1
$\begingroup$

A very simple question about the variance!

I'm interested in the variance of an expression with a random variable inside it.

For an expression of this form:

$a = \sum_{i=1}^{m}(x_i \cdot y_i - z_i)$

where $y_i$ is a random variable with an expectation $\mathbb{E}(y_i)$, would the expectation of the whole function simply be:

$\mathbb{E}(a)=\sum_{i=1}^{m}(x_i \cdot \mathbb{E}(y_i) - z_i)$

Since variance is $Var(a)=\mathbb{E}(a^2)-(\mathbb{E}(a))^2$, I assume I can simply plug this expression (if correct) into this variance formula. Is this correct?

Thanks!

$\endgroup$
1
$\begingroup$

If $x_i$ and $z_i$ are deterministic constants for each $i = 1, \ldots, m$, and $m$ is also a known constant--that is to say, $y_i$ for $i = 1, \ldots, n$ are the only random variables involved, and they are independent--then the variance of $a$ is better calculated as $$\operatorname{Var}[a] \overset{\text {ind}}{=} \sum_{i=1}^m x_i^2 \operatorname{Var}[y_i].$$ This is because $z_i$, being deterministic, has zero variance. If the $y_i$ are not independent, then the variance of $a$ will require knowing the covariances between each $y_i$, $y_j$ for $1 \le i \ne j \le m$ in addition to the variances of each $y_i$.

$\endgroup$
  • $\begingroup$ Aha, I wasn't aware of the rule about constants being squared in the variance! Thanks. I'm still learning ;) $\endgroup$ – Sprog Dec 31 '15 at 0:33

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.