enter image description here

I've tried a log-transformation but data becomes left-skewed.

In general, when data is distributed like this, what is the next best transformation to try if you want to normalize the data?

  • 6
    $\begingroup$ Please explain why you need to transform this dataset to look Normal. (Such efforts are usually unnecessary.) $\endgroup$
    – whuber
    Apr 8, 2022 at 20:26
  • $\begingroup$ Computing the inverse of the normal CDF (cumulative distribution function) on the given data might give you close-to-normal distribution. The reason is that your distribution looks a little bit like a uniform distribution in [0,1]. In fact, if this were the case and if we call u a value measured in your distribution and set x = F^-1(u), where F(.) is the standard normal CDF, we have: Pr(X <= F^-1(u)) = Pr(F(X) <= u) = u = F(x), because U has been assumed to be uniform. In R you can get the inverse of the standard normal CDF of u by running qnorm(u). $\endgroup$
    – mastropi
    Apr 11, 2022 at 14:50
  • 1
    $\begingroup$ @mastropi thank you. makes total sense and it worked. $\endgroup$ Apr 11, 2022 at 15:56
  • 1
    $\begingroup$ "Scaling" and "transforming to normality" are completely different operations. $\endgroup$
    – whuber
    Apr 11, 2022 at 16:31
  • $\begingroup$ @whuber maybe i did (or didn't) phrase this right, but for clustering isn't one of the ways to make variables comparable, which im lead to belive is a necessary step in clustering, to normalize numeric variables? $\endgroup$ Apr 11, 2022 at 16:47


Your Answer

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