Release 6

The real voyage of discovery consists not in seeking new landscapes, but in having new eyes. Marcel Proust

Introductory Guide
JMP, A Business Unit of SAS SAS Campus Drive Cary, NC 27513

JMP Introductory Guide, Release 6 Copyright 2005, SAS Institute Inc., Cary, NC, USA ISBN 1-59047-815-0 All rights reserved. Produced in the United States of America. For a hard-copy book: No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, or otherwise, without the prior written permission of the publisher, SAS Institute Inc. For a Web download or e-book: Your use of this publication shall be governed by the terms established by the vendor at the time you acquire this publication. U.S. Government Restricted Rights Notice: Use, duplication, or disclosure of this software and related documentation by the U.S. government is subject to the Agreement with SAS Institute and the restrictions set forth in FAR 52.227-19, Commercial Computer Software-Restricted Rights (June 1987). SAS Institute Inc., SAS Campus Drive, Cary, North Carolina 27513. 1st printing, August 2005 JMP, SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA and other countries. indicates USA registration. Other brand and product names are registered trademarks or trademarks of their respective companies.

Contents

Contents
JMP Introductory Guide Credits and Acknowledgments 1 Introducing JMP
What You Need to Know . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Learning About JMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Using Tutorials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Searching in the Help . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Learning About Statistical and JSL Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Using the Context-Sensitive Help . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Learning JMP Tips & Tricks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Using This Book in Combination with Other Included Books . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Conventions Used in this Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Step 1: Start JMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Step 2: Open a JMP Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Step 3: Learn About the Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Specifying the Values Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Data Table Cursor Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Selecting Rows and Columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Step 4: Select an Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Casting Columns Into Roles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Step 5: View the Output Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Graphs and Charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Statistical Tables and Text . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Step 6: Save the JMP Output Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 A Practice Tutorial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Open a Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Select an Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Spin the Cowboy Hat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Experiment with the Display Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

Creating a JMP Data Table

Starting a JMP Session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 Creating Rows and Columns in a JMP Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

ii

Contents

Add Columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Set Column Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Add Rows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Entering Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Plotting Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Document the Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24 25 25 26 26 29 30

Summarizing Data
Look Before You Leap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Grouping Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating Statistics for Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Charting Statistics from Grouped Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Charting Statistics for Two Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Finding a Subgroup with Multiple Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparative Scatterplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . What Has Been Discovered? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Finding the Best Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 34 35 36 38 39 40 41 42 43

Looking at Distributions
Look Before You Leap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Displaying Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding Histograms of Nominal and Ordinal Variables . . . . . . . . . . . . . . . . . . . . . . . . Understanding Histograms of Continuous Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Learning About Report Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Reports for Continuous Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Frequency Table for Ordinal or Nominal Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Adding a Computed Column . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating Subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 47 48 49 51 52 52 52 55 56

Comparing Group Means

Look Before You Leap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Graphical Display of Grouped Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Choose Variable Roles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Show Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fit Means Option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fit Quantiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison Circles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Quantify Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Analysis of Variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 59 60 60 61 63 64 65 66

Contents

iii Contents

Mean Estimates and Statistical Comparisons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

Analyzing Categorical Data

Look Before You Leap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Open a Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Address the Research Question . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Modify the Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Contingency Table Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cast Variables Into Roles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Contingency Table Mosaic Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 71 71 72 75 75 76 79

Regression and Curve Fitting

Look Before You Leap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Open a JMP File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Select an Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Choose Variable Roles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting Models to Continuous Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting the Mean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting a Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding the Summary of Fit Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding the Analysis of Variance Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding the Parameter Estimates Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Excluding Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Journaling JMP Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Examining a Polynomial Fit (Linear Regression) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting a Spline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting By Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 83 83 83 84 85 85 86 87 87 88 88 90 91 93 94

A Factorial Analysis
Look Before You Leap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 Open a Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 What Questions Can Be Answered? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 The Fit Model Window . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 Graphical Display: Leverage Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 Quantify Results: Statistical Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 Analysis of Variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 Summary Reports For The Whole Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Summary Reports for Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

iv

Contents

Exploring Data
Solubility Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . One-Dimensional Views . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Two-Dimensional Views . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Three-Dimensional Views . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Principal Components and Biplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using Colors, Markers, and the Brush Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Multivariate Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 109 110 111 112 116 117 118

10 Multiple Regression
Aerobic Fitness Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fit Planes to Test Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Whole Model Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . More and More Regressors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Interpreting Leverage Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Collinearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 123 124 126 127 128 130 130

References Index

Credits

Credits and Acknowledgments

Origin JMP was developed by SAS Institute Inc., Cary, NC. JMP is not a part of the SAS System, though portions of JMP were adapted from routines in the SAS System, particularly for linear algebra and probability calculations. Version 1 of JMP went into production in October, 1989. Credits JMP was conceived and started by John Sall. Design and development were done by John Sall, Chung-Wei Ng, Michael Hecht, Richard Potter, Brian Corcoran, Annie Dudley Zangi, Bradley Jones, Craige Hales, Chris Gotwalt, Paul Nelson, Xan Gregg, Jianfeng Ding, Scott McQuiggan, and Wenjie Bao. In the SAS Institute Technical Support division, Ryan Gilmore, Wendy Murphrey, and Toby Trott provide technical support and conduct test site administration. Statistical technical support is provided by Craig DeVault, Duane Hayes, Elizabeth Edwards, Kathleen Kiernan, and Tonya Mauldin. Nicole Jones, Jim Borek, Kyoko Keener, Hui Di, Laura Lancaster, Joseph Morgan, Wenjun Bao, Fang Chen, and Hugh Crews provide ongoing quality assurance. Additional testing and technical support is done by Noriki Inoue, Kyoko Takenaka, and Masakazu Okada from SAS Japan. Bob Hickey is the release engineer. Special thanks to Mark Bailey and Lori Rothenberg from SAS. The JMP manuals were written by Ann Lehman, Lee Creighton, John Sall, Bradley Jones, Erin Vang, and Meredith Blackwelder, with contributions from Annie Dudley Zangi and Brian Corcoran. Creative services and production was done by SAS Publications. Melanie Drake implemented the help system. Thanks also to Georges Guirguis, Warren Sarle, Gordon Johnston, Duane Hayes, Russell Wolfinger, Randall Tobias, Robert N. Rodriguez, Ying So, Warren Kuhfeld, George MacKensie, Bob Lucas, Warren Kuhfeld, Mike Leonard, and Padraic Neville for statistical R&D support. Acknowledgments We owe special gratitude to the people that encouraged us to start JMP, to the alpha and beta testers of JMP, and to the reviewers of the documentation. In particular we thank Michael Benson, Howard Yetter (d), Andy Mauromoustakos, Al Best, Stan Young, Robert Muenchen, Lenore Herzenberg, Ramon Leon, Tom Lange, Homer Hegedus, Skip Weed, Michael Emptage, Pat Spagan, Paul Wenz, Mike Bowen, Lori Gates, Georgia Morgan, David Tanaka, Zoe Jewell, Sky Alibhai, David Coleman, Linda Blazek, Michael Friendly, Joe Hockman, Frank Shen, J.H. Goodman, David Ikl, Lou Valente, Barry Hembree, Dan Obermiller, Jeff Sweeney, Lynn Vanatta, and Kris Ghosh. Also, we thank Dick DeVeaux, Gray McQuarrie, Robert Stine, George Fraction, Avigdor Cahaner, Jos Ramirez, Gudmunder Axelsson, Al Fulmer, Cary Tuckfield, Ron Thisted, Nancy McDermott, Veronica Czitrom, Tom Johnson, Cy Wegman, Paul Dwyer, DaRon Huffaker, Kevin Norwood, Mike Thompson, Jack Reese, Francois Mainville, and John Wass.

vi

Credits and Acknowledgments

We also thank the following individuals for expert advice in their statistical specialties: R. Hocking and P. Spector for advice on effective hypotheses; Robert Mee for screening design generators; Grey Piepel, Peter Goos, and J. Stuart Hunter for advice on design of experiments; Jason Hsu for advice on multiple comparisons methods (not all of which we were able to incorporate in JMP); Ralph OBrien for advice on homogeneity of variance tests; Ralph OBrien and S. Paul Wright for advice on statistical power; Keith Muller for advice in multivariate methods, Harry Martz, Wayne Nelson, Ramon Leon, Dave Trindade, Paul Tobias, and William Q. Meeker for advice on reliability plots; Lijian Yang and J.S. Marron for bivariate smoothing design; George Milliken and Yurii Bulavski for development of mixed models; Will Potts and Cathy Maahs-Fladung for data mining; Clay Thompson for advice on contour plotting algorithms; and Tom Little, Blanton Godfrey, Tim Clapp, and Joe Ficalora for advice in the area of Six Sigma. For sample data, thanks to Patrice Strahle for Pareto examples, the Texas air control board for the pollution data, and David Coleman for the pollen (eureka) data. Translations Erin Vang coordinated localization. Noriki Inoue, Kyoko Takenaka, and Masakazu Okada of SAS Japan were indispensable throughout the project. Special thanks to Professor Toshiro Haga (retired, Science University of Tokyo) and Professor Hirohiko Asano (Tokyo Metropolitan University). Finally, thanks to all the members of our outstanding translation teams. Past Support Many people were important in the evolution of JMP. Special thanks to Jeffrey Perkinson, David DeLong, Mary Cole, Kristin Nauta, Aaron Walker, Ike Walker, Eric Gjertsen, Dave Tilley, Ruth Lee, Annette Sanders, Tim Christensen, Jeff Polzin, Eric Wasserman, Charles Soper, Yusuke Ono, and Junji Kishimoto. Thanks to SAS Institute quality assurance by Jeanne Martin, Fouad Younan, and Frank Lassiter. Additional testing for Versions 3 and 4 was done by Li Yang, Brenda Sun, Katrina Hauser, and Andrea Ritter. Also thanks to Jenny Kendall, John Hansen, Eddie Routten, David Schlotzhauer, and James Mulherin. Thanks to Steve Shack, Greg Weier, and Maura Stokes for testing JMP Version 1. Thanks for support from Charles Shipp, Harold Gugel (d), Jim Winters, Matthew Lay, Tim Rey, Rubin Gabriel, Brian Ruff, William Lisowski, David Morganstein, Tom Esposito, Susan West, Chris Fehily, Dan Chilko, Jim Shook, Ken Bodner, Rick Blahunka, Dana C. Aultman, and William Fehlner.

1
Introducing JMP
JMP uses an extraordinary graphical interface to display and analyze data. JMP is software for interactive statistical graphics and includes: a data table window for editing, entering, and manipulating data a broad range of graphical and statistical methods for data analysis an extensive design of experiments module options to highlight and display subsets of data a formula editor for each table column to compute values as needed a facility for grouping data and computing summary statistics special plots, charts, and communication capability for quality improvement techniques tools for printing and for moving analyses results between applications a scripting language for saving and creating frequently used routines This introductory chapter gives basic information about using JMP.

1 Preliminaries

1
Contents
What You Need to Know . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Learning About JMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Using Tutorials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Searching in the Help . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Learning About Statistical and JSL Terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Using the Context-Sensitive Help . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Learning JMP Tips & Tricks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Using This Book in Combination with Other Included Books . . . . . . . . . . . . . . . . . . . . . . . . . 7 Conventions Used in this Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Step 1: Start JMP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Step 2: Open a JMP Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Step 3: Learn About the Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Specifying the Values Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Data Table Cursor Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 Selecting Rows and Columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Step 4: Select an Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Casting Columns Into Roles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Step 5: View the Output Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Graphs and Charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Statistical Tables and Text . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 Step 6: Save the JMP Output Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 A Practice Tutorial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Open a Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 Select an Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Spin the Cowboy Hat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Experiment with the Display Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1 Introducing JMPWhat You Need to Know

3 1 Preliminaries

What You Need to Know

Before you begin using JMP, you should be familiar with: Standard operations and terminology such as click, double-click, Ctrl-click and Alt-click (Command-click and Option-click on the Macintosh), Shift-click, drag, select, copy, and paste. How to use menu bars and scroll bars, how to move and resize windows, and how to manipulate files in the desktop. If you are using your computer for the first time, consult the reference guides that came with it for more information. Minimal statistics. Even though JMP has many advanced features, you only need a minimal background of formal statistical training. All analysis include graphical displays with options that help you review and interpret the results. Each analysis also includes access to help windows that offer general help and some statistical details.

Learning About JMP

If you are familiar with JMP, you might want to know only whats new. The JMP New Features document gives a summary of general changes and additions. To learn more about JMP, use the recommendations in the following sections.

Using Tutorials
JMP provides three types of tutorials: Beginners Tutorial The beginners tutorial steps you through the JMP interface and explains the basics of how to use JMP. It is accessible through JMPs Tip of the Day window, which appears when you start JMP. To start the tutorial from the Tip of the Day window, click Enter Beginners Tutorial. Or, start the tutorial by selecting Help (View on the Macintosh) > Tutorials > Beginners Tutorial. Specific Analysis Tutorials Tutorials that step you through creating an analysis in JMP are found under Help (View on the Macintosh) > Tutorials. Tutorials describe how to create a chart, compare means, how to design an experiment, and more. JMP Introductory Guide Book The JMP Introductory Guide is a collection of tutorials designed to help you learn JMP strategies. If you did not receive a printed copy of this book, view the .pdf file by selecting Help > Books > Introductory Guide. By following along with these step-by-step examples, you can quickly become familiar with JMP menus, options, and report windows.

Searching in the Help

You might want help on a specific topic, and you want to search JMPs online help for that topic. The main menu bar contains a Help menu, which provides the appropriate searching capabilities:

1 Introducing JMPLearning About JMP

Windows and Linux Help Menu

Macintosh Help Menu

On Windows and Linux, the Help > Contents, Help > Search, and Help > Index commands access the JMP help system. The help system provides navigable online JMP documentation. On the Macintosh, the Help > JMP Help command displays a list of JMP help items with search capabilities and a table of contents.

Learning About Statistical and JSL Terms

The Help > Indexes command displays the following sources for your reference: Statistics Index Accesses references that give definitions of statistical terms. Once you are in the Statistics Index window, click the Topic Help button to go to the place in the online help that describes the highlighted topic. Click the Script button to view or hide the script associated with the highlighted topic. Click Launch to run the script that corresponds to the item you have highlighted in the list. Click Example to view an example of the highlighted topic.

1 Introducing JMPLearning About JMP

5 1 Preliminaries

list of topics

description

JSL value/script

JSL Operators Index Presents a list of JSL operators, such as Sin, Cos, Sqrt, and Abbrev Date that you would use when writing JSL. Highlight an operator name to see a description of the operator appears in the window on the right. Click the Topic Help button to locate the topic in the online help. Object Scripting Index Presents a list of JSL objects. These are scriptable JSL building blocks. Highlight an object name and messages the object recognizes appear in the window on the right. DisplayBox Scripting Index Presents a list of the elements that make up a JMP report. These elements are the JSL building blocks with which you build output. Highlight a Display Box and available messages for each object appear in the window on the right.

Using the Context-Sensitive Help

To use JMPs online help system, select one of the following methods: Select Help from analysis construction windows (as shown in Figure 1.1) and report windows.

1 Introducing JMPLearning About JMP

Figure 1.1 Help Is Available

Select the help tool ( ) from the Tools menu and click a place in a data table or report on which you need assistance (Figure 1.2). Context-sensitive help tells about the items in the area you clicked. Figure 1.2 Use the Help Tool for Context-Sensitive Help

In some reports, make a small circle with your cursor to reveal information about the item in the area. Figure 1.3 Making a Circle with the Cursor Displays Help

In some menus, hold the cursor on menu items to reveal information about the menu item. Figure 1.4 Display a Description of Menu Items

1 Introducing JMPConventions Used in this Book

7 1 Preliminaries

Learning JMP Tips & Tricks

When you first start JMP, you see the Tip of the Day window. This window provides tips about using JMP that you might not know. To turn off the Tip of the Day, un-check Show tips at startup. To view it again, select Help (View on the Macintosh) > Tip of the Day. See Tip of the Day Window of JMP User Guide for details. Also use the JMP Quick Reference Guide to learn more advanced commands in JMP. View this document by selecting Help > Books > JMP Quick Reference Card.

Using This Book in Combination with Other Included Books

The book you are reading now is the JMP Introductory Guide. See the following manuals for further documentation of JMP: The JMP User Guide has complete documentation of all JMP menus, an explanation of data table manipulation, and a description of the formula editor. There are chapters that show how to do common tasks such as manipulating files, transforming data table columns, and cutting and pasting JMP data, statistical text reports, and graphical displays. The JMP Statistics and Graphics Guide gives documentation of the Analyze and Graph menus. It documents analyses, discusses statistical methods, and describes all report windows and options. The JMP Design of Experiments covers the DOE menu, the experimental design facility in JMP. The JMP Scripting Guide is a reference guide to the JMP scripting language (JSL) that lets you automate action sequences. If you did not receive printed copies of these books, view the .pdf files by selecting Help > Books.

