Using HGLM As A New Approach For Cognitive Diagnosis Model DIF Analysis

This study attempts to apply HGLM model for cognitive diagnosis model DIF analysis, research is divided into two parts, the first part of the model DIF test how cognitive diagnosis first explored, developed by using the theoretical framework of the method; the second part compared it with MH, Wald methods, the researchers used five independent variables, including the matching variables, sample size, DIF type, DIF size, DIF proportion of questions, the main test indicators are type I errors and statistical power.Results of the study showed that:1, Statistical tests show HGLM under KS have the best effect,and HGLM model' power is superior to MH, slightly worse than the Wald method.2, The HGLM exist the expansion of type I error when using the total score and the ability to match the values.3, HGLM method in a non-uniform DIF case significantly better than the uniform DIF control type I error.4, Three methods are non-uniform DIF test when there was greater statistical power drops, MH method decreased the most.5, For type I errors and statistical power, there DIF size is the most important factor, followed by the sample size, and finally DIF questions proportions.HGLM method has advantages include model is not only based on DINA, can be subject to multiple DIF detection, can be applied to multi-level data structures, can be detected between the groups of the plurality of DIF. The disadvantage of this method lies in the use of more complicated, parameter estimation longer time, so if you need to do DIF test cognitive situation diagnosis, and the use of KS as a matching variable, or the presence of a multi-layer data structure, the need for more populations were DIF detection, this method is more suitable.