Comparison of five imputation methods in handling missing data in a continuous frequency table

Abstract

Missing data are sometimes inevitable that could affect the overall results of research. Sometimes missing data that occurs in data render the continuous frequency table incomplete, and hence the need to estimate them to arrive at valid results. Thus to estimate the missing data, it is appropriate to use one of the scientific imputation methods reported in the literature. This study aims to compare five different missing data imputation methods, mean imputation, median imputation, k nearest neighbors, sample imputation, and multiple imputations by using chained equations (MICE). The five imputation methods are compared using four real datasets. Nine different percentages of missingness are introduced completely at random into the datasets. The statistical metric, root-mean-squared error (RMSE), is used to assess the performance of the methods. Results show that the multiple imputations by using chained equations (MICE) outperformed the other imputation methods. The mean and k nearest neighbor (KNN) performed better relative to sample and median imputation methods. The five imputation methods’ performance is independent of the dataset and the percentage of missingness.