A New, Old Method for Assessing Measurement Error in Both Univariate and Multivariate Morphometric Studies

Publisher Website
Google Scholar
Abstract
A new approach to assessing percent measurement error (%ME), using an old statistical technique (Model II Analyses of Variance and Covariance: ANOVA/ANCOVA), was developed using both weight and linear measurements made three times each on 87 freshwater snails (Cipangopaludina chinensis; Vivipariidae). Variability of each measurement, as well as covariability between pairs of measurements, was partitioned into among- and within-snail (i.e., measurement error) components. The %ME varied by two orders of magnitude across the ten measurement variables considered. Shell weight had the lowest %ME (0.059%), while body whorl height had the highest (3.88%). There was a low correlation between %ME and the among-snail coefficient of variation for each variable (r = −0.28; P > 0.20). Within-snail correlations between pairs of measurement variables were uniformly low (rwithin ≤ 0.20), while among-snail correlations were generally very high (r = 0.74 to 0.99) due to the large size range of snails in the sample. In any morphometric study, if %ME is found to be high for a particular variable, it should either be deleted from the study or remeasured a number of times on every individual to be included in the dataset.