Date:         Sun, 5 Oct 2008 10:04:15 -0700
Reply-To:     nchapinal@YAHOO.COM
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         nchapinal@YAHOO.COM
Organization: http://groups.google.com
Subject:      t-test vs. Wilcoxon when samples are small.
Comments: To: sas-l@uga.edu
Content-Type: text/plain; charset=ISO-8859-1

Hi,

I have a too simple a dataset that I do not know how to analyze it.
I ran a trial with 9 groups of pigs (that is, 9 pens). 3 groups were
under treatment 1 and 6 groups under treatment 2. So, n=3 and n=6. The
variables I want to study are mean number of aggressions per pig an
hour, percentage of the aggressions that are around the feeder,
percentage of aggressions with clear winner, and percentage of
aggression of high intensity.

If I had a bigger n, I would plot the residuals to see if the data is
normal, and the variance homogeneous and they I would decide what kind
of test or transformation I should use. In theory, frequencies and
proportions are not normal data.

However, which such a small sample, I do not think it makes sense to
plot the residuals.

Should I be using t-test or Wilcoxon (proc npar1way wilcoxon)?
If I should be using proc npar1way, what p-value should I be looking
at? The p value for the Kruskal Wallis test?

Thanks a lot!