Journal of Experimental Marine Biology and Ecology 277 (2002) 197 – 198 www.elsevier.com/locate/jembe
Book review
Experimental Design and Data Analysis for Biologists Garry P. Quinn and Michael J. Keough, Cambridge University Press, Cambridge; ISBN 0 521 00976 8 (paperback), GBP 29.95; ISBN 0 521 00976 8 (hardback), GBP 75.00 At last, a book that provides a readable introduction to nuances of statistical methods and analysis. Quinn and Keough have written a wonderful book that is packed with lots of practical advice about statistical analyses. Their goal was to strike a balance between how analyses are done versus how to use them correctly, and I think they have achieved that balance. Quinn and Keough cover all the standard material found in introductory books on univariate statistics and provide several chapters on multivariate methods. I found the 19 chapters well organized and informative. More technical information is covered in boxes outside of the main body of text and each chapter ends with a section of general issues and hints for analysis, which are short and listed with bullet points. Data for the examples come from real experiments, and the raw data are available on the web as text and Excel files so the reader can run their data in his or her favourite statistical package. While Quinn and Keough suggest their target audience is biologists, nearly all of the examples are from population and community ecology. The book begins with four chapters that introduce the scientific method, probability distributions, estimation, hypothesis testing and graphical exploration of data. These chapters go well beyond the basics, and Quinn and Keough discuss a wide range of topic including Popperian falsification, maximum likelihood estimation, jackknife and bootstrap methods, Bayesian inference and hypothesis testing, decision errors, significance levels for multiple testing, meta-analysis and censored data. The book closes with a wonderful chapter on presentation of results. The heart of the book is univariate statistics, and almost 60% of the book is devoted to regression, correlation, analysis of variance and analysis of covariance. The two chapters on regression cover linear and multiple regression, provide a discussion of diagnostics, transformations and stepwise methods and introduce a number of less well-known methods such as smoothing with splines and locally weighted regression (i.e., loess regression). And, just to top things off, there is a chapter on logistic regression and generalized linear models and a chapter on the analysis of frequencies. The four chapters on analysis of variance cover not only all of the familiar ground such as factorial designs and repeated measures but also provide good discussions of the use of Type I versus III sums of squares and the pros and cons of restricted versus unrestricted approaches to mixed models. I particularly liked the discussion of restricted and unrestricted models because this is a source of confusion for biologists, yet I know of
PII: S 0 0 2 2 - 0 9 8 1 ( 0 2 ) 0 0 2 7 8 - 2
198
Book review
only one other introductory statistics book that addresses the problem. Restricted and unrestricted models give different estimates of the expected mean squares for mixed models. This is a source of confusion because nearly all introductory textbooks assume the restricted model, yet most modern statistical packages (e.g., SAS, JUMP) implement the unrestricted model. Quinn and Keough discuss multivariate analysis in a little less than 90 pages. There is a brief introduction to multivariate analysis, multivariate analysis of variance and discriminant analysis, principal components, correspondence analysis, multidimensional scaling and cluster analysis. Experimental Design and Data Analysis for Biologists covers a wide range of topics, and the book has a little something for everyone and probably not enough in many places for most readers. In many places, I marvelled at the range of material covered but then wondered why more was not said. Yet, in all fairness to Quinn and Keough, I do not think anyone could have done any better given the range of topics. The discussions of the mechanics of statistical methods are too brief for the book to be used as the textbook for a first course in statistics, but the book is perfect as a reference source or for a graduate level course in statistics. I showed the book to a number of graduate students, and all of them thought it would be a good textbook for a graduate level course. I plan to test the hypothesis and use Experimental Design and Data Analysis for Biologists as the textbook when I teach advanced statistics this coming year. Peter S. Petraitis Department of Biology, University of Pennsylvania, Philadelphia, PA 19104-6018, USA E-mail address:
[email protected] Tel.: +1-215-898-4207; fax: +1-215-898-8780
