I'm currently studying 'Estimation of population correlation coefficient from sample correlation coefficient. Source: https://newonlinecourses.science.psu.edu/stat501/node/259/

Test statistic is like below.

$t = r*\sqrt(N-2)/(\sqrt(1-r^2)$

N is the size of sample and r is correlation coefficient of the sample. The statistic t is known to follow Student t-distribution.

According to the source above, it adopts two-tailed test with below null/alternate hypothesis.

H0: population correlation coefficient $\rho$ = 0

H1: population correlation coefficient $\rho$ != 0; $\rho$ > 0 or $\rho$ < 0.

To me, H1 above makes sense because I can't find any directional/sign information from the test statistic.

But today, my colleague said to me that I should adopt one tailed test. This is because we have a background knowledge that the population correlation coefficient should be always smaller than 0.

So, here're my questions.

  1. Does the test statistic above give me any information about sign of $\rho$?

  2. In case I have a background knowledge; let's say, sign of $\rho$ is always smaller than 0, then, which direction do I need to take with one-tailed test? I'm asking this because t-distribution is not symmetric around 0.

  3. In case I don't have a background knowledge, sign of $\rho$, then, what is the correct way to design H0 & H1? to me, two tailed one is the most tempting one because I don't need to care about sign, but it doesn't give me any information about correlation or anti-correlation.

Thanks in advance.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.