Joel L Pick, Nyil Khwaja, Michael A. Spence, Malika Ihle, Shinichi Nakagawa 

We often quantify the rate at which a behaviour occurs by counting the number of times it occurs within a specific, short observation period. Measuring behaviour in such a way is typically unavoidable but induces error. This error acts to systematically reduce effect sizes, including metrics of particular interest to behavioural and evolutionary ecologists such as R2, repeatability (intra-class correlation, ICC) and heritability. Through introducing a null model, the Poisson process, for describing the frequency of behaviour, we give a mechanistic explanation of how this problem arises and demonstrate how it makes comparisons between studies and species problematic, because the magnitude of the error depends on how frequently the behaviour has been observed as well as how biologically variable the behaviour is. Importantly, the degree of error is predictable and so can be corrected for. Using the example of parental provisioning rate in birds, we assess the applicability of our null model for describing the frequency of behaviour. We then survey recent literature and demonstrate that the error is rarely accounted for in current analyses. We highlight the problems that arise from this and provide solutions. We further discuss the biological implications of deviations from our null model, and highlight the new avenues of research that they may provide. Adopting our recommendations into analyses of behavioural counts will improve the accuracy of estimated effect sizes and allow meaningful comparisons to be made between studies.