I encountered the following sentence while reading a textbook and could not figure out why.

If $x$ and $y$ are jointly normal, then $y$ is a linear (affine) function of $x$: $y = a+bx$.

If someone could explain this to me, that would be great.


If $X$ and $Y$ are jointly normal,

$$ \bigl( \begin{array}{cc} x \\ y \\ \end{array} \bigl) \sim N(\bigl( \begin{array}{cc} \mu_{x} \\ \mu_{y} \\ \end{array} \bigl), \bigl( \begin{array}{cc} \sigma_{x}^2&\sigma_{xy} \\ \sigma_{xy}&\sigma_{y}^2 \\ \end{array} \bigl) ) $$

then the conditional distribution of $Y$ conditional on $X$ is

$$ Y|X \sim N(\mu_{y} + \frac{\sigma_{y}}{\sigma_{x}} \rho(x-\mu_{x}), (1-\rho^2)\sigma_{y}) $$

where $\rho=\frac{\sigma_{xy}}{\sigma_{x} \sigma{y}}$. Then the conditional expectation of $Y$ given $X=x$ has the form you described,

$$ E(Y|X)=\mu_{y} + \frac{\sigma_{y}}{\sigma_{x}} \rho(x-\mu_{x})=a+bx $$

where $a=\mu_{y}-\frac{\sigma_{y}}{\sigma_{x}} \rho \mu_{x}$ and $b=\frac{\sigma_{y}}{\sigma_{x}} \rho$.

| cite | improve this answer | |
  • $\begingroup$ I thought that the author meant the random variable $y$ not the conditional expectation $E(Y|X)$ in the expression $y=a+bx$. Could you construct an example where $x$ and $y$ are jointly normal, and $y$ is not equal to $a+bx$? Thanks a lot. $\endgroup$ – Emx Dec 11 '12 at 7:12
  • 1
    $\begingroup$ Any nondegenerate binormal distribution would serve that purpose, Dave. The simplest example would be a pair of independent standard normal variates. $\endgroup$ – whuber Dec 11 '12 at 7:35
  • 1
    $\begingroup$ @caburke There are several typos in your answer. The conditional mean is $\mu_y + \cdots$ and not $\mu_x + \cdots$ and your final value for $a$ is incorrect too, since you used $\mu_x$ instead of $\mu_y$ and simplified the RHS. $\endgroup$ – Dilip Sarwate Dec 11 '12 at 17:37
  • $\begingroup$ @DilipSarwate Thanks! The errors have been corrected. $\endgroup$ – caburke Dec 11 '12 at 19:35
  • 1
    $\begingroup$ @caburke One more typo. In your first displayed equation, the mean vector should be $(\mu_x,\mu_y)$ (transposed), $\endgroup$ – Dilip Sarwate Dec 12 '12 at 3:09

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.