Non-linear terms

Evaluation of approaches for accommodating interactions and non-linear terms in multiple imputation of incomplete three-level data

Three-level data is common in medical research, as is missing. While multiple imputation (MI) is widely used to handle missing data in such studies, its validity depends on the appropriate tailoring of the imputation model to the substantive analysis. This means all the key features of the substantive analysis such as non-linear relationships, interactions and multilevel features should be appropriately accommodated in the imputation process. This paper evaluates a number of MI approaches that may be used for imputing three-level data when the substantive analysis model contains interactions and non-linear terms using both a simulation and a case study.