3

I have a fractional response regression model, which is estimated with R's glm with family parameter equal to quasibinomial. Now, I want to perform model selection via lasso on the model.

Is there any way to tweak R's glmnet to perform lasso on quasibinomial family?

Or any other package?

DavidH
  • 178
  • 5
  • 3
    Can I ask why you want to do this? Doesn't the quasibinomial just estimate the dispersion parameter from the data and use it as a correction to the standard errors of the model coefficients? It doesn't alter the model coefficients so the lasso path wouldn't change, would it? – Gavin Simpson Jun 26 '15 at 21:41

0 Answers0