I have two quantitative variables, one with negative and positive values (let's call it A), and the other with solely positive values (let's call it B). I would want to do the correlation between these two variables. These values are those of individuals (one value per individual). I would like to know if this makes sense to correlate directly with the two variables, or to look the correlation between A and B in two groups : one where individuals have negative values for A and the other group where individuals have positive values for A.

Thanks for your help !

  • $\begingroup$ Give us some context, what are these variables, what do they represent and their values. $\endgroup$ Commented Sep 14, 2018 at 9:43
  • $\begingroup$ I would want to know if there is a correlation between a IAT (Implicite Association Test) score (decimal value) and a percentage which represents the case percentage where an individual practices hand hygiene. The individual's IAT score can be negative or positive. And so the percentage is always positive, of course. $\endgroup$
    – C.Mag
    Commented Sep 14, 2018 at 9:59
  • $\begingroup$ This could make sense, but it's usually a bit of a dead end. I agree with @user2974951: giving us a data example would help mightily. $\endgroup$
    – Nick Cox
    Commented Sep 14, 2018 at 10:19
  • $\begingroup$ There's nothing inherently wrong with correlating two variables where one has some negative values and the other doesn't, but without knowing why you might separate them, it's hard to be definitive. $\endgroup$
    – Peter Flom
    Commented Sep 14, 2018 at 10:48
  • $\begingroup$ Yes, I think too it's not wrong to correlate these variables. In my case, it would seem that there are two groups of individuals. One group with individuals who have an positive IAT score and another group with individuals who have an negative IAT score. I checked the correlation between the case percentage where an individual practices hand hygiene and the IAT score in the two groups. Which I have trouble understanding is that, when I check globally this correlation, there is a negative correlation. But, when I check this correlation in the two groups, there isn't correlation... $\endgroup$
    – C.Mag
    Commented Sep 17, 2018 at 9:39

1 Answer 1


Correlation is invariant under translations of the two variables: $$ corr(A,B) = corr(A+k_1,B+k_2). $$ So, it is not meaningful to calculate two separate correlations (unless units with positive and negative values of $A$ have a particular meaning, and you want to consider two separate subpopulations).


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.