I fitted a probit model in R using the glm function where my dependent variable is a binary variable and my indepentent variables are also binary and categorial variables. I also fitted a heteroscedastic probit regression model using the hetglm function from the glmx package which extends the link function to account for heteroscedsticity.
When I use the models as predictors however the extended model does not perform better. Now I want to check my data for heteroscedasticity and implement some kind of simulation study to test "how much heteroscedasticity" is needed in the data that the extended model performs better.
But here is the point where I can't quite wrap my head around. How do I even check for het. when I just have binary and categorial variables. I also added a picture where I plotted the Residuals against the Fitted values, but this is also hard to understand for me. The second picture are the residuals of one variable, which make it seem as if there is het. present.
