Reviews of Books and Teaching Materials

Click to increase image sizeClick to decrease image size

Elementary Statistics in Social Research, 12th edition.

Jack Levin, James Alan Fox, and David R. Forde. London: Pearson Education Inc., 2014, xii + 596 pp., $194.00 (H), ISBN-13: 978-0-20-584548-4.

This is a standard textbook for statistics classes offered to nonmajors. It covers a relatively wide range of introductory topics in statistics for the students with decent mathematics background. It is good for a one-semester course. The new 12th edition adds discussion of meta-analysis in Chapter 1 along with new introductions to the coefficient of variation (CV) in Chapter 4, the one sample test of means in Chapter 7, and the calculation of effect size in Chapter 7. The Mann–Whitney and Kruskal–Wallis tests covered in Chapter 9 have been given new discussions. Additionally, Chapter 12 now includes expanded discussion of nonsampling error, standard error, partial correlation, and muticollinearity. Some examples and exercises have also been updated.

The basic organization of the book follows that of earlier editions. It consists of 13 chapters and five appendices. After an introductory chapter, the book is divided into five parts. Part I focuses on descriptive statistics and Part II moves to basic probability. Part III continues with statistical testing and decision making and Part IV discusses associations among variables. Finally, Part V considers the application of the topics covered in the book.

The strength of the book is the step-by-step examples given at the conclusion of each topic, which are extremely helpful to the students. A good number of exercises appear at the end of each chapter. Adding a chapter covering topics such as testing hypotheses on one sample mean and one sample proportion would make this book a great textbook for statistics courses designed for social sciences, psychology, political sciences, and similar areas. This book could also be used as a reference source by nonstatistician researchers.

Morteza Marzjarani

Saginaw Valley State University

Essential Statistical Inference: Theory and Methods.

Dennis D. Boos and L. A. Stefanski. New York: Springer, 2013, xvii + 568 pp., $99.00 (H), ISBN: 978-1-4614-4817-4.

Essential Statistical Inference: Theory and Methods is based on an advanced inference course taught by its two authors at North Carolina State University. It is framed as a textbook for second-year graduate students in statistical theory. Only basic knowledge of the theories of calculus, probability, and statistical inference are required, as the authors lead the reader through classical and modern topics without relying on measure theory.

In Part I of the book, which at 17 pages is by far the shortest part of the volume, introductory material on the role of modeling in statistics is supported by a consulting example. Notation is introduced, and a short review of necessary probability theory reinforces the students’ knowledge of convergence of random variables. The importance of computer simulation is emphasized through an example in R. Finally, an example of a mean and variance model is provided to illustrate these tools.

The second part of the book, entitled “Likelihood-Based Methods,” consists of three chapters. It begins by introducing likelihood and describing how a likelihood is constructed. Through several real-life examples ranging from a historical hurricane study to a problem about fetal lamb movement, the reader is walked through the process of building a discrete, continuous, or mixed likelihood. More complex cases including nonparametric maximum likelihood estimators are also discussed. Next, regression models are examined through the likelihood framework, with descriptions including measurement error models, generalized linear mixed models, and survival models. Marginal and conditional likelihoods are introduced through the Neyman–Scott Problem. Next, the information matrix is defined and carefully studied in detail with several examples. Methods for maximizing the likelihood are also introduced, including the expectation–maximization algorithm. The proof of uniqueness of the maximum likelihood estimator is given as an appendix, and a description of exponential families is also provided. The next chapter introduces the Wald, score, and likelihood ratio test statistics. It describes confidence regions as well as testing in regression problems. A key advantage of this text is that it also includes more modern material on model adequacy and nonstandard testing cases, including scenarios where the null hypothesis is on the boundary of the parameter space. Finally, this section ends with a description of Bayesian inference, introducing the reader to priors, a few examples, and Monte Carlo estimation of a posterior distribution.

Part III of the book describes large sample approximations through an introduction to the basics, including convergence of random variables (in more detail than the approximately page-long introduction in Chapter 1) and a set of tools for proving large sample results. The approximation theory is demonstrated in examples, and the emphasis on approximation by averages is clear and extensive. The importance of such approximations is emphasized, and a modern description of the influence curve as a Gateaux derivative provides more advanced readers with a gentle introduction to the general framework crucial for semiparametric modeling. The Lindeberg–Feller central limit theorem is also given in an optional appendix. The authors then proceed to explain the large sample behavior of maximum likelihood estimation from a fascinating historical perspective. Details of the proofs are provided with careful description of the necessary assumptions.

