So, this is a two part question. I have made a multilinear regression model in R. The model has been ridged with lm.ridge() with the $\lambda$ that minimizes the GCV prediction error. This was done by the following code:

# package{MASS}
models.ridge <- lm.ridge(formula, data = data, lambda = seq(0, 1.5, 1/100))

# Package{broom}
g <- glance(models.ridge)

# performs ridge with the "optimal" ridge constant obtained from g$lambdaGCV
final.ridge <- lm.ridge(formula, data = data, lambda = g$lambdaGCV)

Now i want to bootstrap a confidence interval for the coefficients obtained from the ridge. However, doesn't the introduced bias on the estimators ruin the credibility of a confidence interval?

Also, I'm having trouble with finding a good way to summarize my new ridged model. Like the one you get from: summary(lm.fit) Which requires an lm object and not a lm.ridge object.


I use the glmnet package to do ridge regression. This post addresses how to do ridge regression with glmnet, why one could prefer using glmnet instead of the mass implementation, and how to use the boot library to find the bootstrapped CI for the coefficients. p.s.: This answer would be better suited as a comment but I do not have enough reputation yet.


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.