Date:         Thu, 6 Sep 2007 11:55:52 -0700
Reply-To:     Paige Miller <paige.miller@KODAK.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Paige Miller <paige.miller@KODAK.COM>
Organization: http://groups.google.com
Subject:      PROC NPAR1WAY One-Sided Tests
Comments: To: sas-l@uga.edu
Content-Type: text/plain; charset="iso-8859-1"

I am confused by the description of the one-sided tests in PROC
NPAR1WAY for the case where we compare two means via simple rank
tests.

When I learned statistics, you specified a null hypothesis (in this
case, that the means of two groups are equal) and an alternative
hypothesis. If the alternative hypothesis was that the means of two
groups were not equal, then you had a two-sided test. If the
alternative hypothesis was that group 1 mean was less than group 2
mean, you had a one-sided test. There is another one-sided test that
can be specified in the opposite direction by having the alternative
hypothesis be that group 2 mean is less than the group 1 mean.

However, although PROC NPAR1WAY reports one-sided tests, it doesn't
let you specify which direction the alternative hypothesis points. The
documentation for PROC NPAR1WAY very unhelpfully explains that the
direction of the test depends on whether the computed statistic z is
greater than zero, or not. This makes no sense to me. The direction of
the alternative hypothesis does not depend upon the computed
statistic. Even forgetting that, they do not specify what alternate
hypothesis is tested when z>0, and I am reluctant to assume that the
alternate hypothesis tested is that mean of group 1 > mean or group 2
or vice versa, because I don't know how they pick which level is group
1, and they don't say that when z>0 they are testing group1 mean >
group2 mean (as it could be that they are testing group2 mean > group1
mean when z>0).

Someone please explain this to me. Thanks.

--
Paige Miller
paige\dot\miller \at\ kodak\dot\com