Key note 2: Stef van Buuren
Title: Recipes for multilevel imputation
Multiple imputation of multilevel data is one of the hot spots in statistical technology. In this lecture I will survey the state of the art in the field, bringing together advances from various active research groups.
Accepted methods for independent data break down for multilevel data since these do not preserve the proper group structure. In order to impute multilevel data, we need to know the form of the complete-data multilevel model, the size and number of clusters, the reasons for the missingness, and match the imputation model to the complete-data model.
I will summarize the steps needed for multilevel imputation as two recipes, one recipe for incomplete level-1 variables, and another recipe for incomplete level-2 variables. The recipes can be prepared with the help of the mice, mitml, micemd and miceadds packages in R.