A model with four risk classes was selected on the basis of the BIC, AIC and model interpretation using data for the entire sample (high-risk and normative samples, combined). However, the classes were not invariant across sex, based upon likelihood ratio tests comparing models which restricted item parameters to be equal and models which allowed item parameters to vary by sex (p<.001). Therefore, data from boys and girls were analyzed separately. The number of latent classes was reassessed separately for each sex, and four latent class models remained the best model for each sex based on the same criteria. When each of the three demographic groups within each sex was analyzed separately (e.g., urban African American, urban European American, rural European American), four latent class models were still preferred in all cases and the interpretations remained the same within sex. This finding suggested that it would be reasonable to combine all locality-race groups into a single measurement model separately for each sex, but allow the proportion of children in each class to vary among the locality-race groups.
