1
$\begingroup$

Is there a recommended n to test for Spearman's rank correlation? I have 2 lists of around 10,000 items. I read that Spearman's rank correlation is fine between 10-30 , can I use spearman to compare ranked list of 100 or even 1000?

Thanks

$\endgroup$

1 Answer 1

4
$\begingroup$

Certainly you can. The only onerous part is computing all the ranks, but that's still pretty easy with ten thousand observations.

The questionable part would be why you'd want to test with 10000 pairs. Surely with almost any kind of non-artificial data, you're going to reject - few correlations are so small that they won't fall outside the limits for that. For example - at those sort of sample sizes you'll reject a correlation of around 0.02 as being significantly different from 0!

Surely the size of the correlation is more meaningful than whether you can reject it at such a big sample size?

$\endgroup$
2
  • $\begingroup$ OK. If the list is so long as I mentioned 10,000 or even 20,000 ,and when I compare the ranking of 30 items there is no spearman correlation , does it make sense to compare the ranking of 200 or 300 items? or usually users care about only top ranked items? $\endgroup$
    – tnaser
    Commented Mar 11, 2013 at 15:26
  • $\begingroup$ I'm not sure what you're asking. If you have thousands, why only relate a few hundred (unless those are important a priori)? I can't speak to what users 'care about', doubly so when I have no context at all. $\endgroup$
    – Glen_b
    Commented Mar 11, 2013 at 21:19

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.