Morphological variability or inter-observer bias? A methodological toolkit to improve data quality of multi-researcher datasets for the analysis of morphological variation.

Abstract

The investigation of morphological variation in animals is widely used in taxonomy, ecology, and evolution. Using large datasets for meta-analyses has dramatically increased, raising concerns about dataset compatibilities and biases introduced by contributions of multiple researchers. We compiled morphological data on 13 variables for 3073 individual mouse lemurs (Cheirogaleidae, Microcebus spp.) from 25 taxa and 153 different sampling locations, measured by 48 different researchers. We introduced and applied a filtering pipeline and quantified improvements in data quality (Shapiro-Francia statistic, skewness, and excess kurtosis). The filtered dataset was then used to test for genus-wide sexual size dimorphism and the applicability of Rensch's, Allen's, and Bergmann's rules. Our pipeline reduced inter-observer bias (i.e., increased normality of data distributions). Inter-observer reliability of measurements was notably variable, highlighting the need to reduce data collection biases. Although subtle, we found a consistent pattern of sexual size dimorphism across Microcebus, with females being the larger (but not heavier) sex. Sexual size dimorphism was isometric, providing no support for Rensch's rule. Variations in tail length but not in ear size were consistent with the predictions of Allen's rule. Body mass and length followed a pattern contrary to predictions of Bergmann's rule. We highlighted the usefulness of large multi-researcher datasets for testing ecological hypotheses after correcting for inter-observer biases. Using genus-wide tests, we outlined generalizable patterns of morphological variability across all mouse lemurs. This new methodological toolkit aims to facilitate future large-scale morphological comparisons for a wide range of taxa and applications.