Interrogating Random and Systematic Measurement Error in Morphometric Data

Measurement error is present in all quantitative studies, and ensuring proper biological inference requires that the effects of measurement error are fully scrutinized, understood, and to the extent possible, minimized. For morphometric data, measurement error is often evaluated from descriptive statistics that find ratios of subject or within-subject variance to total variance for a set of data comprising repeated measurements on the same research subjects. These descriptive statistics do not typically distinguish between random and systematic components of measurement error, even though the presence of the latter (even in small proportions) can have consequences for downstream biological inferences. Furthermore, merely sampling from subjects that are quite morphologically dissimilar can give the incorrect impression that measurement error (and its negative effects) are unimportant. We argue that a formal hypothesis-testing framework for measurement error in morphometric data is lacking. We propose a suite of new analytical methods and graphical tools that more fully interrogate measurement error, by disentangling its random and systematic components, and evaluating any group-specific systematic effects. Through the analysis of simulated and empirical data sets we demonstrate that our procedures properly parse components of measurement error, and characterize the extent to which they permeate variation in a sample of observations. We further confirm that traditional approaches with repeatability statistics are unable to discern these patterns, improperly assuaging potential concerns. We recommend that the approaches developed here become part of the current analytical paradigm in geometric morphometric studies. The new methods are made available in the RRPP and geomorph R-packages.