Is there an extension of Multivariate regression for binary outcome or GSEM for two outcomes?

[Marzieh Rohani-Rasaf]

Hello everyone,
I am doing an analysis where I have two binary dependent and correlated variables(two binary outcomes) and the independent variables are also correlated. Independent variables affect both outcomes
I wanted to use the multivariate model in SEM(Structural models 12) but there the dependent variables must be continuous and it does not accept binary link function, and my next solution was to use GSEM which also does not accept two dependent variables. of course I don't have a latent variable. Is there a solution? Can you provide a model for this analysis considering the correlation of variables?

[Joseph Coveney]

. . . I have two binary dependent and correlated variables(two binary outcomes) . . . Independent variables affect both outcomes. . . . Is there a solution? Can you provide a model for this analysis considering the correlation of variables?

Have you considered biprobit?

help biprobit

[Marzieh Rohani-Rasaf]

Thank you, It seems that in this model, the correlation of independent variables or their collinearity is not considered.
What if a third variable is entered as dependent?
I ran two GSEM models for each outcome variable to see the impact of the outcome variables on each other and summed up the direct and indirect path. Isn't that right?

[Joseph Coveney]

It seems that in this model, the correlation of independent variables or their collinearity is not considered.

Try cloning one of the predictors (help clonevar) and see whether biprobit omits one as collinear.

Are the coefficients of independent variables adjusted for the correlation of two dependent variables?

Compare the coefficients obtained with biprobit with corresponding coefficients obtained with fitting two separate probit models.

What if a third variable is entered as dependent?

search mvprobit

I ran two GSEM models for each outcome variable to see the impact of the outcome variables on each other and summed up the direct and indirect path. Isn't that right?

I don't follow what you're doing there, but if I wanted to fit a bivariate binomial regression model, then biprobit would seem to me to be the most straightforward approach.

[Marzieh Rohani-Rasaf]

Thank you for your attention