Mike posted on Thursday, January 12, 2006 - 12:19 pm
Are there any potential problems associated with using standardized continuous data and MLMV estimation? I've been told that MLMV was designed for unstandardized data but it is unclear to me what problems could occur.
Thank you for a wonderful product and a helpful forum.
You should not use standardized data for growth models or multiple group analysis or with models that are not scale free. If you have a scale-free model, then using standardized values with MLMV should be fine.
Mike posted on Thursday, January 19, 2006 - 6:57 pm
Thanks very much for your reply. Your helpful forums are just one of the reasons that I suggest Mplus to my peers.
I apologize, I'm a novice to SEM and am not quite sure what a scale-free model is. In our model, we're testing whether there are significant covariances between factors. These factors each have three indicators. We're using ULI constraints to scale the factors. In some cases, we're also testing whether higher order factors (without indicators) can account for some of these relations.
Do these models fall under the definition of scale-free?
A scale free model is one for which the same chi-square is obtained when analyzing the correlation or covariance matrix. Models with restrictions on the parameters across observed variables are not scale free, for example, a model in which the residual variances of observed variables are held equal. See the Bollen SEM textbook for a discussion of this issue.