Multicollinearity and lassologit

[Franz Czyborra]

Hello Stata-Forum users,

I'm currently working on my master thesis using the lassologit command of the lassopack extension for my data. I wanted to know if anyone of you knows how lassologit or the lasso2 commands deal with multicollinearity issues. Do the models omit variables? I just didn't find information in the help file or in existing topics here.

Thanks for your attention.

best regards,
Franz Czyborra

[Achim Ahrens]

If there are groups of highly correlated predictors, the lasso will tend to pick only one from each group and omit the others. This behaviour is one of the motivations for the elastic net (i.e. mixing ridge and lasso type penalty). We explain this in our working paper as well (see here).

Zou, H., & Hastie, T. (2005). Regularization and variable selection via the elastic net. Journal of the Royal Statistical Society. Series B: Statistical Methodology, 67(2), 301–320. https://doi.org/10.1111/j.1467-9868.2005.00503.x