The behavior of an animal, with or without interaction with its environment, can generally be decomposed into a set of
mutually excluding activity categories such as play, exploration, or rest. Thus, the data consist of one or several series of
successive activities which can be used to answer the following questions: Does the probability of observing a particular
activity depend on the preceding observed activities? Do some particular patterns of successive activities appear more often than expected by chance only? Which external events influence the behavior of this animal?
This article covers several related methods used to answer these questions. At the base is the need to compute and analyze a transition probability, the probability of observing a particular activity in relation to the past. A Markov chain is a stochastic answer to this kind of problem, when lag-sequential analysis refers to a less formalized approach aimed at the identification of the most usual patterns of activities. We review here the basic principles of these two approaches and detail some advances such as the formulation of sequence analysis as log-linear models, the use of mixtures to obtain parsimonious models, the concept of hidden Markov models incorporating latent information, and the use of these models for clustering purpose.