Game of Phones: A Song of Daily-reports with Missing Observations and the Models that Analyze Them
W. Scott Comulada, Dr.P.H.
UCLA Department of Psychiatry & Biobehavioral Sciences
Cell-phone based ecological momentary assessment (CEMA) is an exciting new data collection method. Illicit drug use, sexual behavior, and other HIV-transmission behaviors are reported on a daily basis and in the moment through an electronic device that is already embedded in individuals’ daily routines. The benefits of EMA are realized without the burden of an additional data collection tool. With the capacity to track behaviors at great frequency comes an unwelcome companion of traditional longitudinal studies: missing data. In particular, missing data that is dependent on data not collected by the researcher is problematic, referred to as data not missing at random (NMAR). For example, individuals may not fill out scheduled CEMA on days they are using drugs. Standard statistical models do not adjust for NMAR data and lead to incorrect conclusions. In treatment settings, NMAR patterns overlap with engagement issues and are of interest in their own right. Yet NMAR models have been underutilized due in part to the inability to implement them in standard statistical software packages. Implementation issues have been addressed with the advent of JAGS and other readily available Bayesian software that accommodates NMAR models. A greater awareness of NMAR modeling options in the research community is needed. Towards this goal, I will give an introduction to missing data issues and NMAR models for CEMA data. JAGS code will be provided. Models will be illustrated on CEMA data from pilot studies on youth in outpatient drug treatment and HIV-positive adults. This is joint work with Robert Weiss.
W. Scott Comulada is an assistant professor-in-residence in the UCLA Department of Psychiatry and Biobehavioral Sciences and a member of the CHIPTS methods core. He is part of a cross-disciplinary team of behavioral interventionists, computer scientists, and statisticians that collaborate on mobile phone-based research projects in the behavioral sciences. Dr. Comulada is also a statistician with extensive experience in the analysis of longitudinal and social network data.