Brief summary:
The main idea of this article consists of employing the interaction into MIMIC model to detect uniform and nonuniform DIF simultaneously. The interaction term multiple the latent trait by the group variable in formula. Parameter estimation was conducted by Mplus and LR test for DIF analysis.
Simulation study was conducted in several conditions. The false-positive rate is higher in MIMIC-interaction model, but hit rate is the most high except the condition of ordinal items and nonuniform DIF. Estimates of mean difference is also estimated well and almost well in parameter estimation under MIMIC-interaction model.
Further study could be done on nominal indicators, assumption violations, and unequal variances suggested by author.
Questions:
1. The formula for interaction was written by mistake. βz is right?
2. What does it mean when both uniform and nonuniform DIF were detected?
3. LR test for DIF is restricted under several strict assumptions. Is it a popular method in empirical analysis?
4. LMS estimation seems not a good choice for MIMIC-interaction model, how about in the MCMC algorithm? To see if the false-positive rate could reduce or not.
5. MIMIC-interaction model can detect uniform and nonuniform DIF item rather than other method. But more assumptions are included in itslef at the same time? It seems hard for practician to choose among these methods.
Further ideas:
1. Anchor items approach can be employed in this study for completeness.
2. The condition of ordinal items and nonuniform DIF should be examined more specifically under MIMIC-interaction model. However, it is a trivial work.