The Effect of Missing Data Handling Methods and its Volume on Detecting Differential Item Functioning of Test Items

Abstract

This study aimed to investigate the effect different methods for handling missing data and its volume on detecting differential item functioning (DIF) for items of test according to the gender variable using the generalized Mantel Haenszel test (GMH). To achieve the aims of the study, the responses of Jordanian eighth grade students to the items of the mathematics test in the International Study (TIMSS, 2015) were obtained. The sample of the study consisted of 2261 students, who responded to four test booklets (1, 7, 9, 13). Several methods were used for handling missing data (missing data as nominal, corrected item mean substitution, expectation- maximization algorithm, multiple imputation, and chained equations), GMH was also used to detect DIF depending on the gender variable after each treatment method. The effect of the volume of missing data (5%, 10%) on the percentages of questions that showed DIF was also shown. The results indicated that there were no statistically significant differences at (α = 0.05) between the ratios of items that showed DIF across different missing data processing methods. In contrast, the five methods agreed on the existence of a statistically significant relationship at (α= 0.05) between the presence of DIF in favor of females in the field of algebra and in favor of males in the field of numbers. As for the volume of missing data, the findings revealed that there were no differences in the percentage of questions that showed DIF in favor of males or females due to the volume of missing data of 5% or 10%.

Key words: missing data, Generalized Mantel Haenszel Test, (TIMSS, 2015), Differential Item Functioning, Missing Data Handling Methods.