1
$\begingroup$

I have already log-transformed the dependent variable but there is still heteroscedasticity in the residual-fitted plot. What one usually does in situations like this? My current regression technique in linear regression.
Fitted vs residulas

The data I am experimenting with is from a Kaggle competition: Santander Value Prediction

$\endgroup$
  • 1
    $\begingroup$ Would you please post a link to the raw "un-transformed" data? $\endgroup$ – James Phillips Aug 2 '18 at 12:49
  • $\begingroup$ If you have not yet done so, I suggest adding an offset term to the fit. $\endgroup$ – James Phillips Aug 4 '18 at 21:29
1
$\begingroup$
  1. you could try other transformations. In R, https://stat.ethz.ch/R-manual/R-devel/library/MASS/html/boxcox.html looks for an optional power transformation

  2. If you can't remove the heteroscedasticity by a suitable transformation, you have to fit a model that accounts for changes in the variance. In R, this can be done with the weights option in the gls function.

$\endgroup$

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.