Much of the field of animal behaviour rests on experimental studies of the
responses of animals to different classes of stimuli. Playback
experiments, which compare responses of animals to tape recordings of
different sounds, are a prime example. Consequently, the proper
design of these experiments is central to the scientific study of animal
behaviour. Over a decade ago, a discussion of the design of
behavioural experiments focused on the problems of pseudoreplication
(Kroodsma 1989a, b, 1990; Searcy 1989; McGregor et al. 1992; Weary &
Mountjoy 1992), and a recent paper has reviewed subsequent progress in
avoiding pseudoreplication in experimental studies of bird song (Kroodsma
et al. 2001).
My objective here is ... to expand the discussion of behavioural
experiments. To this end, I identify some compromises any
experimenter must make in justifying the biological independence of
subjects, the external validity of conclusions, the multiple use of
exemplars and subjects, and the effects of sample size on unsuspected
bias. These compromises make it less clear that any one design is
universally optimal. In some circumstances, it is appropriate to
use each exemplar more than once and to test each subject more than once.
Although there is no ideal experimental design, I propose that
there is an ideal way to report a behavioural experiment, one that
explicitly identifies the compromises involved.