Date:         Wed, 22 Oct 2003 11:21:04 -0400
Reply-To:     Peter Flom <flom@NDRI.ORG>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Peter Flom <flom@NDRI.ORG>
Subject:      Re: Missing data in Proc Mixed
Content-Type: text/plain; charset=US-ASCII

At 07:49 22/10/03 -0600, William Dudley wrote:
<<<
I am responding to a reviewer about the use of proc MIXED in a repeated
measures ANOVA with treatment and control.

We think that we are employing an Intent To Treat method because Proc
mixed does not employ listwise deletion as does proc GLM.  However the
reviewer thinks that Proc Mixed must be substituting values.  I notice
that most authors simply say that because Proc Mixed using ML
estimation,that it employs all of the data and cases are not eliminated
if they
provide at least one data point.
>>>

I think the reviewer is correct, at least in substance if not in
detail.

As far as I know, Intent to treat refers to the design of the study,
that is, you assign people to Treatment or Control, and
then, even if someone in the Treatment group doesn't 'do' the
treatment, he/she is still part of the Treatment group.  Thus,
you'd be measuring 'what really happens' instead of 'what the treatment
does'; it seems to me to be an attempt to
increase external validity at the cost of internal validity.

What PROC MIXED does is different, and is, indeed one of the strengths
of HLM.  It uses all the data that's available.  Thus, if you schedule 4
followups, and a person goes to only 1 of them, it uses the information
from that one followup.  Whether this means it 'substitutes values'
depends, I suppose, on what you mean by 'substitutes values'; but I
don't think it could be called an 'intent to treat' model.

HTH





Peter L. Flom, PhD
Assistant Director, Statistics and Data Analysis Core
Center for Drug Use and HIV Research
National Development and Research Institutes
71 W. 23rd St
www.peterflom.com
New York, NY 10010
(212) 845-4485 (voice)
(917) 438-0894 (fax)