Medical Biostatistics, Second Edition by Abhaya Indrayan

International Statistical Review (2009), 77, 2, 300–328 doi:10.1111/j.1751-5823.2009.00085.x

Short Book ReviewsEditor: Simo Puntanen

Time Series Analysis With Applications in R, Second EditionJonathan D. Cryer, Kung-Sik ChanSpringer, 2008, xiv + 491 pages, € 69.95 / £ 55.99 / US$ 84.95, hardcoverISBN: 978-0-387-75958-6

Table of contents

1. Introduction 9. Forecasting2. Fundamental concepts 10. Seasonal models3. Trends 11. Time series regression models4. Models for stationary time series 12. Time series models of heteroscedasticity5. Models for nonstationary time series 13. Introduction to spectral analysis6. Model specification 14. Estimating the spectrum7. Parameter estimation 15. Threshold models8. Model diagnostics 16. Appendix: An introduction to R

Readership: Later year undergraduates, beginning graduate students, and researchers andgraduate students in any discipline needing to explore and analyze time series data.

Chapters 1–2 give examples of time series, and introduce the mathematics needed for workingwith the variance-covariance structure of time series models. Chapter 3 compares stochasticwith deterministic trends, introduces regression methods, and discusses residual analysis.

Chapters 4–11 expound ARIMA processes, starting with autoregressive and moving averageprocesses in Chapter 4, and successively adding further refinements and technicalities in laterchapters. An appendix to Chapter 11, on Forecasting, has a very brief discussion of state spacemodels.

Chapter 12 describes the ARCH and GARCH models for heteroscedasticity that are popularfor use with financial time series.

Chapters 13 and 14 describe the use of spectral analysis. Chapter 15 describes thresholdmodels.

Chapters 11–15 are new to this second edition. (The first edition appeared in 1986.)An appendix lists R commands that can be used to reproduce the analyses and simulations.

These, and the data sets, are also available from the book’s web site. Users of the R system willfind it easiest to obtain the data sets from the R package TSA. This package has a number ofadditional functions that are tailored to make it easy to reproduce the results in the text.

This is a careful and staged introduction both to the mathematical theory and to practicaltime series analysis. Simulation, and associated plots, are used throughout as an aid to intuition.There is extensive detailed comment on practical issues that arise in the application of the

C© 2009 The Authors. Journal compilation C© 2009 International Statistical Institute. Published by Blackwell Publishing Ltd, 9600 Garsington Road,Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.

SHORT BOOK REVIEWS 301

methodologies. Exercises at the end of each chapter mix theory, simulation and data analysis,with a bias towards data analysis.

John H. Maindonald: [email protected] for Mathematics & Its Applications

Australian National UniversityCanberra ACT 0200, Australia

Bayesian Methods: A Social and Behavioral Sciences Approach, Second EditionJeff GillChapman & Hall/CRC, 2007, 752 pages, £ 46.99 / US$ 73.95, hardcoverISBN: 978-1-58488-562-7

Table of contents

1. Background and introduction 9. Basics of Markov chain Monte Carlo2. Specifying Bayesian models 10. Bayesian hierarchical models3. The normal and Student’s t models 11. Some Markov chain Monte Carlo theory4. The Bayesian linear model 12. Utilitarian Markov chain Monte Carlo5. The Bayesian prior 13. Advanced Markov chain Monte Carlo6. Assessing model quality Appendix A: Generalized linear model review7. Bayesian hypothesis testing and the Bayes’ factor Appendix B: Common probability distributions8. Monte Carlo methods Appendix C: Introduction to the BUGS language

Readership: The book will be very suitable for students of social science, for example politicalscience, who wish to analyze data using modern Bayesian methods.

One is tempted to think of this book as essentially consisting of two parts, Chapters 1 through 7on basics of Bayesian Analysis, and Chapters 8 through 13 on MCMC and Hierarchical BayesianAnalysis. However, this simplistic view is not wholly correct, since MCMC calculations are usedthroughout the first six chapters at least by way of illustration. But this conceptual division intotwo parts would help a reader as well as a teacher using it as a text.

Chapter 1 is a beautifully written introduction to the whole subject of Bayesian Analysis.Along with some more material from the other chapters, it could serve as a crash course onBayesian Analysis. I have a small caveat about Section 1.7, where the scientific aspects of SocialScience are discussed. There are three more issues that could have been discussed here: (1)the difficulties of replicability, (2) the hazards of trying to model human behavior (collective orindividual), and (3) the near impossibility of controlled experiments, where some factors leadingto variation are controlled at fixed levels.

Chapter 2 provides the first of several interesting examples, mostly from Political Science.This one concerns the duration of cabinets of some chosen European countries. The posterioranalysis is simple but interesting.

Examples in the later chapters include French Labor Strikes, Ancient Chinese Conflicts, andthe ongoing Afghan war, where the analysis identifies the point of time when the Taliban beganto wrest the initiative from Afghan Government and its Allies.

Chapter 7 on Bayesian Testing is somewhat weak compared with other chapters, due in partto the lack of consensus among Bayesians about sharp nulls. Gill uses Bayes Factors but notfor sharp nulls. He tries to avoid testing sharp nulls by reporting a credibility interval for theparameter, but then notes if the interval contains zero or not, which is a form of testing advocated

International Statistical Review (2009), 77, 2, 300–328C© 2009 The Authors. Journal compilation C© 2009 International Statistical Institute

302 SHORT BOOK REVIEWS

by some Bayesians. He wavers between dismissing sharp nulls completely useless and helplesslyusing them in bread and butter problems of variable selection in linear and generalized linearmodels.

The second part has a lot of material on MCMC, some of which seems a bit too technicaland not strictly relevant for potential users of the book. But they could serve a useful purpose ifsome readers are willing to supplement this material with more standard texts on MCMC likeCasella and Robert. Though the importance of Hierarchical Bayesian Analysis and MCMC isexplained at different places, the chapter on Hierarchical Bayes is good but surprisingly short.Hopefully, it will grow in future chapters.

The book has a few typos or what appear to be typos. For example, the Jeffreys prior isdefined without a mention of a determinant and it isn’t easy to realize we are looking at adeterminant, not a matrix. The reference list is carefully compiled, it will be very useful fora well-motivated reader. Altogether it is a very readable book, based on solid scholarship andwritten with conviction, gusto, and a sense of fun.

Jayanta K. Ghosh: [email protected] of Statistics, Purdue University

West Lafayette, IN 47909, USA

Morphometrics with RJulien Claude with contributions from Michel Baylac, Emmanuel Paradis, and Tristan StaytonSpringer, 2008, xviii + 316 pages, € 44.95 / £ 35.99 / US$ 59.95, softcoverISBN: 978-0-387-77789-4

Table of contents

1. Introduction 6. Statistical analysis of shape using modern2. Acquiring and manipulating morphometric data morphometrics3. Traditional statistics for morphometrics 7. Going further with R4. Modern morphometrics based on configurations of Appendix A: Functions developed in this text

landmarks Appendix B: Packages used in this text5. Statistical analysis of outlines

Readership: Beginning graduate students, and researchers and graduate students in any disciplineneeding to explore and analyze morphometric data.

Chapter 1 begins with a short introduction to geometric morphometrics. A key idea is landmark,a position that is comparable across different organisms or other objects of study. For biologicalorganisms, it should reflect homology arising, usually, from a similar developmental origin. Themethodologies that are described here build on the pioneering work of D’Arcy Thompson. Shapesare mapped to a grid, with changes between objects described by mathematical transformations.Simple changes may for example include one or more of distending, flattening and shearing. Anintroduction to the R system occupies the major part of the chapter.

Chapter 2 begins with a brief section on collecting and organizing morphometric data, thenmoves on to data acquisition and manipulation with the R system. The final two sections discussmissing values and measurement error issues.

Chapter 3 is a very brief introduction to the exploratory analysis of multivariate data such asare used in morphometrics. It ends with brief overviews of principal components analysis, lineardiscriminant analysis, MANOVA, clustering, and related techniques.



Chapter 4 is an account of methods for describing and modeling shape changes, based onthe use of homologous landmarks. It starts with the truss network approach of Strauss andBookstein. These are somewhat akin to the truss networks that may be used in the constructionof, for example, bridges. Remaining sections describe Procrustes and other superimpositionmethods, the use of thin-plate splines, distance matrix analysis, and angle-based methods.

Chapter 5 has sections on splines and other curve-fitting methods, Fourier analysis, andeigenshape analysis.

Chapter 6 describes methods for visualizing shape change. Largely, this is a continuationof Chapter 3. An interesting challenge is to combine morphometric data with phylogeneticinformation that makes evolutionary connections.

