BOOK REVIEWS 3213

GEE is not mentioned at all in this chapter,which is a serious omission because it leavesthe impression that GEE is not useful for linearmodels; the sandwich variance estimate is onlybriefly mentioned.

Chapter 3 introduces linear random coefficientmodels in which coefficients for explanatoryvariables are allowed to vary randomly acrossclusters. The interpretation of models is againdone with great care. However, assumptions aboutvariances and covariances are again underempha-sized although there is an interesting advancedsection on modeling heteroskedasticity.

Models for binary responses are covered inChapter 4. The discussions of the interpretationof latent response models and comparisons ofmarginal (‘population averaged’) with conditional(‘subject-specific’) models are excellent. Mate-rial on adaptive quadrature and empirical Bayesprediction is good, although it might be difficultto follow for many readers who do not alreadyhave a basic understanding of these subjects. GEEis introduced here for the first time in the bookbut is not given sufficient attention.

The final four chapters of the book cover specialmodels, including ordinal responses (Chapter 5),counts (Chapter 6), nested models with three or

more levels (Chapter 7), and crossed randomeffects (Chapter 8). These chapters, like the earlierones, focus on random effects models. The authorscontinue to do a very nice job of discussingthe interpretation of the models and illustratingmethods for model fitting using Stata.

In summary, this is a useful text for researcherswho need to learn how to model multi-level data, especially for those interested inapplying random effects models. It is hopedthat the authors will produce an improvedfuture edition that can better serve as a stand-alonetext. To achieve this goal, what is needed isadditional background material, more completecoverage of modeling assumptions, and a morebalanced presentation of marginal and conditionalmodels.

BRIAN LEROUXDepartment of Biostatistics

School of Public Health andCommunity Medicine

University of WashingtonSeattle, WA, U.S.A.

(DOI: 10.1002/sim.3225)

3. MARKOV CHAIN MONTE CARLO: STOCHASTICSIMULATION FOR BAYESIAN INFERENCE (2ndedn). Dani Gamerman and Hedibert F. Lopes,Chapman & Hall/CRC, Boca Raton, FL, 2006.No. of pages: xvii +323. Price: $69.95. ISBN10:1-58488-587-4, ISBN13: 978-1-58488-587-0

The second edition of this book by Gamermanand Lopes is well written and builds on thefirst edition, published almost a decade ago. Theintention of the second edition is to update theprevious edition with the developments in the fieldof Monte Carlo Markov chain (MCMC) researchthat have occurred during this period. The authorsnote that these advances have largely been in thespace of methodological and application settings.As such the second edition is aimed at focusingon assessing and presenting pros and cons ofthe different options available when simulation isperformed with an MCMC algorithm.

The addition of an associated Web sitewww.ufrj.br/MCMC is a valuable resource that

contains many R scripts, allowing the readers toquickly and easily test different approaches on theirdesired models with minimal effort. Coupling thiswith the depth of examples and references provided,this text provides an excellent first graduate text onMCMC methods.

New topics covered include spatial statisticalmodels, valuable in many areas of modern statisticsfrom weather modeling, computer vision modelingto epidemiological model analysis. Additionally,model assessment and selection, noteably missingfrom the first text, is now covered. This includesrelevant criteria developed for model selectionapproaches such as AIC, BIC and DIC. In additionto these topics, many sections have been updatedto reflect new literature in the field and substantialincreases in practical examples are provided.

A consequence of the broad target audienceintended for this text is that several theoretical andmore advanced recent methodological insights havenot been included in detail. These include alter-natives such as adaptive MCMC, slice samplingand auxiliary approaches. Additionally, approaches

Copyright q 2008 John Wiley & Sons, Ltd. Statist. Med. 2008; 27:3211–3215

3214 BOOK REVIEWS

such as population Monte Carlo and alternativessuch as SMC samplers are only mentioned inpassing. Having said this, the target audience is,however, well served with many references topapers and other texts all of which may act asexcellent companions to this text. Hence the aimto allow non-statisticians to easily pick up andapply the methods presented is achieved and theseomissions do not detract from the books intendedaudience since appropriate references are providedfor the more inquisitive reader.

Chapter 1 deals with introductory stochasticsimulation ideas including inversion samplingfor both discrete and continuous distributions. Ineach example, relevant models commonly usedare presented. A brief section on multivariateanalysis is presented before finishing with rejectionsampling and adaptive rejection sampling. Chapter2 introduces the fundamentals of the Bayesianparadigm and then continues with conjugatemodels followed by brief sections on hierarchicaland dynamic models. The chapter concludes withsections on spatial models including GaussianMarkov random fields and Gaussian random fieldsand a section on model assessment or goodness-of-fit analysis. Although these sections are brief theyprovide extensive reference to relevant literature.This coupled with the examples provides ampleinformation for developing knowledge of theseareas.

It is important to always question how wella given simulation procedure is performing, andoften a problem that arises is what would be a fairand accurate comparison benchmark. The authorsaddress this question in Chapter 3 where theypresent a chapter on approximate methods of infer-ence which can act as a benchmark for further anal-ysis of performance of the simulation algorithmsto be presented in Chapters 5–7.

Chapter 4 is largely concerned with fundamentalMarkov chain theory at an introductory level; formore advanced coverage, the reader is directedto appropriate texts. This chapter is divided intotwo sections: discrete state space models wheretheory for classification of a state space, stationarydistributions, birth and death processes and limit

theorems are presented and then concludes with asection on continuous state spaces.

It would be fair to say that for the more expe-rienced reader, Chapters 5–7 make up the keychapters in this text. They comprehensively coverGibbs sampling with many examples and graph-ical explanations provided. This is coupled withthe code in some places for ease of application.Topics such as convergence diagnostics, reparam-eterization and blocking are all covered with valu-able insights included to help avoid pitfalls thatmay arise in blind applications. Then Chapter 6covers Metropolis Hastings algorithms and the vari-ants thereof. These include symmetric proposals,random walk chains, independent chains and hybridalgorithms. At the end of this section there are afew comprehensive examples that can be workedon by the reader for a deeper understanding of themethodology presented.

To conclude Chapter 7 presents some moreadvanced topics, which will be valuable tools forreaders already familiar with basics of MCMC.It covers model adequacy, model selection andvariable selection. Topics such as importancesampling, annealed importance sampling andbridge and path identities are covered in the book.The book concludes with a good introduction totrans-dimensional samplers covering both productspace formulations and Reversible Jump MCMCalgorithms.

The book is certainly another fine addition on theliterature on MCMC and should be used by anyoneinterested in gaining a solid foundation in MCMCmethods and algorithms. In addition, several valu-able resources are provided to help develop readersalready partly familiar with MCMC approaches.The extensive set of examples and references makeit a valuable text for a graduate course on MCMC.

GARETH PETERSDepartment of Mathematics and Statistics

University of NSW, SydneyAustralia

(DOI: 10.1002/sim.3240)

4. INTRODUCTION TO NONPARAMETRIC REGRES-SION. Kunio Takezawa, John Wiley & Sons Inc.,Hoboken, New Jersey, 2006. No. of pages: xviii +538. Price: $123.00. ISBN: 978-0-471-74583-9

This is a compendious and occasionally labo-rious introduction to many different brandsof nonparametric regression. It is a transla-tion from Japanese of a longer text in two

Copyright q 2008 John Wiley & Sons, Ltd. Statist. Med. 2008; 27:3211–3215