Journal 1: Get 4 reviewers instead of customary 2-3 ('naughty topic'). Rejected. Journal 2: This paper belongs to Journal 1, submit there. Rejected. Seems legit.
-
I'll have it sent to you. But TL;DR version is that item-level MCV analyses requires a few corrections to the method used for test-level data, and if one doesn't apply these, one gets uninterpretable results depending on relationship between difficulty and loading.
We applied the corrections as a test case to a large vocabulary dataset (75 items) with fertility data. Found that the usual results hold: selection appears to be on general factor variance.