Date: Fri, 10 Sep 2010 18:16:16 +0100
Reply-To: John Whittington <John.W@MEDISCIENCE.CO.UK>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: John Whittington <John.W@MEDISCIENCE.CO.UK>
Subject: Re: ESTIMATE statement in Proc MIXED
In-Reply-To: <81F8139F381BE844AE05CA6525FF2AAE01EDC18D@tpwd-mx9.tpwd.sta
te.tx.us>
Content-Type: text/plain; charset="us-ascii"; format=flowed
At 11:39 10/09/2010 -0500, Warren Schlechte wrote:
>I agree. There are situations where the LOCF could have merit.
>
>One caution. If you can, you might want to use LOCF with some
>variability (i.e., normal noise around a reading, variance of all
>upper/lower LOCF, etc.). Otherwise you could end up reducing the
>overall variance in the data and falsely increasing the sensitivity to
>detect treatment effects.
Indeed so - the danger you describe is one of the (several) concerns that
is often raised in relation to LOCF. As you say, introducing some noise
seems like a reasonable way to combat that problem - and I would imagine
that it is probably best to base the magnitude of that noise on the
remainder of the dataset.
I suppose my biggest problem is that I usually function in the atmosphere
of a highly regulated environment (overseen by drug regulatory agencies),
who love to be able to 'tick boxes' in relation to the use of 'standard
techniques'. Whilst one can often come up with 'clever (and reasonable)
ideas' as to how to handle situations like those we've been discussing, it
is often an uphill struggle to convince that such approaches are
acceptable, no matter how 'reasonable' they are!
Kind Regards,
John
----------------------------------------------------------------
Dr John Whittington, Voice: +44 (0) 1296 730225
Mediscience Services Fax: +44 (0) 1296 738893
Twyford Manor, Twyford, E-mail: John.W@mediscience.co.uk
Buckingham MK18 4EL, UK
----------------------------------------------------------------
| 
