Hello,
It has been found that SEM analysis using polychoric correlations + maximum
likelihood estimator produces incorrect test statistics and standard errors
(e.g., Flora, D. B., & Curran, P. J. (2004). An Empirical Evaluation of
Alternative Methods of Estimation for Con?rmatory Factor Analysis With
Ordinal Data. Psychological Methods, 9(4), 466-491).
Standard errors can be dealt with by using bootstrap estimations. But, I was
wondering how the fit indexes in the SEM package were estimated when the
observed variables are categorical (ordinal or/and binary) and when one is
using polychoric correlation (hetcor in polycor package) in combination with
ML. Does SEM use the Satorra-Bentler rescaled Chi-square for instance?
I look in previous posts but couldn't find what i am looking for (hope i
looked well enough!).
Cheers,
Dorothee