variables and of nested designs is fundamental to even begin to sort through that literature.
This book cannot undertake the former, deriving the necessary models, but it can, and does,
address the latter by building a foundation under both fixed and random designs and nest-
ing. I have tried to build similar foundations for other topics, for example, more modern
graphical devices and resampling statistics, where I can do that without dragging the reader
deeper into a swamp. In some ways my responsibility is to try to anticipate where we are
going and give the reader a basis for moving in that direction.Changes in the Seventh Edition
This seventh edition contains several new or expanded features that make the book more
appealing to the student and more relevant to the actual process of methodology and data
analysis:- I have continued to respond to the issue faced by the American Psychological Associa-
tion’s committee on null hypothesis testing, and have included even more material on
effect size and magnitude of effect. The coverage in this edition goes well beyond that in
previous editions, and should serve as a thorough introduction to the material. - I have further developed discussion of a proposal put forth by Jones and Tukey (2000) in
which they reconceived of hypothesis testing in ways that I find very helpful. However,
I have also retained the more traditional approach because students will be expected to
be familiar with it. - I have included new material on graphical displays, including probability plots, kernel
density plots, and residual plots. Each of these helps all of us to better understand our
data and to evaluate the reasonableness of the assumptions we make. - I have updated some of the material on computer solutions and have adapted the discus-
sion and displays to SPSS Version 16. - There is now coverage of the Cochran-Mantel-Haenszel analysis of contingency tables.
This is tied to the classic example of Simpson’s Paradox as applied to the Berkeley grad-
uate admissions data. This relates to the underlying goal of leading students to think
deeply about what their data mean. - I have somewhat modified Chapter 12 on multiple comparison techniques to cut down
on the wide range of tests that I previously discussed and to include coverage of
Benjamini and Hochberg’s False Discovery Rate. As we move our attention away from
familywise error rates to the false discovery rate we increase the power of our analyses
at relatively little cost in terms of Type I errors. - A new section in the chapter on repeated measures analysis of variance replaces the pre-
vious discussion of multivariate analysis of variance with a discussion of mixed models.
This approach allows for much better treatment of missing data and relaxes unreason-
able assumptions about compound symmetry. This serves as an introduction to mixed
models without attempting to take on a whole new field at once. - Data for all examples and problems are available on the Web.
- I have spent a substantial amount of time pulling together material for instructors and
students, and placing it on Web pages on the Internet. Users can readily access additional
(and complex) examples, discussion of topics that aren‘t covered in the text, additional
data, other sources on the Internet, demonstrations that would be suitable for class or for
a lab, and so on. Many places in the book refer specifically to this material if the student
wishes to pursue a topic further. All of this is easily available to anyone with an Internet
connection. I continue to add to this material, and encourage people to use it and critique it.
xviii Preface