Adequate statistical methods to reduce the number of animals used in behavioural experiments: the analysis of the behavioural transitions.

In ethological and behavioural toxicological studies, elaborate behavioural patterns shown by the animals under well-established experimental paradigms or naturalistic conditions are routinely observed and split into single behavioural items. Subsequently, these items are analysed in terms of their frequencies and/or durations. Behavioural observations are usually videotaped and scored by dedicated softwares, which collect the sequences of behavioural items together with frequencies and durations. So far, the Cox proportional hazards model, a method originally developed for the analysis of time-to-event data, has been employed for the analysis of the time-structure of behaviour, but its usefulness has been limited because of difficulties in including random effects in the model. Recent developments in mixed models for the analysis of time-to-event data may overcome this limitation and improve the analysis of behavioural patterns. Data from social interactions in mice on the effects of exposure to chlorpyrifos, a widely used organophosphorous pesticide, are presented to illustrate the use of these new statistical methods. Our results suggest that the study of behavioural sequences may highlight the role of the investigated conditions (treatments, genetic condition, social status) in setting behavioural organisations. In addition, the refinement of statistical methods by time-structured analysis provides more detailed information from an experimental data set, thus contributing to the reduction of the number of animals used in this field of the life sciences.