The fourth part of this text describes inference based on misspecified and partially specified models through the estimating equation framework. A transparent introduction to the basic approach is provided, with several simple examples including the delta method and the instrumental variables approach to measurement error. The connection to the influence curve is provided giving a modern flavor to this section. M-estimators are also described in the context of regression, for use in nonlinear modeling techniques and generalized estimating equation models. Asymptotic theory for M-estimators is given in more detail in an appendix. The topic of hypothesis testing under misspecification is also described, and generalized Wald, score, and likelihood ratio tests are studied.

The last part of the book describes computation-based methods and is unique in this category of text. The first chapter of this section describes Monte Carlo simulation, and emphasizes how to analyze, interpret, and present such results. Next, the authors describe the jackknife and its connection to the influence curve. They give several examples and consider cases of dependent data and nonidentically distributed samples. Several technical results concerning the jackknife are also described. Chapter 11 introduces the bootstrap for variance estimation, confidence intervals, and hypothesis testing. A short description of bootstrap asymptotics is given, and several different approaches for bootstrap confidence intervals are considered and compared. The last chapter of the book describes permutation and rank tests, with a treatment of the Wilcoxon–Mann–Whitney statistic, a set of approximations, and their comparison in a pharmacological study. Optimality properties of permutation testing are described briefly in the context of local power and Pitman efficiency. Rank methods for analysis of variance (ANOVA), distributional symmetry, randomized complete block designs, and contingency tables are also studied.

Throughout this well-written textbook, the authors engage the reader by marrying historical descriptions of central questions in classical statistics with modern techniques and approaches. Maple and R code are provided for examples and experiments to reinforce the importance of the fundamental ideas presented. The breadth of scientific examples motivates students interested in applied problems. Connections highlighted between topics and across chapters link the frameworks in this book and provide emphasis on important concepts that reach across areas of theoretical, computational, and applied statistics. The exercises at the end of each chapter are insightful and ideal for homework assignments. This book will surely become a widely used text for second-year graduate courses on inference, as well as an invaluable reference for statistical researchers.

Russell T. Shinohara

University of Pennsylvania

Negative Binomial Regression, 2nd edition.

Joseph M. Hilbe. New York: Cambridge University Press, 2011, xvii+ 553 pp., $95.00 (H), ISBN: 978-0521-19815-8.

This book is intended to be a comprehensive reference for negative binomial regression, including model specification, estimation, and inference for a wide range of negative binomial regression models. The book enumerates 22 different varieties of negative binomial models covered in the book, such as traditional negative binomial regression, truncated and censored models, zero-inflated and hurdle models, mixed effects and multilevel models, and bivariate negative binomial models. This second edition devotes a large portion of the updates to computational tools (primarily, R); it also includes many new methods and models including, for example, coverage of risk and odds ratios, dispersion, finite mixture models, quantile count models, and missing values in the predictors. Many examples of each type of model with corresponding R, SAS, and STATA code are provided.

The book is well written, easy to follow, and very comprehensive. Because of this, it would work well as an introduction to negative binomial regression or as a resource for almost all types of negative binomial regression models, and is accessible for a master’s-level statistician. While it would not work as a text for a generalized linear models course, it does include a lot of the basics of estimating and fitting generalized linear models and compares negative binomial regression to Poisson and binomial regression. The book author encourages the reader to follow the book sequentially; however, for those familiar with generalized linear models and negative binomial regression, they would easily be able to read any chapter.

While the book gives a nice discussion of model comparison and diagnostics, I would have liked to have seen a bit more on parameter interpretation and inference throughout the book. Furthermore, the Bayesian chapter is quite sparse and the chapter on dependent data was restricted to models currently available in software packages and did not include approaches to model more general temporal and spatial dependence. However, as is, this book serves as a good resource for a large variety of negative binomial regression models.

Candace Berrett

Brigham Young University

Overdispersion Models in SAS.

