Modeling non-random missingness in experience sampling research on substance use

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): For over half a century, psychologists have theorized a link between daily emotion (e.g., stress, negative affect) and substance use. More recently, risk and protective factors including coping strategies and cognitive expectancies have been identified as potential moderators to the emotion-substance use relationship. Recent advances in multilevel statistical modeling techniques and experience sampling methodology (e.g., diary studies) have resulted in a flurry of research applications designed to test a variety of emotion-substance use relations at the intra-individual level. This important development allows a more nuanced understanding of the etiology of substance use that was not possible with inter-personal studies. However, diary studies may be especially prone to nonignorably missing data (i.e., the most troubling kind of missing data) for a number of reasons. First, the sensitive, and sometimes criminal, nature of the measures makes disclosure somewhat risky. Second, ecological assessments of substance use rely on self reports from intoxicated or "high" individuals. Nonignorable missingness leads to biased inferences regarding the relationship between emotion, substance use, and moderators. Recently, researchers in the area of clinical trials have utilized latent class pattern mixture models (LCPMMs) to obtain unbiased parameter estimates even in the presence of nonignorably missing data. LCPMMs have worked in this context by accounting for conditional dependencies between dropout patterns and outcome trajectories with latent class variables. Within-class estimates are aggregated to obtain unbiased overall estimates. While promising, LCPMMs have not yet been applied to experience sampling datasets. The proposed project has three specific aims. The first is to conduct a thorough review of the characteristic types and patterns of missingness in experience sampling datasets which examine substance use. The second aim is an extension of the LCPMM framework to accommodate these types and patterns of missing data. The final aim is to more rigorously test the self medication hypothesis by reanalyzing two datasets that were previously analyzed under questionable assumptions about the missing data mechanisms. PUBLIC HEALTH RELEVANCE: The proposed project will make unique substantive and quantitative contributions. Substantively, this project will reliably measure the effects that day-to-day emotional fluctuations have on substance use behaviors and the role of potential risk and protective factors in this process. This knowledge will reveal new ways to effectively design and implement interventions to prevent substance abuse. The substantive analysis will provide a vehicle for demonstrating and disseminating quantitative advances.