I used to use simple (multivariable) logistic regression plus stepwise regression to select important predictive variables. Though, I saw many critics on this method. Thus, I plan to use LASSO and check whether it gives similar results compared to the stepwise regression.

I am primarily using the glmnet package from R.

After the cv.glmnet, I got the lambda.min and lambda.1se respectively. I am still fine at this point that I probably should choose the lambda.1se for the lambda.

Please comment on above workflow as I wonder whether I have any misunderstandings.

The main question is: I get the coefficients corresponding to lambda.1se by coef(cvfit, s = "lambda.1se"). How should I interpret these coefficients? I acknowledged that some of the unimportant variables have been dropped, thus only important variables are kept.

Should I put these selected (important) variables back into the conventional logistic regression? It is because, I am studying epidemiology, which I am interested in whether these variables are independently associated with the outcome. I am not sure whether I can deduce similar conclusion statement from LASSO's output. Can we get the OR with 95% CI for each coefficient?



Your Answer

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

Browse other questions tagged or ask your own question.