Conventions Used in this Book

Conventions used in this manual were devised to help relate written material to information that appears on-screen: The .jmp extension follows file names on the PC. When you installed JMP, a folder named Sample Data was also installed. On the Macintosh, JMP sample data files have the same name, but show without an extension. Reference to names of JMP files, data tables, variable names, and items in reports appear in Helvetica to help distinguish them from surrounding text. Special information, warnings, and limitations are noted in sentences beginning with the bold word Note. Reference to menu names (File menu) or menu items (Save command) appear in Helvetica bold font. The notation to select a command from a menu is sometimes written as File > New, meaning select the New command from the File menu.

1 Introducing JMPStep 1: Start JMP

Words or phrases that are important or have definitions specific to JMP are in italics the first time they appear.

Step 1: Start JMP

Start a JMP session by double-clicking the JMP application icon. Your initial view of JMP will be a menu bar, a tool bar, the Tip of the Day window, and the JMP Starter window (Figure 1.5). Figure 1.5 First View of JMP (Windows)

To start the online tutorial, click Enter Beginners Tutorial. Or, click the Close button to close the window and follow the tutorials in this book.

Step 2: Open a JMP Data Table

There are several ways to open a data table: Selecting File > New (or clicking the New Data Table button on the JMP Starter window) creates and displays a data table with an empty data grid. First, add rows and columns, then type in or paste in new data. For details, see Creating New Data Tables in JMP User Guide. Selecting File > Open (or clicking the Open Data Table button on the JMP Starter window) presents a file selection window (Figure 1.6) with a list of existing tables. Select a file and click Open. For details, see Opening Existing JMP Files in JMP User Guide.

1 Introducing JMPStep 3: Learn About the Data Table

9 1 Preliminaries

Figure 1.6 The Open Data File Window

Step 3: Learn About the Data Table

Opening or creating a data table creates a data grid and table information panels, like the ones shown in Figure 1.7. The counts of table rows and columns appear in the corresponding panels to the left of the data grid. In the data grid, a row number identifies each row, and each column has a column name. Rows and columns are sometimes called observations and variables in an analysis. Figure 1.7 A Data Table

The JMP data table window is a flexible way to prepare data. Using it, you can accomplish a variety of table management tasks, such as: Editing the value in any cell

10

1 Introducing JMPStep 3: Learn About the Data Table

Changing a columns width by dragging the column line Hiding columns temporarily, or deleting columns permanently Adding rows, or rearranging the order of rows Adding columns, or rearranging the order of columns Selecting a subset of rows for analysis and saving that subset for further use Sorting or combining tables For details, see Entering and Editing Data in JMP User Guide.

Specifying the Values Type

The small icon to the left of the column name in the columns panel is an icon that can be clicked. Use it to declare the modeling type of the values in the column. JMP uses three modeling type to determine how to analyze the columns values: Continuous ( Ordinal ( Nominal ( ) ) ) Values are numeric measurements. Values are ordered categories, which can have either numeric or character values. Values are numeric or character classifications.

Modeling types are changeable depending on how you want to look at your data. For example, a variable like age should be specified continuous to find the mean (average) age, but nominal or ordinal to find frequency counts for each age value. The default modeling type is nominal for character values and continuous for numeric values. To assign a different modeling type to your variables: 1. Click the icon next to the variable name. 2. Select the appropriate modeling type. For details, see Specifying Data Types and Modeling Types in JMP User Guide.

Data Table Cursor Forms

As you move the cursor around the data table, it changes forms. Its shape gives you information about performable actions. The following sections describe the different cursor forms. Arrow Cursor The cursor displays as a standard arrow when it is anywhere in the table panels to the left of the data grid, except when it is on a red triangle icon ( ) or diamond-shaped disclosure button ( on Windows and on the Macintosh) or when it is in the upper-left corner of the data gridthe area where rows and columns are deselected (as shown here).

1 Introducing JMPStep 3: Learn About the Data Table

11 1 Preliminaries

Click to deselect column Click to select column or double-click to edit column name Click to deselect rows I-Beam Cursor The cursor is an I-beam when it is over text in the data grid or highlighted column names in the data grid or column panels. To edit text in the data grid: 1. 2. 3. 4. Click the cell you wish to edit. The cell highlights. Click again next to any character to mark an insertion point. The I-beam deposits a vertical blinking bar. Use the keyboard to make changes.

To edit a column name: 1. Click the column name to highlight the column. 2. Press the Enter key to change the I-beam cursor to an insertion point. 3. Use the keyboard to make changes. Large Cross Cursor The cursor becomes a large cross when moved into a column or row selection area. When moved over a column name, you can edit the name. To do so, click the column name and begin typing. The cross cursor can also be used to select rows and columns. To select a column, click the area above the column name. See the next section, Selecting Rows and Columns, p. 12, for a detailed explanation of selecting rows and columns. Double Arrow Cursor The cursor changes to a double-arrow cursor when positioned on a column boundary or on a panel splitter. Dragging the double-arrow cursor changes the column width or the panel size.

Drag to change width of column

12

1 Introducing JMPStep 3: Learn About the Data Table

Hand Cursor The cursor changes to a hand when you move the mouse over a red triangle icon ( ) or diamond-shaped disclosure button ( on Windows and on the Macintosh). Click the red triangle to reveal the menu and select a menu icon. Click the diamond icon to open or close a panel.

Selecting Rows and Columns

Select rows and columns in a JMP data grid by highlighting them, as explained in Figure 1.8 and shown in Figure 1.9. For additional details, see Selecting Rows and Columns, p. 92 of JMP User Guide. Figure 1.8 Ways to Select Rows and Columns
Action Instructions

Highlight a row Highlight a column Extend a selection of rows or columns Make a discontiguous selection

Click the space that contains the row number. Click the background area above the column name. Or, click the column name in the columns panels to the left of the data grid. Shift-click the first and last rows or columns of the desired range. Ctrl-click (Command-click on the Macintosh) the desired selections.

Figure 1.9 Select Rows and Columns selected column

selected rows

1 Introducing JMPStep 4: Select an Analysis

13 1 Preliminaries

Step 4: Select an Analysis

There are a variety of analyses available through the Analyze and Graph menus in the main menu. An alternate way to access these analyses is through toolbar buttons and selections in the JMP Starter window. Selecting an analysis in the Analyze or Graph menus produces graphs, charts, plots, and/or tables. For example, to see a histogram of columns in the data table you have open, select Analyze > Distribution. Then, complete the window and click OK.

Casting Columns Into Roles

After you select an analysis from the main menu, a window appears that asks you to cast columns into roles. For example, if you select Analyze > Fit Y by X from the main menu, you will see the window in Figure 1.10. Figure 1.10 Fit Y by X Window

The JMP analysis methods are like stages or platforms for variables to dramatize their values. Each analysis requires information about which variables play what roles in an analysis. The most typical variable roles are: Y, Response studied. Identifies a column as a response or dependent variable whose distribution is to be

X, Factor Identifies a column as an independent, classification, or explanatory variable whose values divide the rows into sample groups. Weight Freq
By

Identifies a numeric column whose values supply weights for each response.

Identifies a numeric column whose values assign a frequency to each row for the analysis. Identifies a column that is used to create a report consisting of separate analyses.

14

1 Introducing JMPStep 5: View the Output Report

Step 5: View the Output Report

After you have cast columns into their roles, JMP provides output reports that include graphics and text. For more detail than is presented below, see Working With Output Reports, p. 161 of JMP User Guide.

Graphs and Charts

JMP reports are usually filled with graphs, charts, plots, and other graphical displays that show your results. For example, if you select Analyze > Distribution and assign several columns the Y, Response role in the Distribution window that appears, you create a report containing a graphical display of each column assigned the Y, Response role. For the example shown in Figure 1.11, the Distribution command produces graphical displays which include: Histograms of both the brand and speed columns An outlier box plot of the continuous variable speed Figure 1.11 Distribution Histograms and Outlier Box Plot

Histogram Display Options

Outlier Box Plot

To enhance the default graphical displays that show your results, JMP provides options that you can add to them. These options are found by clicking the red triangle icon beside a report name. For example, the red triangle icon next to the histogram name lists available report options (Figure 1.12). For practice, try selecting different combinations of these options and watch the effect they have on the displays and reports.

1 Introducing JMPStep 5: View the Output Report

15 1 Preliminaries

Figure 1.12 Options for Nominal or Ordinal Variable in an Distribution Analysis

Statistical Tables and Text

In addition to the graphs, charts, plots, and other graphical displays in a report, JMP can also include text tables in a report. The kinds of tables given depend on whether a variable is continuous or categorical (ordinal or nominal). When you installed JMP, a folder named Sample Data was also installed. In that folder is a file named Typing Data.jmp. Using the Distribution command to produce a distribution report creates text tables for the speed and brand variables (Figure 1.13). In this example: For a continuous variable, JMP displays a Quantiles table and a Moments table. For nominal and ordinal variables, JMP displays a Frequency table showing the total sample frequency, category frequencies, and associated probabilities.

16

1 Introducing JMPStep 6: Save the JMP Output Report

Figure 1.13 Statistical Reports

JMP also gives you the ability to change the appearance of these tables. For details, see Formatting Report Tables in JMP User Guide.

Step 6: Save the JMP Output Report

To save a report just as it appears in the report window, select File > Save. Another way to save is to duplicate it in a separate window titled Journal. Then you can append other reports to it or manipulate it in the journal window. To do this, select Edit > Journal. Then, select File > Save As. For more detail than is presented below, see Saving Using the Journal Command, p. 109 of JMP User Guide.

1 Introducing JMPA Practice Tutorial

17 1 Preliminaries

Report window

Journal window

A Practice Tutorial
Before you begin the tutorials in the following chapters of this book, complete this brief practice tutorial that is a short guided tour through a JMP analysis. Follow the steps to see a three-dimensional spinning plot.

Open a Data Table

Open the file called Cowboy Hat.jmp to begin a JMP session. When you installed JMP, a folder named Sample Data was also installed. In that folder is a file named Cowboy Hat.jmp. The data table shown in Figure 1.14 appears.

18

1 Introducing JMPA Practice Tutorial

Figure 1.14 Cowboy Hat Data Table

This data table has three numeric columns and two row state columns. Columns x and y are x- and y-coordinates, and z is created using the function
z = sin x + y
2 2

Select an Analysis
To plot the three columns of information from the Cowboy Hat data table: Choose the Spinning Plot command from the Graph menu.

Select the x, y, and z columns from the column selector list on the left side of the window, and click Y, Columns, as shown in Figure 1.15. Figure 1.15 Spinning Plot Column Selection Window

1 Introducing JMPA Practice Tutorial

19 1 Preliminaries

These column names now appear in the list on the right side of the window. Click OK. The spinning plot appears. Initially, the data points look like a two-dimensional plot because the z dimension is projected onto the x-y plane.

Spin the Cowboy Hat

Select the hand tool from the tool bar.

Position the hand tool on the cowboy hat spinning plot, press the mouse button, and move the hand about. The cowboy hat moves in three dimensions.

To have the plot spin itself: Press the Shift key, and give the plot a push with the hand tool. To stop the spinning plot, click again in the spinning plot frame. You can also control the plots spin using the control panel icons. The six directional arrows control the spin of the three-dimensional image. The plot spins in the direction indicated by the arrow. The spin continues as long as you click and hold the arrow button. Shift-click an arrow to make the plot spin continuously. Click anywhere in the display to stop the spin.

20

1 Introducing JMPA Practice Tutorial

up/down (pitch) zoom in and out left/right (yaw)

increase/decrease spin speed return axes to home position

clockwise and counterclockwise (roll)

access the display options described in the next section

Experiment with the Display Options

The display options, shown here, tailor the spinning plot. By default, the background is black and the axes are visible. Change these conditions by clicking the red triangle icon and selecting an option. Each option switches on and off alternately when it is selected.

Complete the experiment with the spinning cowboy hat by trying out a few of the display options.

2
Creating a JMP Data Table
This lesson evaluates a new drug developed to lower blood pressure. Data were recorded over a six-month period for the following treatment groups: 300 mg dose 450 mg dose placebo control Following are the mean monthly blood pressure for each group, recorded in a journal. This lesson shows how to enter data values into the data table and to create a single neat and informative line chart that shows the study results. Objectives Create rows and columns in a data table, one at a time and in groups. Enter data into JMP. Create a chart using the Chart command. Rescale axes in a plot. Animate a plot.

2 Data T able

Blood Pressure Study Month March April May June July August Control 165 162 164 162 166 163 Placebo 163 159 158 161 158 158 300mg 166 165 161 158 160 157 450mg 168 163 153 151 148 150

2
Contents
Starting a JMP Session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating Rows and Columns in a JMP Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Add Columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Set Column Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Add Rows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Entering Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Plotting Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Document the Report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 23 24 25 25 26 26 29 30

2 Creating a JMP Data TableStarting a JMP Session

23 2 Data T able

Starting a JMP Session

Double-click the JMP application icon to begin a JMP session. Use the New command in the File menu to create an empty data table like the one shown here.

The data values for this project are blood pressure statistics collected over six months and recorded in a notebook page as shown in Figure 2.1. Figure 2.1 Notebook of Raw Study Data

Blood Pressure Study Month March April May June July August Control 165 162 164 162 166 163 Placebo 163 159 158 161 158 158 300mg 166 165 161 158 160 157 450mg 168 163 153 151 148 150

Creating Rows and Columns in a JMP Data Table

JMP data tables have rows and columns, sometimes called observations and variables in statistical terms. The raw data in Figure 2.1 are arranged as five columns (treatment groups) and six rows (months March through August). The first line in the notebook names each column of values. These names can be used as column names in the new JMP table.

24

2 Creating a JMP Data TableCreating Rows and Columns in a JMP Data Table

Add Columns
First create the number of rows and columns that are needed. Select Cols > Add Multiple Columns, which prompts for the number of columns to add, where to add them, and which type of columns to add.

Specify five new columns. The default column names are Column 1, Column 2, and so on, but these can be changed at any time in the columns panel or at the top of the column in the data table. So, the next step is to type in meaningful names. Use the names from the data journalMonth, Control, Placebo, 300 mg, and 450 mg. Edit a column name by clicking it in the data grid or in the columns panel. Begin typing once the name is highlighted.

Click once to select

Then, begin typing

2 Creating a JMP Data TableCreating Rows and Columns in a JMP Data Table

25 2 Data T able

Set Column Characteristics

Columns can have different characteristics. By default, they contain numeric data. However, the values for month in this example are character values. To change the Month column from numeric to character: Highlight the column by clicking either in the area at the top of the column or the area beside its name in the columns panel. Select the Cols > Column Info to display the window in Figure 2.2. Change Month to a character variable, as shown in Figure 2.2, by clicking the box beside Data Type. The Column Info window is also used to change other column characteristics and to access the JMP formula editor for computing column values. Figure 2.2 Change Data Type

Add Rows
Now add new rows to the table: Choose Rows > Add Rows. Specify six new rows. Alternatively, double-click anywhere in the body of the table to automatically fill it with new rows up through the position of the cursor. Select File > Save to name the table BP Study.jmp and save it. The data table is now ready to hold data values. To summarize the table evolution so far, you: Began with a new untitled table. Added enough rows and columns to accommodate the raw data. Tailored the characteristics of the table by giving the table and columns descriptive names. Changed the data type of the Month column to accept character values.

26

2 Creating a JMP Data TableEntering Data

Entering Data
To enter data into the data table, type values into their appropriate table cells. Type the values from the study journal (Figure 2.1) into the BP Study.jmp table as shown here. When entering data into the data table: Edit the cell value by moving the cursor into a data cell and double-clicking. The cursor becomes a flashing vertical bar. Correct a mistake by dragging the text entry bar across the incorrect entry and typing the correction over it. Press the Return or Enter key on the numeric keypad or the Tab key to move the highlight one cell to the right. Press Shift-Tab or Shift-Return/Enter on the numeric keypad to move the highlight one cell to the left.

Click to highlight

Begin typing

The completed data table

Plotting Data
When working with the Analyze and Graph menu commands, you tell JMP which columns to work with and what to do with them. This section shows how to plot the months across the horizontal (x) axis and the columns of blood pressure statistics for each treatment group overlaid on the vertical (y) axis. Select Graph > Chart.

2 Creating a JMP Data TableEntering Data

27 2 Data T able

The window in Figure 2.3 appears. Assign x and y roles and choose the type of chart. This example specification is for a bar chart, with data (as opposed to statistics) as chart points. Assure that the default choice Vertical is selected from the chart type drop-down list. Select the four continuous variables in the Select Columns list. Click the Statistics button and select Data from the drop-down list. Enter Month as the Categories (X). Click OK. Figure 2.3 Creating the Bar Chart

Choose chart type

Choose y-axis information

JMP displays an overlaid bar chart of the data.

Reorder the Values By default, the month values in this chart do not appear in a logical order. To further enhance the report, reorder the months so they are in chronological order rather than alphabetical order.

28

2 Creating a JMP Data TableEntering Data

In the data table, highlight the Month column and select Cols > Column Info. Select Value Ordering from the Column Properties drop-down menu. Value ordering information appears on the right, as shown in Figure 2.4. The list in the window contains values in the order JMP will display them in reports. Use the Move Up and Move Down buttons to change the order of the months. Figure 2.4 The Value Ordering Window

Click OK. The properties icon ( ) now appears next to the column name in the data tables column panel, indicating the column contains a property. In the analysis report, click the red triangle and select Script > Redo Analysis. Rescale the Plot Axis By default, y-axis scaling begins at zero and the overlay chart looks like the one shown here. But, to present easy-to-read information, the y-axis needs to be rescaled and the chart needs labels. Double-click the y-axis area, which accesses the Axis Specification window (Figure 2.5). This window gives you the ability to: Set the minimum and maximum of the axis scale. Specify the tick mark increment. Request minor tick marks. Request grid lines at major or minor tick marks. Format numeric axes. Use either a linear or log-based scale.

