The Stata Journal (2004)

4, Number 3, pp. 350–353

Review of A Handbook of Statistical Analyses

Using Stata by Rabe-Hesketh and Everitt
Nicholas Winter
Department of Government
White Hall
Cornell University
Ithaca, NY 14853
nw53@cornell.edu

Abstract. The new edition of the book by Rabe-Hesketh and Everitt (2004) is
reviewed.
Keywords: gn0013, introductory, gllamm, Stata texts

1 Introduction
The complete set of Stata manuals fills 13 volumes and weighs just over 26 pounds
(almost 12 kg). Gone are the days when one could read the entire documentation in
a long day at the beach and attain a relatively complete knowledge of the program.
Thus, there is considerable demand for reasonably sized guides that provide a point of
entry that is more user friendly and less intimidating than the manuals. A Handbook of
Statistical Analysis Using Stata, 3rd ed. by Sophia Rabe-Hesketh and Brian Everitt is
a newly revised version of one such resource. This book is positioned between the more
introductory text by Hamilton (2004) and more advanced, book-length treatments of
specific modeling strategies using Stata, such as Long and Freese (2003) on regression
modeling with categorical dependent variables or Cleves, Gould, and Gutierrez (2004)
on survival analysis.

2 Overview of the book

After giving a brief introduction to the basics of the program, this book covers a broad
set of intermediate and advanced techniques in Stata. It includes chapters on descriptive
statistics, regression, ANOVA, logistic regression, generalized linear models, descriptive
analysis of longitudinal data, random-effects models, generalized estimating equations,
basic epidemiology, survival analysis, maximum likelihood, principal components, and
cluster analysis—all in less than 300 pages!
Each chapter includes a brief introduction to the relevant statistical model and
then illustrates an analysis using that technique with a substantive problem and a real
dataset. The book includes complete Stata syntax for each command in the analysis,
interspersed with explanations of the Stata features and analytic techniques being em-
ployed, along with the output that Stata produces. The datasets are available from the

c 2004 StataCorp LP

gn0013
N. Winter 351

Internet, so readers can download them and duplicate or extend the analyses as they
read. (Oddly, the book does not include directions for loading the datasets directly into
Stata from the Internet; the web page simply instructs users to download them as a
package to their local computer.)
The chapter on random-effects models is one of the longest and illustrates this ap-
proach well. The chapter begins with a very brief discussion of random-effects models,
including random-intercept and random-coefficient models, as well as mixed and multi-
level models. Using a dataset on postnatal depression, the authors fit a random-intercept
model using xtreg and illustrate various postestimation tasks, including graphical anal-
ysis and the use of lincom. Then they turn to gllamm, a user-written program that
fits generalized linear latent and mixed models. They first refit the equivalent model
and then extend it to include random slopes. Each estimation result is followed by
a set of further analyses that illustrate ways to use Stata’s postestimation facilities,
including an analysis of residual normality using kdensity, plots of individual-specific
response curves, and a plot of predicted values with a confidence interval. The chapter
concludes with an analysis, using xtlogit and gllamm, of a dichotomous dependent
variable; this section includes both a brief theoretical discussion and analysis of various
types of predicted probabilities.
The major changes in this new edition of the book include the addition of two new
chapters (on random effects and cluster analysis) and the updating of the graphics
throughout the book to version 8 syntax and output.

3 Strengths
The best characteristic of the book is the way it demonstrates the flow of an analysis in
Stata. As the analysis in each chapter unfolds, readers see the ways that Stata’s data
management, graphics, model estimation, and postestimation facilities flow together,
always in the context of a particular statistical model, substantive problem, and dataset.
The book also shows off various tips and tricks for using Stata that the novice or
intermediate Stata user might not be aware of or habitually use. A reader working
through the analyses will pick up along the way a quick view of a wide range of tools,
including several different uses of foreach and forvalues; the use of tabstat as an
alternative to table; the preserve/restore sequence; the use of collapse; the use of
reshape; and more. This is where a hands-on book of this sort is at its strongest; using
the manuals is a particularly difficult way to learn these sorts of things, if only because
one does not know that a particular tool or approach would be helpful and, therefore,
does not know to look it up.
Another nice feature (I would have appreciated more of them) is the occasional
set of discussions on differences between modeling strategies; for example, xtreg and
gllamm are compared directly, and random effects-style approaches are contrasted with
generalized estimating equations for longitudinal data (page 183). For anyone relatively
new to longitudinal analysis, for example, these sorts of comparisons could be quite
helpful.
352 Review of Rabe-Hesketh and Everitt

