The main idea is the DIF item can be contributed to the nuisance second dimension of latent trait. It could result from using some external strategies not related to the primary construct. It argued that the nuisance dimension may only exist for those people in the DIF latent class. The evidence was proposed by using model-data fit indexes to find out the most appropriate model. The most suitable model can be used as an intermediate to explain the source of DIF, thus it is an exploratory strategy. Finally two real data sets were analyzed and fit by several models.
Qs:
1. How large value of the discrimination parameter should be regarded as an evidence of DIF?
2. In speed case, it is somewhat weird that speediness is one kind of DIF. It is intuitive that time limit would lead to high guessing behavior during the test. It looks like the source is not resulted from the unfairness of the item for the two groups. How can we attribute it to speediness? It is none of DIF’s business.
3. The proposed model can only be used to explain the possible source of DIF rather than detecting DIF items. It may be not useful in empirical setting. We can’t still tell what it is.
4. DIF free then DIF may a prerequisite to be done first. It is still a challenge to pick up the good anchor item to locate possible DIF items.
5. Although the author’s idea is great, I can feel he always tries to induce me to believe his reasons. It is hard to believe that the speediness is related to DIF. When I go back to check the definition of DIF, they seems no matched in definition.