2 Creating a JMP Data TableEntering Data

29 2 Data T able

Figure 2.5 Axis Specification Window

In this example, the plotted values range from about 145 to 175. Enter these figures into the Axis Specification window for Minimum and Maximum. Also, the increment for the tick marks needs to change from 50 to 1. Enter a 1 in the Increment box and click OK. Tip: The magnifier tool ( ), found in the Tools menu and the cursor tool bar, can also be used to change the scale of graphs. Drag the magnifier diagonally across the points of interest to see the chart automatically adjust. Double-click the plot frame to reset the plot to its original scale. Click the edge of the graph and drag is to the right to increase its width. Change the name of the axis from Y to Blood Pressure: Click the axis name in the report. Type the new name in the resulting edit box.

Document the Report

The chart also needs a title and other documentation to make it easy to interpret. The annotate tool ( ) places text on the report. Refer to Figure 2.6 see where to place the following steps. Select the annotate tool. Click and drag in the report to create a text box. Release the button and enter the text for the title Comparison of Treatment Groups. Click outside the annotation to quit editing the text.

30

2 Creating a JMP Data TableChapter Summary

Repeat to enter the footnote XYZ Blood Pressure Study 2005. Note: Double-click any report title bar to edit the text on the bar. Figure 2.6 Line Chart with Modified Y-Axis, Titles, and Footnotes

Chapter Summary
A study was done to evaluate the effect of a new drug on blood pressure. To complete this analysis, you: Used the New Data Table command in the File menu to create a new JMP table. Created the appropriate number of rows and columns for the data. Typed the data into the empty data grid. Used the Chart command in the Graph menu to request a bar chart of blood pressure measures over time. Ordered the values in chronological order so they would appear properly in the chart. Tailored the chart with a specific axis scale and axis name, and added a plot title and footnote with the annotate tool.

3
Summarizing Data
The hot dog is a questionable item on a school cafeteria menu because of its reputation as an unhealthy food, possibly classified in the junk food category. Many students feel this is unpatriotic and are upset. This lesson examines the hot dog as a menu item, but not before looking into the multitude of brands available. The data shows information about cost, nutritional ingredients of concern, and taste preference for 54 hot dog brands. This information is sufficient to provide a summary of hot dog statistics and to identify the brands that are: most nutritious least costly best tasting The taste, cost, and nutritional variables used in this chapter are an enhancement of data from Moore, D. S., and McCabe G. P., (1989), Introduction to the Practice of Statistics, and Consumer Reports (1986). The brand names were changed to fictional names, and the taste preference labels correspond to a taste preference scale. Objectives Find and mark subgroups of data Produce scatterplots using the Fit Y by X command and use them as discovery tools Label individual points in plots Produce and plot summary statistics

3 Summarizing

3
Contents
Look Before You Leap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Grouping Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating Statistics for Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Charting Statistics from Grouped Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Charting Statistics for Two Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Finding a Subgroup with Multiple Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparative Scatterplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . What Has Been Discovered? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Finding the Best Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 34 35 36 38 39 40 41 42 43

3 Summarizing DataLook Before You Leap

33 3 Summarizing

Look Before You Leap

When you installed JMP, a folder named Sample Data was also installed. In that folder is a file named Hot Dogs.jmp. Open the Hot Dogs.jmp file to see the data shown in Figure 3.1. Figure 3.1 Hot Dogs Data Table

The Hot dogs.jmp table has the following information: The columns called Type, Calories, Sodium, and Protein/Fat (an index ratio of protein to fat) give information about nutrition. The Type column has values Meat, Poultry, and Beef. Cost information is in columns $/oz (dollars per ounce of hot dog) and $/lb Protein (dollars per pound of hot dog protein). Three categories of taste are coded Bland, Medium, and Scrumptious in the Taste column. To get a feel for these data, use the Distribution command. Select Analyze > Distribution to see the window in Figure 3.2. Select all the variables except Product Name and click the Y, Columns button. Click OK. Figure 3.2 Distribution Command

Examine the resulting report to see the distributions and levels of each variable.

34

3 Summarizing DataGrouping Data

Grouping Data
Of course, health is a primary concern of a school cafeteria. It is interesting to see if type of hot dog plays a role in healthfulness. In particular: Which type of hot dog has the fewest calories? Is the amount of sodium different in the three types of hot dog? Which hot dogs have the highest protein content? Which hot dogs taste good and are healthy? To address these issues, the data need to be grouped into hot dog type and taste preference categories with summary statistics computed for each group. The Summary command in the Tables menu groups data and computing summary statistics. The Summary command creates a summary table. This table summarizes columns from the active data table, called its source table. The Hot Dogs.jmp table is the source table in this example. A summary table has a single row for each level (value) of a specified variable. Select Tables > Summary. Select Type and click the Group button to see the window as shown in Figure 3.3. Click OK. Figure 3.3 Summary Window

3 Summarizing DataGrouping Data

35 3 Summarizing

The Hot Dogs By (Type) summary table (Figure 3.4) appears in a new window. The Type column lists hot dog type and the NRows column gives the frequency of each type in the source table. A summary table is not independent of its source table. It has these characteristics: When rows are highlighted in the summary table, their corresponding rows highlight in the source table. The summary table is not saved when closed. Select File > Save As to specify a name and location for the table. Figure 3.4 Summary Table for Type of Hot Dog

Creating Statistics for Groups

Next, expand the summary table with columns of statistics. Summary tables have an additional command in the columns panel called Add Statistics Column. Use this command to add statistical summary columns to the table at any time.

To follow along with this example, do these steps: Click the red triangle icon and select the Add Statistics Column command from the columns panel on the left side of the screen. Select Calories, Sodium, and Protein/Fat in the column selector list of the window, as shown in Figure 3.5. Click the Statistics button and select Mean. You should now see new column names in the Statistics list. Click OK.

36

3 Summarizing DataGrouping Data

Figure 3.5 Summary Window

The new columns of statistics are displayed in the Hot Dogs By (Type) table (top table of Figure 3.6). Repeat the previous steps to create a second summary table of Hot Dogs by Taste to look at health factors and hot dog tastiness. The Hot Dogs By (Taste) summary table shows average calories, sodium, and protein-to-fat ratio for each taste category (bottom table of Figure 3.6). Figure 3.6 Summary Statistics for Hot Dog Groups

Charting Statistics from Grouped Data

The summary tables in Figure 3.6 show the summary statistics in tabular form, but bar charts are better for visual comparison. The Chart command on the Graph menu can also summarize data and then create charts of the summarized data. Make sure the Hot Dogs.jmp table is active. Select Graph > Chart.

3 Summarizing DataGrouping Data

37 3 Summarizing

Assign variable roles as shown in Figure 3.7. Charts like those below should appear. Figure 3.7 Charting Data

Click to choose a statistic

Remove
Overlay

checkmark Change to
Horizontal

Click to add highlighted columns as the x-variable

38

3 Summarizing DataGrouping Data

It appears that poultry hot dogs have fewer calories on average than the other two hot dog types. Also note that the poultry hot dogs have slightly more sodium. The most visible difference is that the protein-to-fat ratio appears much higher in poultry hot dogs. Repeat the above steps using taste to produce another set of bar charts. It may not be surprising to see from the bar charts that hot dogs rated as bland tasting have (on average) more calories, more sodium, and a lower protein-to-fat ratio. It can be seen from the data table that scrumptious tasting hot dogs have the lowest average calories and sodium content. However, medium tasting hot dogs have the highest protein-to-fat ratio, and they compare well with respect to the other nutritional factors.

Charting Statistics for Two Groups

Next, it is useful to know the frequency of the three taste responses for each type of hot dog. Make sure the Hot Dogs.jmp table is active and use Tables > Summary again and select both Type and Taste as grouping variables. This produces the table shown here. There is one row for each taste response within each type of hot dog. The Nrows column lists the frequency in the source table of each type-taste combination.

3 Summarizing DataFinding a Subgroup with Multiple Characteristics

39 3 Summarizing

Select the Graph > Chart command, with both grouping variables as Categories (X) and the Nrows column with the y (statistics: data) role. This produces the chart shown here. In this example, there are side-by-side charts that show the frequency for each taste within each type of hot dog. Note: Graph > Chart can also be used to directly chart data grouped by two variables; the data dont have to be grouped first by Tables > Summary. To label each bar with the frequency it represents: Right-click the bars and select Label > Label by Value. The chart shows that the poultry hot dogs excelled in nutrition factors and that most people find them medium-tasting. However, because the sodium content appears slightly high in some poultry brands, more investigation is needed.

Finding a Subgroup with Multiple Characteristics

Continue the search for the ideal hot dog. Add special markers to the summary table, Hot Dogs by (Type, Taste), that identify each type of hot dog. In the Hot Dogs by (Type, Taste) summary table: Shift-click or click and drag over the medium and scrumptious beef rows (2 and 3) to select them.

40

3 Summarizing DataFinding a Subgroup with Multiple Characteristics

Use the Markers command in the Rows menu to assign them the Z marker. Deselect those rows. Shift-click or drag the medium and scrumptious meat rows (5 and 6) to select them, assign them the Y marker, and deselect them. Shift-click or drag the medium and scrumptious poultry rows (8 and 9), assign them the X marker, and deselect them. The type-taste summary table now looks like the one shown here, and the corresponding rows in the
Hot Dogs.jmp table are marked likewise.

Comparative Scatterplots
Now, examine the relevant variables with scatterplots to identify specific points (brands). The Fit Y by X command in the Analyze menu produces scatterplots when both the x and y are continuous numeric variables. The following scatterplots graphically show the relationship of cost and the nutritional factors together. Click the Hot Dogs.jmp source table to make it active. Select Analyze > Fit Y by X. Make your selections in the window, giving $/lb Protein the y role and both $/oz and Protein/Fat the x role. Click OK. This produces $/lb Protein by $/oz and a $/lb Protein by Protein/Fat scatterplots. Click the red triangle icon and select Group By. Choose Type as the grouping variable from the list of variables in the Grouping window. Repeat this action for the $/lb Protein by Protein/Fat scatterplot. For both plots, click the red triangle icon and choose Density Ellipses > .90 to make a density ellipse visible. Repeat to complete a similar Fit Y by X analysis with Calories as y and Sodium as x.

3 Summarizing DataFinding a Subgroup with Multiple Characteristics

41 3 Summarizing

These commands produce the $/lb Protein by $/oz, the $/lb Protein by Protein/Fat, and the Calories by Sodium scatterplots shown in Figure 3.8 and Figure 3.9. The 0.90 ellipses in the scatterplots show the shape of the bivariate response for each type of hot dog. The special markers identify the taste and type of each point. Figure 3.8 Scatterplots Comparing Cost, Taste, and Nutritional Factors

To further identify and highlight points of interest: Select the brush tool ( ) from the Tools menu. Press the Alt key (Alt-shift on Linux and Option on Macintosh) and drag the brush in the lower left quadrant of the Calories by Sodium scatterplot, as shown in Figure 3.9. These points represent brands with both low sodium and low calories. The highlighted points of these healthiest brands also highlight in the other scatterplots. Figure 3.9 Select Low Sodium and Low Calorie Brands

What Has Been Discovered?

The costs of meat and beef brands range from low to high. However, it is not surprising to see the tight low-cost cluster of poultry brands (X-marked) at the lower left of the $/lb Protein by $/oz scatterplot. The highlighted points include poultry brands, one meat brand, and one beef brand. The selected beef point (Z-marked) is in the upper-right corner of the plot, which places it in the most expensive cate-

42

3 Summarizing DataFinding a Subgroup with Multiple Characteristics

gory. The single meat point (Y-marked) is more costly than the poultry brands but less than the beef brands. A bigger surprise appears in the $/lb Protein by Protein/Fat scatterplot. As the protein-to-fat ratio increases, the cost per pound of protein stays about the same. Further, the poultry brands not only cost the least but also contain the most protein. Most of the selected points are in the three highest protein categories. The density ellipses on the Calories by Sodium scatterplot show clearly that the poultry brands have about the same range of sodium content as the meat and beef brands, but many poultry brands have fewer calories.

Finding the Best Points

Now there is sufficient information to identify several hot dog brands as possible cafeteria menu items. Click the Hot Dogs.jmp table to make it active. Note that the icon beside the Product Name column in the Column panel means that is has been designated as a Label column. The poultry (X-marked) brands are acceptably economical, and some of them have high protein content. Few meat or beef brands compared well. Select the Arrow tool from the Tools menu. Click inside the Calories by Sodium scatterplot to deselect all points. Shift-click to highlight the two poultry brands (X-marked) with the least calories and lowest sodium content. Again Shift-click to highlight the lone meat point (Y-marked) that has the least sodium of all brands, is low in calories, has a moderate protein count, and is average in price. Select Rows > Label/Unlabel to display the brand names of highlighted points (Thin Jack Veal, Calorie-less Turkey, and Estate Chicken), as shown in Figure 3.10. Figure 3.10 Labeling Ideal Points

As a final step, use Analyze > Fit Y by X to look again at the two scatterplots that compare costs.

3 Summarizing DataChapter Summary

43 3 Summarizing

Select Analyze > Fit Y by X. Assign $/lb Protein as Y. Assign both $/oz and Protein/Fat as X. Click OK. The plot to the left in Figure 3.11 shows that the Estate Chicken brand is the most economical of the three labeled brands (showing $/oz as continuous). The plot to the right indicates that the Calorie-less Turkey brand is in the group with the highest proportion of protein (showing Protein/Fat as nominal). Figure 3.11 Winning Hot Dog Brands.

Chapter Summary
This lesson examined different hot dog brands for a cafeteria menu. A JMP table has data for 54 brands of hot dog showing type of hot dog, taste preference, nutritional factors, and cost factors. To find the ideal hot dog, we: Created a summary table that group the data by hot dog type and by taste preference within each hot dog type. Used Graph > Chart to chart summary statistics and identify the subset of hot dog brands that are both the most nutritious and the best tasting. Assigned different markers to each type of hot dog. Used Analyze > Fit Y by X to see scatterplots that compare cost factors and nutritional factors. Selected the points representing the lowest cost, most nutritious, and used the Label/Unlabel command in the Rows menu to identify the Calorie-less Turkey brand as a possible cafeteria hot dog. See Summarizing Columns in JMP User Guide for details about the Summary command. The chapters Bivariate Scatterplot and Fitting and Bar, Line, and Pie Charts in the JMP Statistics and Graphics Guide show scatterplot and bar chart examples.

4
Looking at Distributions
The students in a local school are participating in a health study. This lesson summarizes basic information about the students for the school systems health care specialists. The data collected include age, sex, weight, and height. To document the sample of participating students and identify any students with unusual characteristics who may need special attention, we will need to view summaries of the data. This lesson produces reports with graphs and short, straightforward explanations. Objectives Use the distribution analysis to explore several variables at once. Produce reports of moments, quantiles, frequencies, and proportions. Use the formula editor to compute a columns value. Create a subset of a data table.

4 Distributions

4
Contents
Look Before You Leap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Displaying Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding Histograms of Nominal and Ordinal Variables . . . . . . . . . . . . . . . . . . . . . . . . Understanding Histograms of Continuous Values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Learning About Report Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Reports for Continuous Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Frequency Table for Ordinal or Nominal Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Adding a Computed Column . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Creating Subsets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 47 48 49 51 52 52 52 55 56

4 Looking at DistributionsLook Before You Leap

47 4 Distributions

Look Before You Leap

The first step in this analysis is to become familiar with the data in the Big Class.jmp file. Looking at the information in the JMP data table helps us decide which summary charts and tables to use in the health report. Open the Big Class.jmp data table to see the data table shown in Figure 4.1. Figure 4.1 Big Class.jmp Data Table

The file contains the name, age, sex, height, and weight for each student participating in the health study. The data table is in order by age, and sex is ordered within each age group. Even though there are only five columns of information, these variables address the following questions: How many boys and how many girls are there? How old are they? What is the average height and weight of the students? Are there any students drastically younger or older than the average age? Are there any students whose height or weight might signal the need for special attention?

Displaying Distributions
To summarize the data: Select the Distribution command from the Analyze menu.

48

4 Looking at DistributionsDisplaying Distributions

In the window that appears, select the age and sex columns as Y, Columns.

Click OK. The frequencies table that appears shows that the class of 40 contains 18 girls and 22 boys.

Understanding Histograms of Nominal and Ordinal Variables

After selecting Analyze > Distribution and completing the window, you will see a window that displays histograms for analysis variables. The histogram for ordinal or nominal variables like age and sex has a bar for each level (value) of the variable. Figure 4.2 Histogram of the age and sex Variables

4 Looking at DistributionsDisplaying Distributions

49 4 Distributions

Displaying Mosaic Plots In histograms of ordinal and nominal variables, you can display a mosaic plot by clicking the red triangle icon in the variables title bar and selecting Mosiac Plot. A mosaic plot (shown in the figure to the right) visualizes the proportion of each ordinal or nominal level within the sample. It has a section for each level of the variable, where the size of the section is proportional to the corresponding groups size. Think of a mosaic plot as a bar chart with its bars stacked end to end.

