5
$\begingroup$

I have two correlations (genetic correlations actually), their standard errors, and their z-statistics (but I don't know the sample sizes). Is it possible to do a significance test for the difference between these (genetic) correlations?

$\endgroup$
5
+100
$\begingroup$

Yes, this is possible. The reason is because if you have $r$ and $SE_r$, you do indeed have the sample sizes: $$SE_r = \sqrt{\frac{1-r}{n-2}}$$ or $$n = \frac{1-r}{SE_r^2}+2$$

In your OP, you mention "z-statistics". I believe you are referencing the Fisher's z-transformation (as if you were referring to a test statistic for the correlation, it would be a t-ratio). This is important, because the Fisher transform is needed for the hypothesis test to compare the two correlations: $$H_0 : \rho_1 = \rho_2$$

To conduct the test, transform the correlations to Fisher z: $$z_i = \frac{1}{2} \ln \left(\frac{1+r_i}{1-r_i}\right)$$ Using the standard error for the differences: $$s=\sqrt{\frac{1}{n_1-3} + \frac{1}{n_2-3}}$$ you can use the standard normal distribution to find the P-value for $$\zeta = \frac{z_1-z_2}{s}$$ (This last value is just a z-score, but as we've used z for something else, I called it zeta instead.)

$\endgroup$

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.