Date: Fri, 30 Sep 2005 16:33:28 +0200
Reply-To: Marta García-Granero
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Marta García-Granero
Organization: Asesoría Bioestadística
Subject: Re: proportions and ANOVA
Content-Type: text/plain; charset=ISO-8859-15
ZB> Hello, I would like to test a yes/no question in an ANOVA design (2 btw
ZB> subject IVs and one within subject IV). Can ANOVA handle proportions
ZB> even those the assumptions are likely to not hold?
Your dependent variable: is it the binary no/yes answer for each
subject or the proportion of people answering yes?
If it's the proportion (a given data cell is formed by the proportion
of people), you can try to normalise the distribution (binomial) using
and arcsin transform, although the interpretation of the results is
more complicated (in term of effect sizes) and back-transforming the
summary statistics (means and SD or SE) is not possible.
If the DV is the binary answer, then you'll have to rely on ANOVA
robustness (you don't mention the sample size, an important issue
here, because parametric methods tend to be asymptotically robust).
If you only had a WS IV, you could have used Cochran test, instead.