I was thinking, since $X, Y$ are from $N(0,1)$ and they are independent, then

$X - 2Y$ has a distribution of $N(0, 5)$. Then $X-2Y > 0$ has probability of $1/2$.

The above seems correct to me, though it seems like then $X>nY$ would have probability of $1/2$. That seems a bit wrong. Did I get anything wrong?

  • $\begingroup$ What seems 'a bit wrong' there? Are you thinking about the conditional probability perhaps? ($P(X>nY|Y)$ ... that's not the probability in question) $\endgroup$
    – Glen_b
    Apr 5, 2015 at 23:43
  • $\begingroup$ If I understood you right the results $\frac{1}{2}$ seems not intuitive for you. But even in case if n is large Y is positive with probability $\frac{1}{2}$ (and negative with probability $\frac{1}{2}$). Although |X| is unlikely to be larger than |nY|, the probability without absolute values is reasonalby $\frac{1}{2}$. $\endgroup$
    – Lan
    Apr 7, 2015 at 20:56

1 Answer 1


With a bivariate standard normal (i.e. iid standard normal), the probability of lying on one side of a line through the origin is $\frac{_1}{^2}$ no matter what the slope of the line is.

enter image description here

This follows, for example, from the rotational symmetry of the bivariate distribution about $O$, since we could rotate the problem to one of considering $P(X'\gt0)$ in rotated coordinates.

Indeed, considering the use of affine transformations means it must be $\frac{_1}{^2}$ much more generally -- the argument will apply to any bivariate normal where both variances are greater than 0.

  • 1
    $\begingroup$ Thanks, I just found my conclusion a bit counterintuitive, but your diagram makes all this clear to me. $\endgroup$ Apr 6, 2015 at 0:12
  • 4
    $\begingroup$ If $X$ and $Y$ are zero-mean jointly normal random variables (not necessarily independent), then $X-aY$ is a zero-mean normal random variable and thus $$P\{X > aY\} = P\{X-aY > 0\} = \frac 12.$$ Independence and variances have nothing to do with the matter: all that is needed for the above result to hold is that the variables be jointly normal and that the means be zero. (Trivial exception when $X-aY$ equals $0$, that is, it is a degenerate normal random variable a.k.a. a constant which happens when $X$ and $Y$ are perfectly correlated and $\sigma_X = a\sigma_Y$). $\endgroup$ Apr 6, 2015 at 2:48
  • $\begingroup$ Thanks Dilip, your comment is of course completely correct - I was starting with the conditions as given and attempting to give some motivation for the result the OP had already derived. $\endgroup$
    – Glen_b
    Apr 6, 2015 at 2:54

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.