332 – Recent Advances in Analysis with Missing Data
Imputation for Non-Normal Multivariate Continuous Data Using Copula Transformation
Zhixin Lun
Oakland University
Dealing with missing data problems for skewed data is one of the most difficult tasks in imputation since most of data augmentation methodologies assume multivariate normality. The performance of imputation and the accuracy of parameters inference become questionable when the violation of above assumption occurs. One approach to solve the normality violation is to apply normalizing transformation prior to the imputation phase. However, this approach may introduce new problems such as altering dependence structure among random variables. This article describes the multiple imputation approach based on the Copula transformation, which we use to effectively transform multivariate non-normal data into normal. We compare the performance of the Copula transformation method with traditional normality-based multiple imputation approaches through real non-normal multivariate datasets. We demonstrate that our approach significantly mitigates the impact of blind assumption of multivariate normality for the non-normal multivariate data under the scenario when the data are missing completely at random (MCAR).