0
$\begingroup$

Someone asked me questions about his outcomes from Xlstat. Since I do not know this software at all, here is the question:

How are p-values for Pearson correlation coefficients calculated ($H_1: \rho \neq 0$)? Does Xlstat just calculate Steiger's z-test?

I can not find this info. The Xlstat help does not say anything about that.

EDIT: Or does it do a permutation test? I guess not??? EDIT2: The question is the same for Spearman correlation

$\endgroup$
  • $\begingroup$ Ups, thanks @Nick Cox for correcting my numerous typos! :-| Maybe I had too much coffee :-( $\endgroup$ – Michael Mar 4 '14 at 10:45
  • $\begingroup$ Note that the correlation being non-zero is usually the alternative hypothesis. I edited that. $\endgroup$ – Nick Cox Mar 4 '14 at 11:24
  • $\begingroup$ @ Nick: Yeah, another typo... Sorry for being so sloppy :-( $\endgroup$ – Michael Mar 4 '14 at 13:08
1
$\begingroup$

Answers at several levels spring to mind here, for your friend in some cases. I have never used Xlstat and have only a vague idea of what it is. However, the generic question here is of some interest and concern.

  1. Don't use any software which does not allow this kind of question to be answered. As I understand it, even Excel itself allows calculations of correlations, although I haven't tried examining their documentation. If this seems a counsel of perfection, consider that high quality statistical software is available for free in the form of R.

  2. Any company selling software has an obligation to its customers to provide technical support; quite what that includes or excludes will vary, but asking the company will be a test of their quality and seriousness. Depending on the age and style of the software, online help may be supplemented with printed manuals.

  3. A way of asking is to provide worked examples of small dataset(s) and the correlations and P-values that Xlstat provides. Then others might be motivated to compare with results from other software.

  4. As I understand it, Steiger's z test is for a highly specific correlation problem; I guess, but I guess confidently, that use of such a test is highly unlikely in a program like this.

$\endgroup$

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.