Identifying Cyclical Patterns of Behavior Using a Moving-Average, Data-Smoothing Manipulation

For some individuals, rates of destructive behavior change in a predictable manner, irrespective of the contingencies programmed. Identifying such cyclical patterns can lead to better prediction of destructive behavior and may allow for the identification of relevant biological processes. However, identifying cyclical patterns of behavior can be difficult when using traditional methods of visual analysis. We describe a data-manipulation method, called data smoothing, in which one averages the data across time points within a specified window (e.g., 3, 5, or 7 days). This approach minimizes variability in the data and can increase the saliency of cyclical behavior patterns. We describe two cases for which we identified cyclical patterns in daily occurrences of destructive behavior, and we demonstrate the importance of analyzing smoothed data across various windows when using this approach. We encourage clinicians to analyze behavioral data in this way when rates vary independently of programmed contingencies and other potentially controlling variables have been ruled out (e.g., behavior variability related to sleep behavior).