The MIMIC Model As A Method For Detecting DIF: Comparison With Mantel-Haenszel And Logistic Regression

Now, more and more experts in domestic and overseas focus on the problem of fairness in psychological and educational testing. So, the study of DIF has been the hottest topic in psychological and educational domain.In this paper, we will discuss three questions. Firstly, we will introduce the development and application of DIF (Differential Item Functioning) theory. Secondly, we also introduce three DIF(Differential Item Functioning) detecting methods—MH Method, Logistic Regression Method, MIMIC Method using in this paper. Finally, A simulation study is used in this study, through the power and I type of error rate, comparing the three DIF (Differential Item Functioning) detecting methods.In this paper, with manipulation of 2PLM model, the same number of examinees in focal and reference groups, the same mean abilities between the reference and focal groups, levels of DIF differences between the reference and focal groups, comparing the three DIF (Differential Item Functioning) detecting methods. Results indicate:1,The DIF detection methods have a signficant impact on the type of I rate. In three methods, The MIMIC method have the lowest the type of I rate. The Logistic regression method have the moderate the type of I rate. The Mantel-Haensze method have the highest the type of I rate.2,As for the Mantel-Haenszel method, when the DIF value is 0.55, the power of detection is the highest, when the DIF value is 0.85, the power of detection is in the middle, when the DIF value is 0.25,the power of detection is the lowest.3,As for the Logistic regression method, with the increasing of DIF value, the power of detection is becoming increase. When the DIF value is 0.85, the power of detection is the highest, when the DIF value is 0.55, the power of detection is in the middle, when the DIF value is 0.25, the power of detection is the lowest.4,As for the MIMIC method, when the DIF value is 0.55, the power of detection is the highest; when the DIF value is 0.85, the power of detection is in the middle; when the DIF value is 0.25, the power of detection is the lowest. 5,When the DIF value is 0.25, the power of detection of MIMIC is the highest ,the I type of rate is the lowest; the power of detection of Logistic regression is in the middle, the I type of rate is in the middle; the power of detection of Mantel-Haenszel is the lowest, the I type of rate is the highest.6,When the DIF value is 0.55, as for Logistic regression and MIMIC, the power of detection is almost same, comparison of the Mantel-Haenszel, the power of detection slightly lower. But, the I type of rate of Logistic regression is the highest, secondly is Mantel-Haenszel, the I type rate of MIMIC is the lowest.7,When the DIF value is 0.85, as for Logistic regression and MIMIC, the power of detection is almost same, comparison of the Mantel-Haenszel, the power of detection slightly lower. But, the I type of rate of Logistic regression is the highest, secondly is Mantel-Haenszel, the I type rate of MIMIC is the lowest.