J. G. Morel and N. K. Neerchal. Cary, NC: SAS Institute Inc., 2012, xii + 393 pp., $69.95 (P), ISBN: 978-1-60764-881-9.

Overdispersion problems are common in data collected by pharmaceutical industries, healthcare research organizations, sampling survey organizations, etc. In Overdispersion Models in SAS, the authors present many good examples from different application areas to illustrate how to model and analyze overdispersed binomial, count, and multinomial data. Theoretical concepts and implementation in SAS are integrated nicely throughout the book.

Chapter 1 introduces the definition and common characterization of overdispersion. Sample SAS codes are given for simulating and analyzing Bernoulli data with overdispersion and the exponential family is briefly reviewed.

Chapters 2 and 3 review the generalized linear model (GLM), maximum likelihood estimation (MLE) for GLM, and quasi-likelihood functions for GLM with overdispersion. Several examples are used to illustrate how to fit these models and diagnose the model fitting, using SAS procedures GLIMMIX and GENMOD.

Chapters 4 and 5 focus on overdispersed binomial data. The authors introduce four distributions: beta-binomial, random-clumped binomial, zero-inflated binomial, and generalized binomial for modeling overdispersion. Although the maximum likelihood equations of these models can be difficult to obtain due to the complexity of the likelihood equation, the authors illustrate how SAS procedure NLMIXED can be easily used to fit these models. Two types of goodness-of-fit tests are considered for testing no overdispersion under these four different overdispersed binomial distributions. Extension of goodness-of-fit tests when covariates are present is also discussed.

Chapter 6 covers generalized linear overdispersed model (GLOM) including negative binomial, zero-inflated Poisson, zero-inflated negative binomial distributions, and some hurdle models for overdispersed count data. The authors present examples to demonstrate how SAS procedures GENMOD, GLIMMIX, NLMIXED, and COUNTREG can be used to provide MLEs for parameters in these models.

Chapter 7 considers analysis of more complex multinomial response data. Two likelihood-based models, Dirichlet–multinomial (DM) and random-clumped multinomial (RCM) distributions, are discussed for overdispersed multinomial responses. SAS procedures LOGISTIC, NLMIXED, IML, and module DirMult_RCMult are used to fit these two models and perform goodness-of-fit tests. For DM and RCM distributions, the calculation of MLE for parameters in the models using a Fisher information matrix can be computationally challenging. In Chapter 8, the authors discuss a two-stage procedure using an approximate information matrix to derive MLE for the parameters. Macro DM_RCM_Exact_Approximate is developed for running the two-stage procedure.

Chapter 9 presents how generalized estimating equations (GEE) are used to fit the overdispersion data. Specifically, GEE-type binomial marginal models, Poisson marginal models, and multinomial marginal models are used to analyze overdispersed binomial, count, and multinomial data, respectively. SAS procedures GENMOD, GLIMMIX, COUNTREG, and SURVEYLOGISTIC are used to fit those models. Chapter 10 reviews generalized linear mixed model (GLMM) and how overdispersion interacts with GLMM. Four generalized linear overdispersion mixed models (GLOMM), corresponding to beta-binomial, random-clumped binomial, zero-inflated Poisson, and zero-inflated negative binomial models with random effects, respectively, are discussed.

In summary, this book provides a comprehensive discussion of issues that arise when fitting models for overdispersed data with SAS. The book’s target audience is statisticians with a bachelor or masters degree and adequate data-analysis experience. The example data and SAS code used in the book are available online: https://support.sas.com/publishing/authors/morel.html, which will benefit readers who wish to reproduce the examples in the book or to use native SAS procedures or the authors’ macros/modules in their own research. The book’s discussion of theoretical aspects of models for overdispersed data is quite concise. Readers who are interested in foundational issues related to GLM, GLMM, GEE, categorical data analysis, exponential families, or link functions should refer to other texts, such as those by McCullagh and Nelder (1989), Agresti (1996), Hilbe (2007), Hardin and Hilbe (2013), or Stroup (2013).

Disclaimer: This review reflects the views of the reviewer and should not be construed to represent FDA’s views or policies.

