CFE 2019: Start Registration
View Submission - CMStatistics
Title: Bayesian nonparametric methods for longitudinal outcomes missing not at random Authors:  Antonio Linero - Florida State University (United States) [presenting]
Michael Daniels - University of Florida (United States)
Abstract: The setting of a longitudinal outcome subject to nonignorable missingness is considered. This requires the specification $f(y,r)$ for the joint model of the response $y$ and missing data indicators $r$. We argue that the obvious Bayesian nonparametric approaches to joint modeling which have been applied in the literature run afoul of the inherent identifiability issues with nonignorable missingness, leading to posteriors with dubious theoretical behavior and producing questionable inferences. As an alternative, we propose an indirect specification of a prior on the observed data generating mechanism $f(y_{obs}, r)$, which is fully identified given the data. This prior is then used in conjunction with an identifying restriction to conduct inference. Advantages of this approach include a flexible modeling framework, access to simple computational methods, flexibility in the choice of ``anchoring'' assumptions, strong theoretical support, straightforward sensitivity analysis, and applicability to non-monotone missingness.