Differentiating Intermittent Missingness from Dropout in Dynamic Structural Equation Models

Intensive longitudinal data (ILD) are increasingly popular, but the dense nature of data collection often increases the prevalence of missing data. ILD have an increased risk for being missing not at random (people do not respond because of what their response would have been) and – unlike traditional panel data – a high incidence of intermittent missingness (people return to the study after missing a prompt). Some methods have been developed to address these situations, but limitations are that either they do not (a) differentiate intermittent missingness from dropout or (b) permit disaggregating effects into within- and between-person components. This paper discusses how to integrate missing data methods that differentiate types of missing data into the dynamic structural equation modeling framework, which naturally permits disaggregating effects, in addition to extensions for multivariate outcomes. An application to an intensive longitudinal intervention with a large percentage of missing data is provided.