Introduction to Multiple Imputation With a Principle Component for Nonnormal Categorical Data

• In Event: Poster Session for Quantitative Methods and Statistical Theory

Abstract

Multiple imputation (MI) is widely used technique to handle missing data, but it is often cumbersome when many variables are involved in the dataset. Using a single principle component in MI procedure as auxiliary variable (i.e., PCA) instead of all related variables has been suggested as an efficient alternative process. The purpose of the current study was to examine the performance of the PCA approach when the incomplete variable is non-normal categorical. The result of A Monte Carlo simulation showed that the PCA approach performs similarly or outperforms the conventional all-inclusive approach in general. Specifically, the PCA approach provided comparably unbiased estimate when the incomplete variable was not symmetrically distributed and the sample size was small.

Author

• Youngmin Kim, Texas Tech University