Behavioural effects of endocrine disrupting chemicals on laboratory rodents: statistical methodologies and an...

Appropriate behavioural tests and adequate statistical tools may help to establish the ED properties of a given compound by pointing out the alterations of selected behavioural endpoints. Frequently, laboratory collected data consist of frequencies and/or durations of specific items, and the analysis of variance (ANOVA) technique is performed to assess whether the investigated factors affect these behavioural endpoints. Moreover, when numerous aspects of behaviour are investigated simultaneously, Principal Component Analysis (PCA), a multivariate technique, may be very useful to reduce the overwhelming number of correlated original variables to a few orthogonal artificial variables (factors). Continuous Time Markov Chain (CTMC) models may be applied to analyse the time structure of a behavioural pattern when data consist of sequences of events and the time points at which they occur. Moreover, the Cox Proportional Hazard Model, a methodology originally developed for the analysis of failure time data, may help to evidence the effects of a given treatment on behavioural sequences when the assumptions of CTMC models are not fully satisfied. Analyses on data from mice of the outbred CD-1 strain (controls in a study of toxicity and exposed to PCB during development) are presented as examples to show how adequate statistical analyses and appropriate behavioural tests may reveal relevant effect of treatments otherwise not easily detected.