Consider a set of $100$ people, for each person we have 4 values which are: weight, height, age and the value of a medical exam, which will be called "A" for simplicity. I want to compute the correlation between weight and A, height and A and age and A.

Should I use one or two sided Pearson coefficient?

I computed the Pearson coefficient with the excel function (which I imagine is 1-sided) and the highest coefficient (all coefficients are >0) was the one of weight-A, about 0,077. But then I computed the 2-sided Pearson value with the help of this site http://www.wessa.net/rwasp_kendall.wasp#output and the coefficient of weight-A is the lowest (all coefficients are positive), the highest was the one of age-A, about 0,578.

What does it mean? Which coefficient should I consider?

Thank you

  • $\begingroup$ Is this related to a question asked earlier today, Correlation indexes in medicine? If so, please don't re-ask your question multiple times. $\endgroup$ – gung - Reinstate Monica Jun 19 '16 at 23:05
  • $\begingroup$ Your question unclear. How is a correlation intended to be "one-sided" or "two-sided"? $\endgroup$ – Glen_b Jun 20 '16 at 0:47

Pearson correlation coefficient it's just one coefficient - the "one sided" or "two sided" are different ways of testing hyphotesis. This correlation estimator is suitable for continuous variables. In this case, if the value of a medical exam is measured in a continuous scale, then it is correct to use Pearson.

But if the results of the exam are coded in an ordinal scale, like "low", "medium" and "high"; then you should use Kendall coefficient.

In any case, when interpreting a correlation you should mention first its direction - positive or negative, and then its intensity, based on the p-value provided in your link (if it's larger than 0.05 then it is not significant) or based on your previuos knowledge on the subject.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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