Dependence and sensitivity in regression models for longitudinal responses subject to dropout

In longitudinal studies, subjects may be lost to follow-up and present incomplete response sequences. When the mechanism that leads to exit the study is non ignorable, a possible route is to define a model that accounts for potential dependence between the longitudinal and the dropout process. This model should have, at least, two major features: (i) it should (simply) reduce to an ignorable missing data model, when some conditions are met; (ii) the nested structure should give the way to measure sensitivity of parameter estimates to assumptions on non ignorability. In this work, we discuss random coefficient based dropout models and review measures of local sensitivity