Studies frequently have missing values for both the dependent and independent variables. Multiple imputation (MI) techniques are approaches to replacing missing values so that complete-data analysis methods may be employed. Often, mixed models are used to analyze data with missing independent variable values, but in mixed models, having a large number of independent variables with missing values will exclude a significant number of data vectors from the analysis, creating a major problem when attempting to draw meaningful conclusions from an analysis. Our goal was to use multiple imputation techniques to impute missing values for independent variables in a longitudinal study setting, and then to use the complete data matrix to fit mixed models independently for each of the imputed datasets. In this paper, we present a method to combine multiple estimates and inferential statistics generated from multiply imputed datasets using the same mixed model, which has not previously been done. Additionally, we are incorporated two sets of variance-covariance matrices for each imputed set and also adjusted degrees of freedom. In the example we compared estimates using complete and multiply imputed data using mixed models. We conclude that in some situations it is desirable to use multiple imputation techniques and mixed models together to draw conclusions.
Keywords: Multiple imputations, mixed models, missing values, vectors