Date:         Fri, 30 Dec 2011 20:28:48 -0500
Reply-To:     R B <ryan.andrew.black@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         R B <ryan.andrew.black@GMAIL.COM>
Subject:      Re: group matched regression
In-Reply-To:  <CAHCrizOd6Bx4t_Vf9nvQykErdH6ZBEDkiHFFBLk=Lt9ohv_o2g@mail.gmail.com>
Content-Type: text/plain; charset=ISO-8859-1

Kevin,

It's certainly possible to account for correlation among "matched" subjects
by specifying a RANDOM statement as follows:

random int / subject=matched_group_id;

where "matched_group_id" is a variable that indicates which group each
subject is matched to.

Ryan

On Wed, Dec 28, 2011 at 1:20 PM, Citam <citam.sasl@gmail.com> wrote:

> I have matched subjects based on a series of variables.  Each
> resulting group must have at least one subject from the test group and
> one from the control group, but most have multiple patients.  Does a
> random-effects model seem the most appropriate statistical tool for
> such data?  Besides the matching, this should be a conventional
> analysis, i.e. no stratified sampling.
>
> For example, I was considering using the MIXED procedure with the
> RANDOM statement.  I would appreciate any comments or references.
>
> Thank you,
>
> Kevin
>