Assessment of Morphometric Variation in Natural Populations: The Inadequacy of the Univariate Approach

Abstract

Systematists often attempt to avoid the problem of correlated characters by establishing an arbitrary number of variables that must be significant before groups are considered distinct. The appropriateness of this approach has not been evaluated empirically in the biological literature. We analyzed morphometric data for 27 species of bats from the Brazilian Northeast. Twenty-two mensural characters (12 cranial, 10 external) were analyzed for iterlocality and secondary sexual variation in each species using ANOVA and MANOVA. The univariate and multivariate analyses showed little correspondence; no predictable relationship between the number of characters exhibiting significance for a particular treatment effect in the uivariate analyses and the significance level for that treatment in the multivariate analysis was discernible. Small sample sizes or disparate sample sizes do not contribute to this phenomenon. Results ranged from: 11 of 12 characters significant using ANOVA and nonsignificance in the MANOVA; to no character significant using the ANOVA with significance in the multivariate analysis. Because MANOVA utilizes rather than ignores correlations among characters, it is the correct statistical test for evaluating overall group differences. We show that even conservative interpretations of the univariate results can lead to erroneous systematic conclusions.