Detecting cyclicity in social interaction.

This article reviews spectral and cross-spectral analytic methods for detecting cyclicity, cross-cyclicity, and lead-lag relationships in continuous data derived from the observation of dyadic interaction. It is found that lead-lag relationships can be assessed using the phase spectrum. Spectral analytic methods are then generalized to categorical observational data, and it is shown that by these methods one can derive the classical information theory definition of social communication and its distribution statistics. Researchers who study social behavior are discovering that there are occasions when cyclical patterns characterize dyadic interaction, and thus they are searching for statistical techniques that can detect these cycles. The spectral analysis of time-series records was briefly suggested by Luce (1970) as a useful technique for the study of biological rhythms such as heart rate, respiration, REM sleep, and other cyclic biochemical and physiological processes. However, spectral analysis is not widely known to behavioral scientists, and it has yet to be used in the study of social interaction. A recent exception is the work of Hayes and Cobb (Note 1), who observed couples living in a laboratory setting, analyzed cycles of talk and silence using spectral analysis of time-series records, and related an observed cycle to human circadian rhythms. Researchers who study dyadic social interaction are also interested in the bivariate case in which two time-series records are obtained, one from each of the two interacting organisms; the research question often involves