Multiple Imputation in Longitudinal Path Modeling Using WinBUGS

• In Event: Issues in Clustered Data Structures

Abstract

This paper presented a practical multiple imputation method under the general Gibbs sampling framework to handle missing data in longitudinal datasets. Based on a path causal model with MAR assumption, the proposed method imputes missing longitudinal data through innovative program WinBUGS. Analysis was conducted on the imputed data obtained by proposed approach and the parameters were recovered well with smaller standard deviations than the other two currently available approaches being examined. The new approach conducting multiple imputation in causal path analysis can handle missing data with a wide range of distributions in the longitudinal study with unbalanced design.

Authors

• Qiu Wang, Syracuse University

• Yi Pan, University of North Carolina - Chapel Hill