An embedded and charset-unspecified text was scrubbed... Name: not available URL: <https://stat.ethz.ch/pipermail/r-help/attachments/20110505/7a15bb1e/attachment.pl>
cross-correlation table with subscript or superscript to indicate significant differences
2 messages · yoav baranan, David Winsemius
On May 5, 2011, at 12:40 PM, yoav baranan wrote:
Here is an example for my earlier question.
Say you have a 3x3 correlation matrix:
corrs <- matrix(c(0.25,0.32,0.66,0.14,0.24,0.34,0.44,0.34,0.11),
nrow=3, ncol=3, dimnames = list(c('varA','varB', 'varC'),
c('varA','varB', 'varC')))
And another matrix for the sample size of each correlation:
sizes <- matrix(c(44,68,313,142,144,207,201,100,99), nrow=3, ncol=3,
dimnames = list(c('varA','varB', 'varC'), c('varA','varB', 'varC')))
corrs looks like this:
varA varB varC
varA 0.05 0.14 0.44
varB 0.32 0.24 0.34
varC 0.66 0.57 0.50
sizes:
varA varB varC
varA 44 142 201
varB 68 144 100
varC 313 207 99
i.e., the correlation between variables A and C was 0.66 with sample
size of 313. (I got these tables from rcorr).
Why not offer the result of dput() on the result from rcorrs cone on 3 variables. Then we should have the necessary building blocks for your original request. This way we do not have the matrix of p-values. And as the Posting Guide clearly says ...Please post in plain text. In your case it is particularly annoying because I do not get the filtered version but rather you html version and the text is almost unreadable at a font size of 10 in whatever font it is specifying!
What I want to do is to compare the correlations in each row
(probably using r.test), and then create a correlation table with
subscripts or superscripts indicating the significance
"group" (again: correlations with different superscripts, in the
same row, are significantly different from each other).
Something like this:
varA varB varC
varA 0.05b 0.14b 0.44a
varB 0.32a 0.24a 0.34a
varC 0.66a 0.57ab 0.50b
At the moment we have no way of determining what values should be post- pended with which letters.
Of course, I don't have a 3x3 table. I have about 20 tables of at least 7x7 each, so this is why I'm looking for methods to automate the process. Thanks, Yoav
CC: r-help at r-project.org From: dwinsemius at comcast.net To: ybaranan at hotmail.com Subject: Re: [R] cross-correlation table with subscript or
superscript to indicate significant differences
Date: Thu, 5 May 2011 12:17:25 -0400 On May 5, 2011, at 10:48 AM, yoav baranan wrote:
Hi, I wonder whether the following is possible with R, and whether anyone has done that and can share his/her code with me. I have a correlation matrix, and I want to create a correlation table
that I
can copy to Microsoft Word with a superscript above each correlation, indicating significant differences in the same row. That is, when correlations in the same row do not share
superscript,
it means that they are significantly different from each other. thanks,yoav [[alternative HTML version deleted]]
An example with data and the desired result might help focus the discussion. This shows how to set up an example showing how extract the row numbers from a correlation matrix with absolute values above 0.5 but less than 1 (to exclude the trivial cases).
snipped
David Winsemius, MD West Hartford, CT