Applied Linear Models with SAS - 1st Edition

Visit the link below to download the full version of this book:

https://medipdf.com/product/applied-linear-models-with-sas-1st-edition/

Click Download Now

Applied Linear Models
with SAS

Daniel Zelterman
Yale University
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore,
São Paulo, Delhi, Dubai, Tokyo

Cambridge University Press

The Edinburgh Building, Cambridge CB2 8RU, UK

Published in the United States of America by Cambridge University Press, New York

www.cambridge.org
Information on this title: www.cambridge.org/9780521761598
© Daniel Zelterman 2010

This publication is in copyright. Subject to statutory exception and to the

provision of relevant collective licensing agreements, no reproduction of any part
may take place without the written permission of Cambridge University Press.
First published in print format 2010

ISBN-13 978-0-511-77476-8 eBook (EBL)

ISBN-13 978-0-521-76159-8 Hardback

Cambridge University Press has no responsibility for the persistence or accuracy

of urls for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
 Contents

Preface page ix
Acknowledgments xiii

1 Introduction 1
1.1 What Is Statistics? 1
1.2 Statistics in the News: The Weather Map 4
1.3 Mathematical Background 6
1.4 Calculus 7
1.5 Calculus in the News: New Home Sales 9
1.6 Statistics in the News: IMF Loans and Tuberculosis 11
1.7 Exercises 13

2 Principles of Statistics 21
2.1 Binomial Distribution 21
2.2 Confidence Intervals and the Hubble Constant 25
2.3 Normal Distribution 26
2.4 Hypothesis Tests 30
2.5 The Student t-Test 34
2.6 The Chi-Squared Test and 2 × 2 Tables 42
2.7 What Are Degrees of Freedom? 47
2.8 SAS, in a Nutshell 49
2.9 Survey of the Rest of the Book 51
2.10 Exercises 52
3 Introduction to Linear Regression 58
3.1 Low-Birth-Weight Infants 58
3.2 The Least Squares Regression Line 59
3.3 Regression in SAS 63
3.4 Statistics in the News: Future Health Care Costs 65
3.5 Exercises 66
v
vi Contents

4 Assessing the Regression 75

4.1 Correlation 75
4.2 Statistics in the News: Correlations of the Global Economy 77
4.3 Analysis of Variance 78
4.4 Model Assumptions and Residual Plots 81
4.5 Exercises 84
5 Multiple Linear Regression 90
5.1 Introductory Example: Maximum January Temperatures 90
5.2 Graphical Displays of Multivariate Data 94
5.3 Leverage and the Hat Matrix Diagonal 96
5.4 Jackknife Diagnostics 99
5.5 Partial Regression Plots and Correlations 102
5.6 Model-Building Strategies 105
5.7 Exercises 110
6 Indicators, Interactions, and Transformations 120
6.1 Indicator Variables 120
6.2 Synergy in the News: Airline Mergers 127
6.3 Interactions of Explanatory Variables 128
6.4 Transformations 132
6.5 Additional Topics: Longitudinal Data 137
6.6 Exercises 138
7 Nonparametric Statistics 150
7.1 A Test for Medians 150
7.2 Statistics in the News: Math Achievement Scores 153
7.3 Rank Sum Test 155
7.4 Nonparametric Methods in SAS 156
7.5 Ranking and the Healthiest State 157
7.6 Nonparametric Regression: LOESS 160
7.7 Exercises 163
8 Logistic Regression 169
8.1 Example 169
8.2 The Logit Transformation 170
8.3 Logistic Regression in SAS 173
8.4 Statistics in the News: The New York Mets 177
8.5 Key Points 178
8.6 Exercises 179

9 Diagnostics for Logistic Regression 187

9.1 Some Syntax for proc logistic 188
9.2 Residuals for Logistic Regression 190
vii Contents

9.3 Influence in Logistic Regression 193

9.4 Exercises 197
10 Poisson Regression 204
10.1 Statistics in the News: Lottery Winners 204
10.2 Poisson Distribution Basics 204
10.3 Regression Models for Poisson Data 206
10.4 Statistics in the News: Attacks in Iraq 208
10.5 Poisson Regression in SAS 209
10.6 Exercises 215
11 Survival Analysis 225
11.1 Censoring 225
11.2 The Survival Curve and Its Estimate 227
11.3 The Log-Rank Test and SAS Program 232
11.4 Exercises 235
12 Proportional Hazards Regression 237
12.1 The Hazard Function 237
12.2 The Model of Proportional Hazards Regression 239
12.3 Proportional Hazards Regression in SAS 241
12.4 Exercises 243

