I know the implication of non-normal residuals is uncertain statistical tests because the SEs are inefficient. Can I apply the Newey-West to calculate standardized Standard Errors for an OLS regression in response to non-normal residuals? I have outliers that cannot be captured by the OLS regression using a prescribed set of potential drivers.

Will the Newey-West correction give me appropriate P-values, even if the model does not suffer from heteroskedasticity or serially-correlated residuals?

  • $\begingroup$ The Newey-West SEs are for heteroscedasticity &/or autocorrelation, not for non-normality. If you don't have enough data to trust the central limit theorem, you may want to bootstrap or move to a nonparametric approach. $\endgroup$ – gung Jan 6 '16 at 21:51
  • $\begingroup$ Box-Cox transformation? $\endgroup$ – JohnK Jan 6 '16 at 22:10
  • $\begingroup$ @JohnK That will change the model from positing a linear relationship between the variables to one with a non-linear relationship. $\endgroup$ – Glen_b Jan 7 '16 at 2:06
  • $\begingroup$ @Glen_b Yes but it will bring the response variable closer to normality which is often desirable. There is no free lunch. $\endgroup$ – JohnK Jan 7 '16 at 10:08
  • $\begingroup$ Is there any Method for adjusting Standard Errors and P-Values from a model with non-normal resides so that statistical tests are reliable? (Without respecifying the model). $\endgroup$ – david_i Jan 7 '16 at 23:16

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.