2

If one variable is normal distributed and the other is non normal distributed, which kind of correlation do we use?

I am not sure which correlation is right to use.

cardinal
  • 24,973
  • 8
  • 94
  • 128
user34729
  • 21
  • 1
  • 2
  • 5
    The Pearson correlation coefficient doesn't require the variables to have a certain distribution. The only connection I see between the case you are describing and correlation is that, if the variables are uncorrelated, it does not mean they are independent (which is simply the general case). – means-to-meaning Nov 12 '13 at 22:37
  • My variables are age which is normally distributed and number of activities which non normally distributed. I used Spearman rho cause it doesn't require any certain distribution. The coefficience is a negative number (-.316) does it mean anything specific as for the correlation? – user34729 Nov 12 '13 at 22:55
  • I think it would be better if you would either edit and amend your original question, or, since this seems to be an entirely different question, post this separately. I only answered you with a comment since I think that the fact that your distributions are different doesn't matter in the first place and you can use the most popular coefficient - Pearson, or the non-parametric - Spearman. – means-to-meaning Nov 12 '13 at 23:09
  • 2
    The big distinction between Spearman and Pearson correlation isn't any distributional assumption (neither assumes any, unless perhaps you're performing a test), but between linear correlation (Pearson) and monotonic association (Spearman). If you don't think the association is linear, then Spearman may make more sense. – Glen_b Nov 13 '13 at 00:29

1 Answers1

1

Partially answered in comments:

The Pearson correlation coefficient doesn't require the variables to have a certain distribution. The only connection I see between the case you are describing and correlation is that, if the variables are uncorrelated, it does not mean they are independent (which is simply the general case). – means-to-meaning

The big distinction between Spearman and Pearson correlation isn't any distributional assumption (neither assumes any, unless perhaps you're performing a test), but between linear correlation (Pearson) and monotonic association (Spearman). If you don't think the association is linear, then Spearman may make more sense. – Glen_b

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467