A continuous-time mixture latent-state-trait Markov model for experience sampling data: Application and evaluation

In psychological research, statistical models of latent state-trait (LST) theory are popular for the analysis of longitudinal data. We identify several limitations of available models when applied to intensive longitudinal data with categorical observed and latent variables and inter-and intraindividually varying time intervals. As an extension of available LST models for categorical data, we describe a general mixed continuous-time LST model that is suitable for intensive longitudinal data with unobserved heterogeneity and individually varying time intervals. This model is illustrated by an application to momentary mood data that were collected in an experience sampling study (N = 164). In addition, the results of a simulation study are reported that was conducted to find out (a) the minimal data requirements with respect to sample size and number of occasions, and (b) how strong the bias is if the continuous-time structure is ignored. The empirical application revealed two classes for which the transition pattern and effects of time-varying covariates differ. In the simulation study, only small differences between the continuous-time model and its discrete-time counterpart emerged. Sample sizes N = 100 and larger in combination with six or more occasions of measurement tended to produce reliable estimation results. Implications of the models for future research are discussed.