Abstract:
|
In this paper, we are interested in nonignorable missing data mechanism where the probability of nonresponse depends on the outcome. We consider a selection model for nonignorable nonresponse in logistic regression. Expressions for the bias in parameter estimates are derived in a simple case. Further, we propose a sensitivity analysis to study changes in parameter estimates under different assumptions. We adopt a Bayesian framework as it offers a flexible approach for incorporating different missing data mechanisms. Our modelling strategy is illustrated using survey data from the 45 and Up Study.
|