Understanding Histograms of Continuous Values

If your variables are continuous, the histogram look slightly different than if the variables are nominal or ordinal. To see the difference, look at the distribution of the continuous variables height and weight in the sample of students. Click the Big Class.jmp data table to make it the active window. Again choose the Distribution command from the Analyze menu. Designate the height and weight columns as Y, Column variables. Click OK. Histograms are displayed for height and weight, as shown in Figure 4.3. Figure 4.3 Histograms of height and weight

Both height and weight appear to have approximately normal (bell-shaped) distributions, but notice the extremely high weight value. It will be examined more closely later. It is important to present data in the best possible form. Sometimes it is worthwhile to experiment with the shape of a histogram by changing the number of bars or altering their arrangement on the axis. To adjust the histogram bars: Select the hand from the graph cursor tool bar.

50

4 Looking at DistributionsDisplaying Distributions

Position the hand on the bars and press the mouse button to grab the plot. Move the hand to the left to increase the bar width and combines intervals (see Graphs and Charts, p. 14). The number of bars decreases as the bar size increases. Move the hand to the right to decrease the bar width, showing more bars. Move the hand up or down to change the boundaries of the bins. The height of each bar adjusts according to the new number of observations within each bin. Using Outlier Box Plots Available by default in histograms with continuous variables, the outlier box plot (see Figure 4.4) is a schematic that shows the sample distribution and allows identification of points with extreme values, sometimes called outliers. You can display and hide an outlier box plot by clicking the red triangle icon in the variables title bar and selecting Outlier Box Plot. The ends of the box are the 25th and 75th quantiles, also called the quartiles. The difference between the quartiles is the interquartile range. The line across the middle of the box identifies the median sample value. The lines extending from each end of the box are sometimes called whiskers. The whiskers extend from the ends of the box to the outermost data points that fall within the distance computed as quartile 1.5*(interquartile range). Points beyond the whiskers indicate extreme values that are possible outliers. To label a point, click the point to highlight it, and then select Rows > Label/Unlabel. The bracket along the edge of the box identifies the shortest half, which is the most dense 50% of the observations. Figure 4.4 Outlier Box Plot 25th percentile Interquartile range 75th percentile

possible outliers

shortest half

Using Quantile Box Plots In histograms whose variables are continuous, you display a quantile box plot by clicking the triangle icon in the variables title bar and selecting Quantile Box Plot. A quantile box plot shows the location of preselected percentiles, sometimes called quantiles, on the response axis. The median shows as a line in the body of the box. The ends of the box locate the 25th and 75th quantiles. The number of other quantile lines depends on the available space. The accompanying text report lists the data values for each of the standard quantiles.The box also contains a means

4 Looking at DistributionsLearning About Report Tables

51 4 Distributions

diamond. The two diamond points within the box identify the 95% confidence interval of the mean. The line that passes through the two diamond points spanning the box identifies the sample mean. Looking at the quantile box plot and means rectangle together helps see if data are distributed normally, as shown in Figure 4.5. If data are distributed normally (bell shaped), then the 50th quantile and the mean are the same and other quantiles show symmetrically above and below them. Figure 4.5 Quantile Box Plot and Quantiles Table

means diamond

Learning About Report Tables

JMP produces tables of statistical summaries along with graphs. The tables that JMP produces depend on whether a variable is continuous, ordinal, or nominal. Click the red triangle icon and select Display Options to reveal tables that are not available by default. The diamond-shaped disclosure button ( on Windows and Linux and at the top-left of each report opens and closes it. on the Macintosh)

52

4 Looking at DistributionsAdding a Computed Column

Reports for Continuous Variables

The figure to the right is the table JMP gives along with histograms for the continuous variable height. The Quantiles table displays the maximum value, minimum value, and other values for selected quantiles. The Moments table displays the mean, standard deviation, and other summary statistics. Note: You can also obtain more moments by clicking the red triangle icon on the variables title bar and choosing Display Options > More Moments.

Frequency Table for Ordinal or Nominal Variables

The report for nominal and ordinal variables has a different table from those produced for continuous variables. The table JMP gives along with histograms for the nominal or ordinal (categorical) variables sex and age has frequency tables that show these items: Level lists each value of the response variable. Count lists the number of rows found for each level of a response variable. Prob lists the probability of occurrence for each level of a response variable. The probability is computed as the count divided by the total frequency of the variable, shown at the bottom of the table. The following two statistics are not displayed by default. Right-click (Ctrl-click on the Macintosh) the table and select the Columns menu to reveal them. StdErr Prob lists the standard error of the probabilities. Cum Prob contains the cumulative sum of the column of probabilities.

Adding a Computed Column

Be on the alert for any unusual subjects, such as students who have extreme height or weight values. A good indicator of extreme values is the ratio of weight to height. To examine the ratio of weight to height, create a new column called ratio, computed as weight divided by height. To do this:

4 Looking at DistributionsAdding a Computed Column

53 4 Distributions

Click the Big Class.jmp data table to make it the active window. Select Cols > New Column. To create the new column of weight-to-height ratios, complete the New Column window as in Figure 4.6. Type the new name, ratio, in the Column Name area. The default data type is Numeric and is correct as is. The modeling type is Continuous and is correct as is. Click the drop-down menu beside Format and set the format for ratio in the data grid to Fixed Dec with two decimal places. Click the New Property button and select Formula, as shown in Figure 4.6, to compute values for the new column. Figure 4.6 New Column Window

Construct the formula that calculates values for the ratio column as follows: Highlight the empty term in the formula and select weight from the list of column names in the upper-left corner of the formula editor. Press the divide () key on the formula editor keypad. With the empty denominator term highlighted, select height from the list of column names.

54

4 Looking at DistributionsAdding a Computed Column

When the formula is complete, click Apply or OK on the formula editor, or just close its window. The new column called ratio is now in the Big Class data table as shown here. Its values are the computed weight-to-height ratio for each student.

Now look at the distribution of the ratio variable. Select Analyze > Distribution and assign the new column (ratio) to the Y, Columns role. Click OK. One way to identify subjects that have extreme values is to highlight histogram bars for the highest and lowest values. To highlight more than one bar, press the Shift key and click the desired bars. Select the bars for the two lowest and one highest bar (Figure 4.7).

4 Looking at DistributionsAdding a Computed Column

55 4 Distributions

Figure 4.7 Histogram of Ratio with Bars Highlighted at Extreme Values

The highlighted bars in the histogram represent a ratio either greater than or equal to 3.5 or less than 1.5. The corresponding points automatically highlight in the data table and in all other reports generated from the Big Class data table.

Creating Subsets
Looking in the Big Class data table allows examination of the selected rows, but scrolling through a large data grid can be tedious. For the final report to the health researchers, include a separate list containing only the highlighted studentsthose with extreme values. To do this, use Tables menu commands to create new data tables or modify existing tables. Select Tables > Subset or click the red triangle icon in the tables panel and select Subset. Click OK to accept the default choices presented in the window. This creates a new data table that has only the selected rows and columns from the active data table. The new data table, shown in Figure 4.8, contains only the students that have extreme weight-to-height ratios. By default, the table is named Subset of Big Class. Change the name by clicking the existing name (Subset of Big Class) in the panel located on the top left side of the window. The table can be saved, exported for use in another application, or printed.

56

4 Looking at DistributionsChapter Summary

Figure 4.8 Data Table Containing a Selected Subset

Chapter Summary
In this chapter, the demographic and vital data of students participating in a health study were summarized. The profile was completed using the Distribution command and the data management features of the JMP data table. The Distribution command displayed histograms and box plots or stacked bar (mosaic) bar charts for each variable assigned the role of response variable (y). Using display and text report options to look more closely at the data, the following actions were completed: Adjusted the number of bars and the scale of the histograms Produced supporting statistical reports showing moments and quantiles of numeric variables and frequencies and proportions of nominal and ordinal variables Created a new column in the data table computed as a function of existing columns Highlighted histogram bars to identify a subset of rows in the data table Created a new data table from a subset of highlighted rows Graphs and text reports can be printed directly from JMP. Graphs and reports can be copied to a JMP journal or into other applications to complete a report for the school system health care specialists. See the chapter Univariate Analysis in the JMP Statistics and Graphics Guide for more information about distributions.

5
Comparing Group Means
In keeping with a recent corporate policy to modernize operations, all the typewriters in the typing pool are to be replaced with modern word processors. The typists are eager for this change and willingly participated in a study to help decide what kind of equipment to buy. The company selected three different brands of machine to test. These machines were randomly assigned to three groups of typists with comparable typing skills. The typists completed typing tests and recorded their words-per-minute scores. This lesson finds out if the typing scores are significantly better on any one brand of machine than on the others. Objectives Use the Fit Y by X command to produce plots and analyses appropriate for a one-way analysis of variance. Use JMPs interactive capabilities to examine differences among groups. Produce text reports to display differences among groups.

5 Group Means

5
Contents
Look Before You Leap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Graphical Display of Grouped Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Choose Variable Roles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Show Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fit Means Option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fit Quantiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Comparison Circles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Quantify Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Analysis of Variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mean Estimates and Statistical Comparisons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 59 60 60 61 63 64 65 66 66 67

5 Comparing Group MeansLook Before You Leap

59 5 Group Means

Look Before You Leap

The first step is to become familiar with the data. The typing test scores are in a JMP file so that they can be reviewed and the kind of analysis determined. When you installed JMP, a folder named Sample Data was also installed. In that folder is a file named Typing Data.jmp. Open the file Typing Data.jmp The Typing Data table appears in the form of a data grid, as shown here.

The data table has columns named brand and speed. The modeling type for each column shows to the left of each column name in the columns panel. The character variable brand has nominal ( ) values and the numeric variable speed has continuous ( ) values. There are 17 rows that represent typing scores for 17 typists. However, the number of participants in the groups differs because some of the scheduled participants did not show up for the study. Perhaps other statistics for the groups differ also. In particular: Is the mean (average) typing speed the same for each brand? Do any one of the three brands of word processor stand out from the others? Does it make a difference as to which brand the typists use?

Graphical Display of Grouped Data

Comparing the mean typing scores of each word processor brand involves analyzing two variables, so use the Fit Y by X command from the Analyze menu. Selecting Fit Y by X allows you to perform: Categorical analysis when both x and y have nominal or ordinal values Analysis of variance when x is nominal or ordinal and y has continuous values, as in the example shown here Logistic regression when x is continuous and y has nominal or nominal or ordinal values Regression analysis when both x and y have continuous values.

60

5 Comparing Group MeansGraphical Display of Grouped Data

Choose Variable Roles

To discover if typing speed is related to (dependent on) a brand of word processor, follow these steps: Choose Analyze > Fit Y by X. Select brand as X, Factor and speed as Y, Response (see Figure 5.1). Click OK. The plot shown in Figure 5.2 appears. Figure 5.1 The Fit Y by X Window

Selecting Fit Y by X and completing the window produces a statistical analysis appropriate for the variable roles (x and y) and the modeling type (continuous and nominal or ordinal) of each variable. Y, Response identifies a response (dependent) variable. X, Factor identifies a classification (independent) variable. The next step is to choose an analysis that investigates if there is a statistical difference between the group mean values.

Show Points
Each of the typing test scores is plotted for each brand of word processor. Note that the distance between tick marks on the brand axis is proportional to the sample size of each group. The mean typing score for the total sample is shown as a horizontal line across the plot.

5 Comparing Group MeansGraphical Display of Grouped Data

61 5 Group Means

Figure 5.2 Oneway Analysis for scores by brand

It is easy to see at a glance that most participants who used the SPEEDYTYPE machines typed faster than the others.

Fit Means Option

Now look at more graphical information about the distribution of typing scores. Click the red triangle icon on the title bar and select Means/Anova. This produces the appropriate analysis of variance reports. It automatically turns on mean diamonds, which draws a 95% means diamond for each group, as shown in Figure 5.3. You can also turn the Mean Diamonds option on and off from the Display Options submenu.

62

5 Comparing Group MeansGraphical Display of Grouped Data

Figure 5.3 Example of Means Diamonds

The illustration in Figure 5.4 illustrates the means diamond. The means diamond has a line drawn at the mean (average) value of words-per-minute for each brand of word processor. The upper and lower points of the means diamond span a 95% confidence interval computed from the sample values for each machine. The width of each diamond spans the distance on the horizontal axis proportional to the group size. Overlap lines within each diamond are drawn at ( 2 2 )CI . For groups with equal sample sizes, the marks that appear not to overlap indicate that two group means could be significant at the 95% confidence interval. Figure 5.4 Means Diamond with X Axis Proportional Option Turned On (Left) and Off (Right) overlap marks

95% confidence interval

group mean

total response sample mean

The mean scores of the REGAL and WORD-O-MATIC word processors appear to be nearly the same, but note that the SPEEDYTYPE mean is much higher (Figure 5.3).

5 Comparing Group MeansGraphical Display of Grouped Data

63 5 Group Means

Fit Quantiles
The next logical step is to check the distribution of points within each group. This gives a better idea of the spread of the values and shows the distance of extreme values from the center of the data. Click the red triangle icon and select Quantiles. When you select the Quantiles command, JMP automatically overlays a quantile box plot on each group of typing scores, as shown in Figure 5.5. JMP also displays the report in Figure 5.5, which lists the standard percentiles for each word processor. The median (50th percentile) is the typing speed that divides the sample in half. This means that 50% of the typists had speeds greater than the median, and the other half had lower speeds. Figure 5.5 Fit Quantiles Option

Figure 5.6 illustrates the quantile box plot. The median, or 50th quantile, shows as a line in the body of the box. The top and bottom of the box represent the 75th and 25th quantiles, also called the upper and lower quantiles. The box encompasses the interquantile range of the sample data. The 10th and 90th quantiles show as lines above and below each box. Looking at the quantile box plot and the means diamond together helps show if data are distributed normally within a group. If data are normally distributed (bell shaped), the 50th percentile and the mean are the same and the other quantiles are arranged symmetrically above and below the median.

64

5 Comparing Group MeansGraphical Display of Grouped Data

Figure 5.6 Quantiles Box Plot 90th percentile 75th percentile 50th percentile 25th percentile 10th percentile

sample mean group mean

The quantile box plots (Figure 5.5) show a difference in variation of scores across the three groups. The scores in the REGAL group cluster tightly around the mean score but the WORD-O-MATIC scores show much more variation. However, even with this variation among the groups, the SPEEDYTYPE brand still appears to promote the best performance.

Comparison Circles
To complete the typing data inspection: Click the red triangle icon and choose
Compare Means > All Pairs, Tukey HSD.

This option produces statistical reports (discussed later) and automatically draws a set of comparison circles to the right of the plot that provides a graphical test of whether the mean typing scores are statistically different. Comparison circles for the three word-processor groups are shown in Figure 5.7. The center of each circle is aligned with the mean of the group it represents. For the Students t-test, the diameter of each circle spans the 95% confidence interval for each group. Whenever two circles intersect, the confidence intervals of the two means overlap, suggesting that the means may not be significantly different. Whenever two circles do not intersect, the group means they represent are significantly different. Click the SPEEDYTYPE comparison circle. This graphically illustrates that the SPEEDYTYPE machine is statistically better than the other machines. The comparison circles highlight to show the statistical magnitude of the difference between typing scores. Circles for groups that are statistically the same have the same color.

5 Comparing Group MeansQuantify Results

65 5 Group Means

Figure 5.7 Comparison Circles

The comparison circle for the SPEEDYTYPE brand does not intersect with either of the other two. The REGAL and WORD-O-MATIC brands are statistically slower than SPEEDYTYPE but do not appear different from each other. A later section, Mean Estimates and Statistical Comparisons, p. 66, discusses the multiple comparison tests the comparison circles represent.

Quantify Results
Now, examine the report beneath the plot that consists of several tables. The Summary of Fit table, shown here, summarizes the typing data distribution with these statistics: Rsquare (R2) quantifies the proportion of total variation in the typing scores resulting from different word processors rather than from different people. Rsquare Adj parameters. adjusts R2 to make it more comparable over models with different numbers of

Root Mean Square Error (RMSE) is a measure of the variation in the typing scores that can be attributed to different people rather than to different machines. Mean of Response Observations is the mean (average) of all the typing scores. is the total number of scores recorded.

66

5 Comparing Group MeansQuantify Results

Analysis of Variance
When you select the Means/Anova command from the red triangle icon in the title bar, JMP gives you a standard analysis of variance table. If there are only two group levels, the report also includes a t-test table. Note that the value of the F-probability (Prob>F) for the Analysis of Variance is 0.0004. This implies that differences as great as seen in this typing trial are expected only four times in 10,000 similar trials if the word processors did not really promote different typing performances. The Analysis of Variance table has the following information: Source lists the sources of variation: brand, Error, and C. Total. (SS for short) identifies the sources of variation in the typing scores. DF is the degrees of freedom associated with the three sources of variation. Sum of Squares C. Total is the corrected total SS. It divides (partitions) into the SS attributable to brand and the SS for Error. The brand SS is the variation in the typing scores explained by the analysis of variance model, that hypothesizes the word processors are different. The Error SS is the remaining or unexplained variation. Mean Square F Ratio is the sum of squares divided by its associated degrees of freedom. is the model mean square divided by the error mean square.

