7.1 Introduction

Multiple imputation of multilevel data is one of the hot spots in statistical technology. Imputers and analysts now have a bewildering array of options for imputing missing values in multilevel data. This chapter summarizes the state of the art, and formulates advice and guidelines for practical application of multilevel imputation.

The structure of this chapter is as follows. We start with a concise overview of three ways to formulate the multilevel model. Section 7.3 reviews several non-imputation approaches for dealing with missing values in multilevel data. Sections 7.4 and 7.5 describe imputation using the joint modeling and fully conditional specification frameworks. Sections 7.6 and 7.7 review current procedures for imputation under multilevel models with continuous and discrete outcomes, respectively. Section 7.8 deals with missing data in the level-2 predictors, and Section 7.9 summarizes comparative work on the different approaches. Section 7.10 contains worked examples that illustrate how imputations can be generated in mice, provides guidelines on the practical application, written in the form of recipes for multilevel imputation. The chapter closes with an overview of unresolved issues and topics for further research.