Lasso for variable selection - no group option

[Ariel Karlinsky]

Stata's built in Lasso for variable selection (for example dsregress) treats all variables separately and thus may keep some categories of a factor variable but remove others - which is nonsensical.
There might also be groups of variables that only make sense if present together in the model.

A group-lasso (for example gglasso in R https://cran.r-project.org/web/packa...sso/index.html) allows users to specify groups of variables, such that a model either includes both or neither.
Is there anyway to do this in Stata, and if not, are there any plans to implement this in the near future?

[Andrew Musau]

Stata's built in Lasso for variable selection (for example dsregress) treats all variables separately and thus may keep some categories of a factor variable but remove others - which is nonsensical.

Is it really? You will find many cases where only some levels of a categorical variable are kept, depending on the focus. The following is a table from Rodrick's 1998 Journal of Political Economy paper. Of all the world's regions, he keeps indicators for Latin America, East Asia and Sub-Saharan Africa.