How should we interpret estimates of individual repeatability?

Abstract

Individual repeatability (R), defined as the proportion of observed variance attributable to among-individual differences, is a widely used summary statistic in evolutionarily motivated studies of behaviour and (increasingly) physiology. Although statistical methods to estimate R are well known and widely available, there is a growing tendency for researchers to interpret R in ways that are subtly, but importantly, different. Some view R as a property of a data set and a statistic to be interpreted agnostically with respect to mechanism. Others wish to isolate the contributions of “intrinsic” and/or “permanent” individual differences, and draw a distinction between true (intrinsic) and pseudo-repeatability arising from uncontrolled extrinsic effects. This latter view proposes a narrower, more mechanistic interpretation, than the traditional concept of repeatability, but perhaps one that allows stronger evolutionary inference as a consequence (provided analytical pitfalls are successfully avoided). Neither perspective is incorrect, but if we are to avoid confusion and fruitless debate, there is a need for researchers to recognise this dichotomy, and to ensure clarity in relation to how, and why, a particular estimate of R is appropriate in any case.