|
![](http://www.statmodel.com/discussion/icons/csc/1px.gif) |
ML vs. MLR for IRT models |
![Previous](http://www.statmodel.com/discussion/icons/csc/navgen_arrow_left.gif) ![](http://www.statmodel.com/discussion/icons/csc/1px.gif)
|
Message/Author |
|
|
What is the difference between ML and MLR for IRT models (i.e., assuming categorical data and full information estimation)? For continuous data, I understand that MLR is supposed to help adjust SEs and test statistics for non-normality, but if normality is not assumed for categorical data to begin with, then what does MLR have to offer over ML? I apologize if this topic is already addressed elsewhere, but I could not find it. Thanks in advance for any direction you can provide! |
|
|
We know of no large study of this. MLR may be less sensitive to small degrees of model misspecification. |
|
Back to top |
|
|