Suppose I have a linear model $y_i=\mathbf{x}_i\boldsymbol{\beta}+\mathbf{z}_i\boldsymbol{\gamma}+\epsilon_i$, where $\boldsymbol{\gamma}$ is subject to elastic net regularization. Now I have a nested linear regression $y_i=\mathbf{x}_i\boldsymbol{\beta}+\epsilon_i$. Now I want to know whether $\mathbf{z}_i$ contributes significantly. i.e., If both models are linear models, we could do a F-test to compare these two models. Or, I can compare their adjusted $R^2$. What are the equivalents (F-test and adjusted $R^2$) when one of the models involved are under elastic net regularization?

  • $\begingroup$ By nested regression, you mean random effects? Why would you like to do a hypothesis test against regularized model? $\endgroup$
    – Tim
    Commented Apr 2, 2022 at 10:28
  • $\begingroup$ @Tim No. I meant the most basic linear regression. I want to know whether the added variables truly contribute to the model, and in my case, there are so many additional variables that I have to use regularized model. $\endgroup$
    – wdg
    Commented Apr 2, 2022 at 10:41


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.