Chapter 7 extends discussion of the use of R; there are sections on simulation, variousprogramming issues, and interfaces to other systems.

This is a highly useful guide to the literature and to the range of methodologies. A valuablefeature is the inclusion of R code that shows how to implement the methods that are described.The quality of the writing and overview is, in places, uneven.

John H. Maindonald: [email protected] for Mathematics & Its Applications

Australian National UniversityCanberra ACT 0200, Australia

Modern Regression Methods, Second EditionThomas P. RyanWiley, 2009, xix + 642 pages, £ 83.50 / € 104.20 / US$ 125.00, hardcoverISBN: 978-0-470-08186-0

Table of contents

1. Introduction 9. Logistic regression2. Diagnostics and remedial measures 10. Nonparametric regression3. Regression with matrix algebra 11. Robust regression4. Introduction to multiple linear regression 12. Ridge regression5. Plots in multiple regression 13. Nonlinear regression6. Transformations in multiple regression 14. Experimental designs for regression7. Selection of regressors 15. Miscellaneous topics in regression8. Polynomial and trigonometric terms 16. Analysis of real data sets

Readership: Regression practitioners.

This update includes a new Chapter 15 with brief paragraphs on various alternative regressionmethods, e.g., piecewise, semiparametric, quantile, Poisson, negative binomial, Cox, probit,censored and truncated, Tobit, constrained, interval, random coefficient, partial least squares,errors in variables, life data, survey sampling, Bayesian, instrumental variables, shrinkage, meta,CART and multivariate. More exercises have been added “especially at the end of Chapter 1”.The exercises are interesting and thought-provoking throughout. Macros now available on awebsite have been updated to MINITAB 15. If you liked the first edition, you will be pleasedwith this revision also.

Norman R. Draper: [email protected] of Statistics, University of Wisconsin – Madison

1300 University Avenue, Madison, WI 53706–1532, USA



Probability Models for DNA Sequence Evolution, Second EditionRichard DurrettSpringer, 2008, xii+431 pages, € 69.95 / £ 62.99 / US$ 84.95, hardcoverISBN: 978-0-387-78168-6

Table of contents

1. Basic models 6. Natural selection2. Estimation and hypothesis testing 7. Diffusion processes3. Recombination 8. Multidimensional diffusions4. Population complications 9. Genome rearrangement5. Stepping stone model

Readership: All readers interested in Population Genetics in a broad sense.

This is a beautifully written book by a distinguished probabilist on what used to be calledPopulation Genetics, and now, more appropriately, is called DNA Sequence Evolution. This is asecond edition, but the second edition has so much new material that it’s almost a new book.

The preface provides good summary of what the book does in respect of DNA SequenceEvolution. Chapter 1 deals with the classical Fisher-Wright model, based on statistical geneticsrather than DNA sequence, and Kimura’s hypothesis of neutral mutations. This is like the nullhypothesis of Population Genetics and Chapter 2 shows how this can be tested. Chapters 3through 6 provide alternative models including various forms of natural selection via mutation.Chapters 7 and 8 replace the discrete time models by one dimensional and multidimensional(continuous time) diffusion models. Chapter 9 deals with “evolution of whole genomes”, whichis of very recent vintage, but this chapter is also “the least changed” from the first edition.

The book is very clearly and elegantly written. I enjoyed reading what I sampled, namely,Chapter 1 and parts of Chapters 2, 7, 8, and 9. While the book is very well written and accessibleto readers with a minimal background in Biology and some knowledge of basic probabilitytheory and Markov processes, the material is tough for anyone who wants to read all theproofs carefully. Applications of probability theory includes local times, Markov processes withboundaries, Green’s function, and of course many innovative, interesting calculations.

It is strongly recommended to all readers interested in Population Genetics in a broad sense.Moreover parts of it can be used for other courses. For example Chapter 1 can provide additionalrelated material to a Bayesian course that covers Dirichlet processes, and Chapter 7 could be aninteresting supplement to a course on diffusion.

No matter how good a book is, a reader will have a couple of unfulfilled wishes. Mine arethese: It would be nice to have a non-technical last chapter that summarizes the overview of thoseaspects of the subject that were modeled, tested, and confirmed, and those that are in doubt. Theoriginal goal of Fisher, and probably Wright, was to confirm Darwin’s theory of evolution andnatural selection using Mendelian statistical genetics. How is that confirmation viewed today?Secondly, since the book abounds with beautiful theorems and proofs, a short non-technicalguide to the more important ones, as the preface covers the biological contents, would begood.





Cluster Randomised TrialsRichard J. Hayes, Lawrence H. MoultonChapman & Hall/CRC, 2009, xxii + 315 pages, £ 49.49 / US$ 80.96, hardcoverISBN: 978-1-58488-816-1

Table of contents

Part A: Basic Concepts Part C: Analytical Methods1. Introduction 9. Basic principles of analysis2. Variability between clusters 10. Analysis based on cluster-level summaries3. Choosing whether to randomise by cluster 11. Regression analysis based on individual-level data

Part B: Design Issues 12. Analysis of trials with more complex designs4. Choice of clusters Part D: Miscellaneous Topics5. Matching and stratification 13. Ethical considerations6. Randomisation procedures 14. Data monitoring7. Sample size 15. Reporting and interpretation8. Alternative study designs

Readership: Workers in Medical Statistics, Epidemiologists.

A randomised controlled trial is the traditional gold-standard design in what has come to beknown as evidence-based medicine. The participants or units are randomly allocated to thedifferent conditions or treatments under study. In a cluster randomised trial (CRT) the units areso allocated in groups. These groups, or clusters, can be geographical (look for the ‘fried eggdesign’), institutional (schools, organisations, etc.), or even individual people (the units of thecluster being teeth, for example).

The authors point out that the CRT is relatively new and that, although the topic is covered herepretty comprehensively, it is still an active research area. It’s difficult to think of any importantissue or aspect that is not discussed here, and at length and in depth.

The book is divided into four parts: Part A discusses when and why one should use a CRT;Part B shows how to do it; Part C describes data analyses; Part D rounds up some related issues.In particular, Part C shows how the clustering gives rise to specialised statistics. The ‘problem’is correlation, which results from within- and between-cluster variation: in fact, the so-calledintraclass correlation is just based on a ratio of these.

There is no heavy mathematics so the material is accessible to a wide range of readers. Theemphasis is on practical applications, reflecting in some degree the authors’ own work. Particularattention is given to rates and proportions, such data frequently arising in epidemiology. Thestatistical analyses are performed using the ‘Stata’ package, and data are available on thepublisher’s web site.

Martin Crowder: [email protected] Department, Imperial College

London SW7 2AZ, UK



Stochastic Approximation: A Dynamical Systems ViewpointVivek S. BorkarCambridge University Press, 2008, x + 164 pages, £ 35.00 / US$ 70.00, hardcoverISBN: 978-0-521-51592-4

Table of contents

1. Introduction 6. Multiple timescales2. Basic convergence analysis 7. Asynchronous schemes3. Stability criteria 8. A limit theorem for fluctuations4. Lock-in probability 9. Constant stepsize algorithms5. Stochastic recursive inclusions 10. Applications

Target readership areas: Computational statistics, adaptive signal processing, adaptive controlengineering, communication networks, neural networks, reinforcement learning.

The author notes in his Preface that Stochastic Approximation was born in the Statistics literaturebut has grown up in Electrical Engineering. However, it is by no means fully grown: itsdevelopment continues apace in a wide variety of settings – see the readership areas listedabove. The author describes his book as ‘a compact account of the highlights’ of the subjectfor ‘an interested, mathematically-literate reader’. Here, mathematically-literate entails a goodworking knowledge of real analysis and ordinary differential equations, together with familiaritywith probability theory up to martingales.

There are two broad approaches to the subject: statisticians (more precisely, probabilists) comefrom a background in martingale convergence behaviour, whereas (mathematically-inclined)engineers travel a route through ordinary differential equations associated with dynamicalsystems. As suggested by the subtitle of the book, the latter approach is predominant here.

The first chapter gives a gentle, very readable, introduction to the essence of the subject. Butthen, in chapters 2 to 9, one has to get down to work: the material is quite demanding, muchof it comprising theorems and proofs. Chapter 10 describes some applications and Chapter 11contains some appendices on mathematical prerequisites.

There is no hiding the fact that this is a tough subject, not one for the lazy-minded. But, it’sclearly a vital core subject, central to a variety of important applications. So, ‘an interested,mathematically-literate reader’ looking for a rewarding area of work could do a lot worse thanstudy ‘this little book’, as the author modestly refers to it.


