I am working with a student whose data are behaving strangely in a CFA model. The student used a planned missingness design, so we created 10 imputed datasets and ran a CFA. She had translated the items from English into Georgian using multiple translators, but the factor loadings for the items on their latent subscale factors are completely different from what I get in my US samples. I'm thinking that cultural differences between the US and Georgia might be contributing to the unexpectedly low factor loadings.
I am considering specifying the items as ordinal, but that would require me to round the imputed values to the closest whole number. As I understand it, you're not supposed to do that. How would you suggest I proceed?