Yun Wang

The U.S. Food and Drug Administration

Performance Modeling and Design of Computer Systems, Queueing Theory in Action.

Mor Harchol-Balter. Cambridge: Cambridge University Press, 2013, xxiii + 548 pp., $75.00 (H), ISBN: 978-1-107-02750-3.

This book bridges the disciples of statistics and computer science and covers computer system design from a software perspective. It aims to reintroduce queuing theory to graduate computer science curricula, many of which have dropped the subject in recent years.

The book is divided into seven parts with an unequal number of chapters in each. Some chapters in the book have only a few pages, while others have more than 30. Instructors using the book will need to pay attention to the length variation when designing a syllabus.

Part I consists of two chapters that cover introductory topics in queuing theory borrowing terminology covered in Part II. Part II introduces the readers to the topics in probability and statistics needed for understanding queuing theory. This part might be productively covered at the beginning of the course to provide the required background first. A background at the level of calculus and also familiarity with an introduction to probability and statistics would be beneficial here, but the latter is not essential. Chapter 4 covers simulation and generating continuous and discrete random variables. Topics discussed in Chapter 5 include almost sure convergence and convergence in probability, the strong and weak laws of large numbers, time average and ensemble average, and their comparison and the case where last two are equal (ergodic systems).

Part III consists of two chapters. Chapter 6 starts with Little's Law (1961) and its application to both open and closed systems. The author presents these topics in depth providing the readers with a clear and concise understanding. Chapter 7, titled Modification Analysis, returns to the closed systems and focuses on modifying the operational laws considered earlier for such systems. Some asymptotic bounds for closed systems are developed and used to modify the systems to achieve better performance and to answer the question of which design changes results in higher utilization. Comparison of open and closed systems is also included in this chapter.

Part IV consists of Chapter 8 through Chapter 13. In Chapter 8, the reader learns about discrete time Markov chain (DTMC). Chapter 9 is a continuation of Chapter 8 discussing ergodicity for finite and infinite-state Markov chains and the applications of these to “Real-World Examples, Google, Aloha, and Harder Chains.” Chapter 11 discusses the exponential distribution and its relation to the geometric distribution. The memoryless property of this distribution makes it an attractive candidate for continuous time Markov chain (CTMC) and for the M/M/1 queuing pattern defined in Chapter 13. (M/M/m stands for a queuing system where customers arrive according to a Poisson distribution, service time is exponentially distributed, and m represents the number of servers.) The queue is a first-in, first-out (FIFO) and the state space of this queue represents the total number of customers in the queue (including the one currently being served). Chapter 12 is a brief introduction to CTMC and its transition to DTMC. I found it a bit difficult to follow the ordering of topics in this chapter. Perhaps a more detailed explanation would help the reader to better understand this chapter.

Chapter 13 is a key chapter, in which queuing system patterns are presented. In general, a queuing system looks like A/S/m/p/n, where A is the arrival time, S is the service time, m is the number of servers, p is the number of places (both waiting and service places, i.e., the capacity of the system for holding jobs and the system follows FIFO pattern), and n is the number or the size of the population (both waiting and those currently being served). The last two parameters are usually omitted and assumed to be infinite. In this chapter, we learn about what is called Poisson arrivals see time averages (PASTA): that is, what arrivals see at the moment they enter the queue.

Multi-server systems are presented in Chapter 14. The queuing patterns M/M/m and M/M/m/n are discussed here. As mentioned in the book, the case with limited capacity is somewhat easier to analyze, but jobs arriving to such a system will be lost if all servers are busy. Chapter 16 covers time-reversibility and Burke's Theorem (Burke 1956), which are needed prerequisites for Chapter 17. Chapter 17 discusses queuing networks and their simplest case, the Jackson network (1957, 1963). This network is further developed in subsequent chapters to include closed and open networks.

Parts VI and VII include Chapters 20 through 33 and consider more in-depth topics in queuing theory including perhaps the most important one, scheduling. In particular, there is discussion of more general queuing patterns. The arrival and service time distributions previously discussed are extended considering more general forms of distributions. Among the topics discussed are Pareto distributions and phase-type distributions, where general distributions can be represented as mixtures of exponential distributions and consequently allowing to model queuing systems using Markov chains. As expected, the topics discussed in this chapter are more advanced and require a clear understanding of the previous chapters.

