You are not currently logged in.

Access JSTOR through your library or other institution:


Log in through your institution.

Journal Article


Vol. 54, No. 1 (January 2004), pp. 78-79
DOI: 10.1641/0006-3568(2004)054[0078:sbd];2
Stable URL:[0078:sbd];2
Page Count: 5
  • Download PDF
  • Add to My Lists
  • Cite this Item


ADepartment of Statistics, Oklahoma State University, Stillwater, OK 74078

Experimental Design for the Life Sciences. Graeme D. Ruxton and Nick Colegrave. Oxford University Press, Oxford, United Kingdom, 2003. 136 pp., illus. $24.95 (ISBN 0199252327 paper).

Whenever the words “experimental” and “design” are placed together in a sentence, some biologists cringe and brace themselves for complex model statements and tough-to-understand equations. This is unfortunate, mostly because the line between the art of the design of experiments and statistical analysis has blurred. Granted, the two are inextricably linked, and complete understanding of experimental design means understanding the analytical costs or benefits resulting from the design utilized. However, to speak purely of the discipline of the design of experiments does not require complicated equations, and this book serves as proof.

If you are looking for a good textbook for a course in experimental designs, Ruxton and Colegrave's work could not solely serve for that purpose. There are no analyses of variance, no models, no equations, no homework problems. Does this sound strange for a book about statistics? What you must understand is the viewpoint of the authors. Ruxton and Colegrave, University of Glasgow Reader and University of Edinburgh Lecturer, respectively, claim that this is not a statistics book. They allude to statistical tests and their important role in experimentation. They also present an extensive bibliography of books that deal with analytical concepts. However, the authors believe that the subject at hand is not simply difficult equations and mathematics. As they state, experimental design is “about common sense, biological insight and careful planning.” It's a mantra they chant throughout the text. Experimental design, they contend, is the basis for biology.

I think it is vital to understand the authors' goal in writing this book. I believe they have had innumerable conversations and consultations with biologists, and many of them have asked the same questions regarding the way experiments are laid out. Trust me, I know what that is like. It is likely that these conversations are so ingrained in their respective psyches that this book might have seemed to write itself. They approach each topic logically, and with excellent examples and case studies. The vibe that comes across from the book seems to favor the wildlife biologist, but it would be easy for any biologist, regardless of discipline, to extrapolate to his or her field and set of possible experiments. The authors also contrast biology nicely with the physical sciences. Though both of these broad disciplines require good fundamental experimental design, Ruxton and Colegrave argue convincingly that the presence of biological creatures induces a source of variation that requires more care in the design phase.

Who should consider reading this book? I know that there are many scientists out there who are uncomfortable with their knowledge of experimental design, stemming perhaps from a lack of formal training in the subject in graduate school. This book can serve as a primer for these individuals. An important aspect of this book is the inclusion of key definitions. These are always set apart from the rest of the text in boxes. If nothing else, the reader will be better equipped to discuss these issues with a statistician. For a good example, the authors present a nice discussion of sample size, and they define the related terms power and error and how they affect design issues. This is obviously an important step in any design. Keep in mind, however, that the reader will not be able to perform a power analysis based only on the contents of this book. They also stress in this context that a small amount of good data is much better than a whole bunch of bad data.

As another example, Ruxton and Colegrave give a splendid discussion of the importance of controls in experimentation. One of the great fallacies in experimental design is that every experiment must have a control. The authors define explicitly what the difference is between positive and negative controls and state that controls are needed only if they are important in the comparisons with the treatments of interest. This, by the way, is an argument I've made on numerous occasions, an argument I have lost more often than won. The example Ruxton and Colegrave present is that of a new treatment being compared with an existing treatment (positive control), in which case the comparison of the new treatment with no treatment (negative control) is unnecessary.

The subject of actual experimental designs, such as the completely randomized design (or “randomised,” as our friends from across the pond tend to spell it), isn't introduced until page 64. Since the book has only 114 pages, very little of the content is devoted to these issues. But the authors do include books in the bibliography that present these designs in greater detail. And the book does present a nice discussion of factorial arrangements. Main effects and interaction are defined. Strangely, the authors don't use the term simple effect, which usually serves as the antithesis of main effect. They do warn of the dangers of interpreting main effects in the presence of interaction, but not emphatically enough, in my opinion. I have seen many journal articles that present the p-values for the main effects of A and B and then proceed to discuss the significant interaction. This is a practice that should be nipped in the bud, and books like this one can help to nip it.

One item that I would have liked to have seen is the definition of an experimental unit. This would have fit well in the narrative of pseudoreplication, which was otherwise very nicely done. Another quibble I have is the characterization of crossover designs as repeated measures experiments. There are repeated measures studies that are not crossover experiments, and these play a prominent role in science. Readers would have benefited from a discussion of intrasubject correlation and how it is modeled. This subject could have been extended to spatial relationships as well.

When I started reading this book, I found myself questioning whether Ruxton and Colegrave could successfully omit equations from their book. They have managed to do just that. As a teacher of a course in the design of experiments, I have realized that all of the examples presented are worthy of the subject they represent. I enjoyed reading this book, appreciated the unique approach, and have recommended it to my students as supplementary reading. Experimental Design for the Life Sciences has found a permanent place in my library of experimental design material.

Part of Sustainability