0
$\begingroup$

From a sample population lets say 1000 numbers, I know the standard deviation SD. I take the size of the groups n=100 and fill the groups randomly with those 1000 numbers. I find out the average of those groups: a0,a1,…,a10

I want to find out (or approximate) a new standard deviation taking now as a sample the averages of those groups of 100 numbers: a0,a1,…,a10

The data I know at the start is the standard deviation of the whole population of 1000 numbers, and the size 100 of those groups. I need to find out the new standard deviation of the sample a0,a1,…,a10 without knowing ai values

I have tried dividing the standard deviation of the full population by sqrt(n) and I get close, is there a more precises way?

$\endgroup$
2
$\begingroup$

I think you are doing it correctly by dividing the sample standard deviation by $\sqrt{n}$, however, there is a reason why the number is not exactly what you expect.

As the Central Limit Theorem says, if you have a population, independently of its distribution (so, not necessarily normal), the average of a sample will be distributed as a normal distribution. That means that if you sample an infinite number of 100 groups and compute the mean of each sample, these mean values will follow a normal distribution. The standard deviation of the mean estimate (that is, the standard error) will be indeed the average divided by $\sqrt{n}$.

However, if you don't get closer to the expected number, it may be a sampling issue. In that case, your means (and standard deviations) are almost normal, but not quite, and its computation is subject to error. This is precisely what Student's t-distribution models, so that's a way of narrow down your estimates with confidence intervals. Or, increase the number of samples.

$\endgroup$

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.