Abstract:
The objective of this study is to compare the estimation methods for the multiple linear regression model with nonignorable-missing dependent and independent variables. The estimation methods considered in study are EM Algorithm (EM) , K-Nearest Neighbor (KNN) and Predictive Mean Matching (PMM). Data are simulated with three levels of missing proportion of data of 10%, 20%, 30% and three levels of nonignorable missingness of none, medium, high. The average mean square errors (AMSEs) of all methods are compared with the best method will have the smallest value of AMSE. The findings are the followings : i) KNN method performs best when the standard deviation of error is medium and high (30 and 90), ii) EM method performs best especially when the standard deviation of error is small (10), iii) The performances of all estimation methods perform decrease as the standard deviation of errors, the missing proportion, or level of nonignorable missingness increase.