Prob > F is the probability of obtaining a greater F-value if the mean typing scores for the word processors differed only because different people were typing on them rather than because the word processors promoted different scores in any way.

Mean Estimates and Statistical Comparisons

To see the list of means for each group, look at the Means for Oneway Anova table. This table summarizes the scores for each brand and reveals what level of performance to expect. The Means for Oneway Anova table shows the following information: Level lists the name of each group. is the number of scores in each group. is the standard error of each group mean. is the lower 95% confidence interval for the group means. is the upper 95% confidence interval for the group means. Number Std Error

Mean is the mean of each group. Lower 95% Upper 95%

5 Comparing Group MeansChapter Summary

67 5 Group Means

When you select the Compare Means command from the red triangle icon in the title bar, JMP gives several multiple comparison options to statistically compare pairs of groups. This example uses the All Pairs, Tukeys HSD option, which performs a statistical means comparison for the three pairs of means using the Tukey-Kramer HSD (honestly significant difference) test (Tukey 1953, Kramer 1956). This means comparison method compares the actual difference between group means with the difference that would be significantly different. The difference needed for statistical significance is called the LSD (least significant difference). The graphical results show as the comparison circles previously seen in Figure 5.7. The circles centers represent the actual difference in the group means. The corresponding report is the Means Comparisons table, which shows the actual absolute difference between each mean and the LSD. The top half of the report gives information based on a Students t comparison of each pair. The bottom half shows the results of the Tukey-Kramer multiple comparison tests. Pairs with a positive value are significantly different. The Means Comparison table confirms the visual results in Figure 5.7.

Chapter Summary
In this chapter, the difference in mean typing scores for three brands of word processor was summarized using the Fit Y by X command in the Analyze menu. This command was also used to: Plot the typing scores for the three brands of word processor.

 Overlay a means diamond on each group of typing scores to compare the means of each group. Overlay a quantile box plot on each group of typing scores to compare the shape of the distribution of scores in each group. Produce comparison circles to visualize the difference in mean typing scores. Compute and display a one-way analysis of variance table, which confirmed that at least one pair of means is statistically different. Display a table of the group means and standard errors. Display a table showing the multiple comparison statistical test results for group means. Using the selection tool ( ) from the Tools menu, the graphs or tables can be copied and prepared in a report for presentation. The analysis concludes that, in this typing trial, the SPEEDYTYPE word processor produced significantly higher scores than either of the other two brands. See the chapter Oneway Layout of the JMP Statistics and Graphics Guide for a complete discussion of one-way analysis of variance.

6
Analyzing Categorical Data
Survey data are frequently categorical data rather than measurement data. Analysis of categorical data begins by simply counting the number of responses in categories and subcategories. Counting is easy, but interpreting the relationship between categories based on counts is more complex. It requires computing probabilities and evaluating the likelihood of these probabilities compared to expectations. For example, an American automobile manufacturerfeeling the pinch of competition from foreign auto salesneeds a market analysis before proceeding with a multimillion-dollar advertising campaign. A random sample of people is surveyed. The auto manufacturer wants to know each participants age, sex, marital status, and auto information. The auto information consists of the manufacturing country, the cars size, and the cars type, and whether it is a family, work, or sporty car. This information may provide the advertising experts with direction for the upcoming advertising campaign. Who buys what? Objectives Use the Fit Y by X command to compare two variables consisting of categorical data. Use the formula editor to re-code a categorical variable as a numeric variable. Produce and examine graphs and statistics appropriate for the comparison of proportions such as Chi-squared tests and mosaic plots.

6 Categorical

6
Contents
Look Before You Leap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Open a Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Address the Research Question . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Modify the Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Contingency Table Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Cast Variables Into Roles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Contingency Table Mosaic Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 71 71 72 75 75 76 79

6 Analyzing Categorical DataLook Before You Leap

71 6 Categorical

Look Before You Leap

The first step is to become familiar with the data. Begin by reviewing the data to determine the best way to proceed with the market analysis.

Open a Data Table

When you installed JMP, a folder named Sample Data was also installed. In that folder is a file named Car Poll.jmp. Open Car Poll.jmp, as shown here. The Car Poll data were collected from a random sample of people in a specific geographic area. The columns panel shows that Age is a numeric variable and is assigned the continuous ( ) modeling type. The other five columns are character variables with nominal ( ) modeling types.

Address the Research Question

The basic research question asks, Is the response probability for country of manufacture, size of car, or type of car a function of the age, sex, or marital status of the owner? Look at the data table to see what specific relationships lend insight into this question. You are interested in the relationships between the following automobile characteristics and demographics: manufacturing country by age manufacturing country by sex manufacturing country by marital status size of car by age size of car by sex size of car by marital status type of car by age type of car by sex type of car by marital status

72

6 Analyzing Categorical DataLook Before You Leap

Modify the Data Table

You can sometimes obtain better summary information from age groups rather than specific ages. In fact, dividing people into two age groups is often the basis for a valuable broad analysis. So, lets find the median agethe age that divides the sample into two equal age groups. The distribution of a variable and its corresponding quantiles display a good way to form sample groups. Use the distribution of the age column to find a reasonable value of age that divides the sample into two groups. Select Analyze > Distribution. When the Distribution window appears, select age as the analysis column and click OK. JMP displays a histogram with an accompanying outlier box plot, Quantiles table, and Moments table. The Quantiles table, shown here, identifies 30 as the median age.

The next step is to create a new column whose values identify whether a subjects age is greater than 30, or is less than or equal to 30. Select Cols > New Column to display the New Column window, shown in Figure 6.1, which is used to define column characteristics.
Data Type, Data Source, and Modeling Type options define the new columns characteristics. Enter characteristics for the new column as follows:

Type the new name (call it age group) in the Column Name text box. Because the new column has grouping values instead of measurements, select Character from the box beside Data Type.

6 Analyzing Categorical DataLook Before You Leap

73 6 Categorical

Figure 6.1 New Column Window

Click the New Property button and select Formula. You are presented with the formula editor window shown in Figure 6.2. Figure 6.2 Formula Editor Window

column selector list

function selector list

Suppose 0 represents ages greater than the median (30) and 1 represents the ages less than or equal to the median. To create a formula that divides the sample into two groups, follow these steps: Click Conditional in the function selector list and select the If function.

74

6 Analyzing Categorical DataLook Before You Leap

Highlight the expression term, denoted expr. Choose a<=b from the Comparison functions.

. Highlight the left side of the comparison clause and click age in the column selection list.

6 Analyzing Categorical DataContingency Table Reports

75 6 Categorical

Double-click the right side of the comparison clause to obtain a text entry box. Enter 30 for the numeric comparison. Double-click the term denoted then clause. Enter 1 (in double quotes because this column is a character variable). Double-click the term denoted else clause. Enter 0 (with double quotes). The complete equation should look like the one shown here.

Click Apply, OK, or the formula editors close box to fill the new column with calculated values. Tip: Instead of using the buttons in the formula editor, you can double-click the outermost nesting box to create a single text entry box and enter if(age<=30, 1, 0). Then, press Enter (or Return) or click outside the text box, and the formula appears in formatted form.

Contingency Table Reports

The nominal age grouping variable shows the relationship of age to the other nominal variables using contingency tables. To look at combinations of two variables: Choose Analyze > Fit Y by X. JMP does the statistical analysis appropriate for a variables modeling types and role assignments.

Cast Variables Into Roles

Assign analysis roles to variables by choosing an analysis from the Analyze menu and making selections in the window that appears. In this investigation, the country, size, and type columns are dependent response (y) variables; sex, marital status, and age group are independent (x) variables. This example shows how to complete the Fit Y by X window (Figure 6.3). Select the three y variables (country, size, and type). Click the Y, Response button.

76

6 Analyzing Categorical DataContingency Table Reports

Assign the x variables (sex, marital status, and age group) by selecting them and clicking the X, Factor button. Click OK when finished. Figure 6.3 The Fit Y by X Window

Contingency Table Mosaic Plots

If both x and y have either nominal or ordinal values, JMP displays a mosaic plot with accompanying text reports for each combination of columns assigned x and y modeling roles. A mosaic chart has side-by-side divided bars for each level of its x variable. The bars are divided into segments proportional to each discrete level (value) of the y variable. The mosaic chart in Figure 6.4 shows the relationship of marital status to the manufacturing country. The width of each bar is proportional to the sample size. When the lines dividing the bars align horizontally, the response proportions are the same. When the lines are far apart, the response rates of the samples might be statistically different.

6 Analyzing Categorical DataContingency Table Reports

77 6 Categorical

Figure 6.4 Mosaic Plot Axes response rates proportion of married people with Japanese cars

width of x-axis is proportional to number in sample Scroll to see each x variable as it relates to manufacturing country.
Sex and country do not appear to have any relationship at all. The proportion of automobiles from the three manufacturing countries is about the same for each sex.

The country by age group mosaic plot shows that the proportion of American car owners 30 years or over is only slightly greater than the proportion of American car owners under age 30. The most significant relationship is seen between marital status and country. The mosaic plot, shown previously in Figure 6.4, and its supporting Tests table (Figure 6.5), suggest that married people are more likely than single people to own American cars. The Likelihood Ratio and Pearson Chi-squared tests evaluate the relationship between an automobiles country of manufacture and the marital status of owner. If no relationship exists between country and marital status, a smaller Chi-squared value than the one computed in this survey would occur only seven times in 100 similar surveys. Figure 6.5 Table of Statistical Tests for Marital Status By Country

These statistical results reveal that American auto manufacturers might want to direct advertising plans toward married couples. Scroll the report to see the relationship between size of car and each x variable (sex, marital status, and age).

78

6 Analyzing Categorical DataContingency Table Reports

The three mosaic plots indicate no relationship between car size and gender, marital status, or age group. This is seen numerically by looking at the Contingency Tables and the Tests tables beneath each of the mosaic plots (see Figure 6.6). Note that by default, Col% and Row% also appear in the Contingency Tables. Right-click (Ctrl-click on the Macintosh) the table to access the Columns menu to turn columns on and off. The Chi-squared values support the hypothesis that the purchase of large, medium, and small cars is not significantly different across the sex, marital status, and age group factor levels. The Chi-squared probabilities range from 0.06 to 0.30, so you should expect smaller Chi-squared values to occur six to 30 times in 100 similar surveys. It probably makes no difference what size cars appear in advertisements. Figure 6.6 Tables for Relationships with Size of Car

The market survey categorizes cars based on both size and type, where a cars type is work, sporty, or
family.

Scroll to see the plots that show the relationship between type of car and the three x variables. The mosaic plots in Figure 6.7 show that the type of car varies for levels of marital status and age group. As perhaps expected, many of the cars owned by married people are family automobiles, while the largest proportion of cars owned by single people are sporty cars.

6 Analyzing Categorical DataChapter Summary

79 6 Categorical

Figure 6.7 Reports for Type of Car and Marital Status and Age Group

So, American automobile manufacturers may choose to focus advertisements toward married couples buying family-type automobiles. It follows logically that a relationship between age group and type of car also exists because older people are more likely to be married. The graph to the right in Figure 6.7 shows graphically that the proportion of people over 30 years old who own family cars is much greater than those under 30. The small Chi-squared values support the significant difference in proportions. The Chi-squared values of 0.0005 mean that proportions as varied as these are expected to occur only five times in 1,000 similar surveys.

Chapter Summary
This chapter looked at relationships between categorical variables obtained from a survey. The survey recorded age, sex, marital status, and information about the type of automobile owned by a random sample of people in the same geographical area. The auto information included manufacturing country, size, and type of car. Car types were classified as work, sporty, and family. The question Is the size of car, type of car, or manufacturing country related to the age, gender, or marital status of the owner? was investigated. The Fit Y by X command produced nine mosaic charts with supporting statistical summaries that show: No relationship between either sex or age and manufacturing country.

80

6 Analyzing Categorical DataChapter Summary

A significant relationship between marital status and manufacturing country with married people more likely to own American cars than single people. No relationship between sex, age, or marital status and size of car. No relationship between sex and type of car. Significant relationships between marital status and type of car. As might be expected, married people over 30 years old were more likely to own family type cars than younger, single people. The chapter Contingency Tables Analysis in the JMP Statistics and Graphics Guide discusses analyzing categorical data in more detail. For more information about using the formula editor, see the chapter Using the Formula Editor in the JMP User Guide.

7
Regression and Curve Fitting
This lesson demonstrates the interactive regression capabilities of JMP. The data is from Eppright et al (1972) as reported in Eubank (1988, p. 272). The study subjects are young males. The variables in the data table are age (in months) and the ratio of weight to height. A third variable classifies the subjects into two groups based on age. The goal is to describe and model the growth pattern of subjects for the age range given in the data table. Objectives Use the Fit Y by X command to fit least-squares lines to continuous data. Fit polynomial curves and cubic splines to the data set and explore their goodness of fit. Journal and save analysis results. Use the Group By command to fit different lines to certain groups of data.

7 Regression

7
Contents
Look Before You Leap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Open a JMP File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Select an Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Choose Variable Roles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting Models to Continuous Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting the Mean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting a Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding the Summary of Fit Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding the Analysis of Variance Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Understanding the Parameter Estimates Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Excluding Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Journaling JMP Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Examining a Polynomial Fit (Linear Regression) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting a Spline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting By Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 83 83 83 84 85 85 86 87 87 88 88 90 91 93 94

7 Regression and Curve FittingLook Before You Leap

83 7 Regression

Look Before You Leap

The first step is to become familiar with the data. Begin by reviewing the data to determine the best way to proceed with the regression.

Open a JMP File

When you installed JMP, a folder named Sample Data was also installed. In that folder is a file named Growth.jmp. Open Growth.jmp. A partial listing of the Growth.jmp data table is shown here. There are 2 columns and 72 rows. The ratio column contains the average weight-to-height ratio for each age group in the study. The age groups range from 0.5 to 71.5 months. The modeling type for each column is shown to the left of the variable name in the columns panel. Both columns have a continuous modeling type ( ), as needed for a regression analysis in JMP. The purpose of the analysis is to determine if the ratio values are related to (or dependent on) the age values.

Select an Analysis
To fit regression curves: Select Analyze > Fit Y by X. The Fit Y by X analysis does four kinds of analyses, depending on the modeling type of the variable: Regression analysis when both x and y have continuous values, as in this example. Categorical analysis when both x and y have nominal or ordinal values. Analysis of variance when x is nominal and y has continuous values. Logistic regression when x is continuous and y has nominal or ordinal values.

Choose Variable Roles

The Fit Y by X command first displays the Fit Y by X window. Y identifies a response or dependent variable and x identifies a classification or independent variable. To choose variable roles: Highlight ratio and click Y, Response.

84

7 Regression and Curve FittingFitting Models to Continuous Data

Highlight age and click X, Factor, as shown in Figure 7.1. Click OK. Figure 7.1 The Fit Y By X Window

Now investigate if the ratio of weight to height is a function of age.

Fitting Models to Continuous Data

The scatterplot shown in Figure 7.2 is the result of the Fit Y by X analysis. It is easy to see that the growth pattern is not random. A straight line regression is a good baseline fit to compare with other regression curves. Figure 7.2 Scatterplot of ratio by age

7 Regression and Curve FittingFitting Models to Continuous Data

85 7 Regression

When clicked, the red triangle icon on the scatterplot title bar reveals a variety of fitting commands and additional display options. Options include Show Points, fitting commands, and other features. The Show Points command alternately hides or displays the points in the plot. Fitting options can be as simple as fitting a straight line or involved as drawing density ellipses. Fitting options can be used repeatedly to overlay different fits on the same scatterplot. Begin with a simple line and try different techniques after inspecting the initial straight line regression fit.

Fitting Options

Fitting the Mean

Click the red triangle icon and select Fit Mean. This is the baseline fit that hypothesizes that there is no relationship between x and y. All other fits compare to this fit. Since the Fit Mean table is closed by default: Click the disclosure button ( and on the Macintosh). This displays a report that shows: The sample mean (arithmetic average) of the response variable. The standard deviation of the response variable. The standard deviation of the response mean. The error sum of squares for the simple mean model. on Windows/Linux

Fitting a Line
To fit a simple regression line through the data points, Click the red triangle icon and select Fit Line. The regression line minimizes the sum of squared distances from each point to the line of fit. Because of this property, it is sometimes referred to as the line of best fit.

86

7 Regression and Curve FittingFitting Models to Continuous Data

Red triangle icons Each time a fit is selected from the red triangle icon, the regression equation and another red triangle icon for that fit show beneath the scatterplot, as shown here. Click the red triangle icon to reveal commands that show confidence curves and give the ability to save predicted and residual values as new data table columns. The Save Predicteds command saves the prediction equation for the fit with the new column of predicted values. The fit can be removed from the scatterplot at any time with the Remove Fit command.

Understanding the Summary of Fit Table

Clicking the red triangle icon and selecting Fit Line produces a Summary of Fit table, which summarizes the linear fit.

