Date:         Mon, 27 Jun 2005 20:55:52 -0800
Reply-To:     David Neal <afdbn@UAA.ALASKA.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         David Neal <afdbn@UAA.ALASKA.EDU>
Subject:      Re: Friedman test
In-Reply-To:  <BAY21-F1571E24582C7453E7C515093E10@phx.gbl>
Content-type: text/plain; charset=ISO-8859-1; format=flowed

Isn't the assumption of normality based on the errors and not your DV?
Try running the ANOVA model and looking at the residuals.  Chances are
fairly good that your residuals will be close enough to being normally
destirbuted to call it good.  If not, you should probably try a
transformation.

If I remember Friedman's test, it is a nonparametric test for the
difference between treatments in a randomized complete block design.  I
don't think you have a block design here.

David Neal


Laughing Beggar wrote:
> Hi all,
> I have one non-normal continuous dependent variable (DV) and 2 categorical
> independent variables (one is ordinal, the other nominal). If my DV were
> normally distributed Id do a 2-factor ANOVA. It aint, so it looks like I
> have to do a Friedman test. What's the best way to have SAS do this ?
> Cheers
> L_B
> SAS V8.02
> WIN 95 or XP
>
>
> "The beggar laughs in the face of the thief"