A0253
Title: Uncovering clusters and within-cluster variation in time series: Mixture multilevel vector-autoregressive modeling
Authors: Anja Ernst - University of Groningen (Netherlands) [presenting]
Marieke Timmerman - University of Groningen (Netherlands)
Feng Ji - University of California Berkeley (United States)
Bertus Jeronimus - University of Groningen (Netherlands)
Casper Albers - University of Groningen (Netherlands)
Abstract: In the social sciences, experience sampling methodology is increasingly used to analyze individuals' emotions, cognition and behaviors in everyday life. The objectives in the analysis of the resulting intensive longitudinal data are increasingly focused on inter-individual differences. To accommodate inter-individual differences to a great extent, a mixture of multilevel vector-autoregressive modeling is proposed. The model combines multilevel vector-autoregressive modeling with mixture modeling, to identify individuals with similar traits and dynamic processes. This exploratory model identifies mixture components (or clusters) containing individuals with similar overall means, autoregressions, and cross-regressions. Within each component multilevel coefficients allow additionally for a within-component variation on these coefficients of interest. The model is illustrated on emotion data from the COGITO study. The COGITO data contains two samples of individuals from different age groups of over 100 individuals each. Participants' emotions were assessed daily for about 100 days. The advantage of exploratory identifying mixture components and accounting for within-component variation is illustrated in the COGITO data.
