Analysis of Incomplete Longitudinal Binary Data-A Combined Markov's Transition and Logistic Model for Non-ignorable Missingness.

The problem of incomplete data is a common phenomenon in research that involves the longitudinal design approach. We investigate and develop a likelihood-based approach for incomplete longitudinal binary data using the disposition model when the missing value mechanism is non-ignorable. We combined Markov's transition and a logistic regression model to build the dropout process and model the response using conditional logistic regression model. By holding the missingness parameter that is weakly identified constant, we analyzed their effects through a sensitivity analysis as the estimation of parameters in MLE for non-ignorable missing data is not generally plausible. An application of our approach to Schizophrenia clinical trial is presented.