The book follows a straightforward approach and the topics appear in the right order. The book implements a method similar to a live classroom environment where the instructor attempts to get the students involved in class discussion by repeatedly asking questions and then providing answers. This is a good learning approach, as it encourages the students to follow the instructor step by step. However, frequently repeated Q/A's may cause the reader to lose continuity. I felt in some cases there were too many Q/A's, and I lost track of the main point of the topic under consideration.

Overall, this book is a good resource for researchers in the field of designing systems from a performance perspective and also for a graduate course in computer science or a related field such as computer engineering. It could be used as a textbook in a graduate course in a probability and statistics curriculum if queuing theory is a part of such a program. The book would challenge students at the undergraduate level and readers would be better served if they are familiar with topics covered in an introductory course in probability and statistics and calculus.

Morteza Marzjarani

Saginaw Valley State University

Single-Case and Small-n Experimental Designs: A Practical Guide to Randomization Tests.

Pat Dugard, Portia File, and Jonathan Todman. New York: Routledge, 2012, xiii + 290 pp., $49.95 (P), ISBN: 978-0-415-88693-2.

Single-Case and Small-n Experimental Designs is divided into two major sections, one providing information about randomization tests for a variety of single-case and small-n designs (Chapters 1–4, 6, and 7), and the other providing instruction related to the use and development of SPSS and Excel macros for performing the randomization test analyses (Chapters 5 and 8, Appendix 1–3). Each of these major sections is discussed in turn here.

According to the authors: “This book is intended as a practical guide for students and researchers who are interested in the statistical analysis of data from single-case or very small-n experiments.” Making randomization tests for single-case and small-n experiments accessible to a larger number of applied researchers is a worthwhile goal, and we found the book to be generally well organized and written in a manner that achieves that goal. It should be noted, however, that although randomization tests for a variety of designs are discussed, there are others that are not or at least that are not described in accordance with conventional terminology (notably, the multiple baseline design, the most internally valid and popular single-case intervention design). In particular, throughout the book the term “multiple baseline” is used to refer to a replicated A (baseline)–B (intervention) design, rather than to a design in which the B series start points must be staggered from one case to the next (see, e.g., Horner and Spalding 2010). As a consequence, the “multiple baseline” randomization test macros that the authors provide are based on independent intervention start-point randomization for each of the n cases, resulting in the possibility that two or more of the cases could end up with the same intervention start point. Unfortunately, not included in the book’s macro library are the randomization test procedures developed by Wampold and Worsham (1986) and by Koehler and Levin (1998), both of which avoid this problem. In addition, the discussion of “response-guided intervention” is presented as if its use precludes the use of randomization, with the authors seemingly unaware that both randomized and response-guided elements can be incorporated into any given single-case design (Ferron and Jones 2006).

As a hoped-for coherent guide through the whys and wherefores of randomization tests, we believe that the authors failed to deliver. In particular, the book’s second edition has been revised and reorganized considerably from the first edition. Although much instructive information is included in Chapter 7, “Size and Power” (a carryover from the first edition), most of the remaining chapters generally suffer from various shortcomings. In typical methodological/statistical textbooks, a specific topic is presented, accompanied by a discussion, an example with hypothetical or real data, and analysis procedures bearing on that topic. In contrast, the present authors introduce several different small-n and single-case designs in one chapter (Chapter 3), followed by examples of those designs in Chapter 4, followed by macros and analyses related to those designs in Chapter 5. As a result of the consequent cycling back through chapters in time and space, discussion of the designs is disjointed; and because the discussion is generally truncated and obtuse, making sense of it without sufficient background logic could be challenging for a novice reader.

Although selective references are provided at the book’s end (26, as compared to 65 in the first edition), there is no indication throughout the book either whence the specific designs and associated randomization tests come or the rationale underlying their development. This will be disconcerting to readers who may wish to uncover more technical details about specific procedures. In addition to the unconventional use of the term “multiple baseline” mentioned above, the discussion of “order effects” (pp. 29–32), which the authors use to refer to predicted orderings of intervention-condition outcomes (e.g., Levin, Marascuilo, and Hubert 1978), creates terminological confusion. In conventional usage, the term refers to the uncontrolled effects of stimulus-presentation orders on measured outcomes, which can seriously compromise an experiment’s internal validity (e.g., Shadish, Cook, and Campbell 2002).

