 43.14. Use the moment generating function from Exercise 43.13 to
verify that, if X is Gamma with parameters λ, r, then E(X) = r/λ.

Exercise 43.15. Now use the moment generating function from Exercise 43.13
to verify that, if X is Gamma with parameters λ, r, then E(X 2 ) = r(r + 1)/λ2 .

Exercise 43.16. Verify that if X is a Gamma random variable with parameters

λ, r, then Var(X) = r/λ2 .
Chapter 44

Transformations of One or Two

Random Variables

Math is the only place where truth and beauty mean the same thing.
—Danica McKellar
“Obvious” is the most dangerous word in mathematics.
—E. T. Bell

In a toy store, there are balls of many sizes available in a big bin. You randomly
select a ball from the bin. How does the size of the radius affect the volume of
the ball?

44.1 The Distribution of a Function of One Random

Variable

Example 44.1. Choose a ball at random from a large bin at the toy store.
Suppose that the radius (in inches) is Uniformly distributed on the interval
[0, 15]. What is the expected value of the volume of the ball?

We recall that a spherical ball of radius r has volume 34 πr3 . So if X denotes

the radius of the ball, and Y denotes the volume of the ball, then

4
Y = πX 3 .
3
596
 44.1. The Distribution of a Function of One Random Variable 597

Unfortunately, Y is not linearly related to X (i.e., we do not have Y = aX + b).

So we compute directly
 
4 3
E(Y ) = E πX
3
Z 15
4 3 1
= πx dx
0 3 15
4π 15 3
Z
= x dx
45 0
4π 154
=
45 4
= 1125π
= 3534.291736

So the expected volume of the ball is 3534.291736 cubic inches.

A different way to approach the problem above is to compute the density of
Y . Since 0 ≤ X ≤ 15, then 0 ≤ 43 πX 3 ≤ 34 π153 = 4500π. Again, we emphasize
that Y is not Uniform on the interval [0, 4500π], but we do know that Y cannot
be outside this interval. So the density of Y is fY (y) = 0 for y outside the
interval. For Y inside this interval, we have

FY (y) = P (Y ≤ y)
 
4
=P πX 3 ≤ y
3
 1/3 !
3y
=P X≤
4π
3y 1/3
( 4π ) −0
= since X is Uniformly distributed on [0, 15]
15 − 0
3y 1/3
( 4π )
=
15
Now taking a derivative with respect to Y , we get the density of Y :

d
fY (y) = FY (y)
dy
3y
d ( 4π )1/3
=
dy 15
3y −2/3 3
(1/3)( 4π ) 4π
=
15
( 3y )−2/3
= 4π
60π
With the density of Y in hand, we can compute probabilities that involve Y , or
598 Chapter 44. Transformations of One or Two Random Variables

we can compute E(Y ), or Var(Y ), etc. We compute

Z 4500π
E(Y ) = yfY (y) dy
0
4500π 3y −2/3
( 4π )
Z
= y dy
0 60π
3 −2/3 Z 4500π
( 4π )
= y 1/3 dy
60π 0
3 −2/3
( 4π ) (4500π)4/3
=
60π 4/3
= 1125π

which agrees with the value of E(Y ) we had computed above. The advantage of
this method is that we now have the density of Y , so we compute other things
if we want to, like other probabilities involving Y , or the variance of Y , etc.

Remark 44.2. How to find the CDF and density of a function of a

continuous random variable
If X is a continuous random variable and g is a function, then Y = g(X) is
a function of a continuous random variable. Often we know the distribution
(e.g., either the density or the CDF) of X but not the distribution of (e.g.,
either the density or CDF) of Y .
To find the CDF of Y , follow these basic steps:

1. Determine where Y is defined. In other words, find where fY (y) = 0

and where fY (y) > 0.

2. Focus on y in the region where Y is defined. Find FY (y) = P (Y ≤ y) by

using the inverse function g −1 if one exists, or use a similar technique if
g does not have a unique inverse.

3. Differentiate FY (y) with respect to y to get the density fY (y) of Y .

4. If desired, check to see that fY (y) is really a density, i.e., that integrating
fY (y) over all possible y’s gives a result of 1.

Step 1 of Remark 44.2 can be trickier than it sounds.

Example 44.3. Let X be a Uniform random variable on the interval [−15, 15].
Let Y = X 2 . First, we find the density of Y .
 44.1. The Distribution of a Function of One Random Variable 599

The range where X is located is

−15 ≤ X ≤ 15,

and thus
0 ≤ X 2 ≤ 152 = 225,
or in other words,
0 ≤ Y ≤ 225.
So the density fY (y) of Y is 0 for y outside the interval [0, 225]. Notice that,
in particular, even though X can be negative, that Y can never be negative,
because Y is a “square” of a quantity.
Now consider y in the interval [0, 225]. We want to compute

P (Y ≤ y) = P (X 2 ≤ y)

or equivalently
√ √
P (Y ≤ y) = P (− y ≤ X ≤ y)
This line is perhaps unexpected; note that f (x) = x2 does not have a unique
inverse, but we have used something like the inverse of f to find the correct
range for X.
Since X is Uniformly distributed, it follows that
√ √ √
y − (− y) 2 y
FY (y) = P (Y ≤ y) = = .
30 30
Differentiating with respect to Y yields

y −1/2
fY (y) = for 0 ≤ y ≤ 225,
30
and fY (y) = 0 otherwise.
We can perform an easy check to see that this is plausible for the density of
Y (i.e., that we did not make any mistakes). Of course we could get an integral
of “1” and still have a subtle mistake somewhere, but if the integral is exactly
1, then we are relatively sure that we performed the computation correctly. We
compute
225
225
y −1/2 y 1/2 (225)1/2
Z
15
dy = = = =1
0 30 (1/2)(30) (1/2)(30) 15
y=0

With the density of Y in hand, we can compute probabilities; for instance

64
64
y −1/2 y 1/2 (64)1/2
Z
8
P (Y ≤ 64) = dy = = = .
0 30 (1/2)(30) (1/2)(30) 15
y=0
600 Chapter 44. Transformations of One or Two Random Variables

We can also compute the expected value of Y :

Z  y −1/2 
225
E(Y ) = (y) dy
0 30
Z 225 1/2
y
= dy
0 30
225
y 3/2
=
(3/2)(30)
y=0
(225)3/2
=
(3/2)(30)
= 75.

Example 44.4. Let X be a Continuous Uniform (0, 1) random variable.

Let Y = X 3 . First we find the density of Y . Since 0 < X < 1, then also
0 < Y < 1 too. So fY (y) = 0 for y outside the interval (0, 1). Now consider y
inside the interval (0, 1), i.e., consider 0 < y < 1. We have

FY (y) = P (Y ≤ y)
= P (X 3 ≤ y)
= P (X ≤ y 1/3 )
y 1/3 − 0
=
1−0
= y 1/3

So we can differentiate with respect to y to get the density of Y :

d
fY (y) = FY (y)
dy
d 1/3
= y
dy
= (1/3)y −2/3

If in doubt, we can test

Z 1 1
(1/3)y −2/3 dy = y 1/3 = 1,
0 y=0

so fY (y) is a density.
 44.2. The Distributions of Functions of Two Random Variables 601

Now we find the expected value of Y :

Z 1
E(Y ) = (y)(1/3)y −2/3 dy
0
Z 1
= (1/3) y 1/3 dy
0
1
y 4/3
= (1/3)
4/3
y=0
1
= (1/3)
4/3
= 1/4

44.2 The Distributions of Functions of Two Random

Variables
In the previous section, when one random variable Y is a function of random
variable X, i.e., Y = f (X), we investigated how to determine the density of
Y from the density of X. In the present section, we push this concept one
step further. If U, V are random variables that are each functions of another
pair of random variables, X and Y , then we study how to get the joint density
fU,V (u, v) of U and V from the joint density fX,Y (x, y) of X and Y . The
procedure is relatively straightforward every time, but solving several examples
will help improve our understanding.
First consider two random variables X and Y for which the joint density
fX,Y (x, y) is known.
Write U as a function of X and Y , i.e.,

U = g(X, Y )

and write V as a function of X and Y , i.e.,

V = h(X, Y ).

(The functions g and h should already be given, so the above two equations
do not require any work! Some examples of functions g and h are given in the
examples later in this section.)
Compute four partial derivatives of g and h, with respect to x and y:

∂ ∂ ∂ ∂
g(x, y), g(x, y), h(x, y), h(x, y)
∂x ∂y ∂x ∂y

(These partial derivatives are often very straightforward to compute.)

602 Chapter 44. Transformations of One or Two Random Variables

Write the joint density of U and V (we will need to make some adjustments
afterwards, so this notation will look strange mathematically, but we remember
that x and y depend on u and v):

fX,Y (x, y)
fU,V (u, v) =
∂ ∂ ∂ ∂
∂x g(x, y) ∂y h(x, y) − ∂y g(x, y) ∂x h(x, y)

Notice that the left-hand side has u’s and v’s, but the right hand side has
x’s and y’s (because, as we noted, the x’s and y’s depend on u and v). So
we still need to make a substitution. Before we can use such a result, we will
have to solve for X and Y in terms of U and V , and substitute appropriate
combinations of u’s and v’s for every x and y.

Example 44.5. Consider random variables U, V, X, Y such that

U =1−X −Y

and
V = X − Y.
Suppose that the joint density fX,Y (x, y) of X and Y is known. Find the joint
density fU,V (u, v) of U and V .

Notice that, in this example, the new random variables are just linear combi-
nations of the old random variables in this example. (In Example 44.8 we will
see a non-linear example.)
We write
U = g(X, Y ) = 1 − X − Y
and
V = h(X, Y ) = X − Y.
Then we compute four partial derivatives of g and h, with respect to x and y:
∂ ∂ ∂ ∂
g(x, y) = −1, g(x, y) = −1, h(x, y) = 1, h(x, y) = −1
∂x ∂y ∂x ∂y
So we get
fX,Y (x, y)
fU,V (u, v) =
∂ ∂ ∂ ∂
∂x g(x, y) ∂y h(x, y) − ∂y g(x, y) ∂x h(x, y)
fX,Y (x, y)
=
|(−1)(−1) − (1)(−1)|
1
= fX,Y (x, y)
2
 44.2. The Distributions of Functions of Two Random Variables 603

Finally, we need to substitute u’s and v’s instead of x’s and y’s on the right hand
side of the equation. This means that—when possible (as in this problem)—
we can just solve for the values of x and y in the equations analogous to the
definitions of U and V :
U =1−X −Y
and
V = X − Y.
For instance, in this case, if we add U and V , we get U + V = 1 − 2Y , so
Y = 1−U2−V . If we subtract V from U , we get U − V = 1 − 2X, and so
X = 1−U2+V .
So we conclude that
 
1 1−u+v 1−u−v
fU,V (u, v) = fX,Y , .
2 2 2
This is a very general result because we did not even need to specify the actual
joint density fX,Y (x, y) of X and Y ; rather, this solution works regardless of
what the joint density is. A second question,

however, arises: For which u’s and
v’s is the joint density fX,Y 1−u+v
2 , 1−u−v
2 going to be strictly positive? This
involves some understanding of the two-dimensional geometry of the problem.
So we consider some specific cases below.

Example 44.6. Consider independent random variables X and Y that are each
Uniform on the interval [0, 1]. Suppose that

U =1−X −Y

and
V = X − Y.
Find the joint density fU,V (u, v) of U and V .

We have already concluded, in Example 44.5 above, that

 
1 1−u+v 1−u−v
fU,V (u, v) = fX,Y , .
2 2 2
Since X and Y are independent and Uniform on the square where 0 ≤ X ≤ 1
and 0 ≤ Y ≤ 1, then fX,Y (x, y) = 1 on this square, and thus the equation
 
1 1−u+v 1−u−v
fU,V (u, v) = fX,Y ,
2 2 2
reduces to just
1
fU,V (u, v) = ,
2
604 Chapter 44. Transformations of One or Two Random Variables

but the second question (about the geometry of the problem, which we men-
tioned above) still remains: Where is the joint density of U and V defined?
(Where is the joint density of U and V strictly positive?)
To answer this question, we can step around the four sides of the square
where X and Y are defined, and just find the analogous curves for U and V .
Side 1: When Y = 0, then Y = 1−U −V
2 becomes V = 1 − U .
Side 2: When X = 1, then X = 1−U +V
2 becomes V = 1 + U .
Side 3: When Y = 1, then Y = 1−U −V
2 becomes V = −1 − U .
Side 4: When X = 0, then X = 1−U +V
2 becomes V = U − 1.
In the plane for U and V , the four pieces fit together at the places indicated
in Figure 44.1.

y
Y =1
1

X=0 X=1

x
Y =0 1
v
1
X = 1 so Y = 0 so
V =U +1 V =1−U

u
−1 1
Y = 1 so X = 0 so
V = −U − 1 V =U −1
−1

Figure 44.1: Transforming the square 0 ≤ X ≤ 1, 0 ≤ Y ≤ 1 into the U, V

plane.

