84 Dealing with omitted and not-reached items in competence tests: Evaluating approaches accounting for missing responses in item response theory models (Present by Kuan-Yu on 25 Nov 2013)

cw's comments

cw's comments

by LIU CHEN WEI -
Number of replies: 0

1. The author claimed the correlation about 0.2 is significant, but not convincing me.

2. Also, they claimed the latent modeling is more effective than manifest one. So, can we claim the manifest one is more effective if we used another set of manifest variables? It is a weird conclusion.

3. I don't think IRT model can be 『robust' when there is a high correlation, say above 0.5.

4. I remember there is some study claims that the MAR and MNAR is indistinguishable in statistical modeling. Thus, the underlying missing mechanism is always hard to sort out solely depending on observable responses.