Examining the effect of missing data methods on variance analysis (t-Test, ANOVA) parameters under different variables

In the process of scientific researches there could potentially be missing data. It is well known that these missing data can have negative effects to the results of the researches. It is a big impediment for the researchers to have exact results. Process of missing value analysis applications include approachies aiming to find solution to these problems experienced by the researchers. Firstly researchers should find out the factors resulting missing data and define severity of the missing. In this research, missing value applications were compared using value sets produced in different numbers of unit. These set of values were produced in a way that they would have normal distributions in high and low correlation groups having respectively 50, 100, 200, 400 units. Under random conditions, reduced set of values having respectively %5, %10, %20 losses are in the form of Missing Completely at Random (MCAR). In the value sets produced, mean substitution, regression method, expectation-maximization (EM) method were applied, instead of deletion methods. Difference between the results of methods differentiated dramatically in value sets of different volume and different correlations. It is observed that in value sets with low numbers of units (50 unit, 100 units) regression and EM application are usefull on the other hand in value sets with high numbers of units (200 units, 400 units) mean substitution instead of regression method have more consistent results. Also It is observed that in value sets with low correlation deletion method was not an efficent application. Key Words: Missing Value Analysis, Imputation Method, EM, Mean Method, Regression