So we see that the pair of random variables (U ,V ) have a joint density that
is Uniform. In fact, fU,V (u, v) = 1/2, on a diamond-shaped region of the U, V
plane. √
This diamond is a square tilted on its side, and the diamond has sides of
length 2, so the diamond has area 2. Any time that a function is constant on
a region, then the integral over that region is just the constant times the area.
 44.2. The Distributions of Functions of Two Random Variables 605

Thus, in this case, if we integrate fU,V (u, v) = 1/2 over the whole diamond, we
get (2)(1/2) = 1. This check reassures us that the joint density seems to have
been calculated correctly, without mistakes.

Example 44.7. Consider independent, standard Normal random variables X

and Y . Suppose that
U =1−X −Y
and
V = X − Y.
Find the joint density fU,V (u, v) of U and V .
Again, we have already concluded, in Example 44.5, that
 
1 1−u+v 1−u−v
fU,V (u, v) = fX,Y , .
2 2 2
Since X and Y are independent and standard Normal, then the joint density of
X and Y is
1 2 1 2
fX,Y (x, y) = √ e−x /2 √ e−y /2 .
2π 2π
Thus, the equation
 
1 1−u+v 1−u−v
fU,V (u, v) = fX,Y ,
2 2 2
reduces to
 2 !  2 !
1 1 1−u+v 1 1−u−v
fU,V (u, v) = √ exp − /2 √ exp − /2
2 2π 2 2π 2
1−u+v

2 1−u−v 2

!
1 1 1 2 2
= √ √ exp − −
2 2π 2π 2 2
1 1 1
√ √ exp −(u − 1)2/4 − v 2/4


=
2 2π 2π

1 2 1 2
=p e−(u−1) /4 p e−v /4
(2π)(2) (2π)(2)

Thus U and V are independent Normal random variables. In fact, E(U ) = 1

and Var(U ) = 2, and also E(V ) = 0 and Var(V ) = 2; we can read these facts
directly from the setup in the last line of the equation above. In other words, we
can tell that U and V are independent because fU,V (u, v) factors into a product
for which the first part has only u’s and the second part has only v’s.
606 Chapter 44. Transformations of One or Two Random Variables

To put the power of the methods above into their proper context, we point
out that without these methods, we still could also have determined

E(U ) = E(1 − X − Y ) = 1

and
Var(U ) = Var(1 − X − Y ) = Var(X) + Var(Y ) = 2
and
E(V ) = E(X − Y ) = 0
and
Var(V ) = Var(X − Y ) = Var(X) + Var(Y ) = 2
by simply using the equations U = 1 − X − Y and V = X − Y and by using
the knowledge that X, Y are independent. On the other hand, we emphasize
that we could not have determined, a priori, that U and V are independent. To
know that U and V are independent, this calculation of the joint density of U
and V was absolutely necessary.

One more quick comment about the domain where the joint density of the pairs
of random variables is defined: Since the joint density of X and Y is defined
throughout the X, Y plane, then also the joint density of U and V is defined
throughout the U, V plane too.

Example 44.8. Consider random variables U, V, X, Y such that

U = XY

and
V = X/Y.
Suppose that the joint density fX,Y (x, y) of X and Y is known. Find the joint
density fU,V (u, v) of U and V .
Notice that the new random variables are not just linear combinations of the
old random variables in this example.
We write
U = g(X, Y ) = XY
and
V = h(X, Y ) = X/Y.
Then we compute four partial derivatives of g and h, with respect to x and y:
∂ ∂ ∂ ∂
g(x, y) = y, g(x, y) = x, h(x, y) = 1/y, h(x, y) = −xy −2
∂x ∂y ∂x ∂y
 44.2. The Distributions of Functions of Two Random Variables 607

So we get
fX,Y (x, y)
fU,V (u, v) =
∂ ∂ ∂ ∂
∂x g(x, y) ∂y h(x, y) − ∂y g(x, y) ∂x h(x, y)
fX,Y (x, y)
=
|(y)(−xy −2 ) − (x)(1/y)|
y
= fX,Y (x, y)
2x
Finally, we need to substitute u’s and v’s instead of x’s and y’s on the right
y √
hand side of the equation. The fraction 2x is just 2v
1
. Also uv = x2 , so uv = x;
and u/v = y 2 , so u/v = y.
p

So we conclude that
1 √ p
fU,V (u, v) = fX,Y ( uv, u/v ).
2v
Again, this is very general, so we will consider a specific case below.

Example 44.9. Consider independent random variables X and Y that are each
Uniform on the interval [0, 1]. Suppose that

U = XY

and
V = X/Y.
Find the joint density fU,V (u, v) of U and V .

As before, we first refer back to the general situation, discussed in Example 44.8
above. This establishes the fact that
1 √ p
fU,V (u, v) = fX,Y ( uv, u/v ).
2v
Since X and Y are independent and Uniform on the square where 0 ≤ X ≤ 1
and 0 ≤ Y ≤ 1, then fX,Y (x, y) = 1 on this square, and thus the equation
1 √ p
fU,V (u, v) = fX,Y ( uv, u/v ).
2v
reduces to just
1
fU,V (u, v) = .
2v
but, again, one question remains: Where is the joint density of U and V defined?
To answer this question, we can step around the four sides of the square
where X and Y are defined, and just find the analogous curves for U and V .
608 Chapter 44. Transformations of One or Two Random Variables

y
Y =1
1

X=0 X=1

x
Y =0 1
V
8

4 X = 1 so
V = 1/U
3

1
Y = 1, so V = U
X = 0 so 1 U
V = 0, U = 0

Figure 44.2: Transforming the square 0 ≤ X ≤ 1, 0 ≤ Y ≤ 1 into the U, V

plane.
 44.3. Exercises 609

Side 1: When Y = 0, then U = XY = 0 and V = X/Y is undefined. So for

fixed Y = c where c is small and near 0, we see that U = Xc, and V = X/c, so
V = U/c2 (e.g., for Y = 1/100, we have V = 10,000 U ). So in the U, V plane,
the lines near Y = 0 are arbitrarily close to the line U = 0 and V > 0 (shown
as a dashed line in Figure 44.2).
Side 2: When X = 1, then U = XY = Y and V = X/Y = 1/Y , so
V = 1/U .
Side 3: When Y = 1, then U = XY = X and V = X/Y = X, so V = U .
Side 4: When X = 0, then U = XY = 0 and V = X/Y = 0.
In the plane for U and V , the four pieces fit together at the places indicated
in Figure 44.2. The side X = 0 gets completely compressed into the origin.
In this example, the joint density of U and V is not Uniform. We can still
perform a check to make sure that we actually have a density. We integrate
fU,V (u, v) = 1/(2v) over all u’s and v’s depicted in Figure 44.2, and we check
that we get 1. To see this, we compute
1 Z 1/u
1 1
Z Z
1
dv du = ln(v)|1/u
v=u du
0 u 2v 2 0
1 1
Z
= (ln(1/u) − ln(u)) du
2 0
Z 1
= − ln(u) du,
0

and using integration by parts, we conclude

Z 1
− ln(u) du = (u − u ln(u))|1u=0 = 1.
0

So this joint density integrates to 1 over the region where it is defined, which
helps verify that everything was calculated correctly. We strongly encourage
the use of such pictures and double-checks of the transformed joint density.

44.3 Exercises
44.3.1 Practice
Exercise 44.1. Bottle volume. A certain type of cylindrical bottle always
has height 14 cm. During the manufacturing process however, the radius of the
bottom is Uniformly distributed between 2.3 cm and 2.7 cm. (Whatever radius
is chosen for the bottom is automatically given to the rest of the cylinder too.
Once the bottom is fixed, the whole can is manufactured that way.) What is
the probability that the bottle has a volume of less than 275 cm3 ?
610 Chapter 44. Transformations of One or Two Random Variables

Exercise 44.2. Gift box. A gift is chosen at random, with the price Uniformly
distributed between $5 and $12. Each gift needs (additionally) to be placed in
a box that costs $2. What is the expected total price of the gift and its box?

Exercise 44.3. Fabric size. A customer at the fabric store buys fabric that is
40 inches wide, i.e., 1.11 yards wide. The length is cut by the employee at the
store. When she is asked to cut 1 yard of fabric, the actual length is Uniformly
distributed between 0.87 and 1.05 yards. What is the probability that the entire
piece of fabric has area 1 square yard or larger?

Exercise 44.4. Chess board. You want to buy a square chess board from
a local artists’ collective. Since each chessboard is uniquely handcrafted and
you didn’t bring your ruler, you are not sure of the exact dimensions. Let X
be the length of the side of one of the boards, and assume that X is Uniformly
distributed between 12 and 18 inches.

a. What is the expected area of the board that you buy?

b. What is the standard deviation of the area of the board?

Exercise 44.5. Summer temperature. Suppose that the temperature X in

summer, given in Kelvin, is Normally distributed with mean 300K and standard
deviation 10K. Let Y be the same temperature, given in degrees Fahrenheit. So

9
Y = (X − 273) + 32.
5
Find the probability that the temperature is higher than 90 degrees Fahrenheit.

Exercise 44.6. Phone bill. Lucas calls his girlfriend Margaret every day on
his cell phone. The time X, in hours, that they talk in a day is Uniformly
distributed between 0 and 2. Lucas pays $2 per hour that he is on the phone,
plus a flat rate of $5 per day. So Y = 2X + 5 is the size of his bill per day.

a. Find the probability that he spends more than $6 on a given day for his
cell phone service.
b. Find the expected amount that he spends on a given day for his cell
phone service.
c. Find the standard deviation of the amount that he spends on a given day
for his cell phone service.

Exercise 44.7. Cookie dough. If the amount of cookie dough, X, used in

a cookie is Uniformly distributed between 1.7 and 2.6 tablespoons, then the
height Y of the cookie is
3
Y = X 1/3 .
10
Find P (Y > 0.4).
 44.3. Exercises 611

Exercise 44.8. Cheeseburger price. Sally owns a burger restaurant. The

cost of a pound of beef is $3.75. Each cheeseburger contains an amount of beef
that is Uniformly distributed between 0.23 and 0.27 pounds. Sally includes a
slice of cheese on each burger, for 35 cents, and the bun itself costs 33 cents.
Burgers are listed on the dollar menu. Assuming that the only costs of making
a burger are the beef, cheese, and bun, what is the probability that the costs
exceed the price of the burger?

44.3.2 Extensions
Exercise 44.9. Let X be a random variable that is Uniformly distributed on
the interval [0, π/2]. Find the expected value of Y = sin X.
Exercise 44.10. Generalize Example 44.4 as follows:
Let X be a Uniform random variable on the interval (0, 1), i.e., X is Uni-
formly distributed with 0 < X < 1. Let Y = X n where n is any positive integer.
Find E(Y ).
Exercise 44.11. Consider an Exponential random variable X with parameter
λ > 0.
Is it always true that, if a and b are positive constants, then Y = aX + b is
an Exponential random variable too?
If your answer is “yes,” then give a justification (e.g., give an argument in
favor).
If your answer is “no,” then give a very concrete counterexample (e.g., for
at least one specific a and b of your choice, show that Y = aX + b is not
Exponential).

44.3.3 Advanced
Exercise 44.12. Consider the scenario in which

U = g(X, Y ) = X 2

and
V = h(X, Y ) = X + Y.
Suppose the joint density fX,Y (x, y) of X and Y is Uniform on the square where
0 ≤ X, Y ≤ 1. In other words,

fX,Y (x, y) = 1 if 0 ≤ x ≤ 1 and 0 ≤ y ≤ 1,

and fX,Y (x, y) = 0 otherwise.

a. Find
∂ ∂ ∂ ∂
g(x, y), g(x, y), h(x, y), h(x, y).
∂x ∂y ∂x ∂y
612 Chapter 44. Transformations of One or Two Random Variables

b. Now write the joint density of U and V :

fX,Y (x, y)
fU,V (u, v) =
∂ ∂ ∂ ∂
∂x g(x, y) ∂y h(x, y) − ∂y g(x, y) ∂x h(x, y)

using the values derived above, and using the fact that fX,Y (x, y) = 1 in the
region where X, Y is defined. Make sure that, in your expression of fU,V (u, v),
you convert all x’s and y’s to u’s and v’s in an appropriate way.

Exercise 44.13. Returning to Exercise 44.12, we still need to know the region
where U and V can occur in the U, V plane, in other words, we need to identify
the region where the density fU,V (u, v) is relevant.

a. Find equations for the curves in the U, V plane that correspond to the
four lines in the X, Y plane around the box 0 ≤ X ≤ 1 and 0 ≤ Y ≤ 1.
b. Draw the region in the U, V plane where the joint density fU,V (u, v) is of
interest.

Exercise 44.14. Check your calculations in Exercises 44.12 and 44.13 by mak-
ing sure that you get the usual “1” when you integrate the joint density fU,V (u, v)
from Exercise 44.12, over all values u and v in the picture from Exercise 44.13.

Exercise 44.15. Are the variables U and V from the Exercises 44.12, 44.13,
and 44.14 dependent or independent? Justify your answer.
Chapter 45

Review Questions for All

Chapters

Exercise 45.1. Rolling dice. A player rolls a huge bag of 200 dice. Approx-
imate the probability that 40 or more 1’s appear.

