The detection of behavioral state cycles and classification of temporal structure in behavioral states

Previous methods for the analysis of temporal structure in sleep and other state time series have described cycles, rhythms, and semi-Markov chains. Methods, however, have been subjective and arbitrary. We propose an objective system of classification for these series, based on definitions of temporal structure which are consistent with those long used in the analysis of quantitative series. An ordered sequence of statistical tests is described which classifies observed behavioral state time series into four primary categories. The system is illustrated with examples from normal infant sleep. The results show that some infant sleep series are cycles, as previously reported, some are semi-Markov chains, and some are neither. The proposed objective methods promise consistency, clarity, and a richer understanding of behaviors such as sleep.