When estimating the value of theta (ability) from an IRT model using the Expected A Posteriori method, how does Mplus handle records with missing values for 1 or more items? Does it simply use the remaining non-missing items? Is there a proportion for missing items above which it does not make sense to calculate theta (e.g., 50%)? In case it is important, I am using a 2-parameter logistic IRT model with binary items.
Q1. The parameter estimation uses all available data with ML assuming MAR. The ability (factor score) estimation uses the subject's non-missing items.
Q2. That's difficult to say but you do get a SE for the factor score estimate for the person so that can tell you about lack of precision.
Weizi Wang posted on Monday, June 12, 2017 - 4:04 am
Hi, I am wondering how to estimate the theta (ability) using Mplus. Is it the factor score output by the program the estimate of the theta? If so, I am wondering how to output all the factor scores since they are not outputted completely by default.