Rsquare (R2) quantifies the proportion of total variation in the growth ratios accounted for by fitting the regression line. Rsquare Adj parameters. adjusts R2 to make it more comparable over models with different numbers of

Root Mean Square Error (RMSE) is a measure of the variation in the ratio values that is attributable to different people rather than to different ages. Mean of Response Observations is the arithmetic mean (average) of the ratio values. is the total number of nonmissing values.

7 Regression and Curve FittingFitting Models to Continuous Data

87 7 Regression

Note: Notice that the first line of the report is the regression equation, which is editable.

Understanding the Analysis of Variance Table

In addition to producing a Summary of Fit table, clicking the red triangle icon and selecting Fit Line produces an Analysis of Variance table.

The elements of the table give an indication of how well the straight line fits the data points: Source identifies the sources of variation in the growth ratio values (Model, Error, and C. Total). DF records the associated degrees of freedom for each source of variation. Sum of Squares (SS for short) quantifies the variation associated with each variation source. The C. Total SS is the corrected total SS computed from all the ratio values. It divides (partitions) into the SS for Model and SS for Error. The Model SS is the amount of the total variation in the ratio scores explained by fitting a straight line to the data. The Error SS is the remaining or unexplained variation. Mean Square lists the Sum of Squares divided by its associated degrees of freedom (DF) for Model and Error. F Ratio is the regression (Model) mean square divided by the Error mean square. Prob > F is the probability of a greater F-value occurring if the ratio values differed only because of different subjects rather than because the subjects are different ages. In this example, the significance of the F-value is 0.0001, which strongly indicates that the linear fit to the weight/height growth pattern is significantly better than the horizontal line that fits the sample mean to the data.

Understanding the Parameter Estimates Table

In addition to producing a Line of Fit table and an Analysis of Variance table, clicking the red triangle icon and selecting Fit Line produces a Parameter Estimates table.

Term

lists the parameter terms in the regression model. lists estimates of the coefficients in the regression line equation. lists estimates of the standard error of the parameters. is the parameter estimate divided by its standard error.

Estimate Std Error t Ratio

88

7 Regression and Curve FittingFitting Models to Continuous Data

Prob > |t| is the probability of a greater absolute t-value occurring by chance alone if the parameter has no effect in the model. The significant F-ratio in the Analysis of Variance table tells the student that the regression line fits significantly better than the horizontal line at the mean (the simple mean model). However, while the regression line looks like a good fit for age groups above seven months, it does not describe the data well for ages younger than seven months.

Excluding Points
Because the low-age points are the trouble spots for the linear fit, remove them from the analysis and try fitting the model to the remaining values. To highlight these outliers and exclude them from the analysis: Select the lasso tool from Tools menu or toolbar. Drag the lasso around the points to be excluded. Select Rows > Exclude/Unexclude to exclude the selected points. Right-click the selected points and select Markers to assign the X marker to the excluded points. Click the red triangle icon in the title bar and choose the Fit Line command again to see the results of excluding the low-age points. The scatterplot shown here has both regression lines. The low-age points still show on the plot but are not included in the second regression lines computation.

fitted line after excluding points

fitted line for all points

Journaling JMP Results

After completing this part of the exploratory regression analysis:

7 Regression and Curve FittingFitting Models to Continuous Data

89 7 Regression

Choose Edit > Journal. The first time the Journal command is selected during a JMP session, a journal window opens and fills with the graphs and tables from the report window. The open journal file contains all reports from the active report window. Plots can be resized, opened, or closed, as can outlines. This allows for printing of certain parts of the report. Choose Save As from the File menu to save the journal. The window similar to the one shown here prompts for a file name, and appends .jrn to the file name to identify the file type. Leaving a journal file open causes each subsequent use of the Journal command to append results in the active window at the end of the journal contents. Name the journal Regression
Results.

On Windows, change the Save as type to RTF Files (*.RTF) and click Save. On the Macintosh, change the Format type to RTF document and click Save. On Linux, change the Save as type option to *.rtf (Rich Text File) and click Finished. Navigate to the files directory on your system and open the file. The file should open in your default word processor as shown here. Note that the graphics are saved as graphics, and the reports are saved as text tables.

90

7 Regression and Curve FittingFitting Models to Continuous Data

Examining a Polynomial Fit (Linear Regression)

Now, examine a polynomial fit for comparison to the linear fit. A linear regression is simply a polynomial of degree 1. Click the Bivariate Fit of ratio By age report window to assure it is active. Highlight all the excluded rows to include all rows in the analysis. Choose Rows > Exclude/Unexclude include these points in the next analysis. Keeping the rows highlighted, select Rows > Markers and return the marker shape to the default. Click the red triangle icon in the title bar, select Fit Polynomial > 2,quadratic, which allows the fit to have curvature. Remove the line of fit that excluded the lower age groups by clicking the second (modified) regression lines red triangle icon and selecting Remove Fit. Remove the Fit Mean results so that only the polynomial fit and the line fit to all the data points remain. Click the red triangle icon and select Fit Polynomial > 3,cubic to overlay a polynomial curve of degree 3 on the scatterplot.

7 Regression and Curve FittingFitting Models to Continuous Data

91 7 Regression

Again select the Edit > Journal command to append these results to the existing journal. Figure 7.3 Comparison of Linear Fit and Polynomial Fits of Degree 2 and 3

The tables in Figure 7.3 show the R2 value from the Summary of Fit tables for the linear fit, the second degree polynomial fit, and the third degree fit. As polynomial terms are added to the model, the regression curve appears to fit the data better.

Fitting a Spline
Even the polynomial fit of degree 3 does not quite reach the outlying points of the very young subjects. A free-form function that acts as if it smooths the data, such as a smoothing spline, may be better. Use the Remove Fit command on both polynomial fits, so that only the first linear regression line shows on the scatterplot. Click the red triangle icon on the title bar and select Fit Spline three times, with lambda values of 10, 1,000, and 100,000. Lambda is a tuning factor that determines the flexibility of the spline. The Fit Spline command submenu (shown to the left in Figure 7.4) lists lambda values. The three new fits are overlaid on the scatterplot.

92

7 Regression and Curve FittingFitting Models to Continuous Data

Figure 7.4 Comparison of Spline Fits

linear fit = 100,000 = 1,000 = 10

By inspecting the plot, see that the lambda = 10 curve is too flexible and therefore local error has too great an effect on it. The lambda = 100,000 curve is too stiff. It is so straight that it does not reach down to model the lower ages closely. However, the lambda of 1,000 curve fits well. Its shape is not influenced by local errors, and it appears to fit the data smoothly. If a report of these results is needed, journal these results. Select Edit > Journal. The Journal command appends the scatterplot with spline fits and text reports to the open journal file. After journaling the final analyses, the following draft notes about the spline-fitting technique can be added at the bottom of the journal window. Select the annotate tool from the tool bar. Click and drag a large box at the bottom of the report. Add the following text to the box.
"This fitting technique applies a cubic polynomial to the interval between points; the polynomial is joined such that the curve meets at the same point with the same slope to form a continuous and smooth curve. A small enough lambda could make such a curve go through every point, which would model the error, not the mean. A moderate lambda value forces the curve to be smoother, i.e., less curved. This is accomplished by adding a curvature penalty to the optimization that minimizes the sum of squares error."

By comparing various regression fits, notice that both the polynomial fits and the spline fit with moderate flexibility best describe the data. These models show that infants grow most rapidly during the first months of life and that growth rate decreases significantly at approximately 12 months.

7 Regression and Curve FittingFitting By Groups

93 7 Regression

Fitting By Groups
Excluding Points, p. 88 in this chapter, shows how to overlay a linear fit for the whole sample with a linear fit for children over the age of one year. Carry this idea one step further with overlay fits to compare children under the age of one year with children over one year. In the Growth.jmp data table, create a new column called group to act as a grouping variable. Right-click (Command-click on the Macintosh) and select Formula from the menu that appears. Enter the formula shown in Figure 7.5. This assigns the value Babies to each child less than 12.5 months old, and Toddlers to children who are 12.5 months or older. Figure 7.5 Computed Age Grouping Variable

Click the Bivariate report to make it the active window. Clear the Smoothing Spline fits still showing, such as those seen in Figure 7.4, using each fits Remove Fit command. Click the red triangle icon and select Group By to display the window shown here.

94

7 Regression and Curve FittingChapter Summary

Select group, the newly created grouping variable, and click OK. Choose the Fit Line command. With a grouping variable (group) in effect, the overlaid regression lines shown in Figure 7.6 appear automatically. The points that correspond to each regression give a dramatic visualization of the steep growth rate for babies during the first year of life compared to the more moderate growth rate of toddlers and small children age one to five years. Figure 7.6 Regression Lines for Levels of a Grouping Variable

Chapter Summary
To analyze some bivariate data, the Fit Y by X command was used to examine a variety of regression model fits. The task was to model and describe the growth pattern of subjects over a range of ages. You measured growth using the ratio of weight to height and accomplished this task by: Fitting mean to use as a baseline comparison to other regression models and evaluate the fit using statistical text reports. Fitting a straight line as a first guess for a model. Excluding outliers and again fitting a straight line to compare the R2 values given by the Summary of Fit tables for both lines. Fitting second and third degree polynomials to see if they model the growth pattern more realistically. Fitting smoothing splines with lambda values of 10, 1,000, and 100,000 and comparing them with each other and with the linear fit. Clicking the red triangle icon and selecting the grouping facility (Group By) to compare growth rates of babies under the age of one year with toddlers from age one to five years. Using the Journal command to append each of these regression reports and graphs to a journal file.

8
A Factorial Analysis
This lesson examines two treatments of popcorn. The plain, everyday type has been around for years, but researchers claim to have discovered a special treatment of corn kernels. This new process supposedly increases the popcorn yield as measured by popcorn volume from a given measure of kernels. Is this true? If so, how much is the increase? Are these increases the same for all groups of conditions? The special treatment raises the cost of the popcorn, so the increase in yield must be significant enough to warrant the higher costs. The popcorn data used in this chapter and for examples in the JMP User Guide and the JMP Statistics and Graphics Guide are artificial, but the experiment was inspired by experimental data reported in Box, Hunter, and Hunter (1978). Objectives Learn techniques to analyze a designed factorial experiment using the Fit Model command. Evaluate and interpret effects using JMPs interactive graphical techniques. Examine supporting text reports. Evaluate the significance of interaction effects using interaction plots. Save a models predicted values for each observation.

8 Factorial

8
Contents
Look Before You Leap . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Open a Data Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . What Questions Can Be Answered? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Fit Model Window . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Graphical Display: Leverage Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Quantify Results: Statistical Reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Analysis of Variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Summary Reports For The Whole Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Summary Reports for Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 97 97 98 99 101 102 103 103 105

8 A Factorial AnalysisLook Before You Leap

97 8 Factorial

Look Before You Leap

The popcorn yield data are the result of a designed experiment. The same amounts of different kinds of corn were methodically popped under different conditions. First, look at the data to review the results of the popcorn experiment.

Open a Data Table

When you installed JMP, a folder named Sample Data was also installed. In that folder is a file named Popcorn.jmp. Open
Popcorn.jmp.

The Popcorn data table displays in spreadsheet form as shown here. For the experiment, the corn was popped under controlled conditions. Plain popcorn and specially-treated gourmet popcorn were each popped in large or small amounts of oil and in large or small batches. Two trials were done for both kinds of corn under all popping conditions. This experimental design is called a factorial design. The experiment has three factors, usually called main effects, which are: Kind of popcorn (plain or gourmet) Amount of cooking oil (little or lots) Cooking batch size (large or small)

What Questions Can Be Answered?

The appropriate statistical analysis for a factorial design addresses the following questions about the main effects: Is there an overall difference in yield between plain and gourmet popcorn? Is there an overall difference in yield between cooking in lots of oil instead of a small amount of oil? What is the difference in yield between cooking several small batches instead of one large batch? Analysis of a factorial experiment also provides information about the interaction between the main effects as addressed by the following questions: Does the amount of cooking oil have the same effect on both types of popcorn? In other words, is there an interaction effect between popcorn type and amount of cooking oil used? Is there an interaction effect between batch size and type of popcorn? Is there an interaction effect between batch size and amount of oil used?

98

8 A Factorial AnalysisThe Fit Model Window

Are there interaction effects among the three main effects?

The Fit Model Window

Select Analyze > Fit Model. The Fit Model command lets you specify and analyze complex models like the factorial design in this experiment. The Fit Model command displays the Fit Model window shown in Figure 8.1. This window is used to define the type of model, the model response variable, and model effects. To specify the factorial model: Select yield from the Select Columns list. Click the Y button. Select popcorn, oil amt, and batch from the Select Columns list. Click the Macros button and select Full Factorial. This adds all main effects and interactions (crossed effects) to the Construct Model Effects list (Figure 8.1). Further tailor the model by adding effects or removing unwanted effects with the Add and Remove buttons. In this case, remove the three-way interaction term by highlighting popcorn*oil amt*batch and clicking Remove. Figure 8.1 Fit Model Window

8 A Factorial AnalysisGraphical Display: Leverage Plots

99 8 Factorial

Select Effect Leverage from the box beside Emphasis at the top right of the Fit Model window. Click Run Model to estimate the model parameters and view the results.

Graphical Display: Leverage Plots

The Fit Model command graphically displays the whole model and each model effect as the leverage plots shown in Figure 8.2 through Figure 8.5. It is possible to tell at a glance whether the factorial model explains the popcorn data and which factors are most influential. The Whole Model plot to the left in Figure 8.2 shows actual yield by predicted yield values with a regression line and 95% confidence curves. The regression line and the 95% confidence curves cross the sample mean (the horizontal line), which show that the whole factorial model (all effects together) explains a significant proportion of the variation in popcorn yield. There is also a significant difference in yield between the two types of popcorn, as shown in the right-hand leverage plot for the popcorn main effect. The small p-values beneath the plots quantify the significant model fit and popcorn effect. Figure 8.2 Leverage Plots of Actual by Predicted and of Popcorn Effect

In Figure 8.3, the confidence curves for oil amt and the popcorn*oil amt interaction do not cross the horizontal mean line (rather, they encompass the mean line). This shows that neither of these factors significantly affected popcorn yield.

100 8 A Factorial AnalysisGraphical Display: Leverage Plots

Figure 8.3 Leverage Plots for the Oil Amt and Its Interaction with Popcorn

The leverage plots in Figure 8.4 show that the batch size effect (batch) and the interaction between popcorn type and batch size (popcorn*batch) are significant effects. This means that the size of the batch makes a difference in the popcorn yield. Furthermore, the significant interaction means that batch size affects each type of popcorn differently. Figure 8.4 Leverage Plots for Batch and Its Interaction with Popcorn

The two leverage plots shown in Figure 8.5 show that there is no significant interaction between amount of oil and batch size.

8 A Factorial AnalysisQuantify Results: Statistical Reports

101 8 Factorial

Figure 8.5 Leverage Plots for Other Interaction Effects

For more information about interpretation of leverage plots, see the chapters Understanding JMP Analyses and Standard Least Squares: Introduction and the appendix Statistical Details of the JMP Statistics and Graphics Guide.

Quantify Results: Statistical Reports

Because oil amt and its interactions with other effects are not significant, fit the popcorn data again without these effects. The new model should have the significant factors (type of popcorn, batch size), and their interaction term. This approach condenses the statistical reports that show estimates of yield under the different conditions of interest. Use the same Fit Model command as before. Click the Fit Model window to make it the active window. If the window is closed, click the red triangle icon on the report and select Script > Redo Analysis to open a new Fit Model window. Use the following method to specify the two-factor model: From the full factorial model, select unwanted effects listed in the Construct Model Effects box and click Remove. Click Run Model.

102 8 A Factorial AnalysisAnalysis of Variance

Analysis of Variance
The whole model leverage plot in Figure 8.6 shows that the two-factor model describes the popcorn experiment well. Examine the tables that accompany the whole model leverage plot. The Analysis of Variance table that accompanies the whole model leverage plot quantifies the analysis results. It lists the partitioning of the total variation of the sample into components. The ratio of the Mean Square components forms an F-statistic that evaluates the effectiveness of the model fit. If the probability associated with the F-ratio is small, then the analysis of variance model fits better statistically than the simple model that contains only the overall response mean. Figure 8.6 Analysis of Variance for the Two-Factor Whole Model

The Analysis of Variance table shows these quantities: Source identifies the sources of variation in the popcorn yield values (Model, Error, and C. Total). DF records the degrees of freedom for each source of variation. Sum of Squares (SS for short) quantifies the variation in yield. C. Total is the corrected total SS. It is divided (partitioned) into the SS for Model and SS for Error. The SS for Model is the variation in the yield explained by the analysis of variance model, which hypothesizes that the model factors have a significant effect. The SS for Error is the remaining or unexplained variation. Mean Square F Ratio is a sum of squares divided by its associated degrees of freedom (DF). is the model mean square divided by the error mean square.

8 A Factorial AnalysisAnalysis of Variance

103 8 Factorial

Prob > F is the probability of a greater F-value occurring if the variation in popcorn yield resulted from chance alone rather than from the model effects. In this example, the F-value is 0.0001. JMP indicates a significant F-value by placing an asterisk beside it. The low value of this F-value implies that the difference found in the popcorn yield produced by this experiment is expected only 1 time in 10,000 similar trials if the model factors do not affect the popcorn yield.

