Date:         Fri, 26 Aug 2005 10:31:04 -0500
Reply-To:     Sean Hurley <sean.hurley@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Sean Hurley <sean.hurley@GMAIL.COM>
Subject:      Re: MIANALYZE and Proc Mixed contrasts
In-Reply-To:  <BAY103-F8F2AB4B9D141836CAC4ABB0AB0@phx.gbl>
Content-Type: text/plain; charset=ISO-8859-1

On 8/25/05, David L Cassell <davidlcassell@msn.com> wrote:

> If you have the ESTIMATE results, then you ought to have a point estimate
> and a standard error.  (If you don't have both, then use ODS to get them.)
> Feed those in to PROC MIANALYZE using the DATA= option, and tell
> PROC MIANALYZE the name of the point estimate in the MODELEFFECTS
> statement, and the name of the error in the STDERR statement.
>
> Or have you already done that?  If it didn't work, what went wrong?
>

Thanks David. That's what I had originally tried and got various error
messages about variables not being in the dataset, etc. But your
suggestion made me decide to try again, and I found a way to get it to
work (I think). In case anyone else is interested, this is how I did
it:

The ODS output for the ESTIMATE results gives a file with the
following variables (and a couple of others that I've left out):

_imputation_ = imputation number
label = the label given to an estimate in the ESTIMATE statement
estimate = the point estimate for the contrast
stderr =  the standard error for the contrast

So to have MIANALYZE analyze the estimates for each contrast, you need
to use the following code (or something like it):

proc sort; by label;
proc mianalyze data=estoutput;
modeleffects estimate; stderr stderr; by label;
run;

I'll need to check the results for a couple of contrasts to make sure
it's doing everything correctly, but it seems like this should do the
trick.

--
Sean