Hi: If I use TYPE = MIXTURE to fit a log linear model to a cross-classification for three dichotomous variables, is there a way to get the intercept for the model. To take a simple example, if I fit log(Mu) = beta0 + beta1 + beta2 + beta3, I find estimates of beta1 to beta3 but not for beta 0. Thanks, Jamie
Beta0 exists when the model is regarding cell counts and is Log(N) where N is the sample size. The models in Mplus are not about the cell counts but the probabilities (cell count = N * probability) so beta0=0, i.e., there is no such parameter.
Thanks for your response. If use a GLIM to fit a loglinear model to the counts, the intercept depends on how I code the independent variables. I could not come up with a coding system that yields an intercept equal to the log of the count total. At any rate if I use Mplus to fit the loglinear model to the counts, and use the estimated parameters and an intercept caculated as the log of the model_based estimated count in 2 2 2 cell, I can recover the model-based estimates of the proportions.