London SW7 2AZ, UK



Semi-Markov Chains and Hidden Semi-Markov Models Toward ApplicationsVlad Stefan Barbu, Nikolaos LimniosSpringer, 2008, xiv + 226 pages, € 46.95 / £ 42.99 / US$ 59.95, softcoverISBN: 978-0-387-73171-1

Table of contents

1. Introduction 6. Hidden semi-Markov model and estimation2. Discrete-time renewal processes A. Lemmas for semi-Markov chains3. Semi-Markov chains B. Lemmas for hidden semi-Markov chains4. Nonparametric estimation for semi-Markov chains C. Some proofs5. Reliability theory for discrete-time semi-Markov D. Markov chains

systems E. Miscellaneous

Readership: Applied probabilists, theoretically-oriented statisticians, research workers andstudents.

Semi-Markov Processes (or Markov Renewal Processes), combining Markov Processes andRenewal Theory, were introduced in the mid-1950s. There followed much research, mainly in acontinuous-time framework. This book focuses on the less well-developed discrete-time version,a hybrid of Markov Chains and Recurrent Events (as discrete-time renewal is called in FellerVolume I). In the last chapter a further layer is added in which the semi-Markov chain is hidden(unobserved) but gives rise to a process, via a probability law, that is observed.

There are six chapters: in the first the scene is set (notation, definitions, etc.) and an overviewof the subsequent chapters is given; there follows discrete-time renewal theory (Chapter 2) andsemi-Markov chains (Chapter 3); Chapter 4 addresses non-parametric estimation, essentiallyestimates based on proportions of observed events, Chapter 5 covers reliability aspects ofsuch systems, and Chapter 6 introduces the hidden version and its non-parametric maximumlikelihood estimation. There are also some appendices covering certain technical results.

Overall, this is a stimulating book. As it says, the discrete-time framework has been underusedin the past in view of the fact that much real data in practice (I would even say most) is recordedin discrete time. Also, the applications listed, particularly the DNA sequencing, are importantand timely.


London SW7 2AZ, UK



Bayesian Evaluation of Informative HypothesesHerbert Hoijtink, Irene Klugkist, Paul A. Boelen (Editors)Springer, 2008, xii + 361 pages, € 59.95 / £ 53.99 / US$ 79.95, hardcoverISBN: 978-0-387-09611-7

Table of contents

1. An introduction to Bayesian evaluation of 8. The Bayes factor versus other model selectioninformative hypotheses (Herbert Hoijtink, Irene criteria for the selection of constrained modelsKlugkist, Paul A. Boelen) (Ming-Hui Chen, Sungduk Kim)

Part I. Bayesian Evaluation of Informative Hypotheses 9. Bayesian versus frequentist inference (Eric-Jan2. Illustrative psychological data and hypotheses for Wagenmakers, Michael Lee, Tom Lodewyckx,

Bayesian inequality constrained analysis of Geoffrey J. Iverson)variance (Paul A. Boelen, Herbert Hoijtink) Part III. Beyond Analysis of Variance

3. Bayesian estimation for inequality constrained 10. Inequality constrained analysis of covarianceanalysis of variance (Irene Klugkist, Joris Mulder) (Irene Klugkist, Floryt van Wesel, Sonja van Well,

4. Encompassing prior based model selection for Annemarie Kolk)inequality constrained analysis of variance (Irene 11. Inequality constrained latent class modelsKlugkist) (Herbert Hoijtink, Jan Boom)

5. An evaluation of Bayesian inequality constrained 12. Inequality constrained contingency table analysisanalysis of variance (Herbert Hoijtink, Rafaele (Olav Laudy)Huntjens, Albert Reijntjes, Rebecca Kuiper, Paul 13. Inequality constrained multilevel models (BernetA. Boelen) Sekasanvu Kato, Carl F.W. Peeters)

Part II. A Further Study of Prior Distributions and Part IV. Evaluationsthe Bayes Factor 14. A psychologist’s view on Bayesian evaluation of

6. Bayes factors based on test statistics under order informative hypotheses (Marleen Rijkeboer,restrections (David Rossell, Veerabhadran Marcel van der Hout)Baladandayuthapani, Valen E. Johnson) 15. A statistician’s view on Bayesian evaluation of

7. Objective Bayes factors for informative informative hypotheses (Jay I. Myung, Georgehypotheses: “Completing” the informative Karabatsos, Geoffrey J. Iverson)hypothesis and “splitting” the Bayes factors (Luis 16. A philosopher’s view on Bayesian evaluation ofRaul Pericchi Guerra, Guimei Liu, David Torres informative hypotheses (Jan-Willem Romeijn,Nunez) Rens van de Schoot)

Readership: Psychologists and Bayesian statisticians.

Informative hypotheses are best illustrated by typical examples, e.g.,

H1 : μ1 > μ2 > μ3

or

H2 : {μ1 ≈ μ2} > μ3.

The first statement is clear, the second asserts that μ1 and μ2 are nearly equal and both are greaterthan μ3. In applications μ’s are typically means but could be other parameters of interest. Thesehypotheses are being used by psychologists to have more meaningful hypotheses than the usualversions based on a somewhat artificial null hypothesis and its negation. In that approach eachof the above two hypotheses would lead to more than one null and alternative.

Along with this new formulation, the authors of articles in this volume rely on the Bayesianparadigm both because of its logical and philosophical attractions and the ease with which suchinequality constrained “informative” hypotheses can be tested in this paradigm.



This book has been very imaginatively planned and anchored around analysis of challengingproblems of (Clinical) Psychology. Not only the analyses but the design of experiments and thevariables measured are very interesting.

The first problem, commonly known as Multiple Personality Disorder, is that of a person whoclaims to have two or more separate identities, each of which forgets what the other does. Isthis real or simulated? In addition to the patients, there are three control groups, namely, truecontrols, simulators of amnesia, and true amnesiacs. The hypothesis of interest is H1: μcon >

{μamn = μpat} > μsim, where μ = true mean of a memory score. H1 implies the disorder isgenuine. The other hypothesis H2 is that μcon > μamn > {μpat = μsim}, i.e., the disorder issimulated by the patients.

The second problem concerns the effect of “peer evaluation on mood in children high andlow in depression.” The third is about “coping with loss: the influence of gender, kinship andthe time from loss.”

The problems of inequality constrained hypotheses are studied through Bayes factors. Thereis also a study of robustness of the conclusions with respect to choice of priors. Bayesian readerswill recognize at least two well known Bayesian analysts among the authors.

A final part, namely Part IV of the book, sums up and evaluates these analyses by psychologistsand statisticians. There are three evaluations in this part, one by a psychologist, one by astatistician and one by a philosopher.

This is an excellent book for psychologists and Bayesian statisticians. Strongly recommendedfor both categories of readers.



Who Gave You the Epsilon? & Other Tales of Mathematical HistoryMarlow Anderson, Victor Katz, Robin Wilson (Editors)The Mathematical Association of America, 2009, x + 431 pages, £ 45.00 / US$ 65.50, hardcoverISBN: 978-0-88385-569-0

Table of contents

Analysis (10 articles) Algebra and Number Theory (16 articles)Geometry, Topology and Foundations (11 articles) Surveys (4 Articles)

Readership: All interested in the broad development and history of mathematics.

This wonderful book, containing 41 articles, is a sequel to Sherlock Holmes in Babylon. Bothbooks reprint older, high quality articles on the history of mathematics; this volume has topicsfrom the 19th and 20th centuries. It is an absolutely fascinating volume. Do you want to singa song about the history of group theory? Of course you do! See pages 269–270 for wordsand music! How good was Ramanujan, the Indian intuitive mathematician? Let G. H. Hardytell you on pages 337–348. I quote a famous story from p. 344: “I remember going to seehim [Ramanujan] once when he was lying ill in Putney. I had ridden in taxi cab No. 1729 andremarked that the number seemed to me rather a dull one and I hoped it was not an unfavorableomen. ‘No,’ he replied, ‘it is a very interesting number; it is the smallest number expressibleas a sum of two cubes in two different ways’ (1729 = 123 + 13 = 103 + 93).” Did you knowthat epsilon comes from erreur (error)? Read “Cauchy and the origins of rigorous calculus” on



pages 5–13. Did you know (p. 58) that the late well-known statistician I. J. (Jack) Good was AlanTuring’s chief statistical assistant at the U.K. code-breakers’ Bletchly Park in 1942? Well knownchess masters C. H. O’D Alexander, P. S. Milner-Barry and Harry Golombek were also there(see p. 58). I could go on, but you get the idea! This would be a great gift for any statistician ormathematician.

Each of the four main sections has a brief foreword and afterword written by the editors.


