Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Assume I have a gaussian distribution $\mathcal{N}(\mu, C)$ with mean $\mu$ and covariance $C$. I'm drawing $n$ random numbers from this distribution. Let $m$ be the mean of these numbers. Is there some formula that gives the probability that the distance $d = ||\mu - m||$ is at least $x$, i.e. $P(d \ge x)$?

The background here is that in a recent simulation, the results seemed to cluster around a very slightly different point than expected, and I'd like to calculate the probability of this happening by chance.

share|improve this question
"Random vectors" rather than "random numbers" is what I'd have expected you to say, given the context. I'm inclined to doubt there's a closed form for this. –  Michael Hardy Jun 8 '12 at 23:58
Do you mean covariance, that is, is your distribution multivariate normal? Or is it univariate, like weights of randomly chosen people? –  André Nicolas Jun 9 '12 at 0:00
It's actually multivariate and random vectors, but the anomaly only occurs in one dimension and I thought the univariate case would be easier to answer, so I stated the question in terms of a univariate distribution. (as far as I know, covariance and variance are the same for univariate distributions) –  Benno Jun 9 '12 at 0:06
Would the following make sense to you? It is the probability that $|Z|\ge \frac{x\sqrt{n}}{\sigma}$, where $Z$ is standard normal, and $\sigma$ is your standard deviation. The relevant probabilities for the standard normal are available in tables. Many pieces of software also do the calculation. If you have particular numbers I can walk you through the calculation. –  André Nicolas Jun 9 '12 at 0:12

1 Answer 1

up vote 2 down vote accepted

$x=m-\mu$ follows a normal distribution $\mathcal N(0,C/n)$. I don't think there is a closed form for the probability that its norm is at least $d$, but I don't think this is the right statistical test either. You probably should transform your data so that $C$ becomes a unit matrix, and then $\sum_i x_i^2$ will follow a well-known $\chi^2$ distribution: you will easily find a $p$-value to test your hypothesis.

Edit: In the univariate case, this is of course much easier: $$P(\|x\|\le a)=\mathrm{erf} \frac{a}{\sqrt{2C/n}}$$

share|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.