I've employed Elastic net to fit a logistic model with predictors that displayed high degrees of correlation between themselves. I wanted to be able to see which predictors significantly influenced the models predictions, so I employed the selectiveInference package to correct the coefficients and compute corrected CI and p values. Perhaps this question is very naive; since the coefficients of the model have been corrected, can I compute the odds ratio from the value of the coefficients as I would in a regular logistic model? Should this not be feasible, is there any way to quantify the influence (i.e. effect size) of each predictor for a model that has been built in this manner?
Asked
Active
Viewed 96 times
1
kjetil b halvorsen
- 63,378
- 26
- 142
- 467
GCO
- 11
- 2
-
if the predictors are indeed correlated, how do you distinguish one's effect from the other? I am not familiar with what selectiveInference is doing, but essentially you face the same problem when you quantify the effect.. how do you quantify effect in the presence of correlation – StupidWolf Apr 16 '20 at 09:53