Experimental Design and Data Analysis for Biologists. Quinn, G. P., and M. J. Keough. 2002. Cambridge University Press, Cambridge, United Kingdom. 556 pp. $110.00 (hardcover). ISBN 0–521–81128–7. $48.00 (paperback). ISBN 0–521–00976–6.
Conservation biology is evolving into an increasingly quantitative discipline. The emphasis of late has been on simulation and analytical modeling, typically under the rubric of population viability analysis. Somehow, in the rush to equip young conservation biologists with cutting-edge quantitative skills, we have begun to overlook the importance of good old-fashioned data analysis, the fundamentals of experimental design and sampling, and, perhaps most important, the ability to consider a biological question, look at an existing data set, and determine which statistical tests are appropriate.
Recently, I informally surveyed several dozen graduate students enrolled in two major environmental science programs, both of which have a strong emphasis on conservation biology. When asked to weigh the importance of a variety of research skills, the overwhelming majority lamented that they had not received more training in data analysis and statistics. Ironically, most of the students I surveyed had in fact taken one or two introductory statistics courses as undergraduates. In these courses, they were typically exposed to the t test, analysis of variance (ANOVA), the chi-square test, linear regression, and correlation analysis. Although such courses are an important first step, they simply cannot provide today's conservation biologists with the tools they need to deal with real data and real questions in conservation.
The trouble is that, although there are many excellent introductory texts, few books do a good job of covering the more sophisticated techniques in a manner that is accessible to the typical biologist (rather than the typical statistician). Fortunately, such a book now exists in Quinn and Keough's Experimental Design and Data Analysis for Biologists. In the preface, the authors describe their book as a bridge between the nice, neat examples covered in introductory courses and the in-the-trenches application of these techniques to real data. Quinn and Keough have succeeded marvelously in building this bridge.
Experimental Design and Data Analysis for Biologists has several major strengths. First, Quinn and Keough provide exceptionally lucid explanations of several of the more slippery statistical concepts. For example, their explanation of why degrees of freedom are called such is the clearest I have encountered. Second, they deal with topics that most introductory texts avoid, such as maximum-likelihood estimation and Bayesian approaches. Although the overall emphasis is on “traditional” (albeit advanced) techniques, Quinn and Keough provide a useful introduction to the philosophical and practical differences between Bayesian and Popperian hypothesis testing, delving into the advantages and limitations of each. I especially appreciated that discussions of nonparametric approaches, including resampling or randomization methods, are integrated throughout the book instead of being relegated to a single chapter. Besides some of the more standard techniques such as linear regression, single and multiway ANOVA, and some of the variations thereof (e.g., blocked, nested, repeated measures, split-plot), there is a whole chapter devoted to analysis of covariance and excellent sections on logistic regression and log-linear models. Four chapters are devoted to multivariate techniques, including multivariate analysis of variance, discriminant-function analysis, principle components, correspondence analysis, canonical correlation analysis, multidimensional scaling, and cluster analysis.
The worked examples provided throughout the book are derived from the current ecological literature. The use of real ecological examples is key, not only because it is always easier to appreciate the usefulness of techniques when they are applied within one's own discipline, but also because designs and data for natural systems are often messier than those encountered in other disciplines. The authors deal with real designs and real data without simplifying them or glossing over the challenges they present. Moreover, almost all the raw data files are available from the authors' Web site, allowing the reader to try out the techniques and compare their answers.
Despite its many strengths, the book is not perfect. In fact, Hurlbert and Lombardi (2003) disagree with Quinn and Keough on a large number of specific issues. Clearly, there are places where the authors could, in future editions, take more care with terminology and definitions. But there are precious few books available today that can play so well the role of bridge for students and practitioners alike. Further, the authors do an excellent job of pointing the interested reader to key references for each subject, including the more controversial topics.
Readers should be forewarned that this book is not for beginners. The presentation of techniques and concepts assumes a strong grasp of the fundamentals. The authors move quickly through the basics so they can then delve into some of the more advanced statistical topics. Thus, I would not select this book as the primary text for my undergraduate course in biostatistics (a course with no statistics prerequisite), but it would likely be suitable for an advanced undergraduate course that requires prior completion of one or two introductory statistics courses. This book is ideal for a graduate-level course (probably a two-semester course) in any ecology, conservation biology, or environmental science program, and it is an outstanding reference for all current practitioners of ecology or conservation biology who design sampling programs and experiments or analyze data.