On Comparison of Some Imputation Techniques in Multivariate Data Analysis

Abstract

Listwise or pairwise deletion as the method of handling missing data in multivariate data leads to loss of statistical power, biased results and underestimation of standard errors and P-values.Four imputation techniques namely Regression, Stochastic, Expectation-Maximization (EM) and Multiple Imputation (MI) were considered and compared in terms of preserving the original distribution of the (multivariate) data and the relationships among the variables before the techniques were applied. Results show that none of the techniques performed absolutely better than the rest leaving the choice of imputation techniques in any dataset on the objectives of the researcher.

Keyword: Imputation, missing data, Expectation Maximization, Multiple Imputation, Root Mean Square Error.