I have run latent profile analysis in R using TidyLPA. I have obtained 3 profile solution, which I aim to use in subsequent analysis so that the profiles are dependent variable in multinomial logistic regression. Since the observations are classified into the most likelihood profile based on the highest probability, there is classification error that would need to take into account in the subsequent analyses. So called three-step approach is recommended in the literature, and programs such as Mplus can take the classification error into account, but is there any method in R I could utilize now in this “third step”? Perhaps calculate the classification error manually since i have the posterior properties? Using some kind of sophisticated weights in the multinomial regressions?

I have tried reading about effective solutions but without any luck.