4 Limitations
For this sort of book, there is always a trade-off between breadth and depth (and
length). Because the book is relatively short and attempts to cover an extremely wide
set of techniques, there are definite costs in terms of the depth of coverage it can give to
each. The random-effects chapter discussed above is among the longest in the book; at
the other extreme is the 11-page chapter on principal-components analysis. This allows
for only an extremely cursory introduction to each technique and a rather quick glimpse
of Stata’s handling of it. This may be a useful introduction, but clearly researchers who
want to employ any of these methods for their own research would need to use this book
in conjunction with the Stata manuals. They would also be well advised to consult more
advanced texts on any of the methods they are not already familiar with—helpfully,
the authors give many such references.
This combination of extreme breadth and limited depth makes it difficult to know
precisely whom this book is aimed at. It is clearly pitched above the level of the
statistical novice, as the introduction to each analytic technique is quite brief; on the
other hand, the introductions should be superfluous for anyone well versed in these
techniques. Moreover, the data manipulation at the beginning of each chapter may be
confusing to Stata newcomers. For example, in the chapter on logistic regression, the
data are loaded as a frequency table; egen, seq() is used to create the covariates; the
data are then manipulated with reshape and expand; and, finally, the response variable
is dichotomized. All this is done in less than two pages, which means that there can
be little explanation of what each step accomplishes. Perhaps readers would be better
served if the data were provided in a form more suitable for analysis.
In addition, and perhaps inevitably, I took issue with some of the statistical pro-
cedures employed. Specifically, the multiple regression chapter employs stepwise re-
gression, a technique notorious for its problems (see, e.g., discussion by Harrell [2001]).
While the authors do express caveats about stepwise procedures, their inclusion at least
raises questions about the statistical advice they give for other techniques as well. In
addition, they make some choices about data management that could get a novice user
into trouble. To exclude an outlier in the regression analysis, for example, they drop
it from the dataset, rather than excluding it from the analysis with an if condition.
Without a clear explanation that this is what’s going on, new users might think that this
command means “exclude this from analysis”, rather than “drop from the dataset per-
manently”. At the very least, such users might inadvertently save the dataset without
realizing what they were doing.

5 Conclusion
This book would probably be most useful for two types of Stata users. First, it would
be helpful to working researchers or students who have a basic knowledge of Stata but
are interested in learning a bit more about its capabilities, especially in the context of
several analytic techniques with which they may be unfamiliar. For them, the book can
serve as an accessible stepping stone both to the techniques and to modeling in Stata
N. Winter 353

more generally. Second, for working researchers who are new to Stata and want to make
use of several of the techniques listed above, this book can serve as a nice introduction
to Stata in general, as well as its implementation of the approaches.

6 References
Cleves, M., W. Gould, and R. Gutierrez. 2004. An Introduction to Survival Analysis
Using Stata. rev. ed. College Station, TX: Stata Press.

Hamilton, L. C. 2004. Statistics with Stata (Updated for Version 8). Pacific Grove, CA:
Duxbury Press.

Harrell, F. E. 2001. Regression Modeling Strategies. New York: Springer.

Long, J. S. and J. Freese. 2003. Regression Models for Categorical Dependent Variables
Using Stata. rev. ed. College Station, TX: Stata Press.

Rabe-Hesketh, S. and B. Everitt. 2004. A Handbook of Statistical Analyses Using Stata.

3rd ed. Boca Raton, FL: Chapman & Hall/CRC.

About the Author

Nicholas Winter is an Assistant Professor in the Department of Government at Cornell Univer-
sity. His interests include American politics; public opinion, race and gender, and politics; and
methodology. He is a frequent contributor to Statalist and has written several Stata packages,
which may be downloaded from the SSC archive.