Exercise 45.2. Dice game. In a certain game, a player wants to roll five dice
and get as many 1’s as possible. Here is the scheme:
Round 1: The player rolls all five of the dice, and notices how many 1’s
appear.
Round 2: The player sets the 1’s aside and only rolls the dice which did not
show a 1 the first time.
Round 3: The player sets the 1’s aside from rounds 1 and 2 and only rolls
the dice which still did not yet show a 1.
After three rounds, the player is tired so she stops. How many 1’s does she
expect to have at this point?
(Hint: Consider one Bernoulli for each of the dice, so E(X1 + · · · + X5 ) =
E(X1 ) + · · · + E(X5 ).)

Exercise 45.3. Let X and Y be independent continuous random variables that

are each Uniformly distributed in the interval [0, 30]. Let Z denote the larger
of X and Y ; in other words, Z = max(X, Y ).

a. Find the cumulative distribution function FZ (a) = P (Z ≤ a) of the

random variable Z.
b. Find the density fZ (z) of Z.
c. Find the expected value E(Z) of Z.

Exercise 45.4. Cold drinks. There are many kinds of drinks in a large cooler
in the cafeteria. Fifteen percent of them are decaffeinated. After class (five days

613
614 Chapter 45. Review Questions for All Chapters

a week), Alice always needs a drink. She is always running to her next class,
and the cafeteria is busy, so never has time to pick a specific one. She likes
caffeine, so she will be “happy” if she gets caffeine four or more times during the
five day week. What is the probability that Alice is happy this week?

Exercise 45.5. Snowfall. The amount of snow during the winter in a certain
town follows a Normal distribution with mean 15 inches and standard deviation
4 inches. What is the probability that the snow is more than 7 inches during
the winter?

Exercise 45.6. Texas Hold ’Em. Suppose one is dealing a hand of Texas
Hold ’Em poker at a standard 9-man table (each of the 9 people receives 2
cards, and there are 5 additional community cards placed face-up in front of
the dealer). What is the probability that, after these 23 cards have been dealt,
all 4 of the aces have been dealt?

Exercise 45.7. Tennis players. There are 50 people in a tennis club. In

their professional lives, 25 of the people are accordion players, 15 are basketball
players, and 10 are carrot farmers. (People tend to register for the tennis club
in groups since they are friends.) The club is selecting 10 people at random to
go on a tennis tour around the world.

a. What is the probability that exactly 5 carrot farmers are picked?

b. What is the expected number of accordion players to be picked?

Exercise 45.8. Waiting for a bus. A passenger is sitting at the mall waiting
for the bus to arrive. The expected waiting time is 30 minutes, i.e., half an
hour. Let X be waiting time (in minutes) until the bus arrives. What is the
probability that the waiting time is more than 20 minutes?

Exercise 45.9. Laundry room. While doing laundry, it seems that each
student spends between 3 to 15 minutes in the laundry room per week, and the
distribution for each student is assumed to be Uniform on this interval.

a. How long is each student expected to be in the laundry room?

b. What is the variance of the time a student spends in the laundry room?
c. Given that a student has already been in the laundry room for 7 minutes,
how long is the student’s expected total stay?
d. During a seven day week, in a dorm with 1000 students who each do
laundry once, how many minutes do we expect the students to spend altogether
in the laundry room (i.e., what is the expected value of the sum of times, with
the sum over all 1000 students).
e. What is the approximate probability that, among the 1000 students,
the average time spent in the laundry room (per student) exceeds 9.1 minutes
during a week? Assume that the students’ times are independent.
 Chapter 45. Review Questions for All Chapters 615

Exercise 45.10. Flight time. The flight time of a plane flight from Denver
to New York City is Normally distributed, with average flight time of 3 hours
and 16 minutes, and standard deviation of 30 minutes. What is the probability
that such a flight takes 4 hours or longer?

Exercise 45.11. Checkmate. When presented with a certain position in a

chess game, and asked whether a checkmate is possible from that position, only
29% of people will understand the method to obtain a checkmate. If people
are surveyed until finding the first person who sees the potential to make a
checkmate from such a position, how many people do we expect to need to
survey, until such an intelligent player is found?

Exercise 45.12. Presents. In an affluent suburban home, a certain child ex-

pects to get 35 presents for Christmas. Give an upper bound on the probability
that they receive 40 or more presents.

Exercise 45.13. Chalk length. The length of a randomly selected piece of

chalk is Uniformly distributed between 0 to 1.5 inches. Suppose that there are
10 pieces of chalk in a classroom. How many pieces of chalk do we expect to be
an inch or longer?

Exercise 45.14. Waiting for a bus. Joe finds that, when he waits for a
bus, his waiting time is Exponential, with an average waiting time of 6 minutes.
Assume that he waits for one bus in the morning and again for one bus in the
evening.

a. What is the probability that he spends 15 minutes or less at the bus stop
altogether during the day?
b. What is his expected waiting time at the bus stop (mornings and evenings
included) during a 20 day period?
c. What is the approximate probability that he spends 200 minutes or less
at the bus stop during a 20 day period (again, mornings and evenings included)?

Exercise 45.15. Artichokes. People shopping at the grocery store are inter-
viewed to see whether they enjoy artichokes. Only 11% of people like artichokes.

a. How many people does the interviewer expect to meet until finding the
25th person who likes artichokes?
b. What is the variance of the number of people he meets, to find this 25th
person who likes artichokes?

Exercise 45.16. Prison escapes. There are 10,000 prison inmates in a certain
state. Independently of each other, and independent of their behavior on previ-
ous days, assume that, on a given day, a prisoner has probability p = 0.000001
of escaping.
616 Chapter 45. Review Questions for All Chapters

a. What is the probability, on a given day, that none of the prisoners

escapes?
b. What is the probability that the state has n consecutive days without
any escapes?
c. Let X be the number of days until the next escape occurs. What distri-
bution does X have?
d. How many days does it take until we are at least 99% sure that at least
one prisoner has escaped in the state?
e. Focus attention on a specific criminal. What is the probability that he
escapes during a 50-year period?
f. If 10,000 prisoners are imprisoned, each for 50 years, how many prisoners
do we expect to escape?
g. There is a very intelligent criminal named Scar-Foot. Instead of having
probability of 0.000001 of escaping each day, he has probability 0.0001 of escap-
ing each day. What is the probability that Scar-Foot is able to escape during a
50-year period?

Exercise 45.17. Typing monkeys. A group of scientists is testing the learn-

ing ability of monkeys. They want to see if monkeys have any language prefer-
ences. They acquire a monkey that has lived its entire life in Moscow, and they
put the monkey in front of a Russian keyboard, which contains 33 characters.

a. They want to see if the monkey can accurately type the entire first line
of a famous poem written by Alexander Pushkin:

The poem has 21 characters in this first line (the scientists do not take
spaces into account at all). What is the probability that the monkey types all
21 characters, in order, correctly? (He just types the keys independently, at
random.)
b. Next, the scientists put the monkey in front of a 26-character English
keyboard and see if he can type the phrase in English. It happens to also take
21 characters when written in English: “A MAGIC MOMENT I REMEMBER”
(again, they do not take spaces into account). What is the probability that the
monkey types all 21 characters, in order, correctly?
c. What is the ratio of these two probabilities, i.e., how much more likely
is the monkey to type the 21 character phrase correctly in English versus in
Russian? Give the ratio of the two probabilities that were calculated in the
previous two parts of this problem.

Exercise 45.18. Cheap boots. A company has a strange policy: It sells

inexpensive boots, but it only ships one boot at a time, and it does not tell the
 Chapter 45. Review Questions for All Chapters 617

buyer whether he will receive a left or right boot (hence, the need to sell the
boots for an inexpensive price, to still attract buyers). How many boots should
a person buy, to be at least 95% sure of ending up with a complete pair?
Exercise 45.19. Dead pixels. A high-quality computer company sells moni-
tors which rarely have a dead pixel. The number of dead pixels per monitor has
a Poisson distribution with average 0.05 per monitor. If the company sells 30
monitors, and the distribution of pixels is independent from monitor to moni-
tor, what is the distribution of the total number of dead pixels found on all 30
monitors altogether?
Exercise 45.20. Candies. In a large bag of 40 Starburst candies, there are 8
orange, 9 yellow, 12 red, and 11 pink. You only like orange and red. If you take
8 from the bag, what is the probability that at least 5 out of the 8 are ones you
like (i.e., there are 5 or more reds and oranges altogether)?
Exercise 45.21. Study time. Let X be the time (in hours) that Stephen
spends studying on one particular day during “dead week” (the nickname for the
week before final exams). Then X has density 41 e−x/4 . Let Y be the number of
hours that Stephen spends studying for all 7 days during “dead week.” Assume
that the time spent studying on distinct days is independent.
a. Find E(Y ).
b. Find Var(Y ).
Exercise 45.22. Walk time. If the walking time to the dining hall is Expo-
nential with mean 8 minutes, what is the probability that it takes 15 or more
minutes to get there?
Exercise 45.23. Solitaire. A student wins at solitaire about 17% of the time.
How many games does the student expect to play until winning her 3rd game?
Exercise 45.24. Exam scores. On a 50-question Geography exam, the av-
erage score is 25.5 out of 50. The standard deviation of the score is 8. Find
a bound on the probability that a randomly selected student’s score is greater
than 42 or less than 9.
Exercise 45.25. Couples. Consider n pairs of husbands and wives, sitting
randomly in a row of 2n chairs. What is the probability that each person is
sitting beside her/his spouse, and also no two women are adjacent, and no two
men are adjacent (i.e., the sexes are alternating)?
Exercise 45.26. Flower arranging. Each student in a flower arranging class
gets to take home a house plant. Each student has a variety of plants to choose
from, and the selections are made independently (in particular, there is an
ample number of each kind of plant). There is a 20% chance that, when a
student selects her/his plant, the choice will be a peperomia. If there are 60
students in the class, what is the probability that exactly 14 of the students will
take home a peperomia?
618 Chapter 45. Review Questions for All Chapters

Exercise 45.27. Garden mole. My garden is a square plot of land that

measures 10 feet by 10 feet, and it has a mole living somewhere inside. The
location of the mole is Uniformly distributed in the garden. I have a small
tomato area that is a 2 feet by 2 feet square region. What is the probability
that the mole is found in the tomato patch?

Exercise 45.28. Evil scientist. An evil scientist has devised an experiment

that takes one hour to run to completion. It has a 1/12 chance of causing the
world to explode. Otherwise, it fails and the evil scientist immediately resets
the experiment and tries again! (Her attempts are independent and each have
probability 1/12 of succeeding.) Knowing that a nearby hero will find the evil
scientist and foil her plan after 8 hours have passed, what is the probability
that the evil scientist will succeed before the hero catches her, i.e., that the evil
scientist will have a success within her first 8 attempts?

Exercise 45.29. Exam questions. On a probability exam, each of the n

questions are created independently of the others. Each question has a ten
percent chance of involving a Geometric random variable.

a. How many questions on the exam are expected to have a Geometric

random variable involved?
b. What is the variance of the number of questions that have a Geometric
random variable?

Exercise 45.30. Take-home exam. Consider a student who starts a take-

home exam at midnight and expects to take 8 hours to finish the exam. If the
student has 34 hours remaining until the exam must be completed, give a bound
on the probability that the student is able to complete the exam within the 34
hours until it is due for submission.

Exercise 45.31. Pillows. A girl buys pillows to decorate her apartment living
room. There are square and circular pillows in stock at the store where she is
shopping. There are a total of 400 pillows available:

• 50 are square, with down feathers;

• 250 are square, with cotton fluff;

• 20 are circular, with down feathers;

• 80 are circular, with cotton fluff.

What is the probability that, if she grabs a square pillow at random, it is stuffed
with cotton fluff?

Exercise 45.32. Auditions. A string quartet is being assembled. Each player

that auditions is only likely to be successful at the audition with probability
27%. How many musicians are expected to audition until 4 of them are selected
for the quartet?
 Chapter 45. Review Questions for All Chapters 619

Exercise 45.33. Basketball shooting. A basketball player has been devel-

oping his skill during the season, and he now scores 75% of his free throws
successfully. If he shoots 200 free throws this season, how many points should
he be expected to contribute? (Each free throw is worth 1 point.)

Exercise 45.34. Online dating. Approximately 1 in 4 relationships begin

online. Suppose that we interview couples until we find 10 couples who met
online.

a. What is the expected number of couples we will need to interview?

b. What is the variance of the number of couples we will need to interview?

Exercise 45.35. Biased coin. Consider a sequence of independent flips of

a biased coin. The coin is a Head or Tail with probabilities p or q := 1 − p,
respectively.
Let X be the smallest j such that the jth and (j + 1)st flips are both Heads.
Let Y be the smallest j such that the jth flip is a Head and the (j + 1)st
flip is a Tail.
Let Z be the smallest j such that the jth flip is a Tail and the (j + 1)st flip
is a Head.

a. Find P (X < Y ).
b. Find P (X < Z).

Exercise 45.36. Deck of cards. A deck is thoroughly shuffled, and 5 cards

are chosen, without replacement. What is the probability that the selection of
5 cards contains at least one card from each of the 4 suits?