The book provides a library of downloadable SPSS and Excel macros written for a variety of single-case and small-n randomization test applications, along with a wide range of examples that are broken down into easy-to-follow steps. The macros and examples should be of value to applied researchers, with the SPSS macros functioning efficiently. The macros all build the randomization distribution by resampling procedures and although this typically works just fine, there are cases that can be problematic. For example, in the book’s Figure 5.22 on p. 125, a one-tailed p-value of 0.049475 is reported by the authors for an ABA design example. We conducted the same analysis three times using the provided Excel macros (each based on 2000 samples) and in two of those runs the obtained one-tailed p-values exceeded 0.05. (The exact p-value, resulting from the actual data being associated with the most extreme difference in the complete randomization distribution of 21 possible outcomes, is p = 1/21 = 0.0476.) In situations such as this, where the p-value based on the complete randomization distribution is equal (or very close) to the predetermined Type I error probability α, the error introduced by sampling from the randomization distribution may be consequential.

In addition, we found the microcomputer software and associated notation not to be so user friendly. In particular, where output is provided, it is not specifically labeled, difficult to locate, and lacking in detail (e.g., no descriptive statistics are provided). As was just mentioned, the macros all rely on resampling rather than exact testing. We applied the Excel one-way analysis of variance (ANOVA) macro to an example with n = 9 total observations, consisting of 3 conditions, 3 participants per condition, and based on 2000 random samples. We gave up after 4 min of waiting for results. We then ran the authors’ single-case ABA example in Excel; it took 3-1/4 min on one PC and just over 2-1/2 min on a faster PC. An exact test conducted using Gafurov and Levin’s (2014) ExPRT microcomputer program took less than half a second to complete. Surprised by these findings, we ultimately came upon the following cautionary comment on p. 110 of the book: “The slowest [Excel macros] are the multiple baseline AB and ABA designs, and those may take about 20 minutes, depending on the speed of your computer.” It would have been desirable for the software to allow users to select the approach (resampling or exact test) that they prefer. Exact-test approaches are discussed in Kratochwill and Levin’s (in press) forthcoming edited volume.

All of the foregoing commentary leads to the present reviewers’ consensus bottom-line conclusion: Caveat emptor. The book may be of use to a select number of single-case researchers but for a larger number of single-case researchers, Edgington and Onghena’s (2007) text with accompanying CD software will be more useful.

Joel R. Levin

University of Arizona

John Ferron

University of South Florida

Statistical Analysis for Business Using JMP: A Student’s Guide.

Willbann D. Terpening. Cary, NC: SAS Publishing, 2011, xiv + 364 pp., $53.95 (P), ISBN: 978-1-607-644767.

The preface of this book states its intended purpose as “… to offer a theoretically sound introduction to statistics that is appropriate for the undergraduate or introductory MBA level courses in Schools of Business …” The author accomplishes this task quite nicely for students interested in learning introductory statistical methods via the JMP software package. As suggested by the title, instructions for statistical content and JMP are inseparable in this text.

The book may be considered an “essentials” textbook. Emphasis is given to statistical thought, selection of appropriate graphical and numerical methods, and the use of JMP platforms to obtain desired analyses. The assumption that no one does (nor should do) statistical analysis by hand seems to be an underlying foundation of this book. Formulas associated with the statistical methods are provided, as are conceptual discussions. Minimal space is allocated for examples illustrating the calculation of basic summary or test statistics. Students are taken directly into the JMP software platforms with detailed instructions for obtaining desired analytical results.

Statistical methods illustrated in the book are similar to those found in most introductory statistical methods textbooks including: (a) the “Why” of studying statistics, (b) univariate and multivariate graphical methods, (c) univariate statistics such as the mean, variance, moments, quartiles, moment skewness and the coefficient of variation, (d) probability and sampling distributions such as the binomial, Poisson, normal, t, chi-square, and F distribution, as well as a brief discussion of bootstrapping, (e) inferential methods for the mean, median, variance and proportions, (f) methods for multiple variables including tests for two means, two medians, two variances, two proportions, and ANOVA with multiple comparisons, (g) contingency tables with chi-square tests, (h) simple linear regression, (i) logistic regression, and (j) multiple regression models. Covering this much ground in roughly 300 pages of text requires concise and focused discussions and illustrations. The author does not wax poetically about statistical theory.

