2
$\begingroup$

I have a list of values $S$. I want to compare the Spearman correlations of $S$ with $N$ other lists of values $O_k$ ( $k∈[N]$ ). The lists $O_k$ are not independent. Each value $O_k^j$ ( $j∈[|O_k|]$ ) in $O_k$ is correlated with the corresponding value $O_{k+1}^j$ in $O_{k+1}$.

If the lists were independent, I assume I could use standard multiple comparisons corrections to adjust my $\alpha$ for significance testing - but my intuition is that this would be too conservative given the dependences between the lists. Is there a suitable method for setting my $\alpha$ in this case? Does it depend on the nature of the dependence between the $O_k$s?

I found this related question: Comparing two dependent Spearman's rank correlation coefficients but I'm not sure the answer helps me (especially as I can't get behind the linked article paywall...)

PS: apologies for any sloppy notation......

$\endgroup$

1 Answer 1

1
$\begingroup$

Bonferonni correction can indeed be too stringent. Instead you can construct a null-distribution of correlation values between your variables, and look whether you can reject the null-hypothesis when you randomly swap them. Check the following https://youtu.be/I0qoP7yvP-4

$\endgroup$
1
  • $\begingroup$ Thanks @c0s1n3! Thats the sort of thing I was looking for and makes sense. $\endgroup$ Jul 24, 2019 at 13:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.