06-21-2017 04:26 PM
It's well known that when using maximum likelihood estimation for item parameters or for examinee ability, parameter estimates for items and/or examinees that have no variance (ie., all responses are 1 or 0) are not obtainable.
Bayesian estimation addresses this problem, enabling parameter estimation for such items/examinees. Such estimation is permitted in commercial IRT software, such as BilogMG.
How can this issue be addressed with the SAS Proc IRT procedure? Is there an estimator that handles this, or do I have to exclude items and/or examinees without variance?