[Jeff Wooldridge]

Thanks for that find, Alecia. Here I thought one just put in hundreds or thousands of functions and let LASSO work its magic. I wonder if the problem is less severe when the selection is only over control variables, and the goal is causal inference and not prediction. I suspect that's being too optimistic.

[Saad Tabet]

Hello Alecia,

did you controll for collinearity? I asked Dr Mander and he told me to eliminate highly correlated variables. You may use the VIF!
But I still have a question: do you know for what s(#) stands for? Is it the Lamba for the LASSO?
I tried several values for # but without achieving any results (always the same numbers). I wrote the command like this : lars 'varlist', a(lasso) s(#)
Do you or anybody else have a suggestion?
I would relly appreciate it!

Thank you in advance and good luck!
Saâd