In longitudinal studies, subjects may be lost to follow-up and present incomplete response sequences. When the mechanism leading to exit the study is nonignorable, we need to account for potential dependence between the longitudinal and the dropout process. The nonignorable missing data model should have, at least, two major features. First, it should (simply) reduce to an ignorable missing data model, when some conditions are met. Second, this nested structure should give the way to measure sensitivity of model parameter estimates to assumptions about nonignorability. In fact, it is well known and commonly acquired that every nonignorable missing data model has an ignorable counterpart with the same fit to the observed data. Therefore, we may not distinguish between the two, but rather we should be interested in the impact that such assumptions have on parameter estimates. We review random coefficient based dropout models and discuss pros and cons of potential specifications. We also propose a finite mixture approach that satisfy both the features we have described above. To conclude, we review measures of global/local sensitivity and use the finite mixture approach to discuss their main features. The proposal is detailed using a dataset on longevity and cognitive functioning.