Interobserver error in a large scale anthropometric survey

Abstract

The adverse effects of interobserver error on morphometric population comparisons are well documented in the literature. While interobserver error can rarely be avoided, it can be minimized by having a single individual locate and mark relevant landmarks, by limiting the number of observers for each variable, and by reviewing repeated measures data daily to catch and correct measurer drift during data collection. In this study, two pairs of experts participated in interobserver error trials designed to pre-set observer error limits for use in the quality control of a large scale anthropometric survey. Repeatability data were also collected twice daily in the field and reviewed with the measurers. Interobserver errors obtained in the field were lower than those achieved by the experts for 27 of 30 dimensions. These results suggest that establishment of permissible interobserver error in advance of data collection and frequent review of repeated measurements during data collection can reduce the magnitude of interobserver error below that obtained by experts measuring in a laboratory setting. However, even differences of small magnitude can be serios when they are directional, and 17 of 30 dimensions exhibited statistically significant bias between measurers despite all quality control efforts. The magnitudes of interobserver error observed in this study have proven particularly useful in evaluating the biological relevance of statistically significant differences which are of relatively small magnitude.