Date:         Tue, 29 Apr 2003 09:27:31 -0500
Reply-To:     Warren Schlechte <Warren.Schlechte@TPWD.STATE.TX.US>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Warren Schlechte <Warren.Schlechte@TPWD.STATE.TX.US>
Subject:      Correcting for Overdispersion in Group without Replication
Content-Type: text/plain; charset="iso-8859-1"

Here's an odd question, caused by an odd design.  Assume that the unbalanced
problem arises from missing values.

The design has 3 groups of one treatment and 1 group of a comparison
treatment.  The outcome is a percentage.  So, for example, we could get:

Group           Result
A               30%
A               40%
A               20%
B               50%

How could one test whether the results in group B are consistent with those
in group A?

As I see it, the most empirical way of approaching this problem would be to
assume that the variance about group A (with replicates) is the same as the
variance about group B (without replicates).  However, I am unsure how to
code this?

Any help on how I specify that I want to use the variance about group A to
test the null hypothesis that A=B, when the outcome is a percentage would be
appreciated.  Or, if you disagree that this is the most reasonable approach,
please send your suggestions about how you would approach this problem.

Please note, I do understand that the lack of replication is a problem, and
would prefer to have multiple B groups.  However, I also note that under
GLM, when data are normally distributed, the assumption of equal variance
allows such an analysis to proceed without even producing a warning.

Thanks,


Warren Schlechte
HOH Fisheries Science Center
HC7 Box 62
Ingram, Texas 78025
Phone:   830.866.3356
Fax:       830.866.3549