According to wikipedia, pearson correlation is scale and location invariant.

Does scale refer to "variance" and location refer to "mean" ?



In this case, scale and location are more general. Given two random variables $X$ and $Y$, the correlation is scale and location invariant in the sense that $cor(X,Y) = cor(X_{T},Y_{T})$, if $X_{T} = a + bX$, and $Y_{T} = c + dY$, and $b$ and $d$ have the same sign (either both positive or both negative). Note that if $b > 0$ and $d < 0$ (and vice versa), $cor(X,Y) \neq cor(X_{T},Y_{T})$ because the sign of the correlation between the transformed random variables will be inverted.


$$X = 1,2,3,4,5$$ $$Y = 1,2,3,4,5$$ $$cor(X,Y) = 1$$

If $X_{T} = 1 + 2 X$ and $Y_{T} = 2 + 3 Y$ , then $$X_{T} = 3,5,7,9,11$$ $$Y_{T} = 5 ,8, 11, 14, 17$$ $$ cor(X_{T},{Y_{T}}) = 1 $$

But, if $X_{T} = 1 + 2 X$ and $Y_{T} = 2 - 3 Y$, then $$X_{T} = 3,5,7,9,11$$ $$Y_{T} = -1, -4, -7, -10, -13$$ $$cor(X_{T},{Y_{T}}) = - 1 $$

  • $\begingroup$ It seems that the features are the same as those of affine transformation. $\endgroup$ – Lerner Zhang Sep 28 '18 at 0:39

No, scale and location are more general in this case. A scale and location transformation of a variable X is a deterministic function of X defined as Y=f (X)=aX+b. For the correlation coefficient to be scale and location invariant is the same as saying that for a and b real, the correlation coefficient of X and Y will be the same. If you look at the definition of the correlation coefficient, and use definitions related to expectation and variance of a linear transformation of a variable, you can work to that result.


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.