The introductory chapter does a very nice job of delineating traditional “confirmatory” and modern “exploratory” approaches to statistical analyses. No reader should skip this discussion. These ideas are central to the organization of the text as well as the JMP software package. Readers using this text as an introduction to the JMP software platform will be well served by this introduction. Where as many statistical software packages are organized by statistical techniques, JMP menus are organized according to the sequential process of data analysis. This organization may seem awkward for users who view statistics as a collection of various techniques, but it is extremely conducive to the actual process of data analysis.

Chapter 2 is devoted to introducing JMP. This is a somewhat deceptive chapter. On the surface it appears to be exclusively an overview of JMP. However, a significant portion of the chapter is devoted to data creation and manipulation. Examples of combining data tables of various types, concatenation, data joins, and filtering provide an excellent opportunity to discuss the importance of data understanding, cleaning, and management.

An additional strength of the book is the coverage of data visualization methods in Chapter 3. The text leverages the excellent graphical capabilities found in JMP. The usual list of graphical methods such as histograms, stem-and-leaf plots, bar charts, pie charts, and scatterplots are complimented by tree maps, mosaic plots, and bubble plots. The emphasis is on selection and interpretation of appropriate graphs. For example, key characteristics of histograms (shape, center, spread, outliers) are highlighted while discussions of class boundaries and class widths are omitted. JMP’s dynamic graphing capabilities are well illustrated. A more typical discussion of univariate numerical summaries follows in Chapter 4.

Chapter 5 introduces probability, probability distributions, and sampling distributions. The author has chosen to move quickly through these topics with an eye toward inference. Rules of probability such as formulas for P[A and B], P[A or B], P[A given B] as well as extended discussions of independence and mutually exclusive events are omitted. Neither will the reader find examples of Venn diagrams or tree diagrams for assessing probabilities. The approach appears far more utilitarian, introducing probability as a tool necessary for development of confidence intervals and hypothesis tests found in later chapters. Distributions are very briefly introduced then quickly connected to the sampling distributions of statistics that make them so useful (e.g., sample mean and normal or t, sample variance and F, sample proportion and binomial). The chapter concludes with a brief overview of bootstrapping.

Confidence intervals and hypothesis tests for the univariate cases are covered concurrently in Chapter 6, which formally introduces statistical inference. Methods for the sample mean, median, variance, proportion are examined. Chapters 7–11 introduce methods for analyzing multiple related variables. The chapters are separated by the number and type of variable (qualitative, quantitative) under consideration. Such an arrangement reinforces the importance of understanding the impact of data type on selection of appropriate analyses. Chapter 9 is devoted to simple linear regression, correlation, and residual analysis. Chapter 10 provides an overview of logistic regression and multinomial regression. These important topics are missing in many introductory texts. Multiple regression is introduced in the final chapter. Topics include model building, leverage, multicollinearity, and stepwise variable selection methods.

Statistical Analysis for Business Using JMP: A Student’s Guide provides a very nice overview of introductory statistical methods and the JMP software platform. The book should be useful to several different audiences. Students of statistics already having a formal introductory statistical methods course will find an excellent introduction to the navigation and use of JMP. Instructors of introductory methods courses looking for concise coverage of both statistical methodology and a powerful software platform will find the book attractive. Finally, instructors wishing to devote class time to statistical methodology and finding themselves in need of a student guide for JMP will find the book to be an excellent companion. Critics of the book will point to the minimal theoretical content as well as the scarcity of problem sets. Fans of the book will also find a companion website containing useful PowerPoint slides, datasets, and extended problem sets. While this text will not be suitable for everyone, the author has a very clear vision of his final product. The book provides a concise look at introductory statistical methods and guides students through the statistical discovery process via a very powerful statistical software package. For readers looking for such a text, the rating is likely to be “well done.”

B. Michael Adams

University of Alabama