Clustering intensive longitudinal data through mixture multilevel vector-autoregressive modeling

Experience sampling methodology is increasingly used in the social sciences to
analyze individuals’ emotions, thoughts and behaviors in everyday-life. The resulting intensive longitudinal data is often analyzed with the objective to describe
the inter-individual differences that are present within it. To accommodate interindividual differences to a greater extent than previously possible, a mixture multilevel vector-autoregressive model is proposed. This model combines a mixture
model at level 2 (individual level) with a multilevel vector-autoregressive model [1]
that describes the dynamic fluctuations present at level 1 (time-point level). This
exploratory model identifies mixture components of individuals who exhibit similar
overall means, autoregressions, and cross-regressions. Within each mixture component, multilevel coefficients allow additionally for within-component variation on
these vector-autoregressive coefficients. The advantage of exploratory identifying
mixture components and accounting for within-component variation is demonstrated
on data from the COGITO study. This data contains samples of individuals from
disparate age groups of over 100 individuals each.