Abstract:
|
Many traditional longitudinal models treat variance components as nuisance factors. However, in many applications of intensive longitudinal data, variability of predictors within subjects is important in modeling the outcome of interest. Here, we present a combination of multivariate longitudinal modeling, joint modeling of mean and variation, and time-lagged intensive longitudinal methods to assess associations between outcomes and predictor variation. A joint mixed bivariate model will be presented, using both the outcome of interest and a predictor as dependent variables, and including as independent variables measures of variation from the dispersion model. As an example, analysis will be presented for an analysis of feedback-loop associations between marijuana use and craving using data from an Ecological Momentary Assessment study. We will show that both marijuana use, and craving are associated with each other at subsequent times of observation.
|