One assumption of regression analysis is independence of residuals. I checked this assumption and found small autocorrelation (see figure). One remedy would be to incroporate dummy variables for the lags.

But is it always naccessary? Autocorrelation does not affect the estimated coefficients, but the standard errors. The larger the autocorrelation, the larger the impact on standard errors.

Is there a rule of thumb how large the impact would be in my specific case?

enter image description here

  • $\begingroup$ The residuals look marginally correlated. Can you give us some context? Or do you have domain knowledge on the subject? Is correlation expected / not unreasonable for you data? If yes then you should take care of it, if not then ignore it. $\endgroup$ – user2974951 Dec 12 '18 at 7:18
  • $\begingroup$ The regression is used to predict discounts, customers received. I collected information about prices of products, which products they bought, age of customers etc.. I already tried to incorporate dummy variables for each weekday. But this did not fix the autocorrelation. The ACF plot looked exactly the same. $\endgroup$ – Hans Meier Ruth Dec 12 '18 at 15:09

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.