Sources of Extraneous Variation in the Study of Meristic Characters: The Effect of Size and of Inter-Observer Variability

Publisher Website
Google Scholar
Abstract
The repeatability (precision) with which lizard scale counts can be made was assessed using two observers who repeatedly scored six scale characters on the same five adult and five hatchling specimens of Anolis sagrei. The precision with which characters could be scored varied by more than an order of magnitude, and there was congruence between observers as to which characters could be scored precisely and which could not. Variances derived from the repeated counts were always correlated with the character means, thus confounding the relationship between intrinsic character variation and the mean. For one observer, character variances were consistently higher for hatchlings than for adults, a finding that impugns the conclusions of some vertical cohort analyses that reduction in variance in samples of larger (= older) lizards is evidence of stabilizing selection. There was enormous inter-observer variability in the precision with which the characters were scored, and the observers differed in their conception of the characters themselves. Discriminant analyses "discovered" differences in the data that were easily of the magnitude used to justify taxonomic decisions, even though the differences were due solely to inter-observer variability. There was no evidence of skewness in the distribution of the repeated scale counts, and thus no indication that deviations from the mean were biased. Choice of characters and of the individuals who score the characters is critically important. Pilot studies are recommended in order to identify those characters that can be scored with acceptable precision, and those individuals capable of scoring characters precisely. Error variances based upon repeated counts on the same specimens provide a rationale for partitioning character variances into their intrinsic and artificial components.

This publication has 1 reference indexed in Scilit: