Chapter 9. Experimental design

That said, our current discussion of experimental design only tangentially addresses some fundamental topics.  For example, we provide thorough discussion of power analysis for a variety of experimental designs (see later Chapters) to help researchers decide upon the appropriate level of replication (sample size) for their experiment. 

However, we provide little discussion of why experiments require replication, generally (although we elude to this in our discussion of sampling error in earlier chapters).  Likewise, we currently do not discuss how to select appropriate controls for an experiment (and when controls are necessary). 

Therefore, our treatment of experimental design assumes some prior knowledge.  We refer readers to Ruxton & Colegrave’s excellent book, “Experimental Design for the Life Sciences” for further support; as mentioned elsewhere, this book is written to be accessible to undergraduates, but I know faculty members who find this book for themselves. 

Our treatment of experimental design addresses recommendations in recent literature with respect to designing experiments that use animal models.  That said, our content applies to experiments in biology, generally (e.g., physiology, ecology, evolution), beyond the use of animal models.

View examples of this here:

We focus on the following topics:

• Use of randomization.  ‘Randomization’ serves to eliminate bias, and is the most essential assumption in statistical tests.  We discuss not only the allocation subjects to treatments (the most common use of randomization), but also implementing randomization in all aspects of an experiment.
• Selecting materials (subjects) for an experiment to maximize an experiment’s ability to yield generalizable conclusions.  Note that ecologists and evolutionary biologists should, where applicable, also consider sampling methods for field research (i.e., done ‘in the wild’), which we do not discuss here.
• The importance of ‘blinding’ throughout an experiment; studies that lack blinding can yield surprisingly biased results.
• ‘Blocking’ as a means of accounting for unwanted variation in an experiment.  We introduce the techniques to analyze experiments with ‘block’ in Chapters that address analyses of multiple explanatory variables (i.e., Chapters, “Analyzing experiments with more than 1 factor”, “Mixed effects models”, and “Understanding covariates:  …”).
• ‘Covariates’ as a way of dealing with, and understanding additional sources of variation in an experiment.  We address techniques to analyze experiments with covariates in the Chapter, “Understanding covariates:  …”.