Re: [R] Help: partial.cor significance test

From: John Fox <>
Date: Wed 26 Oct 2005 - 22:15:33 EST

Dear Simon,

The population partial correlation rho[12|3...p] is 0 when the regression coefficient beta[2] for x[2] from the regression of x[1] on x[2] ... X[p] is 0. Thus, the usual t-test for a regression coefficient also tests that the partial correlation is 0.

Now, the sample partial correlation r[12|3...p] = t/sqrt(t^2 + dfe)) where dfe = n - p is the degrees of freedom for error and t is the t-statistic for testing that beta[2] is 0, and thus t = sqrt(dfe*r^2[12|3...p]/(1 - r^2[12|3...p])), so it is easy to compute the (unsigned) t-statistics from the partial correlations.

Why one would want to do this, however, is another matter. What would one do with a matrix of 2-sided p-values?


John Fox
Department of Sociology
McMaster University
Hamilton, Ontario
Canada L8S 4M4

> -----Original Message-----
> From:
> [] On Behalf Of sp219
> Sent: Wednesday, October 26, 2005 5:09 AM
> To:
> Subject: [R] Help: partial.cor significance test
> Hi,
> I have been using the partial.cor function in Rcmdr but I was
> wondering if there is any easy way to get statistical
> significance tests (two tailed) along with the partial
> correlation coefficients?
> Simon Pickett
> Simon Pickett
> Centre for Ecology and Conservation Biology University of
> Exeter in Cornwall Tremough Campus Penryn Cornwall TR10 9EZ UK
> Tel: 01326371852
> ______________________________________________
> mailing list
> PLEASE do read the posting guide!
> mailing list PLEASE do read the posting guide! Received on Wed Oct 26 22:33:37 2005

This archive was generated by hypermail 2.1.8 : Thu 27 Oct 2005 - 00:38:54 EST