1300 University Avenue, Madison, WI 53706–1532, USA

An Introduction to Multilevel Modeling Techniques, Second EditionRonald H. Heck, Scott L. ThomasRoutledge, 2008, xi + 268 pages, £ 25.95 / US$ 49.95, softcover (also available in hardcover)ISBN: 978-1-84169-756-7

Table of contents

Introduction 4. Defining multilevel latent variables1. Investigating organizational structures, processes, 5. Multilevel structural equation models

and outcomes 6. Multilevel longitudinal analysis2. Development of multilevel modeling techniques 7. Multilevel models with categorical variables3. Multilevel regression models

Readership: Graduate or advanced undergraduate students as well as researchers for example inthe organizational, educational, behavioral, and social sciences.

This book presents an applied approach to the use of multilevel modeling in exploring varioustypes of hierarchical data structures. A basic knowledge of data analysis and univariate statisticsis assumed.

Since the first edition (2000) of the book, there have been lots of advances both in the theoryand the software. The book covers the newest and most important achievements, but at the sametime gives a nice summary of the historical development of multilevel models. It also addressessome issues for further consideration.

The first four chapters challenge the reader to think about the complexity of multilevel datastructures compared to the usual or traditional single-level data structures and their analysis.Two software programs, HLM and Mplus, are briefly introduced.

The last four chapters present more complex multilevel modeling schemes, including structuralequation models using latent variables, analyzing change processes via longitudinal analysisand utilizing categorical variables e.g. by mixture models. All these approaches are shown tohave much in common, although their paths of development have varied considerably. In thisbook, the approaches are brought under a unified framework for exploring hierarchical datastructures.

The examples worked through using the software mentioned above and with real data areessential and help to understand the points made in the text. The text itself is very readable andthe level of the mathematical detail is not high. Unfortunately the software programs seem abit clumsy and old-fashioned (both input and output). It may be understood as a current stateof the scientific development, and also as a historical tradition, but it would be important to



implement these prominent models and methods in general statistical software packages, suchas R, in order to make them more accessible and more widely used in the future.

Kimmo Vehkalahti: [email protected] of Mathematics and StatisticsFI-00014 University of Helsinki, Finland

Generalized, Linear, and Mixed Models, Second EditionCharles E. McCulloch, Shayle R. Searle, John M. NeuhausWiley, 2008, xxv + 384 pages, £ 95.95 / € 119.60 / US$ 143.50, hardcoverISBN: 978-0-470-07371-1

Table of contents

1. Introduction 9. Marginal models2. One-way classifications 10. Multivariate models3. Single-predictor regression 11. Nonlinear models4. Linear models (LMs) 12. Departures from assumptions5. Generalized linear models (GLMs) 13. Prediction6. Linear mixed models (LMMs) 14. Computing7. Generalized linear mixed models Appendix M: Some matrix results8. Models for longitudinal data Appendix S: Some statistical results

Readership: Statistics students at the upper-undergraduate and beginning-graduate levels,applied statisticians, industrial practitioners, and researchers.

This is the Second Edition of the book by Charles E. McCulloch and Shayle R. Searle, publishedin 2001. “This text is to be highly recommended as one that provides a modern perspective onfitting models to data.” – That was the first sentence of Philip Prescott’s review on the FirstEdition on these ISR Short Book Reviews in 2001. It is a pleasure to agree with this commentand welcome the new revised edition, with the third coauthor John M. Neuhaus.

The very readable style and peaceful and patient explanation of the details and leadingprinciples, is nicely present in this book. The Second Edition has three new chapters andnumerous new and updated examples, and in addition the rest of the book has been lightlyrevised. Citing the Preface: “As before, the emphasis is on the applications of these models andthe assumptions necessary for valid inference. The focus is not on the details of data analysis northe use of statistical software, though we do briefly mention some examples.” All in all, being agreat Shayle Searle book fan, I am happy to keep this book near my desk and recommend it tomy students.

Simo Puntanen: [email protected] of Mathematics and StatisticsFI-33014 University of Tampere, Finland



Biostatistics and Microbiology: A Survival ManualDaryl S. PaulsonSpringer, 2009, x + 216 pages, € 54.95 / £ 49.99 / US$ 69.95, softcoverISBN: 978-0-387-77281-3

Table of contents

1. Biostatistics and microbiology: Introduction 5. Regression and correlation analysis2. One-sample tests 6. Qualitative data analysis3. Two sample statistical tests, normal distribution 7. Nonparametric statistical methods4. Analysis of variance Appendix: Tables of mathematical values

Readership: Microbiologists who wish to carry out a statistical analysis of their own data.

This is a well-written elementary introduction to common statistical tests, estimates, andconfidence intervals, illustrated with data from microbiology. Also included are testing problemsinvolving bioequivalence, for example that a new drug is as effective as an old one. This, being aclaim by a drug company or a microbiologist, is treated as the alternative and the null hypothesisis its negation. The discussion of this difficult problem as well as its one sided versions is alsoboth simple and illuminating.

Each problem and the suggested inference procedure are introduced with a brief but cleardiscussion. This is followed by an algorithmic description of the inference procedure. Thestrength of the book lies in its simple but clear presentation, its examples from microbiology,and the insight of the author gained from working with scientists. For example, if a pilotexperiment for determining the optimal sample size leads to an unacceptably large sample size,Paulson advises the scientist to redo the pilot experiment much more carefully. Most likely, thepilot experiment was unreliable.



The Statistical Analysis of Functional MRI DataNicole A. LazarSpringer, 2008, xiv + 299 pages, € 59.95 / £ 53.99 / US$ 84.95, hardcoverISBN: 978-0-387-78190-7

Table of contents

1. The science of fMRI 9. Bayesian methods in fMRI2. Design of fMRI experiments 10. Multiple testing in fMRI: The problem of “thresholding”3. Noise and data preprocessing 11. Additional statistical issues4. Statistical issues in fMRI data analysis 12. Case study: Eye motion data5. Basic statistical analysis A. Survey of major fMRI software packages6. Temporal, spatial, spatiotemporal models B. Glossary of fMRI terms7. Multivariate approaches8. Basis function approaches



Readership: Statisticians working or wishing to work on statistical problems in the areaof neuroscience, using fMRI data on the brain. Also suitable for cognitive psychologistsand neuroscientists and other readers familiar with graduate level statistics. This is the firstcomprehensive book dealing with statistical analysis of fMRI image data on the brain.

This excellent interdisciplinary text begins with the basic physics behind neuroimaging throughmagnetic resonance (Chapter 1) and then goes on to discuss the technology involved in collectingthe image data (Chapter 2). The images that we see are obtained by inverse Fourier transforms onraw data. The complex procedure for generating data makes it necessary to preprocess it beforestatistical analysis can begin (Chapter 3). A statistician who wishes to participate in designingan experiment, not just analyze the data, needs to know some of these basic facts.

It appears that experiments are basically of two kinds. One may observe a subject or subjectsbefore and after a stimulus is given, and try to identify which areas of the brain are activated.Alternatively, one may try to estimate the response function of the brain to an event like flashinga checkerboard. Chapters 4 through 11 deal with basic issues as well as standard methodslike linear models, spatiotemporal models, wavelets, Bayesian methods, and multiple testing,including the topological and geometric methods of Keith Worsley, whose tragic death cut shorta brilliant career.

The final chapter discusses data from an interesting experiment on eye motion, which isrelevant in the study of schizophrenia, brain lesions, etc., which seem to lead to loss of inhibitionor poor control of unusual eye movements.



Encyclopedia of Quantitative Risk Analysis and AssessmentEdward L. Melnick, Brian S. Everitt (Editors-in-Chief)Wiley, 2008, 4 vols., 2176 pages, £ 725.00 / € 978.80 / US$ 1450.00, hardcoverISBN: 978-0-470-03549-8

Readership: Anyone who needs an authoritative explanation of aspects of risk.

Risk is defined in different ways by different people. To some it means the probability of anadverse event. To others the expected cost of such an event. But however it is defined, risk isuniversal. An encyclopedia of risk must therefore cover material ranging from technical toolsfor risk management and modelling, to material describing risk in different domains, and thisbreadth means that almost certainly some topics will have been omitted.

In view of this, it might also be regarded as unfair to pick particular topics which have beenomitted or skipped over – simply on the grounds that in such a broad area there will alwaysbe such topics. Having said that, I have a particular interest in risk modelling in the personalbanking sector, and was disappointed by the limited coverage of risk in this area – especiallysince it is so topical at present. At first I thought there was nothing on the topic at all, butthen I found one article on retail credit scoring, under the far-from-obvious heading of Risk incredit granting and lending decisions: credit scoring. Perhaps the editors could include furtherdiscussion of this area in a future edition.

