| This dissertation aims to develop imputation techniques to handle intermittently missing data in longitudinal studies. A new mean imputation method is proposed and compared to different existing techniques for non-monotone missingness which generally does not satisfy the assumption missing at random.; Under the proposed method, missing values are imputed through nonparametric models. Mean estimation and parameter estimators obtained by Generalized Estimation Equation (GEE) treating imputed values as observations are proven to be consistent. Simulation studies demonstrate the robustness of the proposed imputation method against underlying distributions.; The proposed imputation method is applied to data from the Wisconsin Diabetes Registry Project and the Current Employment Statistics (CES) database. |