Exercise 45.37. Let X, Y be independent exponential(1) random variables.

a. Compute the conditional probability that X < ln 2, given X − Y ≥ 0. In

other words, find
P ( X < ln 2 | X − Y ≥ 0) .
b. Compute the conditional probability that X < ln 2, but now given X − Y is
exactly 0. In other words, find

P ( X < ln 2 | X − Y = 0) .
this page left intentionally blank
Answers to Exercises

Chapter 1
1.1 Answers will vary. E.g., (1) (0.25, 0.5);
(2) {(x, y) | 0.2 < x < 0.5, 0.36 < y < 0.42}; (3) {(x, y) | x2 + y 2 ≤ 1};
1.3 Answers will vary. E.g., (1) Chris grabs lemon-lime, lemon-lime, orange;
(2) {(x1 , . . . , x4 , orange) | each of x1 , . . . , x4 is lemon-lime or fruit punch};
(3) {(x1 , . . . , xj , orange) | 0 ≤ j ≤ 12; the xj ’s are lemon-lime or fruit punch,
with ≤ 6 of each};
1.5 {(x1 , x2 , . . . , x75 ) | x1 + x2 + · · · + x75 ≤ 400};
1.7 a. 6; S = {(rrbb), (rbrb), (rbbr), (bbrr), (brbr), (brrb)}; b. 26 = 64;
1.9 a. {(x1 , x2 , x3 ) | xj ∈ R>0 }; b. {(x1 , y1 , x2 , y2 , x3 , y3 ) | xj , yk ∈ R>0 };
c. {(x1 , y1 , x2 , y2 , x3 , y3 ) | xj , yk ∈ R>0 ; x1 < x2 < x3 ; y1 > y2 > y3 };
1.11 S = {(x, y) | x, y ∈ R≥0 ; x + y ≤ 2};
1.13 28;
j 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
1.15
|Aj | 1 3 6 10 15 21 25 27 27 25 21 15 10 6 3 1
Chapter 2
2.1 a. 330/27333 = 0.0121; b. 9153/27333 = 0.3349; c. No. The table gives
a partition of the songs, i.e., each song is only included in one genre and is
therefore only counted once. The genres are disjoint subsets of the sample
space; d. 18180/27333 = 0.6651;
2.3 0.83; 2.5 a. 13/30; b. 17/30; c. Answers will vary. E.g., Partition the
shoes into those that are black and brown and those that are another color;
2.7 a. 10.222, 24.799, 5.000, 63.1, 30 inches of rain; b. {x | x ≥ 0}; c. Answers
will vary. E.g., partition the sample space into those values which are less than
the average annual rainfall and those which are greater than or equal to the
average; d. Each possible outcome is in exactly one part of the partition.
2.9 a. 5/12; b. 3/4; c. 1/2; 2.11 (1/101)7 ;
2.13 a. 0.01; b. 0.1; c. 0.11; d. 0.89; 2.15 a. 2/3; b. 1/6;
2.17 P {(P )} = 1/3; P {(G, P ), (Y, P )} = 1/3; P {(G, P ), (G, Y, P )} = 1/3;
P {(Y, G, P ), (G, Y, P )} = 1/3; P {(P ), (Y, P )} = 1/2;
2.19 920; 2.21 a. 27 ; b. 21; c. 7j ; d. P (Aj ) = 7j (1/2)7 ; 2.23 0.47;

2.27 a. If A and B are disjoint, then equality is true. Otherwise, A and B will
have an intersection, so P (A ∪ B) = P (A) + P (B) − P (A ∩ B) < P (A) + P (B);

621
622 Answers to Exercises

b. Yes. P (A∪B∪C) = P (A)+P (B\A)+P (C \(A∪B)) ≤ P (A)+P (B)+P (C).

2.29 15/216 = 5/72 = 0.0694; 2.31 P (A0 ) = 1/6, P (A1 ) = 2/3, P (A2 ) = 1/6;
2.33 P (Bk ) = (k/6)3
Chapter 3
3.1 0.622; 3.3 58/125; 3.5 a. 63/64; b. 27/64; 3.7 Yes. 3.9 0.7273 or 8/11;
3.11 Yes. E.g., P (A) = 0.7, P (B) = 0.4, P (A ∩ B) = 0.28, P (A ∪ B) = 0.82
Chapter 4
4.1 P (B | C) = p2 , P (C | B) = p2 , P (A | B) = 0,
P (B | A) = 0, P (A | C) = (1 − p)3 , P (C | A) = p5 ;
4.3 P (B | A) = 330/9153 = 110/3051 = 0.0361,
P (J | A) = 537/9153 = 179/3051 = 0.0587,
P (R | A) = 8286/9153 = 2762/3051 = 0.9053;
4.5 a. 5/9;

b. 2/5; c. 2/4; 4.7 47/62; 4.9 2/5; 4.11 2/3; 4.13 2/3;
4.15 nj /(2n − 1)
Chapter 5
5.1 0.1633; 5.3 0.6; 5.5 0.765; 5.7 a. 0.5326; b. 0.7176; c. 0.6372;
5.9 a. 3/5; b. 2/5; 5.11 a. 2/15; b. 0.4333; 5.13 1/3; 5.15 5/12;
5.17 a. 0.71; b. 0.29; c. 48/71; d. 1; e. 25/29; f. 25/39
Chapter 6
6.1 0.03 < P (B) < 0.27 < P (D) ≤ 1; 6.3 0.19; 6.5 a.1/5; b. 1/2; 6.7 0.48
Chapter 7
7.1 discrete; X ∈ {0, 1, 2, 3, 4, 5};
7.3 X is continuous; X ∈ [0, 4];
Y is continuous; Y ∈ [0, 100];
Z is discrete; Z ∈ Z≥0 ;
7.5 X, Y, Z are all discrete and take values in Z≥0 ;
7.7 X, Y are both discrete and have possible values {0, 1, . . . , 10};
7.9 X, Y are both continuous and have possible values (0, ∞);
7.11 X is discrete; X ∈ N;
Y is continuous; Y ∈ (0, ∞);
7.13 X is discrete; X ∈ {1, 2, . . . , 10};
Y is continuous; for instance, Y ∈ (0, 0.52) if the half-gallon of milk could hold
up to 0.52 gallons;
Z is discrete; Z ∈ 1, 2, 3, 4;
7.15 X is discrete; Answers will vary depending on the age of students in the
course, e.g., X ∈ {3000, . . . , 44350};
Y is discrete; Y ∈ N;
7.17 a. 10/17; b. 80/221; c. 11/221
 Answers to Exercises 623

Chapter 8

8.1 a. S = {0, 1, 2, 3, 4}; b. P (X = x) = 4

(7/10)x (3/10)4−x for 0 ≤ x ≤ 4;


x



0 for x < 0,
0.0081 for 0 ≤ x < 1,





0.0837 for 1 ≤ x < 2,

c. FX (x) = ;


0.3483 for 2 ≤ x < 3,
0.7599 for 3 ≤ x < 4,





for x ≥ 4

1

x −1 1 2 3
8.3 Mass: ;
P (X = x) 0.6651 0.0121 0.0196 0.3032


 0 for x < −1,
0.6651 for −1 < x < 1,




CDF: FX (x) = 0.6772 for 1 ≤ x < 2, ;
0.6968 for 2 ≤ x < 3,





for x ≥ 3

1


8.5 a. pX (x) = (0.9)x−1 (0.1);

pX (x)

0.1
b.
0.05
x
1 2 3 4 5 6
Px
c. FX (x) = 0.1 + (0.1)(0.9) + · · · + (0.1)(0.9x−1 ) = j=1 (0.1)(0.9
j−1 );

FX (x)
1
0.8
d. 0.6
0.4
0.2
x
1 2 3 4 5 6 7 8 9 10 11 12 13 14
8.7 a. X = {0, 1, 2, 3, 4, 5};
x 0 1 2 3 4 5
b. 2109 27417 9139 2717 143 33 or equivalently
pX (x) 9520 66640 33320 33320 13328 66640

x 0 1 2 3 4 5
;
pX (x) 0.2215 0.4114 0.2743 0.0815 0.0107 0.0005
624 Answers to Exercises


0 if x < 0;

0.2215 if 0 ≤ x < 1;

pX (x)



if

0.6330 1 ≤ x < 2;


0.4 
c. d. FX (x) = 0.9072 if 2 ≤ x < 3;
0.2
if

0.9888 3 ≤ x < 4;


x 

if

1 2 3 4 5 0.9995 4 ≤ x < 5;




if

1 5 ≤ x;


FX (x)
1
0.8
e. 0.6
0.4
0.2
x
1 2 3 4 5
pX (x)
0.25
8.9 a.
x
1 2 3 4
(
bxc/4 for 0 ≤ x ≤ 4
b. FX (x) =
1 for x > 4
FX (x)
1
0.75
c.
0.5
0.25
x
1 2 3 4
8.11 pX (−10) = 0.1; pX (−5) = 0.7; pX (0) = 0.2;
FX (x)
1
0.75
8.13
0.5
0.25
x
1 2 3 4
 Answers to Exercises 625

pX (x)
0.3
8.15 a. 0.2
0.1
x
1 2 3 4 5
b. FX (x) = xj=0 e−2 2j/j! for x ∈ N;
P

FX (X)
1
0.75
c. 0.5
0.25
x
1 2 3 4 5 6
8.17 a. X = {0, 1, 2, 3, 4, 5, 6}; b. pX (x) = 6
(1/13)x (12/13)6−x ;


x
pX (x)
1
0.75
c. 0.5
0.25
x
1 2 3 4 5 6
P6
d. FX (x) = 6 x 6−x


x=0 x (1/13) (12/13)
FX (x)
1
0.75
e.
0.5
0.25
x
1 2 3 4 5 6
FX (x)
1
0.75
8.19
0.5
0.25
x
1 2
626 Answers to Exercises

Chapter 9
9.1 Yes, since pX,Y (1, 1) = 2/10 = (4/10)(5/10) = pX (1)pY (1);
9.3 Yes. This makes intuitive sense since the number of flips to the first head
has no bearing on the number of flips after the first head to the second head.
Indeed, we have pX,Y (x, y) = (1/2)x+y = (1/2)x (1/2)y = pX (x)pY (y).
9.5 pX,Y (0, 3) = 0.342375, pX,Y (1, 2) = 0.464375, pX,Y (2, 1) = 0.174125,
pX,Y (3, 0) = 0.019125, pX,Y (x, y) = 0 otherwise;
9.7 Dependent;
Y \X 1 2 3 4 5 6
9.9 0 0 0 0 3/9 2/9 4/9
1 9/19 6/19 4/19 0 0 0
Chapter 10
10.1 0.76;
10.3 Let X be the number of songs until (and including) the folk song. Then
P (X = j) = 1/868 for 1 ≤ j ≤ 868. So

E(X) = 1P (X = 1) + 2P (X = 2) + · · · + 868P (X = 868)

= (1 + 2 + · · · + 868)(1/868)
= 869/2
= 434.5 songs.

10.5 $45,200; 10.7 5.5; 10.9 1.5; 10.11 14.2857; 10.13 a. 6; b. 6;

10.15 a. 6/13; b. 2/13; 10.17 Claw; 10.19 7
Chapter 11
11.1 0.3 + 0.46 = 0.76; 11.3 150.5; 11.5 7.6923; 11.7 6; 11.9 3.5714;
11.11 a. P (A
Pj∞) = (5/6)
j−1 = E(X )
j
so E(X) = j=1 (5/6)j−1 = 1−5/6 1
= 6;
P∞
b. E(Y ) = j=1 (5/6) j−1 = 6;
11.13 15/8; 11.15 3.5; 11.17 a. 1391; b. 2781; 11.19 9 other students
Chapter 12
12.1 Var(X) = p(1 − p); 12.3 0.96; 12.5 7499.9167;
12.7 a. 15.36; b. 396.12; c. 160.1904; d. 334.68;
12.9 a. 2781; b. 15,465,141; c. 7,731,180;
12.11 −$38; 12.13 0.5761; 12.15 55/64; 12.17 20/3;
12.19 a. 6/13; b. 108/169; c. 72/169; d. $509.23; e. 170.4142;
12.21 a. 0.4267; b. 0.2104; 12.23 19/6; 12.25 35/6;
12.27 a. −3.34; b. undefined; c. −0.5988; d. 3.46; e. 2.1644; f. 1.3484
 Answers to Exercises 627

Chapter 13
pX (x)

0.4
13.1 a. 0.3
0.2
0.1
x
1 2 3 4 5
b. 2.32; c. 1.3029;



 0 for x<1 FX (x)
for



 0.4 1≤x<2 1
0.8

for

0.6 2≤x<3
d. FX (x) = e. 0.6


 0.7 for 3≤x<4 0.4


 0.98 for 4≤x<5 0.2

 x
for

1 x≥5 1 2 3 4 5 6
13.3 a. pX (x) = 4
(13/18)x (5/18)4−x for x = 0, 1, 2, 3, 4;


x
pX (x)
0.4
b. c. 2.8889; d. 0.8958;
0.2
x
1 2 3 4
e. FX (x) = xj=0 4j (13/18)j (5/18)4−j for integers 0 ≤ x ≤ 4;
P

