Date:         Tue, 22 Feb 2005 14:48:51 -0800
Reply-To:     Jason <falciparum@HOTMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Jason <falciparum@HOTMAIL.COM>
Organization: http://groups.google.com
Subject:      Re: Carry-over residuals (for a newbie)
Content-Type: text/plain; charset="iso-8859-1"

My interest is in accounting for the variance due to the prior
treatment; as I understand, this is called carry-over residual.  For
example with the design mentioned above (where columns are people, rows
are periods, and letters are treatments):

A B C
C D B
D C A

Person 1 gets trt A, then trt C, then trt D.  In the C treatment there
may be carry-over from A, and in the D treatment there may be carry
over from C.  Obviously there is no carry-over for trt A because
nothing came before it (as with B for person 2, and C for person 3).
The point of assigning the first row an arbitrary category 'O' is so
they can be grouped for a variance calculation.

Let's say the response variables are the following:

22 11 13
14 11 16
21 13 14

The data in SAS might look like:

Row Col Trt Resid Response
1      1    A   O      22
1      2    B   O      11
1      3    C   O      13
2      1    C   A      14
2      2    D   B      11
2      3    B   C      16
3      1    D   C      21
3      2    C   D      13
3      3    A   B      14

So no row got deleted; I wanted to be able to account for carry-over.
My question is: is this the right way to do it?  Essentially I have
made Carry-over a factor in my design.  So I will have reduced my MSE
degrees of freedom, but hopefully also reduced my MSE if carry-over has
an effect.  Am I approaching this the right way?

>If you have random effects, why are you using PROC GLM,
>instead of PROC MIXED?

Can I get mixed to print out a nice ANOVA table like glm?  Won't glm be
the same if my design is balanced and I use the RANDOM statement?  (BTW
it's not the above design.)

Thanks,
Jason