HighBeam Research - Article - UM Personal World Wide Web Server

advertisement
HighBeam Research
Title: Linear Mixed Models, A Practical Guide Using Statistical Software
Date: 4/1/2008; Publication: Journal of Quality Technology; Author: Jearkpaporn,
Duangporn
Linear Mixed Models, A Practical Guide Using Statistical Software by Brady T. West,
Kathleen B. Welch, and Andrzej T. Galecki with contributions from Brenda W. Gillespie.
Chapman & Hall/CRC, Boca Raton, Florida 2007. 353 pp, $79.95..
THE BOOK TITLE is a perfect description of this textbook. This book is designed as a
reference on the use of procedures for fitting linear mixed models (LMMs) available in
five widely used software packages, including SAS, SPSS, Stata, R/S-plus, and the
hierarchical linear model software (HLM). The authors clearly show the use of
procedures by working through real-world examples. As the authors mention, this book
is not designed to substitute software manuals but rather as a practical guide for applied
researchers and statisticians to appropriately use LMMs for their data analysis
problems, employing procedures available in these software packages.
The book focuses only on LMMs (sometimes known as multilevel models, or
hierarchical linear models) which are defined as parametric linear models for continuous
outcome variables in which the residuals are normally distributed but may not be
independent or have constant variance. Data sets that may be appropriately analyzed
using LMMs include clustered data, repeated-measures studies, and longitudinal data.
This book is designed for readers who have a general familiarity with a basic working
knowledge of matrix algebra, ordinary linear regression, and ANOVA models. Nonlinear
mixed models and generalized LMMs, spatial correlation structures are not discussed in
this book but a few references on these topics are provided. The authors leave the
theoretical treatment of the linear mixed models and the analysis of variance
components out of this textbook with the intention of making it more readable by the
applied researchers. The book thus includes only the primary concepts and notation
with the focus on the software implementation and model interpretation. References to
theoretical treatment are provided for interested readers.
There are seven chapters in this book. The first two chapters are an introduction and
overview of the book and the topic. The remaining chapters illustrate model fitting to
different type of data structures. Statistical software resources and commonly used
abbreviations and acronyms associated with LMMs are presented in the appendices. In
Chapter 1, the authors define LMMs, outline the book contents, and summarizes a brief
history of LMMs both theoretical and software developments. Chapter 2 introduces the
primary concept of the linear mixed model, including key theories underlying LMMs for
clustered, longitudinal, and repeated-measures data, model building strategies, and
model diagnostics. Chapter 2 also presents the notation and model specification used in
the book. I would recommend readers to skim through this chapter prior to reading
Chapters 3 to 7.
Chapters 3 through 7 demonstrate the use of software packages using real-world
examples. Each of these chapters stands alone, but follows a standardized format and
notation. Each begins with the overview of the case study and model specification,
followed by analysis steps, results and interpretations, model diagnostics, outputs from
fitted models, and comparisons across the software procedures. Each chapter closes
with the software notes. In Chapters 3, 5, and 7, the alternative approaches to address
the problem are also discussed.
Chapters 3 and 4 examine a two- and three-level model for clustered data by analyzing
the birth weights of rat pups within litters and the math scores of students within
classrooms, respectively. Chapter 5 shows an example of repeated-measures data
using the rat brain data while Chapter 6 shows the example of longitudinal data using
Autism data. An example with features of both clustered and longitudinal data is
demonstrated thru dental veneer data shown in Chapter 7. Each chapter includes
analysis steps and syntax used, detail tables of estimates and results. The authors also
provide a supporting website www.umich.edu/~bwest/almmussp.html for the most up-todate versions of the syntax and data sets used in this book.
Overall, I believe that this text is a good reference for any practicing statisticians and
researchers who want a basic introduction to the topic and an easy-to-navigate software
reference. The book is also useful for researchers who are in need to compare their
analysis to existing works done using different software packages. Because the basic
concept is well summarized and presented through examples and tables which make it
easy to understand for beginner readers I would recommend this textbook as a special
topic for teaching an advanced undergraduate or introductory graduate course on linear
models.
[Author Affiliation]
Reviewer: Duangporn Jearkpaporn, Diversified Product Group, Wells Fargo, Phoenix,
Arizona 85029.
Copyright American Society for Quality Apr 2008
This document provided by HighBeam Research at http://www.highbeam.com
Download