Online Program

Return to main conference page
Wednesday, January 8
Wed, Jan 8, 8:30 AM - 10:15 AM
Intensive Longitudinal Data

A Shared-Parameter Location-Scale Mixed Model for Non-ignorable Nonresponses in Self-Initiated Event-Contingent Assessments in Ecological Momentary Assessment Data (307796)


Donald Hedeker, The University of Chicago 
*Qianheng Ma, The University of Chicago 
Robin J Mermelstein, University of Illinois at Chicago 

Keywords: Intensive longitudinal data, EMA, self-initiated assessments, event-contingent assessments, nonignorable missingness, shared parameter model, location scale mixed model, variance modelling

In this presentation, we address the issue of non-ignorable nonresponses in the self-initiated assessments in Ecological Momentary Assessment (EMA) studies as well as other mobile health (mHealth) studies. In such studies, subjects are instructed to self-initiate reports on their mood and the environment when they are experiencing events such as behavioral lapses. However, the occurrence of nonresponses in self-initiated event reports is usually unknown and not at random so that existing methods for missing data can be problematic. In particular, nonresponses can be associated with mood, the primary longitudinal outcome, in terms of mood level (location) and stability (scale). Fortunately, in some EMA studies, there is some information about missing reports. Specifically, we propose a shared-parameter mixed model that links the primary outcome model (for mood) and a model for subjects’ responsivity (for missing event reports) by shared random effects, characterizing a subject’s mood location and scale. Compared to more conventional methods via simulations and real data analysis, the proposed model was found to improve the coverage of parameters and provide better fit to the data.