13 Review of Methods 247

13.1 The Appropriate Method 247
13.2 Other Review Questions 249

Appendix: Statistical Tables 255

A.1 Normal Distribution 255
A.2 Chi-squared Tables 257

References 259
Selected Solutions and Hints 263
Index 269
 Preface

Linear models are a powerful and useful set of methods in a large number of settings.
Very briefly, there is some outcome measurement that is very important to us and we
want to explain variations in its values in terms of other measurements in the data.
The heights of several trees can be explained in terms of the trees’ ages, for example.
It is not a straight line relationship, of course, but knowledge of a tree’s age offers
us a large amount of explanatory value. We might also want to take into account
the effects of measurements on the amount of light, water, nutrients, and weather
conditions experienced by each tree. Some of these measurements will have greater
explanatory value than others and we may want to quantify the relative usefulness
of these different measures. Even after we are given all of this information, some
trees will appear to thrive and others will remain stunted, when all are subjected to
identical conditions. This variability is the whole reason for statistics existing as a
scientific discipline. We usually try to avoid the use of the word “prediction” because
this assumes that there is a cause-and-effect relationship. A tree’s age does not directly
cause it to grow, for example, but rather, a cumulative process associated with many
environmental factors results in increasing height and continued survival. The best
estimate we can make is a statement about the behavior of the average tree under
identical conditions.
Many of my students go on to work in the pharmaceutical or health-care industry
after graduating with a masters degree. Consequently, the choice of examples has a
decidedly health/medical bias. We expect our students to be useful to their employers
the day they leave our program so there is not a lot of time to spend on advanced
theory that is not directly applicable. Not all of the examples are from the health
sciences. Diverse examples such as the number of lottery winners and temperatures
in various US cities are part of our common knowledge. Such examples do not need
a lengthy explanation for the reader to appreciate many of the aspects of the data
being presented.
How is this book different? The mathematical content and notation are kept to
an absolute minimum. To paraphrase the noted physicist Steven Hawking, who
ix
x Preface

has written extensively for the popular audience, every equation loses half of your
audience. There is really no need for formulas and their derivations in a book of
this type if we rely on the computer to calculate quantities of interest. Long gone are
the days of doing statistics with calculators or on the back of an envelope. Students
of mathematical statistics should be able to provide the derivations of the formulas
but they represent a very different audience. All of the important formulas are
programmed in software so there is no need for the general user to know these.
The three important skills needed by a well-educated student of applied statistics
are
1. Recognize the appropriate method needed in a given setting.
2. Have the necessary computer skills to perform the analysis.
3. Be able to interpret the output and draw conclusions in terms of the original data.
This book gives examples to introduce the reader to a variety of commonly
encountered settings and provides guidance through these to complete these three
goals. Not all possible situations can be described, of course, but the chosen settings
include a broad survey of the type of problems the student of applied statistics is
likely to run into.
What do I ask of my readers? We still need to use a lot of mathematical concepts
such as the connection between a linear equation and drawing the line on X − Y
coordinates. There will be algebra and special functions such as square roots and
logarithms. Logarithms, while we are on the subject, are always to the base e (=2.718)
and not base 10.
We will also need a nodding acquaintance with the concepts of calculus. Many of
us may have taken calculus in college, a long time ago, and not had much need to use it
in the years since then. Perhaps we intentionally chose a course of study that avoided
abstract mathematics. Even so, calculus represents an important and useful tool. The
definition of the derivative of a function (What does this new function represent?)
and integral (What does this new function represent?) are needed although we will
never need to actually find a derivative or an integral. The necessary refresher to
these important concepts is given in Section 1.4.
Also helpful is a previous course in statistics. The reader should be familiar with the
mean and standard deviation, normal and binomial distributions, and hypothesis
tests in general and the chi-squared and t-tests specifically. These important concepts
are reviewed in Chapter 2 but an appreciation of these important ideas is almost a
full course in itself. There is a large reliance on p-values in scientific research so it is
important to know exactly what these represent.
There are a number of excellent general-purpose statistical packages available. We
have chosen to illustrate our examples using SAS because of its wide acceptance and
use in many industries but especially health care and pharmaceutical. Most of the
examples given here are small, to emphasize interpretation and encourage practice.
These datasets could be examined by most software packages. SAS, however, is
xi Preface

