2
$\begingroup$

Suppose, I am generating data from bivariate normal distribution with means 0 and variances 1 and correlation coefficient 0,0.5 and 1. I need to calculate power for different correlations. I am doing this simulation in R but when I change my correlation coefficient to 0.5, my power is always one. I think I am missing something here. For correlation coefficient 0, i get my power to be around 0.05 which makes sense but for others it is not.

$\endgroup$
  • $\begingroup$ try r = 0.1 see if it give you the power between 0.05 and 1. maybe r = 0.5 results in power is nearly 1. $\endgroup$ – user158565 Oct 24 '18 at 15:45
  • $\begingroup$ Actually I just ran 500 simulations for 0.1 and 0.2. With 0.1 the power is 0.858 and with 0.2 it is 1. There must be something wrong here. $\endgroup$ – Hello Oct 24 '18 at 16:44
  • 1
    $\begingroup$ sample size? Maybe the sample size is so large such that tiny difference from 0 will be detected. $\endgroup$ – user158565 Oct 24 '18 at 16:46
  • $\begingroup$ Maybe, I am using sample size of 1000. $\endgroup$ – Hello Oct 24 '18 at 16:47
  • 1
    $\begingroup$ When sample size > 10000, everything is significant, cited from a professor.So decrease the sample size to 100, to prove that your program is correct. $\endgroup$ – user158565 Oct 24 '18 at 16:51
2
$\begingroup$

The standard error of a Spearman sample correlation under the null of no correlation will be about $1/\sqrt{n}\approx 0.0316$ and the null distribution of the correlation is approximately normal. A correlation near $0.5$ will be about $15.8$ such standard errors from $0$; i.e. with such a large sample size giving a large standardized effect size, you expect to have power essentially at $1$.

$\endgroup$

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.