My nit-picking aside, it is clear that the editors of this four volume work have made a valianteffort to cover as much as possible, and have produced a valuable reference work. It is also clear



that a massive amount of effort went into it – in addition to the two editors-in-chief, it is theproduct of the work of eleven section editors and nearly 360 authors. The publishers, too, are tobe commended: it is a rather beautiful work.

David J. Hand: [email protected] Department, Imperial College

London SW7 2AZ, UK

Statistical DNA Forensics: Theory, Methods and ComputationWing Kam Fung, Yue-Qing HuWiley, 2008, xxii + 241 pages, £ 55.00 / € 66.00 / US$ 110.00, hardcoverISBN: 978-0-470-06636-2

Table of contents

1. Introduction 7. Interpreting mixtures in the presence of relatives2. Probability and statistics 8. Other issues3. Population genetics Solutions to problems4. Parentage testing Appendix A: The standard normal distribution5. Testing for kinship Appendix B: Upper 1% and 5% points of χ 2 distributions6. Interpreting mixtures

Readership: Statisticians and others who wish to learn the principles of DNA matching; advancedstudents, researchers.

This is a volume in the Wiley Statistics in Practice series. It aims to introduce the key ideas ofprobability and statistics which are necessary for the evaluation of DNA evidence.

No background is assumed in probability or statistics – an introductory chapter covers these –although if this was one’s first exposure to such ideas I think the rest of the book would be hardgoing. This chapter begins by introducing the basic laws of probability, and works up throughrelevant statistical distributions to likelihood ratios, estimation, and testing, all in 18 pages. Thisis followed by another introductory chapter on population genetic models, which allow oneto determine the probability of observing given profiles in a specified population. Subsequentchapters focus on the three main applications of DNA profiling: identity testing, determinationof parentage and kinship, and interpretation of mixed DNA stains.

Computer programs are available on an associated website, and the text is illustrated using thissoftware, including screenshots from it, where appropriate. Mathematical proofs are includedwhere necessary, but these are relegated to appendices at the end of chapters, so as not to disruptthe flow of the substantive ideas. Collections of problems are given at the end of each chapter,and solutions are given at the end of the book.

Approaching this from the perspective of a statistician who would like to learn more aboutDNA profiling, I would have appreciated more detail about the background technology: the threeparagraph description in Chapter 1 proved inadequate for me to understand the basic principles,and I had to resort to the web for a more detailed explanation. Having said that, once I had foundsuch an explanation, the remainder of the book was very clear. Overall, it provides an excellentintroduction to the statistics of DNA matching. I would certainly recommend it to anyone who



wishes to gain an in-depth understanding of the statistical ideals and tools which underlie thetechnology. It would make a great text for a course in this topic.


London SW7 2AZ, UK

Forecasting with Exponential Smoothing: The State Space ApproachRob J. Hyndman, Anne B. Koehler, J. Keith Ord, Ralph D. SnyderSpringer, 2008, xiv + 362 pages, £ 33.99 / € 36.95 / US$ 54.95, softcoverISBN: 978-3-540-71916-8

Table of contents

Part I. Introduction 11. Reduced forms and relationships with ARIMA1. Basic concepts models2. Getting started 12. Linear innovations state space models with

Part II. Essentials random seed states3. Linear innovations state space models 13. Conventional state space models4. Non-linear and heteroscedastic innovations 14. Time series with multiple seasonal patterns

state space models 15. Non-linear models for positive data5. Estimation of innovations state space models 16. Models for count data6. Prediction distributions and intervals 17. Vector exponential smoothing7. Selection of models Part IV. Applications

Part III. Further Topics 18. Inventory control application8. Normalizing seasonal components 19. Conditional heteroscedasticity and applications9. Models with regressor variables in finance

10. Some properties of linear models 20. Economic applications: the Beveridge-Nelsondecomposition

Readership: People wanting to apply exponential smoothing methods in their own area ofinterest, as well as for researchers wanting to take the ideas in new directions.

This book seeks to provide a comprehensive discussion of exponential smoothing forecastingmethods from the innovations state-space perspective. The authors integrate the state-spaceapproach to exponential smoothing forecasting into a coherent whole – and have done anexcellent job. This is certainly a book I would recommend to anyone who wishes to obtain asound grasp of the area, or to a PhD student about to begin research in the area.

Exponential forecasting is an intuitively attractive and very widely used approach, with along history. The state space framework is natural structure for developing prediction intervals,maximum likelihood estimates, and procedures for model selection. The description here isaccessible, and at mathematical level which is just right to explain the ideas and methods froma practical perspective, without labouring mathematical niceties.

I particularly appreciated the format of the outline of the book in the preface, which attemptsto cater for a wide readership, and which spells out clearly what parts could be read ‘if youonly want a snack’, or for ‘readers wanting a more substantial meal’, or those who ‘want thefull banquet’. Three concluding chapters then provide ‘after-dinner cocktails’ of applications toinventory control, economics, and finance. Each chapter concludes with exercises

There is an associated website containing data sets, computer code, and additional exercises.Details of public domain R software for the methods described in the book are given.



In summary, this is a perfect introduction to the subject, and an ideal text for an advancedundergraduate or beginning postgraduate course in the topic.


London SW7 2AZ, UK

Observed Confidence Levels: Theory and ApplicationAlan M. PolanskyChapman & Hall/CRC, 2008, xvi + 271 pages, £ 53.99 / US$ 83.95, hardcoverISBN: 978-1-58488-802-4

Table of contents

1. Introduction 5. Nonparametric smoothing problems2. Single parameter problems 6. Further applications3. Multiple parameter problems 7. Connections and comparisons4. Linear models and regression Appendix: Review of asymptotic statistics

Readership: Practicing statisticians.

Given a set of regions of a parameter space, observed confidence levels tell us how confident canwe be that the parameter lies in each of the regions. For example, the book’s opening illustrationdescribes how a hypothesis of bioequivalence of treatments can be examined by estimating theobserved confidence that the parameter lies in a specified interval about the value correspondingto no clinically important difference between treatments.

The basic theory of the single parameter case and the multi-parameter case are described inseparate chapters, because of the significant extra complications in the latter. Later chaptersthen look at the special cases of linear models and nonparametric smoothing methods. Thebook shows how the ideas are related to other statistical concepts, including hypothesis testing,multiple comparisons, attained confidence levels, and Bayesian approaches. It includes manyexamples, from a wide range of different application domains, based on real data. There areexercises at the end of each chapter, apart from the first, and the R code the author used isavailable from his website.

Techniques such as this touch on areas which have stimulated considerable discussion, evencontroversy, in the past. The author says ‘I am willing to let the examples presented in this bookstand on their own and let the readers decide for themselves what their final opinion is of thesemethods.’ The breadth of real examples that the author provides certainly demonstrates that thisis a class of techniques worth considering.


London SW7 2AZ, UK



A First Course in Statistical Programming with RW. John Braun, Duncan J. MurdochCambridge University Press, 2008, x + 163 pages, £ 24.99 / US$ 50.00, softcoverISBN: 978-0-521-69424-7

Table of contents

1. Getting started 5. Simulation2. Introduction to the R language 6. Computational linear algebra3. Programming statistical graphics 7. Numerical optimization4. Programming with R Appendix: Review of random variables and distributions

Readership: Everyone interested in learning statistical programming.

Programming skills are becoming increasingly important in the repertoire of a statistician.For example, in order to apply new methods not yet available in standard software or forsimulation studies. This book is a solid introduction to (statistical) programming and providesthe reader with all programming tools needed to approach typical problems. The programminglanguage the authors suggest for this purpose is R and they assume no prior knowledge of it.The second chapter hence is a general introduction to R which is, however, still worthwhilereading for readers already familiar with R because, for example, the chapter also explainswhat floating point numbers are. The title of the third chapter is a bit misleading because itactually does not teach how to program code that produces graphs but shows how to make use ofinbuilt R functions to obtain graphs. The actual material about programming is contained in thechapters 4–7. Many important and relevant concepts of statistical programming as, for instance,flow control, random number generation, numerical algebra, and optimization are explainedthere. Especially in chapter 4 the authors use interesting problems to illustrate the conceptsin question, but in the subsequent chapters such a motivation is often missing. In my opinionespecially the chapters 6 and 7 might gain a lot by putting the concepts discussed there moreinto a statistical framework and applying them to concrete statistical problems.

Overall, the book is a concise and easy to read introduction to programming in R; however,if used in a course on statistical programming maybe the connection of some of the concepts tostatistical problems might need some more elaboration.

