Dynamic random coefficient based drop-out models for longitudinal responses

We propose a dynamic random coefficient based drop-out model for the analysis of longitudinal data subject to potentially non-ignorable drop-out. The presence of a non-ignorable missingess may severely bias inference on the observed data. In this framework, random coefficient based drop-out models represent an flexible approach to jointly model both longitudinal responses and missingess. We extend such an approach by allowing the random parameters in the longitudinal data process to evolve over time according to a non-homogeneous hidden Markov chain. The resulting model offers great flexibility and allows us to efficiently describe both between-outcome and within-outcome dependence.