FX (x)
1
0.8
f. 0.6
0.4
0.2
x
1 2 3 4
13.5 a. pX (x) = 5−x17

3
20

x / 5 for x = 0, 1, 2, 3;

0
 for x < 0
0.3991 for 0 ≤ x < 1




b. FX (x) = 0.8596 for 1 ≤ x < 2 ;
0.9912 for 2 ≤ x < 3





for x ≥ 3

1

c. 0.75; d. 0.5033;
x -3 1 2.5 7
13.7 a. b. 0.8; c. 17/20; d. 80/97; e. 90.43;
pX (x) 0.03 0.17 0.56 0.24
13.9 a. True. This happens when P (X ≤ x) = 1, or in other words, when all
the mass of the function lies at or below the given value.;
628 Answers to Exercises

b. True. See Remark 8.7; c. False. See Remark 8.5;

d. True. For example, if the sample space contains a discrete group of x’s
(x1 . . . xn ) then for any value xj > xn , FX (xj ) = 1. In such a case there are
infinitely many values > xn .;
e. False. Whenever there is a discrete PMF, the CDF will contain a jump
wherever P (X = x) > 0.;
f. True. This happens whenever P (X ≤ x) = 0.;
g. True. If there are infinitely many numbers x such that P (X ≤ x) = 0
this will be the case. Consider a discrete PMF with values at x1 . . . xn . Then
P (X ≤ x) = 0 for any x < x1 .
h. True. FX (x) > 0 if P (X ≤ x) > 0;
13.11 $4.50;
x 30 45 60
13.13 a. 0.5; b. c. Yes. 0.2 + 0.5 + 0.3 = 1.;
pX (x) 0.2 0.5 0.3
d. P (25 < X < 50) = 0.7; e. 0; f. 1; g. 0; h. 0.3; i. 46.5; j. 10.5;
13.15 a. $300; b. $2227; c. $3600; d. $7715;
13.17 a. E(X 2 ) = 12.8673; b. E(−12X + 10) = −27.44; c. 451.1376;
x 50 100 150 200 250 270 290
13.19 a.
pX (x) 7/28 6/28 5/28 4/28 3/28 2/28 1/28
b.$145.7143; c. $78.5779;
x 0 1 2 3 4
13.21
pX (x) 0.92 0.034 0.020 0.0145 0.0113
13.23 Average price = $78.75; Standard deviation = $21;
13.25 a. 0.0157; b. No. Passengers often come in groups, so either the whole
group would be there or the whole group would miss the flight.;
c. E(X) = $15, 592.50; σX = $294.2257; d. E(X) = $192.50; σX = $32.69
Chapter 14
14.1 a. A “success” occurs if the call is from a family member; p = 0.125;
b. A “failure” occurs if a non-family member calls; q = 0.875; Note: the answers
to a and b may be switched, depending on how the student defines the problem.
c. There is a single trial with a success/failure outcome, p = 0.125;
d. Random variable X indicates whether the call is from a family member; it
can be 0 or 1;
e. p = 0.125; f. p = 0.125; g. 0.125; h. 0.3307;
14.3 a. A “success” is when a student has done the assignment; p = 0.02.; b. A
“failure” is when the student has not done the assignment; q = 0.98; c. There
is a single trial with a success/failure outcome, p = 0.02; d. The X indicates
whether the student Hui chooses has done the assignment; it can be 0 or 1; e.
p = 0.02; f. 0.02;
14.5 a. 7/55; b. 15/55 = 3/11; c. 3/7; 14.7 a. 19.5; b. 6.825;
14.9 a. E(X) = 0.4; Var(X) = 0.24; b. E(Y ) = 0.4; c. $12.80
 Answers to Exercises 629

Chapter 15
15.1 a. A “success” is when a Skittle is purple; p = 0.2;
b. a “failure” is when a Skittle is not purple, p = 0.8;
c. Random variable X gives the total number of purple Skittles found. It can
take on any integer value from 0 to 25;
d. There are n = 25 independent trials (because it was a random sample), with
equal probability of success(p = 0.2) on each trial, and we are counting up the
number of successes (# of purple Skittles);
e. 0.196; f. 0.9726; g. 5; h. 2; i. 32.5 cents; j. 3;
15.3 a. Random variable X gives the number of babies who are not born by
C-section. It can take an integer value from 0 to 9;
b. There are n = 9 independent trials (because it was a random sample), with
equal probability of success(p = 0.8), and we are counting up the number of
successes (# of babies not born by C-section);
c. 0.1762; d. 0.2587; e. 2; f. 7.2; g. 1.44;
FX (x)
0.4
h. 0.2
x
1 2 3 4 5 6 7 8 9
FX (x)
1
0.8
i. 0.6
0.4
0.2
x
1 2 3 4 5 6 7 8 9
15.5 a. 0.8100 ; b. 20; c. Yes. The expected profit is $650; d. 14 boxes;
15.7 9 times 15.9 a. 0.2503; b. 0.7759; c. 0.9274; d. 0.7625;
15.11 a. 0.992; b. 0.936; c. 0.9285; d. 0.9995; e. 0.7273;
15.13 3.2726 × 10−11 ; 15.15 a. −4; b. 194.4; c. 0.01229;
15.17 a. 0.161; b. 1.4; c. 1.302;
15.19 a. 1.75; b. No. Since there is no replacement, they are not independent
Bernoulli trials;
15.21 School B, because E(XA ) = 222.8 < 243.81 = E(XB );
15.23 a. 270; b. 50 50 j 50−j = 0.0007;
P

j=45 j (7/10) (3/10)
15.25 limn−>∞ P (Xn > 0) = 1. Yes because as n grows very large, the proba-
bility (0.4n ) that no student would eat cereal for breakfast approaches 0.
Chapter 16
16.1 a. The variable X is the number of Skittles that come down the line until
the inspector finds the first one that is striped; here, X can be 1,2,. . . .
630 Answers to Exercises

b. The trials are independent, each with equal probability of success (p = 0.05).
The trials are performed until the first success of finding a striped Skittle.

c. 0.0315; d. 0.4312; e. 0.6634; f. 0.6983; g. 20; h. 19.4936;

16.3 a. The variable X is the number of parents you have to ask until you find
one who did not have a baby by C-section; the X can be 1,2,. . . .

b. The trials are independent, each with equal probability of success (p = 0.8).
The trials are performed until the first success of finding parent of a baby not
born by C-section.

c. 0.0016; d. 0.008; e. 0.0079; f. 1.25; g. 0.3125;

pX (x)
1
0.8
h. 0.6
0.4
0.2
x
1 2 3 4 5 6 7 8 9

FX (x)
1
0.8
i. 0.6
0.4
0.2
x
1 2 3 4 5 6 7 8 9

16.5 a. 2.3256; b. 1.7558;

pX (x)
1
0.8
c. 0.6
0.4
0.2
x
1 2 3 4 5 6 7 8 9
 Answers to Exercises 631

FX (x)
1
0.8
d. 0.6
0.4
0.2
x
1 2 3 4 5 6 7 8 9
16.7 a. 13 times; b. 0.7865; c. 0.2367; Binomial; n = 6, p = 0.7865;
16.9 a. 3.5714; b. 17.8571 minutes; c. 0.0374;
16.11 a. 0.065498; b. 0.19988; c. 5;
16.13 a. 2.5 games; b. 3.75 games; c. 0.216;
16.15 a. E(X) = 14.2857; b. Var(X) = 189.7959; c. P (Y = y) = q x−3 p;
16.17 P (X > n) = (7/8)n ;
16.19
There is only one possible way for X = x if X Geometric, but there
are nx ways for X to equal x if X Binomial. Geometric variables can have
an infinite number of values, whereas Binomials can only take values between
0 and n. So if X Geometric then P (XP > x|X >
y) = q x /q y = q x−y , but if

X Binomial then P (X > x|X > y) = nj=x+1 nj pj q n−j / nj=y+1 nj pj q n−j ,

P

which clearly does not have the memoryless property.

Chapter 17

17.1 a. The variable X is the number of Skittles until the 3rd striped Skittle
is found; this X can take values 3, 4, 5, . . . .
b. We must count the number of trials until we get the 3rd success, which is
the 3rd striped Skittle found. Each Skittle is independent and each trial has
the same probability of success (p = 0.05).
c. 0.0089; d. 0.0012; e. 0.9978; f. 60; g. 33.7639;
17.3 a. The variable X is the number of parents you have to ask until you find
the 7th whose baby was not born by C-section. The X can be 7,8,. . . .
b. We must count the number of trials until we get the rth success (r = 7),
each set of parents is independent, and each trial has the same probability of
success (p = 0.8).
c. 0.1409; d. 0.2618; e. 8.75; f. 2.1875;
pX (x)

0.4
g.
0.2

x
6 7 8 9 10 11 12 13 14
632 Answers to Exercises

FX (x)
1
0.8
h. 0.6
0.4
0.2
x
1 2 3 4 5 6 7 8 9 10 11 12 13 14
17.5 0.52765; 17.7 E(X) = 66.6667; σX = 19.4365;
17.9 a. 0.3907; b. 60 minutes;
17.11 a. 0.6405; b. 4/27; Geometric; The number of tries to the next success
does not depend on how many tries it took to get the previous successes.

Chapter 18

18.1 a. The X is a nonnegative integer valued random variable. We are given a

rate (marriages per day) and we are asked to count up the number of marriages
in a particular time frame (5 minutes). The average rate is λ = 35/36 = 0.9722
marriages/5 minutes.
b. e−280 280300 /300! = 0.0115; c. 35/3; d. 3.4157; e. 0.3413;
f. 0.0451; Binomial(24,0.3413);
18.3 a. 0.064998 for a child; 6.3524 × 10−23 for an adult;
b. There is a an average rate and we must count the number of times an
adult/child laughs in a certain period of time (hour).;
c. 0.2094; d. 0.8823;
pX (x)
0.6
e. 0.4
0.2
x
1 2 3 4 5 6
pX (x)
0.2
f. 0.1
x
5 10 15 20 25
FX (x)
1
0.8
g. 0.6
0.4
0.2
x
1 2 3 4 5 6 7
 Answers to Exercises 633

FX (x)
1
0.8
h. 0.6
0.4
0.2
x
5 10 15 20 25
18.5 a. Binomial(n = 1,000,000, p = 1/729,000);
b. P (X = 3) = 1,000,000 3 (728,999/729,000)999,997 ;


3 (1/729,000)
c. 1.3717; d. Poisson(λ = 1000000/729000 = 1.3717); e. P (X = 3) ≈ 0.1091;
18.7 P (X ≥ 1) ≈ 0.86466;
pX (x)
0.2
18.9 a. 0.1
x
1 2 3 4 5 6 7 8 9
FX (x)

0.8

b. 0.6

0.4

0.2

x
1 2 3 4 5 6 7 8 9 10
18.11 a. 0.4562; b. 0.2081; c. 0.5595;
18.13 a. 0.1804; b. 0.3233; c. 0.5581; d. $5; e. $3.54;
18.15 a. 0.1606; b. 0.000003;
c. 0.0362; the 35 defects can occur any time in the 168 hours (there does not
have to be a certain amount each day).
18.17 a. 0.1126; b. 8.8811; Geometric; 18.19 a. 4; b. 0.1563;
18.21 a. 42; b. 0.04388;
18.23 a. P (X = 8) = 60,000 (1/10,000)8 (9999/10,000)59,992 ;


8
b. P (X = 8) ≈ e−6 68/8!; c. 0.1033; 18.25 Yes
Chapter 19
19.1 a. A “success” is a bag of cookies. A “failure” is a bag of potato chips or
pretzels.
b. The random variable X is the number of bags of cookies you get, so X can
be 0,1,2 or 3.
634 Answers to Exercises

c. This is Hypergeometric because we are sampling without replacement and

looking for the number of “desirable” items from a larger population. Parame-
ters: M = 5, N = 18, n = 3.
d. 35/204 = 0.1716; e. 169/204 = 0.8284; f. 5/6; g. 0.7287; i. 7/6; j. 1;
19.3 a. A “success” is finding green Skittles as you eat. A “failure” is not finding
green Skittles.
b. The random variable X is the number of green Skittles you eat, so X can be
0, 1, . . . , 9.
c. We are sampling n = 10 items without replacement from a larger population
of N = 45 items, looking for “desirable” items, of which there are M = 9.
d. 0.3133; e. 3.6024 × 10−8 ; f. 2;
19.5 a. Here, X is the number of DVDs that are truly “Bridget Jones,” so X
can be 0,1,. . . ,20.
b. We are sampling n = 20 items without replacement from a larger population
of N = 50,000, looking for “desirable” items, of which there are M = 40,000.
c. 16; d. P (X = 18) = 40,000

