This paper proposed the MIMIC-interaction model to detect uniform and nonuniform DIF simultaneously. First, the rationale of MIMIC-interaction model was described by introducing the z (group) and theta, as well as parameter estimation methods based on the practical consideration of software. Then, the authors compared MIMIC-interaction, MIMIC without interaction and IRT-LR method for a various simulation conditions and found out that the proposed method having the good Power on DIF detection but inflated Type I error. Also, an empirical example was illustrated. At last, the author revisit the estimation methods for MIMIC-interaction procedure and provided some considerations on future works.
The estimation method XWITH is still not convincing though the author claimed to the practical issues and some correction can be used to control the inflation of Type I error.
At the same time, I notices that from the former 3 weeks, all three papers were all proposed correction methods to control simulation results. I was curious about the mechanic of correction and why they should always after the unsatisfied results were generated.
The anchor items should be chosen using purification procedure if we want a more realistic generalization use of MIMIC-interaction.