Describing Sequences of Behavior

Abstract

A review is given of the methods currently available for analyzing sequences of behavior. Simple flow diagrams based on the frequencies or conditional probabilities of individual transitions are considered to be of restricted usefulness except where the sequence is highly ordered and the different patterns occur at similar frequencies. It is more helpful to compare the data with a random model provided that repetitions of the same behavior, and any transitions which cannot occur, are excluded before the expected number of each type of transition is calculated. Such comparisons are most likely to be helpful if the behavior patterns included are closely related and fall into discrete homogeneous categories. The fact that most behavioral data are unlikely to be stationary is considered to be the main factor limiting this approach. It is suggested that first-order transition analysis and correlative techniques are the best current methods for examining such data. The search for higher-order dependencies is useful only in stationary data, where grouping of acts due to changing causal factors can be assumed to be unimportant. Additional difficulties involved in the analysis of sequences of interaction between individuals are briefly discussed. The major complicating factor here is that the behavior of an individual is likely to be dependent both on that of others and on its own previous behavior. Some ways of improving current techniques to take account of this are put forward. It is emphasized that sequence analysis provides only a description of the behavior under study and that there are dangers in making causal inferences on the basis of such descriptions alone.

This research was financed by a grant from the Science Research Council.