4

Possible Duplicate:
Under what conditions does correlation imply causation?
Can somebody illustrate how there can be dependence and zero covariance?

Or could there still be a relationship? Is it possible that the math just works out this way given the way OLS works but there is some relationship? Perhaps a nonlinear one? Anything?

Or does $R^2=0$ imply that X and Y are completely unrelated?

Community
  • 1
Jack Martin
  • 41
  • 1
  • There's also very closely related information [here](http://stats.stackexchange.com/questions/13314/is-r2-useful-or-dangerous) and [here](http://stats.stackexchange.com/questions/32781/how-can-i-use-the-value-of-r2-to-test-the-linearity-assumption-in-multiple-re). Several other places on the site also discuss the difference between uncorrelated and independent but those were the first three I recall at the moment. – Macro Jan 28 '13 at 23:42
  • "Does this imply that Y and X are unrelated?" The short answer is "No"; first because you're measuring only linear relationship in the mean, and secondly, sample statistics have variation, so it doesn't necessarily imply that there's no linear relationship. It may just imply a small sample! – Glen_b Jan 28 '13 at 23:46
  • And [here](http://stats.stackexchange.com/questions/30156/can-somebody-illustrate-how-there-can-be-dependence-and-zero-covariance/). And [here](http://stats.stackexchange.com/questions/26300/does-causation-imply-correlation/26318). – StasK Jan 29 '13 at 00:21

2 Answers2

9

No, it doesn't mean that they are unrelated. $R^2$ only measures linear relationships. For example, if $X$ is symmetric around zero and $Y = X^2 + \epsilon$, where $\epsilon$ is some error independent of $X$, then the $R^2$ for the regression of $Y$ on $X$ will be zero--even if $\epsilon \equiv 0$, so that there is a perfect deterministic relationship between $X$ and $Y$.

Jonathan Christensen
  • 3,989
  • 19
  • 25
3

One counter-example is if $X$ and $Y$ are phase-offset. The following works in R:

t <- seq(0,2*pi,0.1)
x <- sin(t)
y <- cos(t)
summary(lm(y ~ x))$r.squared
user12719
  • 1,009
  • 1
  • 8
  • 10