Comparing the performance of non-parametric change point detection methods for capturing response concordance.

Response concordance is a key concept in the behavioral sciences. It can be defined as the occurrence of changes in response patterning (change in mean) and/or response synchronization (change in covariation) in a multivariate time series.  Revealing response concordance can be viewed as a change point detection problem, where the number of change points is unknown a priori. To solve this problem, DeCon was recently developed, which detects change points by combining a moving windows approach and robust PCA. Yet, in the literature, several other methods have been proposed  that employ other non-parametric tools: E-divisive,  Multirank and KCP. The relative performance of all these methods for capturing response concordance is still unknown, however. Therefore, we compare E-divisive, Multirank, KCP and Decon, through extensive simulations. Specifically, we use the simulation settings of Bulteel et al. implying changes in mean and in correlation structure and those of Matteson et al.  implying different numbers of (noise) variables. KCP emerged as the best method in almost all settings. However, in case of two or more noise variables, only DeCon performed adequately.