Modifying Conditional Probability in E-step

4 views
Skip to first unread message

smi...@rowan.edu

unread,
Oct 6, 2020, 10:54:28 AM10/6/20
to mirt-package
Greetings, Phil (and others).

I am interested in modifying the expression for conditional probability in the E-step when calculating the updated r and N values, and I am wondering if this is possible in mirt. My situation is that I have a test with a series of nominal items, followed by a group of dichotomous items (for which I'm using 2pl). I would like the group of dichotomous items to have as much influence during the E-step as each of the individual nominal items. I think I could accomplish this by modifying the conditional probability product such that the terms for the dichotomous items are raised to a power equal to the inverse of the number of items (e.g., raised to the 1/5 if there are 5 dichotomous items). Is this possible in mirt? If it matters, I'm using a unidimensional model.

I've been searching the literature, as well as this group, but I haven't been able to find anything similar to this that treats different items differently when expressing the conditional probability. Thanks so much for any information or advice!

Best,
Trevor
Reply all
Reply to author
Forward
0 new messages