Comparison of Listwise Deletion and Imputation Methods for Handling a Single Missing Response Value in a Central Composite Design

Abstract

This research aims to investigate appropriate methods for handling missing data during analysis, which is one of the most challenging tasks for statistical inference. Our motivation is to replace a missing response value in a central composite design (CCD) and its effect on each particular part (factorial, center and axial) in which the value is missing. Statistical software packages generally set listwise deletion as the default method for dealing with missing data, while imputation methods are also widely used. Hence, we compared listwise deletion and mean and regression imputation. Four test functions were used to examine all possible cases of a single missing response in a CCD with two factors. The performances of the methods for handling a missing response value in each of the three parts of the CCD (factorial, center, or axial) were compared in terms of their optimal responses with complete data by using mean-squared error and correlation coefficient values. Regression imputation and listwise deletion provided similar results for handling the missing value in each of the CCD parts (factorial, center, and axial) and were both superior to mean imputation.