Klaus Nordhausen: [email protected] School of Public Health

FI-33014 University of Tampere, Finland



Introduction to Empirical Processes and Semiparametric InferenceMichael R. KosorokSpringer, 2008, xiv + 483 pages, € 69.95 / £ 62.99 / US$ 89.95, hardcoverISBN: 978-0-387-74977-8

Table of contents

Part I. Overview 12. The functional delta method1. Introduction 13. Z-estimators2. An overview of empirical processes 14. M-estimators3. Overview of semiparametric inference 15. Case studies II4. Case studies I Part III. Semiparametric Inference

Part II. Empirical Processes 16. Introduction to semiparametric inference5. Introduction to empirical processes 17. Preliminaries for semiparametric inference6. Preliminaries for empirical processes 18. Semiparametric models and efficiency7. Stochastic convergence 19. Efficient inference for finite-dimensional parameters8. Empirical process methods 20. Efficient inference for infinite-dimensional parameters9. Entropy calculation 21. Semiparametric M-estimation

10. Bootstrapping empirical processes 22. Case studies III11. Additional empirical process results

Readership: Researchers who need to develop inferential tools for relatively complicatedmathematical or statistical modeling problems, statisticians and biostatisticians, advancedstudents.

The main focus of this book is to introduce empirical processes and semiparametric inferencemethods to researchers interested in developing inferential tools for relatively complicatedmathematical or statistical modeling problems. The material is mainly self-contained. Thereforethose with moderate knowledge of probability and mathematical statistics may becomeacquainted with the areas with the aid of the book. The material is divided into three parts.The first part consists of an overview of the topics avoiding basic mathematical definitions andproofs, which are introduced in later parts. The second part introduces foundations of empiricalprocesses and the third part introduces foundations of semiparametric inference. Each part isfollowed by a set of case studies of statistical or biostatistical modeling. The material is structuredin sensible way supporting the learning and understanding of useful and challenging techniquesof empirical processes and semiparametric inference. The book could well be very helpful forthose studying and applying these techniques.

Tapio Nummi: [email protected] School of Public Health




Medical Biostatistics, Second EditionAbhaya IndrayanChapman & Hall/CRC, 2008, 824 pages, £ 55.99 / US$ 99.95, hardcoverISBN: 978-1-58488-887-1

Table of contents

1. Medical uncertainties 13. Inference from proportions2. Basics of medical studies 14. Relative risk and odds ratio3. Sampling methods 15. Inference from means4. Designs of observational studies 16. Relationships: Quantitative data5. Medical experiments 17. Relationships: Qualitative dependent6. Clinical trials 18. Survival analysis7. Numerical methods for representing variation 19. Simultaneous consideration of several variables8. Presentation of variation by figures 20. Quality considerations9. Some qualitative aspects of medicine 21. Statistical fallacies

10. Clinimetrics and evidence-based medicine Appendix 1: Statistical software11. Measurement of community health Appendix 2: Some statistical tables12. Confidence intervals, principles of tests of

significance, and sample size

Readership: Medical and health professionals, advanced and basic level students of health andmedicine, biostatisticians.

The book contains a fairly comprehensive introduction of basic statistical concepts and methodsused in medical and health sciences. The basic idea is to give clear verbal explanations with goodexamples avoiding heavy mathematical definitions or formulations. My opinion is that the authorhas succeeded very well in this challenging task. By the aid of the book it should be relativelyeasy to get the main ideas, and if deeper understanding is needed a list of further references isprovided to each topic. As a marginal weak point I can mention that a very commonly adoptedfree statistical software package R is not mentioned on the list of general purpose statisticalsoftware. However, I can easily recommend this book for persons working on the area of medicalbiostatistics.

Tapio Nummi: [email protected] School of Public Health




Clinical Prediction Models: A Practical Approach to Development, Validation,and UpdatingEwout W. SteyerbergSpringer, 2009, xxviii + 497 pages, £ 53.99 / € 59.95 / US$ 89.95, hardcoverISBN: 978-0-387-77243-1

Table of contents

1. Introduction 14. Estimation with external informationPart I. Prediction Models in Medicine 15. Evaluation of performance

2. Applications of prediction models 16. Clinical usefulness3. Study design for prediction models 17. Validation of prediction models4. Statistical models for prediction 18. Presentation formats5. Overfitting and optimism in prediction models Part III. Generalizability of Prediction Models6. Choosing between alternative statistical models 19. Patterns of external validity

Part II. Developing Valid Prediction Models 20. Updating for a new setting7. Dealing with missing values 21. Updating for multiple settings8. Case study on dealing with missing values Part IV. Applications9. Coding of categorical and continuous predictors 22. Prediction of a binary outcome: 30-day

10. Restrictions on candidate predictors mortality after acute myocardial infarction11. Selection of main effects 23. Case study on survival analysis: Prediction12. Assumptions in regression models: Additivity of secondary cardiovascular events

and linearity 24. Lessons from case studies13. Modern estimation methods

Readership: This book is suitable for those with a basic knowledge of biostatistics and statisticalmodeling. The intended audience includes epidemiologists and applied biostatisticians lookingfor a practical guide to developing and testing clinical prediction models, or health careprofessionals and health policy makers interested in critically appraising a clinical predictionmodel.

Clinical Prediction Models is an excellent practical guide for developing, assessing and updatingclinical models both for disease prognosis and diagnosis. The book’s clinical focus in this eraof evidence-based medicine is refreshing and serves as a much-needed addition to statisticalmodelling of clinical data. The book assumes a basic familiarity with modelling using generalizedlinear models, focussing instead on the real challenges facing applied biostatisticians andepidemiologists wanting to create useful models: dealing with a plethora of model choices,small sample sizes, many candidate predictors and missing data. This is an example-based bookilluminating the vagaries of clinical data and offering sound practical advice on data exploration,model selection and data presentation. Model selection is at the core of the text with in-depthdiscussion of choices of candidate predictors, pre-specified models, models with interactions,stepwise selection methods in linear models, as well as modelling using generalized additivemodels (GAM), fractional polynomials and restricted cubic splines. There are also a few pagesdevoted to more modern selection methods such as Bayesian model averaging (BMA). Thereis an excellent discussion of estimation bias, over-fitting and optimism in prediction modelsmotivating the use of methods to correct for overestimation of model coefficients. Uniformshrinkage methods, penalized maximum likelihood methods, and least absolute shrinkage andselection operator (LASSO) shrinkage for selection are discussed in some detail.

The author considers many interesting examples of clinical data throughout the text, usingdata from rich data sources like the GUSTO-1 and the SMART studies. These data sets are made



available on the book’s website (http://www.clinicalpredictionmodels.org) for the purposes ofpromoting practical experience with modelling.

The author uses simple simulations using a few reproducible R commands to motivate the useof imputation methods and shrinkage. These simple but illuminating illustrations are one of thehighlights of the book and serve as excellent pedagogical tools for motivating good statisticalthinking.

There is some mention of statistical software available to try out the newer estimation methods.The author shows partiality to R software and provides some R code in the book and makesfull programs available of the website. This may be an impediment to some readers wedded tomenu-driven packages.

Teresa Neeman: [email protected] Consulting Unit, Australian National University

Canberra ACT 0200, Australia

Analysis of Messy Data Volume 1: Designed Experiments, Second EditionGeorge A. Milliken, Dallas E. JohnsonChapman & Hall/CRC, 2009, xiv + 674 pages, £ 54.99 / US$ 89.95, hardcoverISBN: 978-1-58488-334-0

Table of contents

1. The simplest case: one-way treatment structure in 14. Using the effects model to analyze two-waya completely randomized design structure with treatment structures with missing treatmenthomogeneous errors combinations

2. One-way treatment structure in a completely 15. Case study: two-way treatment structure withrandomized design structure with heterogeneous missing treatment combinationserrors 16. Analyzing three-way and higher-order treatment

3. Simultaneous inference procedures and multiple structurescomparisons 17. Case study: three-way treatment structure with

4. Basics for designing experiments many missing treatment combinations5. Multi-level designs: split-plots, strip-plots, 18. Random effects models and variance components

repeated measures, and combinations 19. Methods for estimating variance components6. Matrix form of the model 20. Methods for making inferences about variance7. Balanced two-way treatment structures components8. Case study: complete analyses of balanced 21. Case study: analysis of a random effects model

two-way experiments 22. Analysis of mixed models9. Using the means model to analyze balanced two- 23. Case studies of a mixed model

way treatment structures with unequal subclass 24. Methods for analyzing split-plot type designsnumbers 25. Methods for analyzing strip-plot type designs

10. Using the effects model to analyze balanced two- 26. Methods for analyzing repeated measuresway treatment structures with unequal subclass experimentsnumbers 27. Analysis of repeated measures experiments when

11. Analyzing large balanced two-way experiments the ideal conditions are not satisfiedhaving unequal subclass numbers 28. Case studies: complex examples having repeated

12. Case study: balanced two-way treatment structure measureswith unequal subclass numbers 29. Analysis of crossover designs

13. Using the means model to analyze two-way 30. Analysis of nested designstreatment structures with missing treatmentcombinations



Readership: Experimenters and statisticians involved with classic plot experiments and theiranalyses.

