I clearly remember reading (can’t recall where) that we should use population standard deviations (if both sample and population std dev are available) to calculate Sample correlation. Could any of you confirm this?

if you know the population standard deviation, why are you taking a sample?

I have the sample covariance. I have population standard deviations. I also have sample standard deviations. But I do not have the sample or population correlation. I do not have population covariance. I wish to find out the *(linear) relation* between the two samples. So I need the correlation to find out the correlation. Does this make sense or am I missing something?

My point is this…if you are able to know the population std. deviations, you should be able to know everything else about the population. Sounds to me like you have been given a few variables, and are trying to solve for another, but realistically speaking this particular situation will never arise. If you wish to estimate the correlation, use the “known” population std. deviations, but realize that in practice you will either never know the population std. deviations, or if you do, you will have no need for a sample, and therefore, this type of problem will never exist.

Cool… I get your point and was wondering the same myself when I read the text. But truth is that I can’t really recall where and in what context I had read this. Today when I was reading something else, the formula came up and they did not discuss the population volatility at all. So I checked up a few more sources which all used sample std. deviations. And hence the post. Thanks for the confirmation btw.

I have two fair dice except that they always seem to roll doubles when I play backgammon. I think they are psychically connected. To test this I throw the dice 10,000 times and get the following data… I know population means and s.d.'s but not the correlation (which isn’t the right test to do here anyway, but…)