HavilahRenee posted on Tuesday, December 09, 2014 - 11:06 am
Hello, I am testing for uni-dimensionality on IRT data. Most of the items are MC (0/1) and a few are CR (0/1/2/3/4). I have a couple of questions:
1. I specified the data as categorical. Should this type of data be specified as categorical? I ask because I have seen researcher using Mplus with a similar goal. Sometimes the data are specified as categorical and other times not. What is the determining factor?
2. After specifying the data as categorical, I used the MLR estimator because my data are sparse. I just want to verify MLR is indeed a full information estimator.
1. The variables are not continuous. They are categorical. The correct specification is CATEGORICAL. When the split is close to 50/50, how you treat them may not make a big difference. See the Tables starting with Slide 136 of the Topic 2 course handout on the website for further information.
2. The default for MLR is to use all available information with FIML.
HavilahRenee posted on Tuesday, December 09, 2014 - 12:39 pm
Thank you for the rapid response. One follow up question....what if students have a likelihood of getting a test question correct based on an underlying ability estimate that is continuous? Must the data still be treated as categorical or is continuous reasonable?