Examination of the Effects of Different Missing Data Techniques on Item Parameters Obtained by CTT and IRT
The purpose of this study is to examine the effect of different missing data techniques on the item parameters estimated for Classical Test Theory (CTT) and Item Response Theory (IRT) comparatively through simulated and real data sets. For this purpose, data sets with missing completely at random pattern with different sample sizes and missing data rates have been generated. Item parameters based on CTT and IRT are estimated after applying different missing data techniques (listwise deletion, regression imputation and expectation – maximization). Estimated parameters were compared with the parameters of complete data sets. The same procedure was performed on the data of the PM3 subtest in the PISA 2012 application. It is found that higher difficulty and lower discrimination values are obtained with listwise deletion while accurate item difficulties can be obtained with regression imputation and expectation - maximization algorithm, but the same situation is not valid for item discriminations. It is also seen that missing data at a ratio of %2 can lead to serious problems even if they have missing completely at random pattern.