=========================================================================
Date:         Thu, 13 Jul 2006 00:29:18 +0200
Reply-To:     Karl Koch <TheRanger@gmx.net>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         Karl Koch <TheRanger@gmx.net>
Subject:      Which test?
Content-Type: text/plain; charset="iso-8859-1"

Hello group,

I have performed a simple experiment (and got the data). Now it is time to find out if the results are stat. significant. I would prefer a non-parametric test since less assumptions are required to hold.

I have one IV (system) with two possible categories (system A and system B). My null hypothesis is that both systems cause the same effect on participants. The outcome variable are a number of relevance ratings (on a 6 point scale S={1,2,3,4,5,6}) from users that used system A and a number of relevance ratings from the same users that used system B. The number of relevance ratings users did was up to them and can be different for system A and system B. An extract of the data looks like this:

User   System   Rating
1      A        2
1      A        1
1      A        3
1      B        4
1      B        6
2      B        5
2      B        6
2      A        2
2      A        1
2      B        6
...


It is important to point out that half of the users have first used System A and then system B to solve the task. The other half of the users have done the opposed order.

Since both systems have been counterbalanced, I would assume that the Mann-Whitney test should be my choice.

However, I also have used the same people on the test and that would formally ask for a Wilcoxon Matched Pairs Test.

I would like to know what the experts here think about that.

Best Regards,
Karl
--


Der GMX SmartSurfer hilft bis zu 70% Ihrer Onlinekosten zu sparen!