The 1984 first edition had 473 pages (32 chapters) compared with the second edition’s 674 (30),and many of the chapter headings and many segments of text are the same or similar in botheditions. This might give a first impression that the new revision is perhaps not worth investingin. However the exact opposite is true! New to the second edition are “modern suggestionsfor multiple comparisons . . . additional examples of split plot and repeated measures designs. . . and the use of SAS-GLM, . . . , SAS-MIXED and JMP” in various analyses. Every chapterhas been systematically re-written for greater clarity, and added explanatory material has beeninserted throughout. Many new diagrams and redrawn diagrams have been provided; those thatshow how to lay out the experimental designs are just superb and extraordinarily clear. Thereference list has increased from 44 to 99. This revision is highly recommended to those whoplan and analyze experiments of the type described.


1300 University Avenue, Madison, WI 53706-1532, USA

Statistical MisconceptionsSchuyler W. HuckPsychology Press, 2009, xx + 288 pages, £ 19.95 / US$ 32.95, softcover (also available inhardcover)ISBN: 978-0-8058-5904-1

Table of contents

1. Descriptive statistics 9. t-Tests involving one or two means2. Distributional shape 10. ANOVA and ANCOVA3. Bivariate correlation 11. Practical significance, power, and effect size4. Reliability and validity 12. Regression5. Probability Appendix A: Citations for material referenced in the preface6. Sampling Appendix B: References for quotations presented in the sections7. Estimation entitled “Evidence that this misconception exists”8. Hypothesis testing

Readership: Students and teachers of introductory statistics; statisticians wishing to communi-cate statistical theory unambiguously.

This book with its royal purple cover is dedicated (p. v) to “three groups of individuals: thosewho have overcome one or more of their statistical misconceptions, those who will . . . followin the footsteps of the first group, and those whose life’s work includes helping others to castaside false beliefs about statistics.” Those who have overcome statistical misconceptions shouldbe pleased to see the road they travelled laid out in clear terms. Those who will have to facethese misconceptions in the future should also find this book enlightening. I think the last groupare most strongly advised to buy this book, as it will encourage them to think about what theysay in their classes. Those who rely on a firm rule-based approach to statistics will be the mostdisappointed, as most of the rules they have developed will be overturned.



The book is very highly structured, which some readers may find helpful and others mayfind irritating after a while. Each chapter consists of three to six misconceptions. For eachmisconception, there are five parts. Firstly, the misconception is stated in one or two sentences.Then, evidence is given that the misconception exists. This is in the form of quotes from journals,textbooks and websites. All the references are given in an appendix: I hope you do not find yourown work referred to there!

The third part is a section explaining why the misconception is dangerous. These sections areshort yet they manage to explain the dangers in an accessible manner to someone who has studiedan introductory statistics course. Next, the misconception is undone. Sometimes this involvesa graph or picture, sometimes it is simply a page or two of text describing helping readersto confront and then dispel the misconception. Finally, an internet assignment is given. Thisinvariably points to the book’s website, http://www.psypress.com/statistical-misconceptions/.These worked easily when I tested a handful of them, and some connected to applets producedby well-known statistics educators such as Alan Rossman and Beth Chance. Some sections alsoinclude a list of recommended reading.

Now for some details of the misconceptions themselves. Some of them are indeed dangerousmisunderstandings of statistical theory, such as “If the 95% confidence intervals that’sconstructed for one sample partially overlaps the 95% confidence interval that’s constructedfor a second sample, the two samples are not significantly different from each other at α =0.05.”

Some of them are statistical paradoxes that have been widely written about e.g. the birthdayproblem and the Monty Hall problem.

But some of them are not so much misconceptions as simplifications that many educatorswould use to introduce a topic in an unequivocal manner, so as to leave exceptions and specialcases for later study. For instance, “The null hypothesis is always a statement of ‘no difference’”would frequently be used by educators to introduce hypothesis testing. Or, “there are threedifferent measures of central tendency: the mean, the median and the mode”. Of course thereare others, but these three cover a wide variety of situations, and even three may be more thanis needed for an introductory statistics course.

This book has the potential to shake statisticians out of any complacency they have aboutconveying the precise meaning of fundamental statistical theory and methods. Since the aim ofstatistics is to be accurate and precise, this book could usefully find a place on the shelves ofmost statisticians.

Alice Richardson: [email protected] of Information Sciences and Engineering, University of Canberra

Bruce ACT 2601, Australia



Statistical Inference, Econometric Analysis and Matrix Algebra: Festschrift inHonour of Gotz TrenklerBernhard Schipp, Walter Kramer (Editors)Physica-Verlag, 2009, xvi + 434 pages, € 119.95 / £ 108.00 / US$ 189.00, hardcoverISBN: 978-3-7908-2120-8

Table of contents

Part I. Nonparametric Inference 14. Minimum description length model selection1. Adaptive tests for the c-sample location problem in Gaussian regression under data constraints

(H. Buning) (E.P. Liski, A. Liski)2. On nonparametric tests for trend detection in 15. Self-exciting extreme value models for stock

seasonal time series (O. Morell, R. Fried) market crashes (R. Herrera, B. Schipp)3. Nonparametric trend tests for right-censored 16. Consumption and income: a spectral analysis

survival times (S. Leissen, U. Ligges, M. (D.S.G. Pollock)Neuhauser, L.A. Hothorn) Part V. Stochastic Processes

4. Penalty specialists among goalkeepers: a 17. Improved estimation strategy in multi-factornonparametric Bayesian analysis of 44 years of Vasicek model (S. Ejaz Ahmed, S. Nkurunziza,German Bundesliga (B. Bornkamp, A. Fritsch, S. Liu)O. Kuss, K. Ickstadt) 18. Bounds on expected coupling times in a Markov

5. Permutation tests for validating computer chain (J. J. Hunter)experiments (T. Muhlenstadt, U. Gather) 19. Multiple self-decomposable laws on vector spaces

Part II. Parametric Inference and on groups: the existence of background6. Exact and generalized confidence intervals in the driving processes (W. Hazod)

common mean problem (J. Hartung, G. Knapp) Part VI. Matrix Algebra and Matrix Computations7. Locally optimal tests of independence for 20. Further results on Samuelson’s inequality

archimedean copula families (J. Rahnenfuhrer) (R.W. Farebrother)Part III. Design of Experiments and Analysis of 21. Revisitation of generalized and hypergeneralized

Variance projectors (O.M. Baksalary)8. Optimal designs for treatment-control 22. On singular periodic matrices (J. Groß)

comparisons in microarray experiments 23. Testing numerical methods solving the linear(J. Kunert, R.J. Martin, S. Rothe) least squares problem (C. Weihs)

9. Improving Henderson’s method 3 approach when 24. On the computation of the Moore–Penroseestimating variance components in a two-way inverse of matrices with symbolic elementsmixed linear model (R. al Sarraj, D. von Rosen) (K. Schmidt)

10. Implications of dimensionality on measurement 25. On permutations of matrix products (H.J. Werner,reliability (K. Vehkalahti, S. Puntanen, I. Olkin)L. Tarkkonen) Part VII. Special Topics

Part IV. Linear Models and Applied Econometrics 26. Some comments on Fisher’s α index of diversity11. Robust moment based estimation and inference: and on the Kazwini cosmography (O.M. Baksalary,

the generalized Cressie–Read estimator K.L. Chu, S. Puntanen, G.P.H. Styan)(R.C. Mittelhammer, G.G. Judge) 27. Ultimatum games and fuzzy information

12. More on the F-test under nonspherical (P. Sander, P. Stahlecker)disturbances (W. Kramer, C. Hanck) 28. Are Bernstein’s examples on independent

13. Optimal estimation in a linear regression model events paradoxical? (C. St↪epniak, T. Owsiany)

using incomplete prior information 29. A classroom example to demonstrate statistical(H. Toutenburg, Shalabh, C. Heumann) concepts (D. Trenkler)

Selected Publications of Gotz Trenkler

Readership: Researchers interested in statistics and econometrics, both theoretical and applied.