Summary Reports For The Whole Model

Other tables in the Fit Model report provide statistical summaries. The Summary of Fit table shows the numeric summaries of the response for the factorial model: Rsquare (R2) of 0.809 tells the scientist that the two-factor model explains nearly 81% of the variation in the data. Rsquare Adj adjusts R2 to make it more comparable over models with different numbers of parameters. Root Mean Square Error (sometimes called the RMSE) is a measure of the variation in the yield scores that can be attributed to random error rather than differences in the models factors. Mean of Response scores. Observations is the mean (average) of the yield

is the total number of recorded scores.

The F-test probabilities in the Effect Test table tell the scientist that all model effects explain a significant proportion of the total variation. JMP indicates a significant F-value by placing an asterisk beside it. There is also a table that gives the parameter estimates for the model.

Summary Reports for Effects

Now look at the summary tables for each effect in the model. The tables for the main effects are shown here. The Least Squares Means table lists the least squares means and standard errors for each level of the model factors, without considering the interaction between them. In this balanced example, the least squares means are simply the sample means of each factor level.

104 8 A Factorial AnalysisAnalysis of Variance

The nature of the interaction is important in the interpretation of the popcorn experiment. To examine the significant popcorn*batch interaction, Click the red triangle icon from the popcorn*batch title bar and select LSMeans Plot, or click the red triangle icon from the Response yield title bar and select Factor Profiling > Interaction Plots. This command plots the least squares means for each combination of effect levels, as shown in Figure 8.7. Figure 8.7 Interaction Plots

The Least Squares Means table for the popcorn*batch effect tells the whole story. Batch size makes no difference for the plain brand popcorn, but popping in small batches increases the yield in the new gourmet brand. Because the factorial model with two-factors is a good prediction model, save the prediction formula. Click the red triangle icon in the Repsonse yield title bar and select Save Columns > Prediction Formula, as shown in Figure 8.8.

8 A Factorial AnalysisChapter Summary

105 8 Factorial

Figure 8.8 Prediction Formula for Popcorn Yield

This command creates a new column in the Popcorn data table called Pred Formula yield that contains the predicted values for each experimental condition. The prediction formula, shown at the bottom of Figure 8.8, becomes part of the column information. To see this formula: Highlight the new column name (Pred Formula yield). Select Formula from the Cols menu. The prediction formula can be copied to the clipboard using standard cut and paste techniques. Results show that popcorn should be packaged: in small packages so that the yield will be good. in family size packages with smaller packets inside. in family size packages with popping instructions that clearly state the best batch size for good results.

Chapter Summary
In this chapter, a designed experiment evaluated the difference in yield between two types of popcorn. A three-factor factorial experimental design was the basis for popcorn popping trials. The results were analyzed by using the Analyze > Fit Model command. The following results were found: The leverage plots for the factorial analysis of three factors showed one main effect and its associated interactions to be insignificant. A more compact two-factor analysis with interaction adequately described the variation in yield for the popcorn trials.

 The interaction between the two main effects was significant. The Least Squares Means table for the interaction showed how the two types of popcorn behaved under different popping conditions. The new, more expensive gourmet popcorn had better yield than the plain everyday type only if popped in small batches.

9
Exploring Data
Advanced Example with Principal Components
Exploration is the search to find something newthe endeavor to make some discovery. For data analysis, exploratory study is often the most fruitful part of the analytical process because it is the most open to serendipity. Something noticed in a data set can be the seed of an important advance. There are two important aspects of exploration: What is the pattern or shape of the data? Are there points unusually far away from the bulk of the data (outliers)? When exploring data composed of many variables, the great challenge is dealing with this high dimensionality. There can be many variables that have interesting relationships, but its hard to visualize the relationship of more than a few variables at a time. Objectives Use graphical techniques to search for outliers in one, two, three, and higher dimensions. Perform a principal components analysis and examine it graphically. Examine outliers graphically using Mahalanobis distance.

9 Exploring

9
Contents
Solubility Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . One-Dimensional Views . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Two-Dimensional Views . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Three-Dimensional Views . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Principal Components and Biplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Using Colors, Markers, and the Brush Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Multivariate Distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 109 110 111 112 116 117 118

9 Exploring DataSolubility Data

109 9 Exploring

Solubility Data
This lesson examines compounds for those with unusual solubility patterns in various solvents. When you installed JMP, a folder named Sample Data was also installed. In that folder is a file named Solubility.jmp. Data from an experiment by Koehler, Grigorus, and Dunn (1988) are in the Solubility.jmp file. Open Solubility.jmp. There are 72 compounds tested with six solvents, in columns called 1-Octanol, Ether, Chloroform, Benzene, Carbon Tetrachloride, and Hexane. The Labels column in the table should serve as a label variable (Figure 9.1) so when you plot them, the compound names instead of row numbers identify points. Although this is already done for you in Solubility.jmp, you should know that to assign the label role to columns, you should select the columns, then select Cols > Label/Unlabel or click the red triangle in the columns panel and select Label/ Unlabel from the resulting menu. Figure 9.1 Solubility Data Table

Use this menu to assign Label role to a column There are six solvent variables, but there are no six-dimensional graphics. However, it is possible to look at six one-dimensional graphs, 15 two-dimensional graphs, and 20 three-dimensional spinning plots. Using principal components, a representation of higher dimensions can be displayed.

One-Dimensional Views
The Distribution command helps you summarize data one column at a time. It does not show any relationships between variables, but the shape of the individual distributions helps identify the one-dimensional outliers.

110

9 Exploring DataSolubility Data

To begin exploring the solubility data: Choose Analyze > Distribution. Select the six solubility columns and click the Y, Columns button. Click OK. Their histograms, resized and trimmed of other output, are shown in Figure 9.2. Click any histogram bar. That bar, and all other representations of that data, are highlighted in all related windows. To see how outlying values are distributed in the other histograms: Shift-click the outlying bars in each histogram. This identifies the outlying rows in each single dimension. Use the Rows > Markers palette to assign the X marker to these selected rows. The markers appear in the data table and in subsequent plots. Figure 9.2 One-Dimensional Views

To create a new data table that contains only the outlying rows: Use the Tables > Subset command as shown here. Click OK to accept the default settings. Scroll through the new subset table to see the compound names of the one-dimensional outliers.

Two-Dimensional Views
Return to Solubility.jmp. Select Analyze > Multivariate Methods > Multivariate. Highlight all the continuous columns in the table and click the Y, Columns button. Click OK. This displays a correlation matrix and a scatterplot matrix of all 30 two-dimensional scatterplots (Figure 9.3). The one-dimensional outliers appear as Xs in each scatterplot. Note in the scatterplot matrix that many of the variables appear to be correlated, as evidenced by the diagonal flattening of the normal bivariate density ellipses. There appear to be two groups of variables that correlate among themselves but are not very correlated with variables in the other group.

9 Exploring DataSolubility Data

111 9 Exploring

Figure 9.3 Two-Dimensional View

The variables Ether and 1-Octanol appear to make up one group, and the other group consists of the remaining four variables. These two groups are outlined on the scatterplot matrix shown in Figure 9.3. Scan these plots looking for outliers (points that fall outside the bivariate ellipses) of a two-dimensional nature and identify them with square markers using the following steps. Shift-click each outlier. Select Rows >Markers and select the square marker from the palette. Now, both one- and two-dimensional outliers are identified.

Three-Dimensional Views
To see points in three dimensions: Select Graph > Spinning Plot, which opens the three-dimensional Spinning Plot window. Add all six continuous variables to the Y, Columns list. Click OK. After the plot appears, click and drag the X, Y, and Z axis tags to any combination of three variables. The goal is to look for points away from the point cloud for each combination of three variables. To aid in this search: Rotate and examine each three-dimensional plot with the spin controls. Select the hand tool from the tools palette.

112

9 Exploring DataSolubility Data

Click and drag inside the plot to rotate the plot in any direction. Select the arrow tool from the tools palette when the plot is at an angle to see outliers. Figure 9.4 shows two three-dimensional outlying points in the view of Ether by 1-Octanol by Benzene that hadnt been apparent before. To label them: Shift-click these points. Select Rows > Label/Unlabel. Their labels, METHYLACETATE and ACETONE, appear on the plot. Figure 9.4 Spotting Outliers in a Three-Dimensional View

spin controls

Principal Components and Biplots

Because many of the variables in the Solubility.jmp table are highly correlated, there is not a lot of scatter in six dimensions. The scatter is oriented in some directions but is flattened in other directions. To illustrate this: Remove the labels from METHYLACETATE and ACETONE by highlighting them, right-clicking the graph, and selecting Row Labels. Select Graph > Spinning Plot. Add only two highly correlated variables, in this case Ether and
1-Octanol.

Click OK. Click the red triangle icon in the Spinning Plot title bar and select Principal Components. The results are shown in Figure 9.5. Note that because the data are highly correlated, the scatter in the points runs in a narrow ellipse whose principal axis is oriented in the direction marked P1.

9 Exploring DataSolubility Data

113 9 Exploring

Figure 9.5 Two Correlated Variables with Principal Components

To see the greatest variation of the data in one dimension, Rotate the axis so that the first principal component, P1, is horizontal. The technique of extracting these orientations that capture the highest variance is called principal components analysis. Principal components capture the most variation possible in the smallest number of dimensions. Use principal components to explore all six dimensions with the following steps: Select Graph > Spinning Plot. Add all six continuous variables to the Y, Columns list. Click OK. Click the red triangle icon and select Principal Components. The result is the Principal Components table in Figure 9.6. The cumulative percent row (CumPercent) shows that the first three principal components account for 97.8% of the six-dimensional variation. Figure 9.6 Principal Components Text Report

On the spinning plot, principal components appear as additional rays labeled P1, P2, and so on, as shown in Figure 9.7. These rays are the projections of the six-dimensional direction of the principal component in the three dimensions shown. P1 tends to be the longest ray because it shows the direction of greatest variance. P2 appears as the second longest ray and describes the next most prominent direction. The other rays are labeled successively as their variances decrease.

114

9 Exploring DataSolubility Data

Figure 9.7 Principal Component Rays in Three of the Variables

A spinning plot of the first three principal components produces the best three-dimensional representation of the six-dimensional space. Drag the X, Y, and Z axes icons to the spaces labeled Prin Comp 1, Prin Comp 2, and Prin Comp 3, respectively. In this space, CAFFEINE appears as an outlier (see Figure 9.8). In the principal component space, the principal components are the axes and the variables show as rays. A plot showing both variable rays and points in an approximation of a high-dimensional space is called a biplot (Gabriel 1971). The configuration of the variable rays in Figure 9.8, shows how the variables relate. Note that eth and oct seem to dominate one direction, while the other four variables cluster to define the other directions. As more dimensions (variables) are condensed into principal components, the angles between variables become indicators of their correlation. In a factor analysis, these directions are further refined, mapping the variables into clusters called factors. Figure 9.8 Biplot of Variables and Observations

9 Exploring DataSolubility Data

115 9 Exploring

When looking for outliers, it is often revealing to examine the space of the last three principal components instead of the first three. The last principal components define the least popular directions of the scatter (directions with the least variation). If a point is unusual in a multivariate sense, then its prominence in the least popular direction suggests it is an outlier. To use this strategy, Drag the X, Y, and Z axes icons to the spaces labeled Prin Comp 4, Prin Comp 5, and Prin Comp 6, respectively. Rotate the plot to find an outlier. Click the outlier and select Rows > Label/Unlabel. Your spinning plot should show SULFATHIAZOLE as the most unusual value. Most other points should be in a tight cluster near the center. Figure 9.9 Outlier in the Last Principal Components

Another strategy is to plot each variable with the first two principal components and then with the last two. As an example, Drag the X, Y, and Z axes icons to the spaces labeled Prin Comp 1, Prin Comp 2, and hex, respectively. You will see that hexane is on the z-axis with the first principal components plotted on the x- and
y-axes.

116

9 Exploring DataUsing Colors, Markers, and the Brush Tool

Figure 9.10 Hexane and the First Two Principal Components

Drag the X, Y, and Z axes icons to the spaces labeled Prin Comp 5, Prin Comp 6, and CCl4, respectively. The variable Carbon Tetrachloride should be plotted with the last two principal components. Figure 9.11 Carbon Tetrachloride and Last Two Principal Components

Using Colors, Markers, and the Brush Tool

Though a given plot shows two or three dimensions geometrically, aspects of other dimensions can be added by plotting the points with different markers and colors. In addition, using the brush tool ( ) or the lasso tool ( ) shows where points in areas of one plot show up on other plots.

9 Exploring DataMultivariate Distance

117 9 Exploring

Multivariate Distance
The basic concept of distance in several dimensions relates to the correlation of the variables. For example, in a Multivariate scatterplot cell for Benzene by Chloroform (Figure 9.3), HYDROQUINONE is located away from the point cloud. This compound is not particularly unusual in either the x or y direction alone, but it is a two-dimensional outlier because of its unusual distance from the strong linear relationship between the two variables. The ellipse is a 95% density contour for a bivariate normal distribution with the means, standard deviations, and correlation estimated from the data. The concept of distance that takes into account the multivariate normal density contours is called Mahalanobis distance. Though only three dimensions can be visualized at a time, the Mahalanobis distance can be calculated for any number of dimensions. To produce a plot of the Mahalanobis distance: Select Outlier Analysis > Mahalanobis Distance from the menu accessed by the red triangle at the top of the multivariate report. Figure 9.12 shows the Mahalanobis distance by the row number for each data point. To label these points: Select the brush tool ( ) from the tools palette. While holding down the Shift key, drag the brush over the points labeled in Figure 9.12. These are the five points with the greatest Mahalanobis distances. Select Rows > Label/Unlabel.

118

9 Exploring DataChapter Summary

Figure 9.12 Mahalanobis Distance Plot to See Multivariate Outliers

Chapter Summary
In this example, commands from the Analyze and Graph menus were used for data exploration to locate and identify unusual points. The data were first examined in one dimension using the Distribution command and then in two dimensions using the Multivariate command to look for unusual points in histograms and scatterplots. Next, the Spinning Plot command in the Graph menu was used to plot three columns at a time. The technique of principal components was used to summarize six dimensions and to plot principal component rays. The Principal Components table showed that the first three principal components accounted for more than 97% of the total variation. To locate multivariate outliers, each column was plotted with the first two principal components and then with the last two principal components. Finally, the Outlier Analysis command in the Multivariate report produced the Mahalanobis outlier distance plot, which summarizes the points in six dimensions. The multivariate outliers were highlighted and labeled in this multi-dimensional space. See the chapter Correlations and Multivariate Techniques, in the JMP Statistics and Graphics Guide, for documentation and examples of multivariate analyses. Three-Dimensional Viewing in the JMP Statistics and Graphics Guide documents the spinning plot.

10
Multiple Regression
Multiple regression is the technique of fitting or predicting a response by a linear combination of several regressor variables. The fitting principle is like simple linear regression, but the space of the fit is in three or more dimensions, making it more difficult to visualize. With multiple regressors, there are more opportunities to model the data well, but the process is more complicated. This chapter begins with an example of a two-regressor fit that includes three dimensional graphics for visualization. The example is then extended to include six regressors (but unfortunately no seven-dimensional graphics to go with it). Objectives Illustrate the concept of a fitting plane using graphical techniques. Combine data tables using the Concatenate command. Explore a three-dimensional version of a leverage plot.

10 Multiple

10
Contents
Aerobic Fitness Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fitting Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fit Planes to Test Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Whole Model Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . More and More Regressors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Interpreting Leverage Plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Collinearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 123 124 126 127 128 130 130

10 Multiple RegressionAerobic Fitness Data

121 10 Multiple

Aerobic Fitness Data

Aerobic fitness can be evaluated using a special test that measures the oxygen uptake of a person while running on a treadmill for a prescribed distance. However, it would be more economical to evaluate fitness with a formula that predicts oxygen uptake with simpler measurements. To identify such an equation, runtime and fitness measurements were taken for 31 participants who ran 1.5 miles. The participants ages were also recorded. When you installed JMP, a folder named Sample Data was also installed. In that folder is a file named Fitness.jmp. Open Fitness.jmp. The data are shown in Figure 10.1. For purposes of illustration, certain values of MaxPulse and
RunPulse have been changed from data reported by Rawlings (1988, p. 124).

Figure 10.1 Partial Listing of the Fitness.jmp Data File

Investigate Age and Runtime as predictors of oxygen uptake using the Fit Model analysis. Choose Analyze > Fit Model and you should see the window shown here.

122 10 Multiple RegressionAerobic Fitness Data

To specify a multiple regression model with two effects, Highlight Oxy in the Select Columns column. Click the Y button. Highlight both Age and Runtime. Click the Add button to specify them as model effects. Click Run Model. You should now see the tables shown in Figure 10.2. These statistical reports are appropriate for a response variable and factor variables that have continuous values.

