Date:         Mon, 30 Mar 1998 08:08:40 -0500
Reply-To:     Mike Davenport <Mike.Davenport@RICHMOND.PPDI.COM>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         Mike Davenport <Mike.Davenport@RICHMOND.PPDI.COM>
Subject:      Re: AB/BA design
Comments: To: Miss LS Smoljanovic <ehmasls@CS6400.MCC.AC.UK>
Content-Type: text/plain; charset=us-ascii

The model for a 2x2 crossover design is:
 
y = sequence + subject(sequence) + treatments + periods
 
you must use the subject(sequence) term as the error when testing for
sequence effects. A significant sequence effect ( p < 0.10) would indicate
the potiential presence of a carry-over effect. Since the estimate of the
treatment effect contains a sequence component this is an important test.
Period effects can be examined directly (but its a nusiance parameter, so i
don't know why you are interested). Treatments can also be examined
directly (assuming you don't have a sequence effect).  GLM or MIXED will
provide you with the answer you require. i use GLM unless i want to compute
confidence intervals in which case the output tables from MIXED are useful.
 
 
 
mike
 
Miss LS Smoljanovic wrote:
 
> Dear SAS-L members,
>
>   I would like to aske for advice and a little help.
> The problem is following:
>
> a) cross-over trial (design AB/BA)
> b) interested in treatment and period effect among and between subjects
>
> What is the best SAS procedure GLM or MIXED?
> How can I get the results using SAS code for b) part of the problem?
>
> Thanks in advance
>
> Lada