Most behavioral experiments, involving direct observations of sequences of events and the time at which they occur, require detailed time records of behavior. Statistical methods primarily designed to analyze the time structure of behavior give information about effects of the investigated factors on the time structure providing better insight into the behavioral processes. The continuous time Markov chain models are the simplest way to describe the time structure. The parameters of the model represent behavioral tendencies to switch between acts and to start and stop activities, and subsume the information of both the duration and the alternation of each act providing a complete summary of the behavioral pattern. The transition rate, l(AB), from a behavior (A) to another (B) represents the chance per time unit that one behavior is followed by another, so it can be interpreted as the priority of the following behavior while the animal performs the preceding one. The termination rate of state A, l(A), is the chance per unit of time that the item A is interrupted and hence any other item may be shown. The transition probability P(AB) from A to B is the probability to switch from A to B. The relation between these parameters is: P(AB)=l(AB)/l(A). The maximum likelihood estimators are: P'(AB)=N(AB)/N(A), l'(A)=N(A)/S(A) and l'(AB)=N(AB)/S(A) where N(AB) denotes the number of observed transitions from A to B, N(A) the number of bouts of A, and S(A) the time spent in A. By analyzing the transition rates, which contain all the information concerning distributions of the sojourn times and the probabilities of alternations between acts, one can detect, for example, whether an increase in time spent on behavior is due to an increase in the tendency to start the behavior or to a decrease in the tendency to stop the behavior. Such distinctions, which are lost when analyzing only frequency and duration, are obviously important.

These methods have been applied to assess the effect of a toxicant (low (L) and high (H) concentrations) on social-aggressive interaction in adult mice, undergoing a 20-min agonistic encounter with a standard opponent. Behavior was videotaped, and scored using a keyboard event recorder system (The Observer). The scored items were: Attack (ATT), Chase (CHA), Aggressive grooming (AGR), Tail rattling (TRA), Offensive upright posture (OFF), Defensive upright posture (DEF), Submissive posture (SPO), Freezing (FRE), Escape (ESC) and Other (OTH). Non-parametric analysis of variance highlighted significantly higher duration and frequency of ATT in L- than in H-mice. The analysis of the behavioral sequences showed that the concentration modifies also the behavioral patterns of ATT: H-mice less frequently showed both TRA and OFF after an episode of ATT. Also the transitions rates from ATT to both TRA and OFF were lower in the H-mice. Moreover, H-mice showed significantly lower probability and tendency to switch between CHA and ATT, and lower probabilities to switch both from DEF to OTH, and from OTH to ATT. As shown above, the analysis of sequence gives a detailed description of the behavior allowing us to better understand changes in behavioral processes due to treatment.

References

1. Haccou, P.; Meelis, E. (1992). Statistical Analysis of Behavioural Data. An Approach Based on Time-Structured Models. Oxford: Oxford University Press.