Say I have an observable z, which linearly correlated with two other variables x and y, like this:

z = ax + by

If you assume x and y are completely uncorrelated, I would think that the correlation coefficient Rx,z or Ry,z can not become one (or even close to). Is there a theoretical maximum for this correlation coefficient? And if yes, how can you prove it? And what is it? I'm guessing it would depend on a and b.

  • 2
    $\begingroup$ Consider $a = 0$. What's the correlation between $z$ and $y$? $\endgroup$
    – jbowman
    Commented Aug 6, 2015 at 17:19
  • $\begingroup$ Well, assuming b is not zero it would be one. But I'm more interested in the non-trivial cases. $\endgroup$
    – Lu Kas
    Commented Aug 6, 2015 at 17:45
  • 1
    $\begingroup$ $a=0$ isn't "trivial." If you want to exclude it, what would your answer be for $a=10^{-100}$? If you fix $a$ and $b$, then you're asking how to compute the correlation of $z$ in terms of the variances and covariances of $x$ and $y$--which is a standard formula you can readily find in many threads here. $\endgroup$
    – whuber
    Commented Aug 6, 2015 at 18:21


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.