How does using regularization affect inference? If your parameters aren’t MLE then the score (gradient of loglikelihood) isn’t zero. So, it seems like a lot of the assumptions behind the calculation of standard errors and p-values are thrown off. Does anyone have a good reference?
