Statistical Qualification of the Impact of Missing Data in EMA Studies

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): In response to the strong demand for developing appropriate analytic techniques for use with new kinds of data and new approaches to behavioral and social science research, we propose to develop principled and parsimonious statistical measures that are applicable in studies using intensive measurement methods, such as Ecological Momentary Assessment (EMA) methods, to quantify the reliability and validity of empirical findings to nonignorable missingness. Like any study involving human subjects, missing data are common in EMA studies. For example, when studying the question "Are moods just prior smoking different than moods during random background times", there can be a moderate amount of missing data because of study participants' nonresponses to those random prompts. It is often suspected that the missing data caused by such prompt nonresponses are nonrandom in that the prompt nonresponse behaviors are related to contemporaneous mood outcomes and consequently the observed data may be a selected nonrandom subset of a person's background mood even though the planned prompts are random. Such nonignorable missingness needs to be properly accounted for in the analysis of EMA data. However, unlike in more traditional studies, nonignorable missingness in intensive EMA data poses significant new analytic challenges and calls for more general, flexible and robust methods that are applicable in EMA studies to quantify and improve the reliability, validity and usability of the collected data. Thus, the aims of the proposed study are to (1) develop general, robust and tractable statistical measures and accessible software for assessing the impact of missing data on analysis of EMA data, and (2) examine the role of smoking on mood regulation in adolescents while accounting for the impact of nonrandom missingness, using data from our program project grant, "Social and Emotional Contexts of Adolescent Smoking Patterns" (NCI grant #PO1 2CA98262), which established a cohort of adolescents at high risk for the development of smoking and nicotine dependence. This study has the potential to make methodological and substantive contributions to EMA data analysis and understanding the relationship between mood variation and smoking dependence. The principled and simple statistical measures and accessible software to be developed will allow researchers to conveniently quantify the robustness of empirical findings from studies using EMA or other types of measurement-intensive methods to nonignorable missingness for a wide range of data types and models, missing data patterns and mechanisms. These methods can also easily generalize to a variety of cancer-relevant research areas, including studies using other types of new intensive measurements, such as mHealth (mobile heath) studies.