Message/Author 

ljc posted on Tuesday, May 14, 2013  7:17 pm



1. Can you do an IRT/IFA model with only 1, 2 or 3 Likert scale items. What is the minimum number of items required? Is it like CFA in which you use the concepts of just identified/over identified to determine if you have enough items? 2. What is the formula used for outputted thetas using SAVE = FSCORES? Thank you. 


1. The rules are the same as for CFA. You need three factor indicators for a justidentified model. With two factor indicators, the model is not identified unless it borrows information from another factor. 2. See Technical Appendix 11. 

ljc posted on Wednesday, May 15, 2013  1:48 pm



I thought it might be different since 1 Likert item with 5 levels, actually goes into the model as 4 dichotomous items and error variances aren't estimated. I'll stick with a minimum of 3. Thank you. 


I am assuming each Likert item is one variable. This is how I would treat them. 

ljc posted on Wednesday, May 15, 2013  4:23 pm



I thought that was what the CATEGORICAL ARE command did, made the one 5 level Likert item go in as 4 items, so you get 4 intercepts/thresholds and 1 loading for each Likert item. While excluding the CATEGORICAL ARE command would make it CFA so the Likert item goes in as 1 item giving you 1 intercept and 1 loading. 


It does not create four dummy variables. Each Likert item gets one factor loading and the number of categories minus one thresholds. 

Back to top 