capable of handling huge datasets so the skills learned here can easily be used if and
when much larger projects are encountered later.
The reader should already have some familiarity with running SAS on a computer.
This would include using the editor to change the program, submitting the program,
and retrieving and then printing the output. There are also popular point-and-click
approaches to data analysis. While these are quick and acceptable, their ease of use
comes with the price of not always being able to repeat the analysis because of the
lack of a printed record of the steps that were taken. Data analysis, then, should be
reproducible.
We will review some of the basics of SAS but a little hand-holding will prevent
some of the agonizing frustrations that can occur when first starting out. Running the
computer and, more generally, doing the exercises in this book are a very necessary
part of learning statistics. Just as you cannot learn to play the piano simply by
reading a book, statistical expertise, and the accompanying computer skills, can only
be obtained through hours of active participation in the relevant act. Again, much
like the piano, the instrument is not damaged by playing a wrong note. Nobody will
laugh at you if you try something truly outlandish on the computer either. Perhaps
something better will come of a new look at a familiar setting. Similarly, the reader is
encouraged to look at the data and try a variety of different ways of looking, plotting,
modeling, transforming, and manipulating. Unlike a mathematical problem with
only one correct solution (contrary to many of our preconceived notions), there is
often a lot of flexibility in the way statistics can be applied to summarize a set of data.
As with yet another analogy to music, there are many ways to play the same song.
 Acknowledgments

Thanks to the many students and teaching assistants who have provided useful com-
ments and suggestions to the exposition as well as the computer assignments. Also
to Chang Yu, Steven Schwager, and Amelia Dziengeleski for their careful readings of
early drafts of the manuscript. Lauren Cowles and her staff at Cambridge University
Press provided innumerable improvements and links to useful Web sites.
The DASL (pronounced “dazzle”) StatLib library maintained at Carnegie Mellon
University is a great resource and provided data for many examples and exercises
contained here. Ed Tufte’s books on graphics have taught me to look at data more
carefully. His books are highly recommended.
I am grateful to The New York Times for their permission to use many graphic
illustrations.
Finally, thanks to my wife Linda who provided important doses of encouragement
and kept me on task. This work is dedicated to her memory.
The Pennsylvania State University Department of Meteorology supplied the
graphics for the weather map in Fig. 1.1.

DANIEL ZELTERMAN
Hamden, CT
August 25, 2009

xiii
The following figures are copyright The New York Times and used with permission: Figure 1.4
(June 23, 2007); Figure 1.6 (August 15, 2008); Figure 1.7 (August 4, 2008); Figure 1.8 (August
23, 2008); Figure 1.9 (January 8, 2009); Figure 1.10 (October 14, 2008); Figure 4.4 (April 17,
2008); Figure 6.1 (January 17, 2008); Figure 7.1 (June 13, 2007); Figure 10.3 (May 30, 2007). All
rights are reserved by The New York Times.

