Inter-individual differences in multivariate time series: Latent class vector-autoregressive modelling

Abstract

Theories of emotion regulation posit the existence of individual differences in emotion dynamics. Current multi-subject time-series models account for differences in dynamics across individuals only to a very limited extent. This results in an aggregation that may poorly apply at the individual level. We present the exploratory method of latent class vector-autoregressive modelling (LCVAR), which extends the time-series models to include clustering of individuals with similar dynamic processes. LCVAR can identify individuals with similar emotion dynamics in intensive time-series, which may be of unequal length. The method performs excellently under a range of simulated conditions. The value of identifying clusters in time-series is illustrated using affect measures of 410 individuals, assessed at over 70 time points per individual. LCVAR discerned six clusters of distinct emotion dynamics with regard to diurnal patterns and augmentation and blunting processes between eight emotions.

Dynamic system perspectives on anxiety and depression

“…Promising future avenues to study anxious and depressed moods include the integration of complex dynamic processes across levels of resolution via simultaneous top-down and bottom-up approaches (Forbes et al, 2016;Witherington, 2007) and models that integrate within and between person estimates (Adolf et al, 2014;Ernst et al, 2019;Fisher et al, 2018).…”

“…Researchers also questioned the validity of using centrality indices to study mood symptom networks (Bringmann et al, 2018;Forbes et al, 2017;Hallquist et al, 2019), including betweenness and closeness centrality as measures of node importance (Bringmann et al, 2018), and struggled with statistical equivalent network and latent variable models despite their marked conceptual differences (Markman et al, 2018;Molenaar et al, 2007, which suggests that other and more advanced methodologies are required. Future studies might apply more advanced dynamic system techniques (examples in this book and ), dynamic cluster models (Ernst et al, 2019), and more diverse and multimethod measures (e.g, observers, language analyses, interviews), as the almost exclusive usage of self-report measures of adolescents' anxiety and depression goes along with some threats to validity (Eisenberg et al, 2010), and should be interpreted with caution.…”

Insight Into Individual Differences in Emotion Dynamics With Clustering

“…The covariance structure of these models depends on the length of the time series, rendering these models unsuited for time series with a high number of observations and/or unequal length. Recently, probabilistic, adaptive approaches have been proposed for long and unequal time series for the univariate (Michael & Melnykov, 2016) and multivariate case (Ernst et al, 2019).…”

“…Adaptive clustering methods offer many advantages over filtering methods. For instance, some are able to describe different clusters through different time series models (for details, see Ernst et al, 2019). These models, however, are very complex and their iterative estimation procedure makes them slow when studying large data sets and scenarios where various numbers of potential clusters and/or potential time series models are considered.…”

Clustering individuals’ temporal patterns of affective states, hunger, and food craving by latent class vector-autoregression

“…Until recently, it was not possible to determine between-subject variability in within-subject associations in a data driven manner. The development of latent class vector-autoregression (LCVAR) closes this gap by exploring latent classes of individuals (i.e., distinct subgroups within a sample) based on similarities in lagged Level 1 associations, thus based on similar temporal dynamics (28).…”