I am running some data in R that is intended for multiple regression. However my data is not normally distributed. Most of my independant variables are continuous, however one is a catagorical factor containing two groups. The issue is that when looking at my dependent variable in terms of these groupings, there is a strong positive skew with one group and a strong negative skew with another group. I have included histogram images below to get an idea. Is there a transformation that I can apply that can help resolve this type of non-normality and how can I do this in R? I tried a log transformation, square root transformation, cube root transofrmation... no luck.

enter image description here enter image description here enter image description here enter image description here

  • 1
    $\begingroup$ What about the residuals of your regression? Those are what matter. $\endgroup$
    – Dave
    Jun 26, 2020 at 11:32
  • $\begingroup$ Do you know anything about the process generating the data? $\endgroup$
    – user289381
    Jun 26, 2020 at 11:42
  • $\begingroup$ As @Dave commented: Data do not need to be normaly distributed for linear regression or else you could not even think of binary predictors. $\endgroup$
    – Bernhard
    Jun 26, 2020 at 11:58
  • $\begingroup$ Added residials via an edit. This has not been adjusted for outliers. $\endgroup$
    – Guest
    Jun 26, 2020 at 12:56
  • $\begingroup$ I would't worry based on these plots. For tests/confidence intervals, you could try bootstrapping. $\endgroup$ Jun 27, 2020 at 2:36

1 Answer 1


There is an R package called trafo that may give some ideas. I have not used it but it tests model assumptions using residuals (see @dave comment).

The dependent variable and group two look candidates for preprocessing by taking logarithms.

The Q-Q plot is not bad as far as these things go. The heteroscedasity in the residuals can be tested for using the Breusch–Pagan test.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.