Estimating Latent State-Trait Models for Experience-Sampling Data in R with the lsttheory package: a Tutorial

As the popularity of the experience-sampling methodology rises, there is a growing need for suitable analytical procedures. These studies often aim to separate fleeting situation-specific influences from more enduring ones. Latent state-trait (LST) models can make this differentiation. This tutorial discusses multiple-indicator wide-format LST models suitable for experience-sampling data. We outline second-order and first-order model specifications, their advantages and disadvantages, and make the assumptions of first-order specifications explicit. These LST models are very flexible, allowing for various different models and for testing invariance assumptions. However, their specification is tedious and error-prone. This tutorial introduces a new user-friendly browser app and R-function for experience sampling models in the R-package lsttheory. Extending on existing models, the software also allows to add covariates, which can further explain the stable components. Throughout the tutorial, we answer exemplary research questions about well-being in everyday life with data from a five-day experience-sampling study. An autoregressive model with indicator-specific traits was most appropriate for the data and revealed relatively high consistency, implying that well-being depends more strongly on the person than the current situation. Of the Big Five, extraversion, emotional stability and agreeableness are predictive of trait well-being. We conclude with recommendations about model fit and comparisons. \\ A version of this manuscript is in production at Multivariate Behavioral Research, with doi 10.1080/00273171.2025.2454904