GSEM predicted probability calculation

[Necati Celik]

Dear Stata Forum,

I am estimating an SEM model with two latent variables as such:

 gsem (Latent1 Latent2, ) ///
      (Latent2 -> Some_Categorical_Var, family(ordinal) link(logit) ///
      (Some_Continuous_Var -> Latent1, ), difficult latent(Latent1 Latent2) nocapslatent

My goal is to accurately put respondents into the category of the categorical variable with the highest predicted probability, given the value of observed continuous variable. For example, given the SAT score of a college applicant, I want to be able to predict which college major they would be in (hypothetical example). I know how to find the predicted probabilities after gsem.

My question is: How do I find predicted probabilities manually after this model? I need to build a spreadsheet tool where someone can plug in the observations of continuous variables and get the predicted category.

Any insight is greatly appreciated,

Much thanks.

[Phil Bromiley]

Stata provides such predictions - see GSEM Postestimation in the SEM documentation. In the documentation, Stata normally tells you how it calculates things.

I don't understand your model since you seem to have no link between the two equations, but that may just be my lack of experience with GSEM.

Why don't you use multivariate probit or discrim logistic?

[Necati Celik]

Thank you for your response Phil.
I forgot the link between latent factor in the code above. It should have been:

gsem (Latent1 -> Latent2, )

I know how to find predictions using Stata. For the purpose of creating a spreadsheet tool, I need to manually calculate predictions. Not familiar with discrim logistic at all. What would be the advantage of using multivariate probit over ordinal logit? Categories are ordered.