Clustering-Imputation on Partially Observed Ecological Momentary Data

• Activity Number: 423
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Social Statistics Section
• Author(s): Xiaoxue Li*+ and Stewart Anderson and Saul Shiffman
• Companies: University of Pittsburgh and University of Pittsburgh and University of Pittsburgh
• Keywords: Ecological momentary assessment ; imputation ; partially observed covariates ; clustering

Abstract

Ecological momentary assessment (EMA) refers to studies where subjects are measured or recorded instantaneously over time. We investigated one such study involving the instantaneous assessments of behavior, mood, symptoms or activities over time in subjects who were either daily smokers (DS) or intermittent smokers (ITS) at times they either experienced an "smoking" event or are randomly prompted at a time when no "event" happens. In our study, each successive event is assessed according to a probability based on the number of smoking events that were recorded the day before. This probability scheme complicates the analysis of EMA data. Moreover, cigarettes reported tend to be clustered in time among ITS group only. In current analyses, we assume that each assessed event provides an unbiased sample of all events and hence, we weighted each observation by the inverse probability of assessment. However, this assumption may not be true, and even if it is true, the current analyses ignore the fact that cigarettes are clustered in time among ITS. Thus, we propose a way to impute the non-assessed observations that utilized the fact that ITS smoke in bouts.