I would like to test a path model in which two dummy variables are used as mediators. I found that one way to handle this situation is to create a latent categorical variable. I have two types of alcohol provision represented with two dummy variables for parent provision and all other source of provision, with reference category was nondrinkers. I had syntax as below but it didn't work out well. Nobody was assigned to class 3 (intended to be the reference), and entropy was .58 .The frequency of the alcohol provision variable is: 70% nondrinkers, 10% parent provision, and 20% all others. I would like to examine whether means of the outcomes are different across 3 categories of the alcohol provision variable. Would you please guide me how to solve this puzzle?
Don't you want the 3rd class to have +15, +15 so that parents=0 and others=0? Note that the value you specify is a threshold in the logit scale, so that a high value means a low probability of the variable being 1 (i.e. a high prob that it is 0).