Emily Blood posted on Friday, December 03, 2010 - 1:17 pm
I am getting the same value for the regression coefficient of a predictor if I fix the intercept/threshold term at 0 and when I don't when I fit a simple probit model to data I've generated (so I know the true values). This does not happen with a logit model. Is there a reason for this? I would think they should change. If I fit the same model with Splus I get different results. My model is: results for the "z" parameter are the same with this: y on z*0.3; [y$1@0]; and this: y on z*0.3; using THETA parameterization.
I am curious if it is possible for Mplus to yield an intercept in a probit model similar to that given in Stata (or other programs)? I have tried doing this with the input below, but receive an error (without results) saying that the mean estimation for the categorical outcome variable is ignored. Again, I just want to calculate an intercept similar to what is given in Stata so I can compare models. If there is another way, besides what I tried below great!
Categorical are class1 ;
Analysis: Estimator = WLSMV;
Model: ! Model of Social Class class1 ON educ prestg80 paeducM maeducM NApaeduc NAmaeduc age black asian hispanic othrace ;
! First threshold of class1 fixed at 0 [class1$1@0];