The book is dedicated to Professor Gotz Trenkler on the occasion of his 65th birthday. It startswith a short biography of Professor Trenkler, including main scientific achievements as well assome personal details as his passion for chess and tennis.

In the first part one can find the list of contributors – there are 55 authors who published theirpapers in the volume, and 17 of them are also Professor Trenkler’s co-authors.

The volume contains the collection of 29 articles, which are divided to seven sections, whichpresent Professor Trenkler’s interests, such as nonparametric and parametric inference, designof experiments and analysis of variance, linear models and applied econometrics, stochasticprocesses, matrix algebra and matrix computations in its relation to statistics. In the articlespublished in the volume authors refer to the papers of Professor Trenkler, such as e.g. thecommon papers with H. Toutenburg and E. P. Liski (1992), H. Buning (1994), J. Diersen (1996,2001), D. Trenkler (1983), S. Puntanen (2006), J. Groß (1997), and K. Schmidt (2006).

Finally, the list of selected publications of Professor Gotz Trenkler is presented. Gotz Trenkleris the author and coauthor of 8 monographs and 159 scientific articles in well known journalsas well as proceedings of conferences and festschrifts in honor of his colleagues. In his papersProfessor Trenkler presents both theoretical results and applications.

Katarzyna Filipiak: [email protected] of Mathematical and Statistical Methods

Poznan University of Life SciencesWojska Polskiego 28, 60-637 Poznan, Poland

Bayesian Disease Mapping: Hierarchical Modeling in Spatial EpidemiologyAndrew B. LawsonChapman & Hall/CRC, 2008, xviii + 344 pages, £ 49.99 / US$ 79.95, hardcoverISBN: 978-1-58488-840-6

Table of contents

Part I. Background 6. Disease cluster detection1. Introduction 7. Ecological analysis2. Bayesian inference and modeling 8. Multiple scale analysis3. Computational issues 9. Multivariate disease analysis4. Residuals and goodness-of-fit 10. Spatial survival and longitudinal analysis

Part II. Themes 11. Spatiotemporal disease mapping5. Disease map reconstruction and relative A. Basic R and WinBUGS

risk estimation B. Selected WinBUGS codeC. R code for thematic mapping

Readership: Statisticians and epidemiologists interested in spatial and spatiotemporal Bayesianmodeling and analysis and disease mapping.

Lawson begins by building a solid Bayesian background in Chapters 2, 3, and 4, coveringboth basic theory and methods, computation, and residuals and measures of goodness of fit,including AIC, BIC, and DIC. Only the treatment of priors is not as strong as that of the otheraspects of Bayesian analysis. For example the inverse gamma with very small hyperparametersis mentioned following Gelman (Bayesian Analysis, 2006), but it is not pointed out that Gelmanis actually very negative about this particular prior and mentions it as an example of priorsnot to choose for the standard deviation of random effects. He recommends the uniform,



which appears later in this book, and the half Cauchy, a relatively new prior introduced byGelman.

The remaining seven chapters provide a thorough review of modeling relative risk, and itsanalysis and mapping, different types of clustering in spatial (or spatiotemporal) distribution ofdisease, multiple (spatial) scale analysis of disease distribution, and spatiotemporal modeling.Other important topics include establishing relations between geo-referenced data and predictorslike poverty, exposure to hazardous material, etc. Also included are multivariate disease analysisand spatial survival analysis.

Lawson provides well written reviews of many topics and many aspects of those topics arecovered in his reviews. The literature cited is huge and diverse, showing the current importanceof the subjects covered. One can also gain hands training in analysis and visual presentations, sostressed in the book, by following carefully the detailed introduction to R and WinBUGS givenin the book.

Many important data sets used in the book are available at http://www.musc.edu/biometry/people/lawsonab/Data%20and%20Programs.html but do not seem to be available at the sitegiven on p. 5 of the book.

Readers of this book might also consider reading the monograph by Banerjee, Carlin, andGelfand (Hierarchical Modeling and Analysis for Spatial Data, 2004), also published byChapman and Hall. This could be a good companion volume stressing more basic theoryand also models based on Gaussian processes and variograms.



Design and Analysis of Bioavailability and Bioequivalence Studies, Third EditionShein-Chung Chow, Jen-pei LiuChapman & Hall/CRC, 2008, xxii + 733 pages, £ 63.99 / US$ 99.95, hardcoverISBN: 978-1-58488-668-6

Table of contents

I Preliminaries III Population and Individual Bioequivalence1. Introduction 11. Population and individual bioequivalence2. Design of bioavailability studies 12. Statistical procedures for assessment of population3. Statistical inference for effects from a standard and individual bioequivalence

2 × 2 crossover design IV In Vitro and Alternative EvaluationII Average Bioequivalence Bioequivalence

4. Statistical methods for average bioequivalence 13. Assessment of bioequivalence for drugs with5. Power and sample size determination negligible plasma levels6. Transformation and analysis of individual 14. In vitro bioequivalence testing

subject ratios 15. In vitro dissolution profiles comparison7. The assessment of inter- and intra-subject V Other Bioequivalence Studies

variabilities 16. Meta-analysis for bioequivalence review8. Assumptions of outlier detection for average 17. Population pharmacokinetics

bioequivalence 18. Other pharmacokinetic studies9. Optimal crossover designs for two formulations 19. Review of regulatory guidances on bioequivalence

for average bioequivalence 20. Frequently asked questions and future challenges10. Assessment of average bioequivalence for more Appendix A. Statistical tables

than two formulations Appendix B. SAS programs



Readership: Statisticians in pharmaceutical companies and FDA, biostatisticians interested inBioequivalence, Bioavailability and how to test these, and FDA regulations and guidelinesrelating to these very important topics.

Most statisticians, including the present reviewer, are aware that the null and alternativehypotheses for testing Bioequivalence are the opposite of the usual null and alternative fortests of superiority of a new drug over the old drug in the market. Of course the same statisticalphilosophy underlies both cases for superiority tests, the claim for the new drug is the alternative,to be established by producing enough evidence, while its negation is the null and represents asort of status quo that we abandon only when enough evidence is produced.

In the same vein, the claim that a new generic drug is Bioequivalent to the old drug, whosepatent is running out, is the alternative and its negation, namely, that they are different is thenull, representing status quo. This makes the problem very intriguing since we rarely have to testsuch nulls. In this context, it is interesting that Lehmann, in his famous book, Testing StatisticalHypotheses, published in 1959, showed that its usual formulation

H0: absolute value of deviation of population means is ≥ δ

vs

H1: the above absolute value is < δ

has an UMP Test (for given δ > 0 and some level of significance α). Lehmann’s solution camemany years before testing Bioequivalence became a hot problem.

Little do we know that the problem of Bioequivalence is far more complex and remains one ofthe most challenging problems where consensus doesn’t exist on many aspects of what should betested and how. For example, Bioequivalence of population averages or at the level of individualpatients (so switching is possible)? Moreover, there are ethical questions as to whether the testshould be made on patients or healthy volunteers available, since there is no question of the newdrug being superior. At best it can be equivalent. If the test is to be made on healthy volunteers,it can’t be a clinical trial that tests the effects of the drugs. Rather testing on volunteers canonly help decide whether the two drugs have the same sort of pharmaco-kinetic parameters, forexample whether the drugs are available at the site where a drug is needed, the period it’s there,its strength, etc. In other words that would be testing Bioavailability, rather than Bioequivalence,though clearly both are closely related.

Of over seven hundred pages, the book provides an encyclopedic coverage of all these issuesand more. It is divided up into five parts; each part is further subdivided into chapters. Part 1explains what Bioequivalence means, the history of the evolution of this concept as well asthat of Bioavailability, the pharmaco-kinetic parameters that are used to measure it. Part 1 alsodiscusses some basic designs and inference on population averages. Part 2 provides a detaileddiscussion of inference on comparing Bioavailability at the level of population averages. Thereis also a detailed discussion of FDA regulations. These two parts could lead to a good course onBioequivalence and its proxy, namely, Bioavailability.

The remaining chapters are more specialized. Part III deals with tests at the level of individuals,but these individuals are still healthy volunteers and the tests are on Bioavailability. FDArecognizes the importance of such tests but they are still unregulated. Part V provides a reviewof many other related topics, like testing of inhalers, which leave no trace on blood and so requirea different definition of Bioavailability, its measurement and test.

The reviewer notes that after FDA approval, generic drugs usually replace the original, moreexpensive drug. Such data could lead to post approval studies on Bioequivalence directly, withouttaking resort to proxies. Probably such studies are already taking place. If not, the data mentioned



about could be a rich source of information at the level of both individuals and the populationaverage.




Documents

Medical Biostatistics, Second Edition by Abhaya Indrayan