|
Abstract:
|
The presence of missing data can create issues in statistical inference. Multiple imputation is a method of correcting for some of the issues, and multiple imputation of data with both fixed and random effects (or multilevel data) is an important area of recent statistical research. Joint model (JM) imputation and fully conditional specification (FCS) imputation are common multiple imputation methods, and although missing values in the design matrix for the fixed effects has been primarily addressed, missing values in the design matrix for the random effects has received much less concern. This can be considered as a case when the correlation structure is partially unknown and some recent work has investigated the ability of these methods to capture the covariance structure. We assess the performance of JM and FCS imputation models in this setting, and in particular when there is a large amount of missing data for the random effects. Real data examples will also be discussed to illustrate the applicability of missing data imputation methods for multilevel data.
|
