I am using latent profile analysis to identify unique profiles of children based on early academic skills. I have compared 1-, 2-, 3-, 4-, and 5- class solutions. Based on AIC, BIC, and entropy values, the 5-class solution provides the best fit. According to my output, however, one of the classes in this solution has zero cases in it, but still has mean scores and variances for each of the measured variables in the dataset. How is this possible? Thanks in advance for any help you can provide!
During model estimation, all persons are proportionally in all classes. That is why you get some results for the zero class. Also, variances are held equal across classes as the default so other classes contribute to the estimation of those variances. A solution with a class with no observations should not be used.