Methods of Dealing with Missing Data: Advantages, Disadvantages, Theoretical Approaches and Introduction of Applicable Software

In some cases, in data analysis, missingness happens in the observations for different reasons and ways. How to deal with these observations in the data analysis process is very important, especially in the high stack decisions. Previously, the prevalent method to deal with missing data problems was to eliminate them, which resulted in low quality data and consequently, biased results. Today, due to the advances in various fields of science and powerful statistical methods, it is possible to apply imputation methods to replace the missing data with suitable values before modeling the incomplete data. In this paper, the various types of missing data, imputation methods, the assumptions, their advantages and disadvantages are discussed. In this setting, we shall try to provide applied examples using statistical software. Especially an applied example due to 91th TOLIMO test, which was provided by Iranian National Organization of Educational Testing, has been presented (NOET). Comparison of results according to MSE using three methods of multiple imputation, the EM algorithm and the DA algorithm, indicated that the EM algorithm had the best performance for analyzing this data set.