Testing mediation model in stata when the dependent variable is multinomial

[Ishana Balan]

Hi all,

I am testing a mediation model. I want to calculate the direct and indirect effects of the independent variable. There are commands like sgmediation to test mediation for a normal regression. Within a structural equation framework, postestimation commands allow us to test the mediation effect. However, all this is applicable only when the dependent variable is continuous. Is there any way to test mediation when the dependent variable is multinomial? Appreciate any references. Thanks!

[Eric Makela]

Hi Ishana. While I think perhaps your question is a bit outside the bounds of a typical Statalist query (a general, un-reproduceable one of a specific statistical technique with no code or data), you might consider analyzing the multinomial levels independently, if data are sufficient. The idea would be similar to that of quantile regression for a continuous variable, with the potential at least for some good summary statistics and graphics.