10,000
50,000

18 2 / 20 ; e. Binomial(20, 0.8); f. 0.1369;
19.7 a. 120/1771 = 0.0678; b. 150/1771 = 0.0847; c. 135/161 = 0.8385;
d. 2/191 = 0.0105; e. 1.3043; f. 0.8187; g. 0.004591; Binomial;
19.9 10/19 = 0.5263
19.11 a. 116/329 = 0.3526; b. 2.4; 19.13 P (X = 0) = 0.0725; E(X) = 2;
19.15 a. 20/39 = 0.5128; b. 28/55 = 0.5091;
c. No. Part b is based upon a given outcome on the first night, so the probability
would not be the same. A conditional probability is not the same as a non-
conditional probability, unless the events are independent. But here the events
are dependent.
19.17 a. 56/19 = 2.9474; b. 1232/1083 = 1.13758;
c. 770/12597 = 0.0611; d. 770/849 = 0.9069;
19.19 a. 0.01008; b. 13/4

Chapter 20

20.1 a. All 5 outcomes are equally likely (p = 1/5), and you are only selecting
one at random.
b. 1/5;
c. The random variable X is the number of the color that you pick, so X can
be 1, 2, 3, 4, or 5.
d. 4/5;
pX (x)
0.2
e. 0.1
x
1 2 3 4 5
 Answers to Exercises 635

FX (x)
1
0.8

f. 0.6
0.4
0.2
x
1 2 3 4 5 6
20.3 Since Y has the same mass as X + 1, it follows that E(Y ) = E(X) + 1,
and Var(Y ) = Var(X).
Chapter 21
21.1 Binomial; n = 50, counting up total number of broken cones, each with
p = 0.12;
21.3 Poisson; there is a rate λ = 2/minute and a set interval (1 hour);
21.5 Hypergeometric; sampling without replacement, population and sample
sizes, number of successes in population;
21.7 Discrete Uniform; equal probability of success for each outcome;
21.9 Binomial (or Poisson approximation); n = 10000, counting up total num-
ber of undercooked cones, each with p = 0.00005; since n is large and p is small,
the approximation with λ = 0.5 is appropriate.
21.11 a. 0.2; b. Binomial(10, 0.2); 0.6242; c. Geometric(0.2); 0.1024;
d. Negative Binomial(0.2, 3); 15; e. Binomial (7, 0.6242); 0.1698;
f. Discrete Uniform(5); 1/5;
21.13 a. Hypergeometric (M = 10, N = 15, n = 3); 24/91 = 0.2637;
b. Binomial(3, 2/3); 8/27 = 0.2963; c. Geometric(2/3); 2/27 = 0.0741;
d. Bernoulli(2/3); 2/3; e. Bernoulli(2/3); 2/3;
21.15 a. Geometric(0.08); 0.0164; b. Negative Binomial(0.08, 4); 50;
c. Binomial(150, 0.08); 12; d. Bernoulli(0.08); 0.92;
21.17 a. Poisson(30); 0.1755; b. Geometric(0.1755); 5.699;
c. Negative Binomial(0.1755, 4); 0.0504;
21.19 a. Hypergeometric(M = 10, N = 45, n = 5); 0.7343;
b. Binomial approximation to the Hypergeometric(5, 1/36); 0.1314;
c. Geometric(0.1314); 7.6113;
21.21 a. Discrete Uniform(7); 1/7; b. Geometric(1/7); 7;
c. Binomial(20,0.25); 0.1897;
d. Poisson approximation to the Binomial (0.365); 0.0005;
21.23 Hypergeometric(M = 5, N = 18, n = 3); a. E(X) = 15/18; σX = 0.7287;
b. 0.1716; c. Binomial approximation to the Hypergeometric(3, 5/18); 0.1886
636 Answers to Exercises

Chapter 22
22.1 a. 4060; b. 24,360;
22.3 a. 840/34650 = 4/165 = 0.0242; b. 210/34650 = 1/165 = 0.0061;
c. 8/55; d. 6/11;
22.5 a. 0.3087; b. 0.03087;
22.7 a. 29/52 = 0.5577; b. 3/52 = 0.0577 c. 20/52 = 0.3846;
22.9 0.00003; 22.11 1/11; 22.13 0.0738;
22.15 a. 57/616 = 0.0925; b. 39/1496 = 0.0261;
c. No. These are only the cases where all boys or all girls are chosen. There are
several cases where some boys and some girls are chosen.
22.17 0.0002; 22.19 a. 64/425 = 0.1506; b. 3/4;
22.21 pX (0) = 44/120 = 11/30; pX (1) = 45/120 = 3/8; pX (2) = 20/120 = 1/6;
pX (3) = 10/120 = 1/12; pX (4) = 0; pX (5) = 1/120
22.23 7.037; 22.25 a. 0.6644; b. 0.0055; 22.27 17/45;
22.29 a. 1/64; b. 0.0757; c. 0.0002; d. 0.0023;
22.31 a. 35; b. 15/35 = 3/7; c. 5/35 = 1/7; d. 210;
22.33 a. 3,628,800 ways to arrange seating;
b. 181,440 ways to arrange seating with the family sitting together;
22.35 1/19; 22.37 a. 0.0319; b. 0.049; c. 0.1976;
22.39 The probability of not sitting next to each other is n−3n−1 for n ≥ 3, or 0
for n = 2.
22.41 2/15; 22.43 2/3; 22.45 n!n!/(2n − 1)! 22.47 0.1
Chapter 24
24.1 a. continuous; b. discrete; c. continuous; d. discrete;
X is discrete X is continuous
P (X ≥ 2) ??? 0.3
24.3 P (X < 2) ??? 0.7
P (X ≤ 2) 0.7 0.7
P (X = 2) ??? 0
24.5 a. k = 30; b. 53/512 = 0.1035;
for x < 2;

0

24.7 a. 4/5; b. FX (x) = (x − 2)/5 for 2 ≤ x ≤ 7;
for x > 7;

1

c. 3/5; d. 0; e. 3/10; f. 0;
1
fX (x)
g. 0.5
x
2 4 6 8 10
1
FX (x)
h. 0.5
x
2 4 6 8
 Answers to Exercises 637

for x ≤ 4;

0

24.9 a. FX (x) = (x − 4)/6 for 4 < x < 10;
for x ≥ 10;

1


1
FX (x)
b. 0.5
x
5 10
for y < 0

0

24.11 7/8; 24.13 FY (y) = −3y + 4y for 0 ≤ y ≤ 1
4 3

for y > 1

1

24.15 a. 0.06699; b. 1/2; c. 3/4;
24.17 FX (x) = 1 − e−cx for x > 0;
for x < 16

0

24.19 a. FX (x) = x /64 − x/2 + 4 for 16 ≤ x ≤ 24
2

for x > 24

1

b. a = 20
24.21 k = 1/(ln 20 − 7999) = −0.0001; 24.23 1/3; 24.25 0.3233
24.27 Yes. An example of the CDFs of two such random variables is given
below.
FX (x)
1

x
−10 −8 −6 −4 −2 2 4 6 8 10

Chapter 25
25.1 5/9; 25.3 7/9; 25.5 10/27; 25.7 0.0252;
25.9 a. 1/64; b. 7/64; d. 49/64; e. (1+7+7+49)/64=1;
for x < 0, y < 0

0

25.11 FX,Y (x, y) = x y/2 + xy /2 for 0 ≤ x, y ≤ 1
2 2

for x, y ≥ 1

1

25.13 0.9997; 25.15
( 1/4;
0 for w ≤ 0
25.17 FW (w) =
1−e −3w −e −5w +e −8w for w > 0
25.19 0.06699; 25.21 9/25 = 0.36; 25.23 1/16
Chapter 26
26.1 a. Yes. fX,Y (x, y) is defined on a rectangle and can be factored into
fX (x)fY (y).
b. fX (x) = (2/9)(3 − x) for 0 ≤ x ≤ 3; c. fY (y) = (1/2)(2 − y) for 0 ≤ y ≤ 2;
26.3 a. fX (x) = 3x2 for 0 ≤ x ≤ 1; b. fY (y) = 1/(3y ln 2) for 1/2 ≤ y ≤ 4;
638 Answers to Exercises

26.5 a. No; b. fX (x) = (3/2500)x(10 − x)2 for 0 ≤ x ≤ 10 − y;

c. fY (y) = (3/2500)y(10 − y)2 for 0 ≤ y ≤ 10 − x;
26.7 P (X + Y ≤ 4) = 2/9; 26.9 a. Yes; b. 0.7476;
c. 0.7476; No; fX,Y (x, y) 6= fX (x)fY (y) and not defined on a rectangle; Look-
ing at the region where the joint density is defined, it is enough to see if the
variables are dependent (if it is not defined on rectangles). However, the density
being defined on a rectangle is not enough to determine that the variables are
independent. For this to be true, the individual densities must multiply to equal
the joint density.
26.11 a. Yes; b. 1/8 c. 1/64; d. 1/84;
26.13 a. fX,Y (x, y) = 1; b. 7/8;
26.15 a. No; b. fX (x) = 3(1 − x)2 for 0 < x < 1;
c. fY (y) = 6y(1 − y) for 0 < y < 1; d. fZ (z) = 3z 2 for 0 < z < 1;
26.17 a. Yes; b. 25/64; 26.19 63/64; 26.21 a. No; b. 1/8; c. 7/8;
26.23 a. fX (x) = 1/2 sin x for 0 ≤ x ≤ π, 0 otherwise;
b. fY (y) = (1/4) sec2 (y/4) for 0 ≤ y ≤ π, 0 otherwise;
c. 0.2071; 26.25 17/81
Chapter 27
27.1 a. fY (y) = (5/256)y(4 − y)3 for 0 ≤ y ≤ 4;
b. fX|Y (x|y) = 3x2 /(4 − y)3 for 0 ≤ x ≤ 4, 0 ≤ y ≤ 4, x + y ≤ 4;
27.3 a. fY (y) = 1/10 + (3/40)y for 0 ≤ y ≤ 4;
b. fX|Y (x|y) = (3/2)(x2 + y)/(4 + 3y) for 0 ≤ x ≤ 2, 0 ≤ y ≤ 4;
27.5 a. fY (y) = 5e−5y for 0 < y; b. fX|Y (x|y) = e−x+2y for 0 < y < x/2;
27.7 a. fY (y) = (6/7)(1/3 + y + y 2 ) for 0 ≤ y ≤ 1;
b. fX|Y (x|y) = 3(x + y)2 /(1 + 3y + 3y 2 ) for 0 ≤ x, y ≤ 1;
27.9 a. fY (y) = (1/450)(30 − y) for 0 ≤ y ≤ 30;
b. fX|Y (x|y) = 1/(30 − y) for 0 ≤ y, 0 ≤ x, x + y ≤ 30;
27.11 a. fX|Y (x|y) = 1/(3 − y)for 0 < x < 3; b. 1/2; c. 2/5;
27.13 a. fX|Y (x|y) = 2e−2x ; b. 0.1353; c. 0.7534;
27.15 a. fX|Y (x|y) = (2/9)(3 − x); b. 8/9; c. 1/4; 27.17 2/35
Chapter 28
28.1 E(X) = 20.3333; 28.3 E(X) = 3; 28.5 E(X) = 7.5; 28.7 E(X) = 3.1378;
28.9 E(X) = 1; 28.11 a. E(X) = 1/2; b. E(Y ) = 1/7;
28.13 a. E(X) = 1; b. E(Y ) = 2/3; 28.15 E(X) = 1.25;
28.17 a. E(X) = 0; b. E(Y ) = −1/9; 28.19 E(Y ) = 10/3; 28.21 E(X) = 6/5
Chapter 29
29.1 Var(X) = 100/3; 29.3 Var(X) = 25/12; 29.5 Var(X) = 50/9;
29.7 Var(X) = 1/4; 29.9 a. E(X + Y ) = 2; b. Var(X) = 1/2; 29.11 0.2704;
29.13 E(X 2 + Y 3 ) = 2.3; 29.15 a. E(X + Y ) = 5; b. Var(X + Y ) = 13/3;
29.17 9; 29.19 9; 29.21 549; 29.23 $27.55; 29.25 22/49;
29.27 a. E(X) = 35/9; b. Var(X) = 5.7099; 29.29 0.1945;
29.31 a. 162; b. E(X n ) = n!3n
 Answers to Exercises 639

Chapter 30
30.1 a. FX (x) = 2x for 0 ≤ x ≤ 1/2; b. 0.4; c. 0.1; d. 0; e. 1/4; f. 0.1443;
3
fX (x)
2
g.
1
x
0.2 0.4 0.6 0.8 1
FX (x)
1

h. 0.5 median is 1/4

x
−1 −0.5 0.5 1
30.3 a. FX (x) = 1 − 1/x for x ≥ 1; b. 1/3; c. 4/5; d. 0.7619; e. ∞;
fX (x)
1

f. 0.5
x
2 4 6 8 10
1 FX (x)

g. 0.5
x
2 4 6 8 10
25th percentile = 4/3; 50th percentile = 2; 75th percentile = 4;
30.5 k = 4/5; 30.7 a. False; 0 ≤ FX (x) ≤ 1 for all x; b. True;
c. False; 0 ≤ pX (x) ≤ 1 for all x;
30.9 a. False; the derivative of FX (x) is fX (x); b. True; c. True;
d. False; the reverse is true.; e. True;
f. False; the reverse is true since FX (x) is the Cumulative Distribution Function
which is the area under the density.
30.11 a. False; 0 ≤ FX (x) for all x since FX (x) is a probability, and probabilities
are always nonnegative;
b. False; 0 ≤ fX (x) for all x since densities are integrated to get probabilities,
and probabilities are always nonnegative;
c. False; 0 ≤ pX (x) for all x since pX (x) is a probability, and probabilities are
always nonnegative.
640 Answers to Exercises

