Date:         Mon, 31 Jan 2005 22:01:35 -0500
Reply-To:     Richard Ristow <wrristow@mindspring.com>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Richard Ristow <wrristow@mindspring.com>
Subject:      Effect-size measure in crosstab
Content-Type: text/plain; charset="us-ascii"; format=flowed

It's shameful not to know this after being around so long, but what are
recommended effect-size measures for categorical-by-categorical
comparisons - "crosstabs"?

I've got some where the significance by chi-square p<.001; but the
sample size is large, and, at least by eye, the effect is not
remarkable.

I've thought of using reduction in Shannon entropy -- i.e., Shannon
information -- as a measure.(1) It is sample-size independent, and
tests pretty much what chi-square tests: that knowing one of the
variables, tells you anything about the other. I don't think it's a
symmetric measure. And if it's included in the symmetric or asymmetric
statistics available with SPSS Crosstabs, I don't recognize it.

Advice much appreciated!
Richard Ristow

(1) If there are k possible events and even i occurs with probability
p(i),
Shannon entropy = -sum(i=1 to k) p(i)log2(p(i)).