10 Multiple RegressionFitting Plane

123 10 Multiple

Figure 10.2 Statistical Text Reports

Clicking the red triangle icon and selecting Save Columns displays a list of save commands. To save predicted values and the prediction equation for this model: Click the red triangle icon and select Save Columns > Prediction Formula. This command creates a new column in the Fitness data table called Pred Formula Oxy. Its values are the calculated predicted values for the model. To see the columns formula: Right-click the Pred Formula Oxy column name and select Formula. The Formula window opens and displays the formula
88.4356809 + -0.1509571 * Age + -3.1987736 * Runtime

This formula defines a plane of fit for Oxy as a function of Age and Runtime. Click Cancel to close the window and return to the data table window.

Fitting Plane
JMP can show relationships between Oxy, Runtime, and Age in three dimensions with a surface plot. Select Graph > Surface Plot. Add Oxy and Predicted Formula Oxy as Columns, and click OK. Change the Style drop down menu to needles, and view the plot, as shown in Figure 10.3.

124 10 Multiple RegressionFitting Plane

Figure 10.3 Initial View of the Surface Plot of Oxy, Age, and Runtime

Click and drag to rotate the plot so it looks like that in Figure 10.4. Figure 10.4 Observed Points using Age, Oxy, and Runtime with the Predicted Plane of Fit Observed Oxy values

Plane of predicted Oxy values

Fit Planes to Test Effects

The example in the previous section showed a plane fit to the whole model. You can also look at hypothesis tests for each regressor and to test whether the regressors parameter is significantly different from zero. One way to view this test is to evaluate the difference between the current fit and the fit that occurs if the regressor variable is removed from the model.

10 Multiple RegressionFitting Plane

125 10 Multiple

For example, remove the Runtime variable from the model by following these steps: First, make sure Fitness.jmp is the active data table. Select Analyze > Fit Model, set oxy as y and add age as a model effect, and click Run Model. Click the red triangle icon and select Save Columns > Prediction Formula. The new predicted column (labeled Pred Formula Oxy 2) is calculated using the formula
62.4229492 + -0.3156031*Age

To compare this fitted line with the plane in the previous example, Select Graph > Surface Plot. Add Oxy, Pred Formula Oxy, and Pred Formula Oxy 2 as Columns. and click OK. Change the Point Response Column for both Pred Formula Oxy and Pred Formula Oxy2 to Oxy. Change the Style drop down menu for both Pred Formula Oxy and Pred Formula Oxy2 to needles. Change the Surface drop down menu for both Pred Formula Oxy and Pred Formula Oxy2 to Both Sides. Both this grid and the one in Figure 10.4 represent least squares regression planes, but this plane has a slope of zero in the orientation of the Runtime axis. Figure 10.5 shows the plot from an angle. Figure 10.5 Three-Dimensional Plot with Regression Planes Observed Values (x)
Oxy by Age and Runtime regression

plane

Oxy by Age

regression plane

Click and drag to rotate the plot. A rotated view is shown in Figure 10.6. Notice this subset (Age-only model) regression showing as a line instead of a plane. The view is edge-on for Runtime, which eliminates it from the visual model.

126 10 Multiple RegressionFitting Plane

Figure 10.6 Comparison of Three-Dimensional Views Runtime effect is edge-on

Open the Ranges outline (Figure 10.7). Figure 10.7 Changing the Ranges

Click the Oxy button and change the minimum to 32, the maximum to 60, and the increment to 5. Click the Age button and change the minimum to 32, the maximum to 60, and the increment to 5. Your plot shows the bivariate regression plane edge-on and represents the linear combination of the effects fit by the plane.

Whole Model Tests

The leverage plot in Figure 10.8 shows joint test of the Age, Weight, Runtime, RunPulse, and MaxPulse effects in the model. This plot compares the full model with the model containing the intercept that fits the overall response mean only. This leverage plot is formed by plotting the actual observed values on the y-axis and the values predicted by the whole model on the x-axis. The residual for the subset model is the distance from a point to the horizontal line drawn at the sample mean.

10 Multiple RegressionFitting Plane

127 10 Multiple

Figure 10.8 Leverage Plot for the Whole Model (Age and Runtime)

More and More Regressors

Its easy to visualize two regressors predicting a response by using fitting planes. But how can this be done with more regressors when the analysis requires more than three dimensions? In actuality, the fitting, testing, and leverage plot analyses still work for more regressors. Select Analyze > Fit Model. Fill the window with Oxy as Y and add Age, Weight, Runtime, RunPulse, and MaxPulse as model effects. Click Run Model to run the model. In this case, the prediction formula is
Oxy = 101.3 - 0.2123 Age - 0.0732 Weight - 2.688 Runtime - 0.3703 RunPulse + 0.3055 MaxPulse

Look at the significance of each regressor with t-ratios in the Parameter Estimates table or F-ratios in the Effects Tests table (see Figure 10.9). Because each effect has only one parameter, the F-ratios are the squares of the t-ratios, and have the same significance probabilities. The Age variable seems significant, but Weight does not. The Runtime variable seems highly significant. Both RunPulse and MaxPulse also seem significant, but MaxPulse is less significant than RunPulse.

128

10 Multiple RegressionFitting Plane

Figure 10.9 Statistical Tables for Multiple Regression

Interpreting Leverage Plots

The leverage plots for this example multiple regression model allow visualization of the contribution of each effect. First, look at the whole-model leverage plot, shown to the right, of observed versus predicted values. This plot illustrates the test for the whole set of regressors. The Analysis of Variance table in Figure 10.9 shows a highly significant F corresponding to this plot. The confidence curves show the strong relationship because they cross the horizontal line. Now examine the leverage plots for the regressors. Each plot illustrates the residuals as they are and as they would be if that regressor were removed from the model. The confidence curves in the leverage plot for Age in Figure 10.10 show that Age is borderline significant because the curves barely cross the horizontal line of the mean. Note that the significance of the Age effect is 0.03 in the text reports (Figure 10.9), which is only slightly different from the 0.05 confidence curves drawn by JMP. The leverage plot for Weight shows that the effect is not significant. The confidence curves do not cross the horizontal line of the mean.

10 Multiple RegressionFitting Plane

129 10 Multiple

Figure 10.10 Leverage Plots for the Age and Weight Effects

The leverage plot for Runtime shows that Runtime is the most significant of all the regressors. The Runtime leverage line and its confidence curves cross the horizontal mean at a steep angle.

The leverage plots for RunPulse and MaxPulse shown in Figure 10.11 are similar. Each is somewhat shrunken on the x-axis. This indicates that other variables are related in a strong, linear fashion to these two regressors, which means the two effects are strongly correlated with each other. Figure 10.11 Leverage Plots for the RunPulse and MaxPulse Effects

130 10 Multiple RegressionChapter Summary

Collinearity
When two or more regressors have a strong correlation, they are said to be collinear. These regression points occupy a narrow band showing their linear relationship.When a plane is fit representing collinear regressors, the plane fits the points well in the direction where they are widely scattered. However, in the direction where the scatter is very narrow, the fit is weak and the plane is unstable. In text reports, this phenomenon translates into high standard errors for the parameter estimates and potentially high values for the parameter estimates themselves. This occurs because a small random error in the narrow direction can have a huge effect on the slope of the corresponding fitting plane. An indication of collinearity in leverage plots is when the points tend to collapse toward the center of the plot in the x direction. The fitness example shows collinearity geometrically in the strongly correlated regressors, RunPulse and MaxPulse. To examine these regressors, examine Figure 10.12, which shows rotated views of the regression planes. Most of the points are near the intersection of the two planes. When both planes are edge-on, as shown on the right of Figure 10.12, most of the scatter is hidden. From that angle, notice that the fitting plane representing both variables holds no better than the subset plane. The angle of the fitting plane is steep, but the hold is unstable. Geometrically, collinearity between two regressors means that the points they represent do not spread out in x space enough to provide stable support for a plane. Instead, the points cluster around the center causing the plane to be unstable. The regressors act as substitutes for each other to define one direction redundantly. This is cured by dropping one of the collinear regressors from the model. In this case, drop either MaxPulse or RunPulse from the model because both measure essentially the same thing. Figure 10.12 Comparison of RunPulse and MaxPulse Effects

Chapter Summary
Multiple regression uses the same fitting principle as simple regression, but accounting for significance is more subtle. Each regressor opens a new dimension for fitting a hyperplane, and its significance is

10 Multiple RegressionChapter Summary

131 10 Multiple

tested by how much the fit suffers in its absence. When regressors correlate to each other, they are said to be collinear, and they define directions where the fitting hyperplane is not well supported.

References

References
Becker, R.A., and Cleveland, W.S. (1987), Brushing Scatterplots, Technometrics, 29, 2. Belsley, D.A., Kuh, E., and Welsch, R.E. (1980), Regression Diagnostics, New York: John Wiley & Sons. Box, G.E.P., Hunter, W.G., and Hunter, J.S. (1978), Statistics for Experimenters, New York: John Wiley & Sons, Inc. Daniel C. and Wood, F. (1980), Fitting Equations to Data, Revised Edition, New York: John Wiley & Sons, Inc. Draper, N. and Smith, H. (1981), Applied Regression Analysis, 2nd Edition, New York: John Wiley & Sons, Inc. Eppright, E.S., Fox, H.M., Fryer, B.A., Lamkin, G.H., Vivian, V.M., and Fuller, E.S. (1972), Nutrition of Infants and Preschool Children in the North Central Region of the United States of America, World Review of Nutrition and Dietetics, 14. Eubank, R. L., (1988), Spline Smoothing and Nonparametric Regression, New York: Marcel Dekker. Gabriel, K.R. (1982), Biplot, Encyclopedia of Statistical Sciences, Volume 1, Kotz and Johnson editors, New York: John Wiley & Sons, Inc. Hartigan J.A. and B. Kleiner (1981), Mosaics for Contingency Tables, Proceedings of the 13th Symposium on the Interface between Computer Science and Statistics, W. F. Eddy editor, New York: Springer. Hawkins, D.M., (1974), The Detection of Errors in Multivariate Data Using Koehler, Grigorus, Dunn (1988), The Relationship Between Chemical Structure and the Logarithm of the Partition, QSAR, 7. Koehler, M.G., Grigorus, S., and Dunn, J.D. (1988), The Relationship Between Chemical Structure and the Logarithm of the Partition Coefficient, Quantitative Structure Activity Relationships, 7. Leven, J. R., Serlin, R. C., and Webne-Behrman, L. (1989), Analysis of Variance Through Simple Correlation, American Statistician, 43. Mosteller, F. and Tukey, J.W. (1977), Data Analysis and Regression, Reading Mass: Addison-Wesley. Rawlings, J. O., Pantula, S.G., and Dickey, D.A. (1998), Applied Regression Analysis: A Research Tool2nd ed., New York, NY: Springer-Verlag New York Inc. Sall, J. P. (1990), Leverage Plots for General Linear Hypotheses, American Statistician, 308-315. SAS Institute (1987), SAS/Stat Guide for Personal Computers, Version 6 Edition, Cary NC: SAS Institute Inc. Snedecor, G.W. and Cochran, W.G. (1967), Statistical Methods, Ames Iowa: Iowa State University Press.

134

References

Winer, B.J. (1971), Statistical Principals in Experimental Design, 2nd Edition, New York: McGraw-Hill, Inc.

Index

Index
JMP Introductory Guide Symbols
? tool 3, 5 collinear 130
Column Info 25

A
Add Columns 24 Add Rows 25 Add Statistics Column 35

All Pairs, Tukeys HSD 64 analysis methods see platforms Analysis of Variance see Fit Y by X, Fit Model Analysis of Variance table 66, 87, 102 analysis role 13 analysis type see modeling type 10 analyze categorical data 6980 Analyze menu 13 Annotate cursor 29 arrow cursor 10 assign role 27

column name 24 columns 9 Compare Means 57, 64, 6768 Comparison Circles 64 confidence curves 99, 128 confidence interval 51, 62 construct formula 5254 Construct Model Effects 98, 101 continuous 10, 15, 60, 83 Count 52 Cowboy Hat.jmp 17 create subset 5255 crossed effects 98 Cum Prob 52 cursors in data table 10 curve-fitting 83

D
data grid 9 data table create 2130 density contour 117 designed experiment 97 DF 66, 87, 102 disclosure control 51 discrete data see nominal, ordinal Display Options 14 DisplayBox scripting index 5 distance 117 Distribution 72, 110 example 14 documentation overview 7 double-arrow cursor 11 drug experiment example 2130

B
bar chart 27 beginners tutorial 3, 8 biplot 114 box plot 50 BP Study.jmp 25

C
C. Total 66 calculator example 54 Car Poll.jmp 71 categorical analysis see Fit Y by X, Fit Model categorical data 6980 categorical type 10 character column 10 Charts 26, 36 Chi-Square 78 classification variable 13

136

Index

E
enter data 26 Estimate 87 Example button 4 Example button in Index Help 4 Exclude/Include 88, 90 explanatory variable 13 explore data 107118 extend selection 12 extreme values 52, 63

honestly significant difference see Tukey-Kramer HSD hotdog example 3143 Hotdogs.jmp 33

I
I-beam cursor 11 independent variable 13 Index tab on JMP Starter 4 interaction 97, 104 interquartile range 50

F
F Ratio 66, 87, 102 F Statistic 102 Factor role 13 factorial analysis example 95106 fit by groups 93 Fit Line 85, 94 Fit Model 98, 121 fit plane 124 Fit Polynomial 90 Fit Y by X 40, 59, 75 Fit Spline 91 Fitness.jmp 121 formula 53 example 54 prediction 105, 123 F-probability see Prob > F Freq role 13 frequencies see Distribution frequency table 15, 52

J
JMP Starter window 8 Journal 89, 92 journaling analysis results 88 JSL Operators menu item 5

L
Label/Unlabel 50, 109

G
Graph menu 13 Group By 9394 grouped charts see Charts grouped fitting 93 grouping data 34, 38 grouping variable 13 Growth.jmp 83

lambda 91 large cross cursor 11 Launch button 4 Launch button in Help Index 4 Least Squares Means table 103104 Level 52, 66 leverage plot 100, 126 Likelihood Ratio 77 linked table 35 local error 92 logistic regression see Fit Y by X, Fit Model LS Means Plot 104

M
magnifier tool 29 Mahalanobis distance 117 main effect 9798 Markers 88, 110 Mean 66 Mean of Response 65, 86, 103 Mean Square 66, 87, 102 Means Diamonds 50, 6162 Means for Oneway Anova table 66 Means, Anova/t-test 61 median 50 menus tips 6

H
hand cursor 12, 49 Help using online help 3, 5 high dimensionality 107 highlight see Select histogram see Distribution

Index

137 Index

modeling type 10, 60 modify data table 7275 moments see Distribution mosaic plot see Distribution multiple comparison see Compare Means multiple regression 119, 122131 Multivariate 110113

question mark (?) tool 3, 5 Quick Reference Guide 7

R
R2 91 rays 113 references 133 regression analysis 83 regression example 83 regression line 99 regression see Fit Y by X, Fit Model Remove Fit 86, 91 report options 14 rescale axis 28 resize plot 89 Response role 13 role 13, 27, 75 Root Mean Square Error (RMSE) 65, 86, 103 rows 9 Rsquare 65, 86, 103 Rsquare Adj 65, 86, 103 Run Script button on JMP Starter 4

N
New Column 53 computing values 54 New Data Table 8 nominal 10, 15, 60, 75, 83 normal distribution 49, 51 notation used in manuals 7 Number 66 numeric column 10

O
object scripting index 5 Observations 65, 86, 103 Open Data Table 8 operators 5 ordinal 10, 15, 60, 83 outlier 50, 88, 107, 117 outlier box plot see Distribution

S
Save As 89 Save Predicteds 86 Save Prediction Formula 123

P
Parameter Estimates table 87 partitioning 102 pattern in data 107 Pearson Chi Square 77 percentile see quantile plane fit 124 platforms 13 pointer cursor 12 Polynomial fit 90 post hoc see Compare Means Prediction Formula 104 principal components analysis 113 Prob 52 Prob > |t| 88 Prob > F 66, 87, 103

saving a JMP session 16 scatterplot matrix 110 scatterplot see Fit Y by X, Multivariate, Overlay
Plot

Q
Quantile Box Plot 50, 63 Quantile Box Plot see Distribution

Script button 4 select rows and columns 12 selection tool 68 Shift-Tab 26 shortest half 50 Show Points 85 smoothing 91 solubility study 107118 Solubility.jmp 109 Source 66, 87, 102 spin controls 111 Spinning Plot example 17 spline 91 start JMP 23 statistical index 4 statistical summaries see Distribution

138

Index

Std Error 66, 87 StdErrProb 52 Subset 5255, 110 Sum of Squares 66, 87, 102 summarizing data 3143 Summary 34 Summary of Fit table 65, 86, 91, 103 survey data 6980

T
Tab 26 tension 91 Term 87 three-dimensional plots 119131 three-dimensional plots see Spinning Plot tick mark 60 Tip of the day 7 Tools 29 Topic Help button 4 t-ratio 87 t-test 87 tutorial 3, 8 tutorial examples data table 2130 drug experiment 2130 exploratory study 107118 multiple regression 119131 popcorn experiment 95106 regression analysis 83 summarizing data 3143 survey data 6980 typing study 5768 tutorials learning JMP 3 Typing Data.jmp 59 typing study 5768

W-Z
Weight role 13 weight-height ratio example 83 whiskers 50 Whole-model plot 99 X, Factor role 13 Y, Response role 13