Hi all!

Just a quick post to ask if any of you have experience with the coefficient of concordance, Kendall's W. It seems to be an equivalent of the more 'standard' rank correlation coefficient Spearman's rho, but you can use it for comparing (or: testing the level of concordance between) more than two subjects at once.

I had problems in the past having to compute corr coefficients between more than two raters. Nobody told me of the Kendall'W and I did what many studies do: computed Spearman for each pair combination of raters and then averaged it. The problem is, in doing so I got often non significant coefficients.

Using the Kendall's W (I checked right now) I see that I get a overall level of correlation similar to the average of all Spearmans, but with highly significant p value!!

My questions:

- do you know if Kendall'W is an .."authoritative" test? i.e. if it is something well-known and seen as a good stat? (because I only found it in 1 article so far)
- Am I cheating if I use Kendall's W and claim I have correlations? (that's a general issue, I'm new to stats and always impressed by how you can get different 'results' by changing tests!
- I've found two studies that don't use Kendall's W, rather do the other way - averaging corr coefficents of all possible pairs. BUT: they don't report the p values! I've asked my sup and he only said that's odd, since you should always report p values. I've written to one of the authors (maybe unpolite on my side?) to ask about the p values..no answer.

Don't want to waste your time, but in case you have some knowledge / experience with correlations and Kendall's W and the p values issues, it would be great if you could help me illuminate the topic.

Thanks a lot!

*BUMP* Can anyone help Wanderingbit?

I don't really know anything about Kendall's W but a quick search on google scholar has come up with quite a lot of articles that have used it. I'm pretty sure it's well known as it's offered in SPSS :)