It is my understanding that the confidence interval for a Pearson correlation is asymmetric. (Confidence interval for correlation, for example.) r’s command cor.test does give upper and lower bounds that are unequal distances from the statistic estimate. The documentation describes that cor.test uses Fisher’s z transformation.

I need uncertainty for a weighted correlation - hundreds of them, actually. The only weighted correlation command in r I have found so far that provides any uncertainty is wtd.cor. It reports a single standard error for the correlation. In the documentation I do not find a description of how the standard error is derived.

  1. Is there an r command I’m overlooking or some other simple solution?
  2. If not, is there a better option than to write code to create bootstrapped confidence intervals? For example, is there some reasonable (and check-able) guess about how the standard error was derived so that I could back-transform it using Fisher’s z?

I'm not sure whether this question belongs in statsExchange or stackOverflow - seems which one fits best could depend on the answer.



Your Answer

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

Browse other questions tagged or ask your own question.