Many physiologists have treated all variance between individuals as noise, and eliminated it by averaging the responses from several animals. A typical example is a study by Peitsch et al. (1992), who measured the color-receptor wavelength positions of several species of Hymenoptera. They found that differences between species, although slight, exceeded differences between individuals of the same species, and concluded that variance between data from animals of one species was entirely caused by measurement error. This may be correct; however it would also be worthwhile to take the possibility seriously that there might be real (i.e., heritable) variation between individuals. While such variation may be a nuisance for the physiologist trying to extract smooth functions from noisy data, it is a resource for the evolutionary biologist interested in predicting how animals will respond to directional selection.
A more serious (and common) error in physiological work is caused by equating intraindividual variance with interindividual variance. Many authors regarded it is legitimate to take repeated measurements from the same individual animal, and treat these as if they had taken independent measurements from different animals. In fact, numbers of individuals tested are not even available in some behavioral or physiological studies of honeybees; instead, only total numbers of choices or measurements are given. The result is that the numbers of observations, in such studies, is often drastically inflated. It is trivial to most biologists that one cannot obtain a sample size of 150 leaf diameters by measuring 3 leaves 50 times over. Yet, this is precisely what some physiologists do in their data analyses. This is especially dangerous when comparisons between groups of animals are performed. For example, Vorobyev et al. (1999) tested honeybees' ability to detect artificial "flowers" of different colors on a green background. They used the total n of choices (270) as a basis for the conclusion that white was more easily detectable than gray, but the number of individuals tested (which should have been used for statistics) for white flowers was only three! Clearly, within-individual behavior is noisy, and therefore one needs several data points from each animal. However, behavioral variance across individuals can be large, particularly in honeybees whose experience before and between experiments is outside the control of the experimenter. Thus, once each animal's behavior is quantified (if necessary, with several tests), only a single data point per animal may be used for comparison between groups of individuals (see, e.g., Chittka & Thomson 1997; Chittka et al. 1997).
Because interindividual differences were regarded as noise, many authors pooled data from individuals without testing for heterogeneity. This can be hazardous. For example, Scherer & Kolb (1987) tested innate floral color preferences of Pieris butterflies. They found that colors both in the blue and the red part of the spectrum were preferred, and conjectured that a neuronal mechanism summing up the responses from blue and red receptors might drive this behavior. This mechanism is simple and therefore attractive, but there are alternative explanations. In essence, Scherer & Kolb (1987) used average group behavior to deduce the neuronal mechanisms implemented in individuals. When only pooled data are presented, it is equally possible that group behavior is caused by some individuals preferring red, others blue. If this were the case, no single butterfly would use summed inputs from two receptors to search for flowers. Furthermore, pooling the data masks possible sequence effects. Say, for example, that butterflies tend to visit blue first (bypassing red flowers), then red (bypassing the blue flowers it has already found unrewarding); such a pattern would also suggest a different model of neuronal control.
Was this article helpful?