I am anaylsing a data set, which displays a heavy-tailed distribution when examined on a Quantile-Quantile plot. What is (or are) the best transformation(s) to use to correct a dataset with a heavy-tailed distribution?

  • 3
    $\begingroup$ What do you want to correct? What makes you think there's something that needs to be corrected? $\endgroup$ – Kodiologist Jul 22 '16 at 19:13
  • 5
    $\begingroup$ Some distributions are just heavy tailed, like the mass distribution in a cat. My cat would be pretty unhappy if I attempted to "correct" this. $\endgroup$ – Matthew Drury Jul 22 '16 at 19:26
  • 2
    $\begingroup$ I'm with kodiologist -- what's the problem with a heavy tailed distribution? Even if there is a problem, why would you transform rather than do something else? $\endgroup$ – Glen_b Jul 23 '16 at 4:07
  • 3
    $\begingroup$ Do you have data to share, or at least some plots? What do you want to achieve with your analysis, why do you care about normality? $\endgroup$ – Georg M. Goerg Dec 24 '16 at 14:30
  • $\begingroup$ What do you want to acheive for this data? What is your model? Why do you think tails is wrong then you would like to correct it? Did you have any ideas about transformation? It is good idea to ask yourself many questions before asking this will help you to answer your question. $\endgroup$ – Alice Oct 18 '17 at 7:19

You could transform the series with the natural logarithm. Alternatively, some of the literature looking at the determinants of net worth have used the inverse hyperbolic sine transformation. (See Pence 2006) It has the advantage of accommodating zero and negative values.


protected by kjetil b halvorsen Oct 18 '17 at 6:31

Thank you for your interest in this question. Because it has attracted low-quality or spam answers that had to be removed, posting an answer now requires 10 reputation on this site (the association bonus does not count).

Would you like to answer one of these unanswered questions instead?

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