0
$\begingroup$

I have to detect outliers on 3 variables. On the internet I found the mahalanobis distance but I understood I can use it only on multivariate normally distributed data, and my data isn't. So, do you have any suggestions?

And, for you is a correct way detect univariate outliers first, then bivariate and then trivariate outliers?

Thanks in advance for any suggestions :)

$\endgroup$
3
  • $\begingroup$ This is a hard question you are asking, there is no clear answer. But you should definitely detect outliers in all dimensions together and not for each dimension individually, if you must do this. $\endgroup$ Jan 14 at 11:35
  • $\begingroup$ @user2974951 thanks, man. I read many papers and blogs but I've not found any solution. It's my first multivariate outliers analysis :D $\endgroup$
    – simo954
    Jan 14 at 12:39
  • $\begingroup$ stats.stackexchange.com/… $\endgroup$
    – whuber
    Jan 14 at 13:51