Table 2.1: From J. P. Frisby and J. L. Clatworthy, “Learning to see complex random-dot stere-
ograms,” Perception 4(2), pp. 173–78. Copyright 1975 Pion Limited, London.
Figure 2.1: Courtesy of John Huchra.
Table 2.8: From Marcello Pagano and Kimberlee Gauvreau. Principles of Biostatistics, 2E.
Copyright 2000 Brooks/Cole, a part of Cengage Learning, Inc. Reproduced by permission.
www.cengage.com/permissions
Table 2.10: From N. Teasdale, C. Bard, J. Larue, et al., “On the cognitive penetrability of posture
control,” Experimental Aging Research. Copyright 1993 Taylor & Francis, reprinted by permission
of the publisher (Taylor & Francis Group, http://www.informaworld.com).
Tables 5.1 and 6.15: From Frederick Mosteller and John W. Tukey, Data Analysis and Regression:
A Second Course in Statistics, Table 5.1 pp. 73–74 and Table 6.15 pp. 549–51. Copyright 1977
Addison-Wesley Publishing Company, Inc. Reproduced by permission of Pearson Education,
Inc.
Table 6.10: From Douglas Bates and Donald Watts, Nonlinear Regression Analysis and Its Applica-
tions. Copyright 2007 John Wiley and Sons, Inc. Reprinted with permission of John Wiley and
Sons, Inc.
Table 6.11: From James A. Koziol and Donna A. Maxwell, “A distribution-free test for tumor-
growth curve analysis with application to an animal tumor immunotherapy experiment,” Bio-
metrics 37, pp. 383–90. Reprinted with permission of Oxford University Press.
Table 7.1: From A. J. Dobson, Introduction to Generalized Linear Models, 2E. Copyright 2001
Taylor & Francis Group LLC – Books. Reproduced with permission of Taylor & Francis Group
LLC – Books in the format textbook via Copyright Clearance Center.
Table 9.1: From R. G. Miller et al., Biostatistics Casebook 318. Copyright 1980 John Wiley and
Sons, Inc. Reprinted with permission of John Wiley and Sons, Inc.
Table 10.7: From P. J. Antsaklis and K. M. Passino, Introduction to Intelligent and Autonomous
Control, Chapter 2, pp. 27–56; Figure 2 of James S. Albus, “A Reference Model Architecture for
Intelligent Systems Design.” Copyright 1993 Kluwer Academic Publishers with kind permission
of Springer Science and Business Media.
Table 10.11: From Pieter Joost van Wattum et al., “Patterns of response to acute Naxolone infusion
in Tourette’s Syndrome,” Movement Disorders 15 (2000) pp. 1252–54. Reprinted with permission
of Oxford University Press.
Table 11.2: From Jennifer Wheler et al., “Survival of patients in a phase 1 clinic,” Cancer 115 (5),
pp. 1091–99. Copyright 2009 American Cancer Society.
Table 11.4: From Thomas R. Fleming et al., “Modified Kolmogorov-Smirnov test procedures with
application to arbitrarily right-censored data,” Biometrics 36 (1980), pp. 607–25. Reprinted with
permission of Oxford University Press.
Table 12.2: From Nicholas Lange et al., Case Studies in Biometry. Copyright 1994 John Wiley and
Sons, Inc. Reprinted with permission of John Wiley and Sons, Inc.
Table 12.4: From J. M. Krall, V. A. Uthoff, and J. B. Harley. “A step-up procedure for selecting
variables associated with survival,” Biometrics 31 (1975), pp. 49–57. Reprinted with permission
of Oxford University Press.
 1

Introduction

We are surrounded by data. With a tap at a computer keyboard, we have access

to more than we could possibly absorb in a lifetime. But is this data the same as
information? How do we get from numbers to understanding? How do we iden-
tify simplifying trends – but also find exceptions to the rule? The computers that
provide access to the data also provide the tools to answer these questions. Unfor-
tunately, owning a hammer does not enable us to build a fine house. It takes expe-
rience using the tools, knowing when they are appropriate, and also knowing their
limitations.
The study of statistics provides the tools to create understanding out of raw data.
Expertise comes with experience, of course. We need equal amounts of theory (in
the form of statistical tools), technical skills (at the computer), and critical analysis
(identifying the limitations of various methods for each setting). A lack of one of
these cannot be made up by the other two.
This chapter provides a review of statistics in general, along with the mathematical
and statistical prerequisites that will be used in subsequent chapters. Even more
broadly, the reader will be reminded of the larger picture. It is very easy to learn
many statistical methods only to lose sight of the point of it all.

1.1 What Is Statistics?

In an effort to present a lot of mathematical formulas, we sometimes lose track of
the central idea of the discipline. It is important to remember the big picture when
we get too close to the subject.
Let us consider a vast wall that separates our lives from the place where the
information resides. It is impossible to see over or around this wall, but every now
and then we have the good fortune of having some pieces of data thrown over
to us. On the basis of this fragmentary sampled data, we are supposed to infer the
composition of the remainder on the other side. This is the aim of statistical inference.