I have a two-level structural model in which I test the relationships among 5 latent variables at the student level and 3 latent variables at the classroom level. The factor indicators which define the latent variables are categorical.
I also think to check multivariate normality, however, I would not be able to use MAHALANOBIS option of Mplus, since I have categorical outcomes.
For power analysis with categorical outcomes is Monte Carlo study appropriate or such simulation studies specific to continuous variables in Mplus?
There is no assumption of normality of categorical items. Categorical data methodology takes into account floor and ceiling effects.
Monte Carlo studies are appropriate for any scale type.
Utkun Ozdil posted on Tuesday, June 14, 2011 - 10:04 pm
Thanks Linda,, I have a follow up question. Mplus does not save a file for residuals.In order to test the independence of residuals for different levels and the independence of residuals for different units in the same level with respect to the two-level model above can I save residual files in SPSS and use it as a .dat file in Mplus?
Or there is no need to test these assumptions with categorical factor indicators using Mplus?
Wendy Rote posted on Thursday, August 14, 2014 - 12:21 pm
Is it still the case that Mplus is not able to save factor residuals at the individual level? Dyer, Day, & Harper(2014) claim to have saved residuals using Mplus 7.11 for later analysis in SPSS, but I cannot determine how to do so myself.
Wendy Rote posted on Thursday, August 14, 2014 - 4:05 pm
Journal of Family Psychology, Vol 28, Issue 4, pgs 516-528.
Wendy Rote posted on Friday, August 15, 2014 - 12:23 pm
The first author of the paper got back to me about this: To save residuals he sets the residual of a DV/indicator to 0, creates a latent variable with no indicators, and regresses the DV/indicator onto the new latent variable setting the regression weight to 1. The latent variable then equals the residual and you can save the "residual" by saving the factor score of the latent variable with 'Fscores'.