Chapter 31

31.1 a. The density of the defect should be equally weighted throughout the
10-yard section.
b. The X is the actual location of the defect. c. a = 0, b = 10;
d. E(X) = 5 yards; e. σX = 2.8868;
f. fX (x) = 1/10 for 0 ≤ x ≤ 10, 0 otherwise;
fX (x)
0.3
0.2
0.1 x
2 4 6 8 10
for x < 0

0

g. FX (x) = x/10 for 0 ≤ x ≤ 10
for x > 10

1


1 FX (x)

0.5
x
2 4 6 8 10
h. 0.2; i. 0.29; j. 0.4;
31.3 a. 1/4; b. 80; 31.5 4/15; 31.7 1/33; 31.9 a. $46.75; b. $18.0422;
31.11 a. 4.4444; b. 8.025; c. 7.7778; d. 22.84; 31.13 0.2637; 31.15 2;
31.17 E(min (X, Y, Z)) = 2.25; 31.19 $0.21; 31.21 0.075; 31.23 0.19;
31.25 0.6434; 31.27 E(X) = 1; 31.29 FY (y) = 3y 2/100 − y 3/500

Chapter 32

32.1 a. We know the average daily rate of eggs the chickens will lay, and we
waiting for the first egg to be laid (one event). We are measuring the farmer’s
wait in minutes (continuous). b. The X is the amount of time (in minutes)
until the first egg is laid. c. λ = 0.0125/minute; d. E(X) = 80 minutes;
e. Var(X) = 80 minutes;
f. fX (x) = 0.0125e−0.0125x , x > 0
0.015
fX (x)
0.01

0.005
x
20 40 60 80 100 120
 Answers to Exercises 641

1 FX (x)

g. FX (x) = 1 − e−x/200 , x > 0 0.5

x
20 40 60 80 100 120
h. 0.8825; i. 0.2149; j. 0.2435; k. 184.2 minutes;
32.3 a. 0.3679; b. 0.4724; c. 0.4420; d. 0.3912; e. 0.1386 hours;
32.5 0.8047 hours; 32.7 a. 0.2835; b. 0.1353; c. 0.3679; 32.9 $1461.62;
32.11 a. 0.5666; b. 3.0787; 32.13 ln 2/λ; 32.15 0.9965;
32.17 a. pY (y) = e−λy − e−λ(y+1) ;
b. Geometric, where we count number of losses (do not count the win), and
probability of a win on a trial is p = 1−e−λ , and probability of a loss is q = e−λ .
2 2
32.19 a. λ2a2 + λb + c, if X is exponential; b. λb 2 + 8ab
λ3
+ 20a
λ4
Chapter 33
33.1 a. We know the average egg-laying rate, and we are measuring the wait
until the 6th egg. Waiting for just 1 egg would be an Exponential variable.
b. The random variable X is the amount of time (in minutes) until the 6th egg
is laid.
c. λ = 0.0125, r = 6; d. 480 minutes; e. 195.9592;
6
f. fX (x) = 0.0125
5! x5 e−0.0125x , x > 0;
0.003
fX (x)
0.002
0.001
x
200 400 600 800 1,000
j
g. FX (x) = 1 − e−0.0125x 5j=0 (0.0125x) ,x>0
P
j!
1 F (x)
X

0.5
x
200 400 600 800
h. e−360×0.0125 5j=0 (360 × 0.0125)j/j! = 0.7029;
P

i. e−720×0.0125 5j=0 (720 × 0.0125)j/j! − e−780×0.0125 5j=0 (780 × 0.0125)j/j! =

P P
0.0385;
33.3 a. 315 minutes; b. 119.0588 minutes; 33.5 a. 5 hours; b. 1.5811 hours;
33.7 21 minutes; 33.9 0.6916; 33.11 50 minutes; 33.13 0.4232; 33.15 0.2381;
33.17 0.406
Chapter 34
34.1 a. 3/7; b. 0.17496;
34.3 a. It gives the proportion of horses; the density is kx2 (1 − x)2 .
b. α = 3, β = 3; c. k = 30; d. E(X) = 1/2
642 Answers to Exercises

Chapter 35

fZ (z) fZ (z)
0.4 0.4

35.1 a. 0.9292 b. 0.0708

0.2
z z
−2 2 −2 2

fZ (z) fZ (z)
0.4 0.4

c. 0.0708 d. 0.9292
0.2
z z
−2 2 −2 2
e. 0

fZ (z)
0.1
35.3 a. 0.4483
z
−5 5 10

fZ (z)
0.1
b. 0.9998
z
−5 5 10

fZ (z)
0.1
c. 0.3679
z
−5 5 10

fZ (z) fZ (z)
0.4 0.4

35.5 a. 1.645 b. 1.04

0.2
z z
−2 2 −2 2
 Answers to Exercises 643

fZ (z)
0.4

c. 0.935

z
−2 2
35.7 a. 4.31; b. 4.43; c. −2.935 and 6.935;
35.9 a = −0.48; 35.11 0.2389; 35.13 a. 1.464 or less; b. 0.0475; c. 0.095;
35.15 0.9418; 35.17 0.0548; 35.19 0.6826; 35.21 0.0548;
35.23 a. 0.0062; b. 0.3944; c. 0; d. 16.256 ounces;
e. (15.832, 16.168); f. (16.392, ∞)
Chapter 36
36.1 0.3015; 36.3 0.8888; 36.5 0.1685; 36.7 0.2810; 36.9 0.3409; 36.11 0.3778;
36.13 0.7372; 36.15 0.5
Chapter 37
37.1 0.1685; 37.3 0.1271; 37.5 0.0384; 37.7 0.1251; 37.9 0.881; 37.11 0.8098;
37.13 0.3936; 37.15 0.6228; 37.17 0.3156; 37.19 0.7372; 37.21 0.2090;
37.23 0.1841; 37.25 0.0307; 37.27 0.7794; 37.29 a. 0.2177; b. 0.6808; c. 0.6179
Chapter 38
38.1 Normal; 38.3 Bernoulli; 38.5 Geometric; 38.7 Poisson; 38.9 Gamma;
38.11 Hypergeometric; 38.13 a. Binomial(n = 280,000, p = 0.84); b. 0.1515;
38.15 a. 0.3012; b. 0.2019; c. 0.3012; d. 0.5987;
38.17 a. 0.1792; b. 2/5; c. 1/5; 38.19 a. 0.1122; b. 1/8; c. 0.01113;
d. 5 weeks; e. 0.1353; f. 14.78 people;
38.21 a. 30 years; b. 0.05; c. 0.2105;
38.23 a. p =√1/2, n = 1; b. λ √= 2; c. µ = 1/2 = σ;
d. a = (1 − 3)/2, b = (1 + 3)/2; e. not possible;
38.25 a. 0.3125; b. 0.75; c. 0.0935;
38.27 a. 3 students; b. 10 minutes; σ(X) = 10 minutes;
c. 30 minutes; σ(X) = 17.3205 minutes;
d. 0.0111; e. 0.06197; f. 0.3168; g. 1/6; h. 3/5;
38.29 a. 0.2708; b. 0.8546; c. 0.3486;
d. A high-risk driver may cause an accident with a low-risk driver, in which
case they both have an accident, so the assumption of independence is not very
realistic. This assumption allows us to calculate the probability that both types
of drivers have an accident by multiplying the individual probabilities.
38.31 0.4688; 38.33 a. $146.80; b. $7340
Chapter 39
39.1 a. Dependent; b. Cov(X, Y ) = −2.4; c. ρ(X, Y ) = −1;
39.3 a. Cov(X, 1.3X − 10) = 43.3333; b. Corr(X, 1.3X − 10) = 1;
644 Answers to Exercises

39.5 a. Cov(X, a − X) = −V ar(X); b. Corr(X, a − X) = −1;

39.7 a. Cov(X, Y ) = −0.0052; b. Corr(X, Y ) = −0.04545;
39.9 a. Cov(X, Y ) = −1/9; b. Corr(X, Y ) = −1/2;
39.11 a. Cov(X, Y ) = 2.75; b. Corr(X, Y ) = 0.5238;
39.13 a. Cov(X, Y ) = 0.9452; b. Corr(X, Y ) = 0.4795;
39.15 a. Cov(X, Y ) = 1/2; b. Corr(X, Y ) = 0.8944; 39.17 25/64
Chapter 40
40.1 3/4; 40.3 15.42; 40.5 10.89; 40.7 6; 40.9 1.2; 40.11 19 minutes;
40.13 1.5; 40.15 a. 1/5; b. 1/5;
c. Same; Given that the sum of the dice is 8, there is only one possible combi-
nation of rolls where one die shows the value 4. This is the case in which both
dice simultaneously show 4, since if one die shows 4, the other die must show
8 − 4 = 4 as well. Thus painting √
the dice makes no difference to the probability.
40.17 2.5 songs; 40.19 a. 0; b. 3/2; 40.21 9/20; 40.23 3/11 = 0.2727
Chapter 41
41.1 a. 4/5; b. P (6 ≤ X ≤ 18) ≥ 3/4; 41.3 a. 3/4; b. 3/4;
41.5 P (120 ≤ X ≤ 156) ≥ 0.9228; 41.7 3/5;
41.9 a. 5/8; b. P (6 ≤ X ≤ 14) ≥ 0.3398;
41.11 a. P (X ≥ 6) ≤ 0.5444; b. P (|X − 196| ≥ 60) ≤ 1/4;
41.13 P (|X − 30| ≤ 5) ≥ 0.84;
41.15 a. P (X ≥ 1000) ≤ 0.75; b. P (|X − 750| ≥ 250) ≤ 0.16;
41.17 a. 3/4; b. P (|X − 9| ≤ 3) ≥ 0.5556; 41.19 5/7;
41.21 P (|X − 673| ≤ 105) ≥ 0.434;
41.23 a. P (A) ≤ 0.4655; b. P (B) ≥ 0.1045
Chapter 42
42.1 a. fY (y) = 1/20 (1 − y/100)4 for 0 ≤ y ≤ 100, 0 otherwise;
b. 0.3277; c. 0.4718;
42.3 a. fX(1) (x1 ) = 3/20 e−3x1 /20 for x1 ≥ 0;
b. 0.1653; c. 0.7627;
3/2 √
42.5 fX(5) (x5 ) = 5(x5 /27)4 ( x5 /18) for 0 ≤ x5 ≤ 9, 0 otherwise;
42.7 a. fX(1) (x1 ) = 3/4((242 − x1 )/4)2 for 238 ≤ x1 ≤ 242, 0 otherwise;
b. fX(3) (x3 ) = 3/4((x3 − 238)/4)2 for 238 ≤ x3 ≤ 242, 0 otherwise;
c. fX(2) (x2 ) = 6x2 (4 − x2 )/64 for 238 ≤ x2 ≤ 242, 0 otherwise;
42.9 a. fX (x) = 7/2((8 − x)/2)6 for 6 ≤ X ≤ 8, 0 otherwise;
b. fY (y) = 7/2((y − 6)/2)6 for 6 ≤ Y ≤ 8, 0 otherwise;
c. 6:15 PM; d. 7:45 PM; 42.11 a.
8:34:38 AM; b. 0.00002;
42.13 a. P (X(j) = x) = j−1,1,6−j6
(x/9)j−1 (1/9)(1 − x/9)6−j for 1 ≤ j ≤ 6;
b. 0.6488
Chapter 43
43.1 MX (t) = pet /(1 − et (1 − p)); 43.5 MX (t) = (pet /(1 − et (1 − p)))r
 Answers to Exercises 645

Chapter 44
44.1 0.5013; 44.3 0.8333; 44.5 0.3015; 44.7 0.2551; 44.9 2/π;
√ √
44.13 a. y = 0 → v = u; y = 1 → v = 1 + u;
x = 0 → u = 0; x = 1 → u = 1;
2 v
1.5
b. 1
0.5
u
0.5 1
44.15 U and V are independent, since fU,V (u, v) can be factored into fU (u) =
√
1/(2 u) and fV (v) = 1, which are both bonafide densities in the defined regions
√ √
0 ≤ u ≤ 1 and u ≤ v ≤ 1 + u.
Chapter 45
45.1 0.1210;
for x < 0

0

45.3 a. FZ (z) = (z/30) for 0 ≤ x ≤ 30 ;
2

for x > 30

1

b. fZ (z) = 2z/30 for 0 ≤ z ≤ 30, 0 otherwise; c. E(X) = 20;
2

45.5 0.9772; 45.7 a. 0.0161; b. 5;

45.9 a. 9 minutes; b. 12 minutes; c. 11 minutes; d. 9000 minutes; e. 0.1814;
45.11 3.45; 45.13 10/3; 45.15 a. 227.2727; b. 1838.843;
45.17 a. 1.2918 × 10−32 ; b. 1.93 × 10−30 ; c. 149.4; 45.19 Poisson(λ = 1.5);
45.21 a. 28 hours; b. 112; 45.23 17.6471; 45.25 2n!/(2n)! = (n − 1)!/(2n − 1)!;
45.27 0.04; 45.29 a. 0.1n; b. 0.09n; 45.31 5/6; 45.33 150
this page left intentionally blank
Bibliography

[1] Patrick Billingsley. Probability and Measure. Wiley, Anniversary edition,

2012.

