To split behaviour into bouts, log-transform the intervals.

Analysis of behaviour that is displayed in bouts depends crucially on quantitative estimates of bout criteria, that is, the lengths of the shortest intervals between bouts. Current methods estimate bout criteria by modelling the log-transformed (cumulative) frequency distributions of intervals between events. For analysis of feeding behaviour, these models will not result in biologically meaningful quantitative estimates (Tolkamp et al. 1998, Journal of Theoretical Biology194, 235-250). We proposed a method that models the frequency distribution of log-transformed interval lengths instead. Applying this method to a single data set showed that the log-transformed lengths of intervals between feeding events were distributed as two Gaussians. Here we test this model using a data set of 35 171 intervals between feeding that was obtained during an experiment with 38 cows in three dietary treatment groups. No meaningful bout criterion could be obtained for some individuals, which casts doubt on the general validity of the proposed model. Addition of a third log-normal improved the fit of the model and we hypothesized that this third population represents intervals including drinking. In a second experiment, we found the measurements to be consistent with this hypothesis. We obtained meaningful meal criteria for all individuals by fitting either a double, or a triple, log-normal model to the frequency distributions of the lengths of intervals between feeding. These log-normal models appear to be not only more biologically meaningful than log (cumulative) frequency models but also far more flexible. Copyright 1999 The Association for the Study of Animal Behaviour.