Dear Mrs. and Mr. Muthen, I am doing some SEM-analyses based on different response scales (5- and 6-point scales). There are five different constructs. Two constructs are based solely on items with 5-point scale and three constructs are based on items with 6-point scale. In my model I would like to use testhalves to avoid too complex models and too high degrees of freedom. For this purpose, I examined first unidimensionality of my single constructs using MLR-estimator. Everything works fine for my 5 constructs regarding examination of unidimensionality, but now I am note sure, if it matters to analyze the relationship between the latent factors based on the test halves (which are computed using different response scales for the different constructs) without doing some kind of z-standardization. As far as I know it should not be a problem, because I can use the standardized coefficients for interpretation. Is that right? Thanks a lot. Tony
I don't see that it would be a problem because scale differences in the sets of halves are taken into account in the factor model, which will have different intercepts and loadings. Standardized solution should be useable as you say.