Date:         Wed, 30 Jul 1997 04:36:16 -0400
Reply-To:     Staffan Lindberg <slind@MHK.KS.SE>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From:         Staffan Lindberg <slind@MHK.KS.SE>
Subject:      Ex post facto comparisons between x-tab of nominal variables
Content-Type: text/plain; charset="us-ascii"

Could anyone help me with suggestions of an analysis method for the
following problem:
 
I have 4 groups (1,2,3,4 in the variable GROUP). I want to crosstabulate
this with a lot of other nominal variables (with number of values between 2
to 5). For the crosstabulations with significant Chi-squares I want to
pinpoint the groups that contribute the most towards the difference.
 
I guess I can do this by collapsing the group variable into several
dichotomous variables (covering all the combinations) and study the change
in Chi-square. This, however, seems both tedious and clumsy. Is there
another more general method ?
 
Lastly in some cases where the total Chi-square is non-significant, you can
see that the collapse of some categories, would result in a significant
Chi-square. I remember from my elementary statistics courses way back that
fiddling with your tables afterwards was severely frowned upon. Is this
still the prevailing attitude in the statistical community ?
 
Thankful for any comments
 
 
 
=====================================================
Staffan Lindberg
Karolinska Hospital
Magnus Huss clinic
S-171 76 Stockholm
SWEDEN
=====================================================
Ph. at work   Int+46 8 51772936
Ph. at home  Int+46 8 6470383
Mobile ph. Int+46 8 070-5436925
Fax  Int+46 8 326369
e-mail  slind@mhk.ks.se
=====================================================