Abstracts

Modeling non-linear dynamics in psychological time series with multilevel hidden Markov models

Emmeke Aarts

Time series (or intensive longitudinal) data are increasingly used in psychological science, allowing the study of the dynamics of human functioning with unprecedented granularity. A powerful class of models for capturing these dynamics is the Hidden Markov Model (HMM), which captures processes that switch between a set of latent states over time, each associated with a distinct probability distribution. Unlike currently predominant linear models such as the Vector Autoregressive (VAR) model, HMMs can capture more complex behaviors. They can characterize multiple equilibrium states (for instance, manic and depressed states as seen in bipolar disorder) and quantify how the system transitions between them. They can also naturally accommodate skewed or multimodal data distributions commonly observed in intensive longitudinal data. Extending HMMs to a multilevel Bayesian framework enables simultaneous modeling of within-person temporal dynamics and between-person differences, thereby leveraging the full richness of intensive longitudinal data. In this talk, I will introduce multilevel HMMs in an accessible manner, touch upon their implementation in the open-source R package mHMMbayes, and offer empirically based recommendations for researchers applying these models in psychological research.

Latent Markov factor analysis for evaluating measurement model differences in intensive longitudinal data

Leonie Vogelsmeier

When studying intensive longitudinal data (e.g., with Experience Sampling Methodology), drawing conclusions about dynamics of psychological constructs (e.g., well-being) requires the measurement model (MM) to be invariant between subjects and within subjects over time. However, there might be non-invariance in the MM, for instance, due to subject-specific differences and changes in item interpretation or response styles. Latent Markov factor analysis (LMFA) is a dynamic mixture modeling approach specifically developed to explore such MM differences for multiple subjects and time points simultaneously, without requiring prior assumptions about where or how these differences occur. LMFA combines a discrete- or continuous-time latent Markov model (that clusters observations into separate states, according to state-specific MMs) with mixture factor analysis (that estimates the MM for each state). In my talk, I will explain the main principles of LMFA, discuss practical considerations and recommendations for applying the method, and introduce the user-friendly R package lmfa, which allows researchers to investigate MM differences in their own intensive longitudinal data.