When regressing latent class membership on a predictor variable, Mplus now conveniently provides the results for ďALTERNATIVE PARAMETERIZATIONS FOR THE CATEGORICAL LATENT VARIABLE REGRESSION.Ē Iím wondering if someone can provide the formula by which this is done? Iím doing a bunch of simulations and need to calculate these as a function of parameters in the save data file. Thanks! Lee
Try a case with 3 classes and no x's, changing from last to first class as reference class. Subtract from all classes the intercept for the first class. Then in the numerator of the multinomial formula you get exp with transformed arguments using * notation:
class 1: alpha*_1=alpha_1-alpha_1=0 class 2: alpha*_2=alpha_2-alpha_1 class 3: alpha*_3=alpha_3-alpha_1 = -alpha_1,
so those are the new intercept values. Then expand to having an x variable with betas and compare to the output to see if it comes out right.