1. The study only investigated the condition when one DIF item exists. It is not realistic. When many DIF items are possible, purification or DIF-free-then-DIF approach can be readily applied with RFA model.
2. I’m confused with its setting described in page 564, “In conditions with dependent T and V, …”. Why it is called dependency by manipulating the means for each group?
3. It fixes many parameters for the convergence of RFA/LMS. But the results still poorly converged. Is it because the sample size or number of item is small?
4. It says nonuniform bias is more difficult to detect than uniform bias except T and V are correlated. But it seems the false positives is slightly inflated.