There are multiple reasons for applying standardisation/mean centre for predictors before putting them into a regression model. However, in the literature, some people do not do so or even argue against this idea. They say that it is not necessary to do standardisation/mean centre. Actually what are their reasons? I heard that standardisation/mean centre changes the interpretation of main effect, but not the interaction effect. Is it true? Are there any drawbacks to standardisation/mean centre before applying the predictors into a regression model?

The links below are some previous discussions on the benefits of standardisation/mean centre. But in order to be as critical as possible, I'd like to know the backside of standardisation/mean centre.

  • 1
    $\begingroup$ One disadvantage is that you lose the original scaling. That might be important when you read the estimated coefficients. $\endgroup$ – SmallChess Oct 30 '17 at 5:03
  • $\begingroup$ @SmallChess, true, but standardization is normally undone after the estimation, so the coefficients for the original variables are obtained. $\endgroup$ – Richard Hardy Oct 30 '17 at 6:22
  • $\begingroup$ The major benefits and drawbacks of any linear transform will often come down to interpretability, I think, and the downside comes when people are not able to properly interpret a standard deviation. I don’t think it’s necessarily ever wrong to do so - it just isn’t necessary and is often more cosmetic. (Not to downplay the importance of clear reporting.) $\endgroup$ – RickyB Nov 8 '17 at 13:47

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.