2
$\begingroup$

How does one calculate confidence intervals for a binomial mixed-effect model that was fit using the R package glmmLasso? I am interested in the 95% confidence intervals for the fixed effects.

confint throws "no applicable method for 'vcov' applied to an object of class "glmmLasso"", and intervals does not seem to work either.

edit: I see that the recommended approach to calculating standard errors in LASSO is to use Bayesian LASSO, but the packages recommended there do not handle random effects models. Nonetheless, on closer inspection glmmLasso does seem to produce standard errors for the parameters: is there any issue with simply turning these into confidence intervals using the standard approach described in the top answer here for getting confidence intervals from standard errors in logistic regression?

$\endgroup$
1
1
$\begingroup$

The author of the glmmLASSO package has kindly responded to me and shed some light on how the standard errors for the parameters are produced, saying:

"Unfortunately, in general, for a LASSO estimate no direct SEs are available and as far as I know this is still under current research. If in glmmLasso, the default argument final.re = FALSE is chosen, hence, no SEs are reported (and only NAs are shown in the summary). Only if final.re = TRUE is chosen, based on the covariates selected by LASSO, an unregularized (post-LASSO) model is fitted and SEs are reported. However, these have to be treated with much caution, as they are rather wrong, as they ignore the uncertainty involved in the LASSO step."

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.