An Investigation of the Effect of Missing Data on Differential Item Functioning in Mixed Type Tests

In this research, the aim was to examine the effects of Markov Chain Monte Carlo (MCMC), multiple imputation (MI), and expectation maximization (EM), all methods of coping with missing data in mixed type tests including dichotomous and polytomous items, on the differential item functioning (DIF). The study was carried out on a complete data set consisting of the scores of 1160 students who took booklet number 9 in the science test in Trends in International Mathematics and Science Study (TIMSS) 2019 and answered it in full. The conditions to be examined for the effectiveness of the methods were missing data mechanism (MCAR and MAR), DIF level (A, B, and C), and missing data rate (10% and 20%). Data were assigned to the missing data sets created by deleting data at different rates under the missing completely at random (MCAR) and missing at random (MAR) mechanisms over the aforementioned data set. DIF analysis was performed on all the data sets obtained with the poly-SIBTEST method using the MCMC, MI, and EM methods. The results obtained from the complete data set were then compared with the result implications from other data sets for reference. The study showed that the EM and MCMC methods performed better for the C-level DIF than the A and B levels in terms of all conditions examined. MI was observed to be the most successful method in determining DIF in items showing DIF in 10% and 20% MCAR mechanisms. When compared with the complete data set, the three methods showed similar results in the 10% MAR mechanism while MCMC gave the closest results in the 20% MAR mechanism.