

Scaling method for categorical indica... 

Message/Author 

Ai Ye posted on Thursday, February 08, 2018  12:31 pm



Hi there, I have a question of specifying a CFA model, hope to get some help. If we are fitting a CFA model using categorical indicator items, we know that in general there are two choices one can make in scaling the underlying variables. One is to set the first threshold of each indicator to zero and freely estimate the intercept of the equation that links the underlying variable (of each indicator) with the latent factor; the other is to set the intercepts to zero and freely estimate the thresholds. What seemed to be the default in Mplus is the latter. In addition, the mean of the latent factor is set to zero by default. But I wondering if there is a way to use the other scaling method, and to estimate the mean of the latent factor as well as the intercept of categorical indicators. I was able to do this in lavaan by using the following constraints: fix the first threshold of each indicator at zero and freely estimate the remaining thresholds, fix the intercept of the scaling item at zero and freely estimate the intercepts of all nonscaling items. However, Mplus does not allow me to estimate the intercepts. Any though? Thanks, Ai 


This can be done in Mplus by adding a factor behind an item. See the 2016 WuEstabrook article in Psychometrika where they give the Mplus code. 

Ai Ye posted on Monday, February 12, 2018  7:04 am



Thank you so much. Dr. Muthen. I was able to find the paper and follow their lead. That was a very intuitive (and smart) solution. 

Back to top 

