Modelling short-term feeding behaviour

Previous analyses of feeding behaviour have either ignored the bouted structure of that behaviour or used log-survivorship analysis for the estimation of bout criteria. The latter is based on the null hypothesis that feeding behaviour is random within as well as between bouts. We falsify this null hypothesis by analysing how specific feeding events (such as visits to a feeder) by animals of a number species are distributed in time. They all show bouted, but non-random, feeding behaviour with a similar structure. This is caused by satiety at the end of feeding bouts, followed by an increase in the probability of animals starting to eat with time since feeding last. This results in a skewed normal distribution of the population of longer between-feeding, i.e. between-meal, intervals, which can be (almost) normalised by log-transformation of interval lengths. We show that either a Gaussian (for pooled data sets) or a Weibull (for individuals or data pooled across animals with similar feeding strategies) give excellent descriptions of the distribution of this population of logtransformed interval lengths. Depending on species, methodology and individual animal habits there may be one or more population(s) of shorter (i.e. within-meal) intervals. We developed methods that allow estimation of meal criteria even if the number of within-meal interval populations and their shape are not exactly known. Because the structure of feeding behaviour is the same across species, similar models to estimate bout criteria can be applied to data obtained with animals of species as diverse as cows, pigs, chickens, turkeys, ducks and dolphins. We demonstrate how analysis of data subsets, following disaggregation of pooled data, can elucidate underlying mechanisms that would otherwise remain obscure. The developed models can also be applied to other forms of animal behaviour that are affected by satiety-like principles.