[2] Rick Durrett. Probability: Theory and Examples. Cambridge, 4th edition,
2010.

[3] Philippe Flajolet and Robert Sedgewick. Analytic Combinatorics. Cam-

bridge, 2009.

[4] Jim Pitman. Probability. Springer, 1993.

[5] Sheldon Ross. A First Course in Probability. Prentice Hall, 9th edition,
2014.

647
this page left intentionally blank
Index

average, see expected value variance of the sum of (general

case), 519, 520
Bayes’ Theorem, 60–69 variance of the sum of
Bernoulli random variables, (independent case), 378
173–181 continuous random variables, 301
Beta random variables, 438–444 Continuous Uniform random
binomial coefficient, xxii, 183 variables, 391–411
Binomial random variables, counting, 271–298
182–196 covariance, 516–518
expected value, 131 of independent random
variables, 521
Cardano, Gerolamo, 483 symmetry of, 520
Carl Friedrich Gauss, 446 cumulative distribution function
ceiling function, xxii (CDF), 88
Central Limit Theorem, 483–504
Chebyshev inequality, 557–561 DeMorgan’s Laws, 11–13
complementary events, 11 density, see probability density
conditional density, see conditional function
probability density dependent events, 38
function Discrete random variables
conditional mass, see conditional comparison to Continuous
probability mass function random variables, 300–301
conditional probability, 48, 49 expected value of the sum of,
multiplication with, 66 130
satisfies the probability review of, 161–170, 259–269
axioms, 54 variance of the sum of (general
conditional probability mass case), 519, 520
function (conditional variance of the sum of
PMF), 113 (independent case), 154
Continuous random variables discrete random variables, 80
comparison to Discrete random Discrete Uniform random variables,
variables, 300–301 254–258
expected value of the sum of, disjoint, 4
374 disjoint events, 18
review of, 384–388, 505–514 distributive laws, 53

649
650 Index

empty set, 4 independent random variables,

events, 4 107–113
complementary, 11 indicator random variables, 84, see
consisting of a partition, 22 Bernoulli random variables
dependent, 38 expected value of the sum of,
disjoint, 4, 18 133
equally likely, 19–24 for independent events, 110,
independent, 37, 40, 41 113
probability of good before bad, intersection, 8
43–44
subsets of, 4 joint density, see joint probability
expected value, 120 density function
of continuous random joint cumulative distribution
variables, 355–365 function (joint CDF), 103
of discrete random variables, joint mass, see joint probability
120–129 mass function
of functions of continuous joint probability density function,
random variables, 366–383 319–332
of functions of discrete random joint probability mass function
variables, 142–160 (joint PMF), 103
of sums of continuous random
variables, 376–377 Markov inequality, 554–557
of sums of discrete random mass, see probability mass function
variables, 130–141 mean, see expected value
Exponential random variables, memoryless property
412–427 of Exponential random
memoryless property, 417–419 variables, 417–419
minimum of, 419–421 of Geometric random variables,
moments of, 422–423 205–207
Moment Generating Functions,
floor function, xxii 582–593
moment of a random variable, 422
Gamma random variables, 428
Gaussian random variables, see Negative Binomial random
Normal random variables variables, 214–223
generating functions, 582–594 Normal distribution table, 608
Geometric random variables, Normal random variables, 445
197–213 approximation to Binomial,
memoryless property, 205–207 492–496
geometric sum, xxi, 28, 111 approximation to Poisson,
496–500
Hypergeometric random variables, normalizing a.k.a.
241–253 standardizing, 448
sums of, 470–482
inclusion-exclusion, 25–27
independent events, 37, 40, 41 order statistics, 565–581
 Index 651

outcome, 4 review of, 73

partition, 22 sample space, 4

Poisson processes sampling
introduction to, 421 with replacement, 184, 261,
Poisson random variables, 224–240 276–280
approximation to Binomial, without replacement, 8, 242,
231 262, 276–280
sums of, 229 seating arrangements, 16, 35, 47,
poker, 293–295 59, 280–285, 291, 292
Texas Hold ’Em, 614 set notation, 7
probability, 17–36 standard deviation, 146
axioms, 18 subset, 4
conditional, 48, 49 sum of integers, xxi
probability density function (PDF), sum of squares of integers, xxi
302
probability mass function (PMF), transformation of random
88 variables, 596–612
trials, 42
random variables probability of good before bad,
Bernoulli, 173–181 43–44
Beta, 438–444 success or failure of, 173
Binomial, 182–196
Continuous Uniform, 391–411 Uniform random variables
definition of, 78 Continuous, 391–411
Discrete Uniform, 254–258 Discrete, 254–258
discrete versus continuous, union, 8
78–86
Exponential, 412–427 variance, 145
Gamma, 428–437 as the covariance of a random
Geometric, 197–213 variable with itself, 520
Hypergeometric, 241–253 Venn diagram, 52, 53
indicator, 84
waiting times, see Exponential
moments of, 422
random variables
Negative Binomial, 214–223
weighted average, see expected
Normal, 445
value
sums of, 470–482
Poisson, 224–240 Yahtzee, 33, 159, 295–298
transformation of, 596–612
Randomness zombie, 222
 Summary of Named Discrete Random Variables
Name Mass Expected value Variance Parameters What X is When used
Bernoulli pX (1) = p p pq p = prob. 0 or 1 1 success
pX (0) = q succ./trial (no or yes) or failure
n = # trials; 0, 1, 2, . . . , n successes
Binomial n
px q n−x p = prob. (successes) in n trials


x np npq
succ./trial
Geometric q x−1 p 1/p q/p2 p = prob. 1, 2, 3, . . . # trials
succ./trial (trials) to 1st succ.
p = prob.
Negative x−1
q x−r pr qr/p2 succ./trial; # trials


r−1 r/p r, r + 1, . . .
Binomial r = # of (trials) to rth succ.
succ. needed
Poisson e−λ λx/x! λ λ λ rate 0, 1, 2, 3, . . . # events
(events) in period
Hyper- M good, # of good
−M
(Mx )(Nn−x )
geometric nM M N −n
N − M bad; selected


N
(n) N 1− N N −1 nM/N 1, 2, . . . , M
n selected
Discrete 1/N (N + 1)/2 (N 2 − 1)/12 N outcomes 1, 2, . . . , N equally
Uniform likely
 Summary of Named Continuous Random Variables
Name Uniform Exponential Gamma Beta Normal
2 2)
λe−λx (λx)r−1 Γ(α+β)xα−1 (1−x)β−1 e−(x−µ) /(2σ
Density 1/(b − a) λe−λx Γ(r) Γ(α)Γ(β)
√
2πσ 2
fX (x)
Domain a≤x≤b x≥0 x≥0 0≤x≤1 −∞ < x < ∞
Pr−1 (λx)j
CDF (x − a)/(b − a) 1 − e−λx ; 1−e −λx
j=0 j! , convert to std. Normal
FX (x) if r ∈ N Z = (X − µX )/σX
E(X) (a + b)/2 1/λ r/λ α/(α + β) µ
αβ
Var(X) (b − a)2/12 1/λ2 r/λ2 (α+β)2 (α+β+1)
σ2
Param- a, b are λ is average λ is average usually some µ = exp. value;
eters endpoints # of successes # of successes given σ = st. dev.
per time unit per time unit constraints
wait time ’til wait time ’til fraction, practical random
What location in 1st event rth event percentage, variable for
X is an interval (like continuous (like continuous proportion, quantities clustered
Geometric) Neg. Binom.) around avg. value
When spread wait time wait time Bayesian Central Limit Thm.
used evenly until 1st event until rth event statistics and applications

Note: Γ(r) = (r − 1)! if r ∈ N.

 Standard Normal Table
z . 0 . 1 . 2 . 3 . 4 . 5 . 6 . 7 . 8 . 9
0.0 0.5000 0.5040 0.5080 0.5120 0.5160 0.5199 0.5239 0.5279 0.5319 0.5359
0.1 0.5398 0.5438 0.5478 0.5517 0.5557 0.5596 0.5636 0.5675 0.5714 0.5753
0.2 0.5793 0.5832 0.5871 0.5910 0.5948 0.5987 0.6026 0.6064 0.6103 0.6141
0.3 0.6179 0.6217 0.6255 0.6293 0.6331 0.6368 0.6406 0.6443 0.6480 0.6517
0.4 0.6554 0.6591 0.6628 0.6664 0.6700 0.6736 0.6772 0.6808 0.6844 0.6879
0.5 0.6915 0.6950 0.6985 0.7019 0.7054 0.7088 0.7123 0.7157 0.7190 0.7224
0.6 0.7257 0.7291 0.7324 0.7357 0.7389 0.7422 0.7454 0.7486 0.7517 0.7549
0.7 0.7580 0.7611 0.7642 0.7673 0.7704 0.7734 0.7764 0.7794 0.7823 0.7852
0.8 0.7881 0.7910 0.7939 0.7967 0.7995 0.8023 0.8051 0.8078 0.8106 0.8133
0.9 0.8159 0.8186 0.8212 0.8238 0.8264 0.8289 0.8315 0.8340 0.8365 0.8389
1.0 0.8413 0.8438 0.8461 0.8485 0.8508 0.8531 0.8554 0.8577 0.8599 0.8621
1.1 0.8643 0.8665 0.8686 0.8708 0.8729 0.8749 0.8770 0.8790 0.8810 0.8830
1.2 0.8849 0.8869 0.8888 0.8907 0.8925 0.8944 0.8962 0.8980 0.8997 0.9015
1.3 0.9032 0.9049 0.9066 0.9082 0.9099 0.9115 0.9131 0.9147 0.9162 0.9177
1.4 0.9192 0.9207 0.9222 0.9236 0.9251 0.9265 0.9279 0.9292 0.9306 0.9319
1.5 0.9332 0.9345 0.9357 0.9370 0.9382 0.9394 0.9406 0.9418 0.9429 0.9441
1.6 0.9452 0.9463 0.9474 0.9484 0.9495 0.9505 0.9515 0.9525 0.9535 0.9545
1.7 0.9554 0.9564 0.9573 0.9582 0.9591 0.9599 0.9608 0.9616 0.9625 0.9633
1.8 0.9641 0.9649 0.9656 0.9664 0.9671 0.9678 0.9686 0.9693 0.9699 0.9706
1.9 0.9713 0.9719 0.9726 0.9732 0.9738 0.9744 0.9750 0.9756 0.9761 0.9767
2.0 0.9772 0.9778 0.9783 0.9788 0.9793 0.9798 0.9803 0.9808 0.9812 0.9817
2.1 0.9821 0.9826 0.9830 0.9834 0.9838 0.9842 0.9846 0.9850 0.9854 0.9857
2.2 0.9861 0.9864 0.9868 0.9871 0.9875 0.9878 0.9881 0.9884 0.9887 0.9890
2.3 0.9893 0.9896 0.9898 0.9901 0.9904 0.9906 0.9909 0.9911 0.9913 0.9916
2.4 0.9918 0.9920 0.9922 0.9925 0.9927 0.9929 0.9931 0.9932 0.9934 0.9936
2.5 0.9938 0.9940 0.9941 0.9943 0.9945 0.9946 0.9948 0.9949 0.9951 0.9952
2.6 0.9953 0.9955 0.9956 0.9957 0.9959 0.9960 0.9961 0.9962 0.9963 0.9964
2.7 0.9965 0.9966 0.9967 0.9968 0.9969 0.9970 0.9971 0.9972 0.9973 0.9974
2.8 0.9974 0.9975 0.9976 0.9977 0.9977 0.9978 0.9979 0.9979 0.9980 0.9981
2.9 0.9981 0.9982 0.9982 0.9983 0.9984 0.9984 0.9985 0.9985 0.9986 0.9986
3.0 0.9987 0.9987 0.9987 0.9988 0.9988 0.9989 0.9989 0.9989 0.9990 0.9990
3.1 0.9990 0.9991 0.9991 0.9991 0.9992 0.9992 0.9992 0.9992 0.9993 0.9993
3.2 0.9993 0.9993 0.9994 0.9994 0.9994 0.9994 0.9994 0.9995 0.9995 0.9995
3.3 0.9995 0.9995 0.9995 0.9996 0.9996 0.9996 0.9996 0.9996 0.9996 0.9997
3.4 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9997 0.9998
3.5 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998 0.9998
3.6 0.9998 0.9998 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999 0.9999
For a standard Normal random variable Z, these are the values of the cu-
mulative distribution function FZ (z) = P (Z ≤ z), also corresponding to the
area under the density fZ to the left of z.

fZ (z)
0.4
For example, in this graph,
the area under the curve is
P (Z ≤ 0.75) = 0.7734.
z
−2 2