Abstract

It is obvious that the treatment of missing data has been an issue in statistics for some time now, and hence has started gaining the attention of researchers. This paper established the various methods usable in estimating missing values, determined which of the methods is the best in estimating missing values in one-way analysis of variance ANOVA , determined at which percentage level of Missingness is the method best and verified the effect of missing values on the statistical power and non-centrality parameters in one-way ANOVA. The methods examined are Pairwise Deletion PD , Mean Substitution MS , Regression Estimation RE , Multiple Imputation MI and Expectation Maximization EM . Mean Square Errors MSEs , that is variances of the methods were compared. It was found that MS had the least variance at 5, 10, 15, and 25 percent levels of Missingness while EM had the least variance at 20 percent Missingness level. PD method yielded the least statistical power at all the percentage levels of Missingness. Non-centrality parameters increased with increasing percentage level of Missingness and it was also found that at 25 percent level of Missingness after 20 percent , the statistical power started to reduce. EM method was recommended since MS yielded the least MSEs because of its limitations. Meanwhile PD should not be an option while dealing with missing data in one - way ANOVA due to loss of statistical power and possibly increased MSE. Eze, Francis Chkwuemeka | Chukwunenye, Victor Gozie "Comparing Methods of Estimating Missing Values in One-Way Analysis of Variance" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-3 | Issue-2 , February 2019, URL: https://www.ijtsrd.com/papers/ijtsrd18599.pdf Paper URL: https://www.ijtsrd.com/mathemetics/statistics/18599/comparing-methods-of-estimating-missing-values-in-one-way-analysis-of-variance/eze-francis-chkwuemeka

Authors and Affiliations