Summary We introduce a selection model-based multilevel imputation approach to be used within the fully conditional specification framework for multiple imputation. Concretely, we apply a censored bivariate probit model to describe binary variables assumed to be missing not at random. The first...