A Comparison of Estimation Methods for Missing Data in Multiple Linear Regression with Two Independent Variables

In multiple linear regression, if the incomplete values occur in sample, many researchers will use the statistical software packages which delete the cases that deal with missing value. Then they use the ordinary least square (OLS) to estimate parameters in the regression model. However, the validity of this approach is decreased in making inference because the size of the sample reduces. In this paper, when missing values occur in response variable, we perform a simulation study of imputation based procedures and indicate that missing values should be filled by the EM algorithm and regression imputation methods, which have been more efficient than deletion of observations. However, in case of having the large sample sizes and small variances of errors and small proportions of missing values, pairwise deletion methods can be used efficiently. Further, we analyze completed data by ordinary least square method to predict regression coefficients in the regression model. This leads to a result with minimum mean squared errors.