Analytic, Sensitivity and Graphical Methods for Investigating Dropout Data

DESCRIPTION (provided by applicant): Longitudinal data are very common in sociological, behavioral and biomedical researches. The data may come from longitudinal clinical trials, community surveys, family studies or spatial-temporal studies to investigate
some health outcomes. The responses are measured repeatedly over a period of time, and it could be either continuous or discrete. Typically, the interest focuses on the impact of some treatment intervention or the pattern of change in response over time. S
uch data could be very complex when there are multiple levels of data structures. In addition, it is often the case that there exists missing response in the data. In the analysis of longitudinal data, the missing data mechanisms have to be incorporated in
order to derive valid results. In the most severe case, the missing mechanism is not ignorable, i.e. one has to model simultaneously the observed and unobserved outcome variables and the missing indicator. On the other hand, those modeling assumptions are
often not testable, and one has to rely on the sensitivity analysis and graphical methods to study the robustness of the assumptions. We are interested in developing software that incorporates the analytic methods, sensitivity analysis and graphical metho
ds in one software. Such software is not available in the market yet. We will develop a user-friendly system with web and desktop applications. We will also develop algorithms and dynamic graphical methods for the analysis of dropout data and the diagnosis
of modeling assumptions. The software will be useful to biomedical researchers working on sociological, behavioral and biomedical studies with complex data structures. Manuscripts and course packs will be developed to assist practitioners in applying appr
opriate methods and tools in their studies. PUBLIC HEALTH RELEVANCE This project aims at statistical software for the analysis of complex longitudinal data with non-ignorable missing responses. The methods and software will be useful for biomedical studies
, e.g. longitudinal clinical trials. We will develop algorithms, analytic methods and dynamic graphical tools for model fitting, model diagnosis and justification of assumptions.