Latent Trait Pattern-Mixture Mixed-Models for Ecological Momentary Assessment Data

Latent trait pattern-mixture mixed-models (LTPMMM) for Ecological Momentary Assessment (EMA) data are developed in which data are collected in a intermittent fashion. Initial work with intermittent data has used latent class pattern-mixture models. Using Item Response Theory (IRT) models, a latent trait is used to model the missingness mechanism and modeled jointly with a mixed-model for longitudinal outcomes. Both one- and two-parameter LTPMMMs are presented. These new pattern-mixture models offer a unique way to analyze EMA data with many unique response patterns that cannot easily formed into latent classes. Data from an EMA study involving high-school students' positive and negative affect are presented. The proposed models will estimate a latent trait that corresponds to the students' “ability" to respond to the prompting device. One-thousand simulations are performed to test the proposed models across different simulation scenarios. The models are compared to a Missing at Random (MAR) mixed-model and a latent class pattern-mixture model. The proposed models have lower bias and increased efficiency compared to standard approaches. The new models offer a viable alternative to latent class pattern-mixture models previously used with intermittent missing data.