The afternoon pre-conference workshop on “extracting personalised latent dynamics using multilevel hidden Markov models” was taught by dr. Emmeke Aarts. Emmeke is an Associate Professor at the department of Methodology and Statistics, Utrecht University, and expert on hidden Markov models, intense longitudinal data, and multilevel models.

Facilitated by technological advances such as smartphones, smartwatches, and sensors, it has become relatively easy and affordable to collect data on groups of individuals with a high temporal resolution: intensive longitudinal data (ILD). Due to the high sampling frequency, ILD can uniquely be used to study how psychological, behavioral and physiological processes unfold over time at the within-person level, and between person differences herein. When the dynamics over time of interest can be represented by a latent construct consisting of mutually exclusive categories, the hidden Markov model (HMM; Rabiner, 1989; de Haan-Rietdijk et al., 2017) is a promising novel approach. The HMM is a probabilistic, unsupervised, longitudinal machine learning method which uncovers empirically derived latent (i.e., hidden) states and the dynamics between these latent states over time. Utilising the multilevel framework, heterogeneity between individuals is accommodated, facilitating the study of individual specific dynamics and differences herein.

During this pre-conference workshop, I will introduce you to the multilevel hidden Markov model. Using an empirical application on personalised (suicidal) crisis dynamics, the concepts and benefits of using the multilevel HMM will be explained in an intuitive manner. The introduction is followed by a hands-on workshop using the R CRAN package mHMMbayes. The package mHMMbayes can accommodate categorical, continuous and count data. We will work in a small group so that you can optimally benefit from the workshop. After the workshop, you will be able to use multilevel hidden Markov models to extract personalised latent dynamics in your next project.

Examples of extracting personalised latent dynamics using the multilevel hidden Markov model can be found in e.g., this paper on bipolar disorder using continuous input data, this paper on nonverbal communication using categorical input data, this paper on neural spiking based behavioural event states using count data input, and the tutorial vignette and model estimation vignette of the R CRAN package mHMMbayes.