Sparse group Lasso for multinomial logistic regression

[Gil Martin]

Hi everyone,

I'm planning to do a multinomial logistic regression and would like to use the sparse group lasso for variable selection. There's a package in R, and I wanted to confirm with the STATA community that no such package exists in STATA? Or any sparse group functions implemented in existing standard lasso packages?

Thanks.

Best wishes,

Gil

[Matthew J. Baker]

Gil Martin --

I don't think there is a package that does this directly. In the course of a project, I have had to program up some of this sort of thing in the context of treatment models with binary treatments and outcomes. I have a github site up that has all my code and data, and describes how this code was mocked up if you are interested.

It is here: https://github.com/mbaker21231/CivilWarSurgeryPublic

Hope that helps!

Matthew J. Baker

[Ralph Crott]

Similar question : I have an ordered clinical outcome of 4 increasing severity levels and would like to use a lasso approach for independent variables selection given that I have a low observations/variables ratio of around 10. Is there a Stata package to apply Lasso with ordered logistic regression or a way to do this in Stata ?

[Gil Martin]

@Matthew J. Baker

Thanks for your answer and help. I might end up having to do the variable selection in R!

Best wishes,

Gil