Date: Wed, 11 Apr 2007 14:10:49 -0400
Reply-To: Sigurd Hermansen <HERMANS1@WESTAT.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Sigurd Hermansen <HERMANS1@WESTAT.COM>
Subject: Re: Identifying a sample design
In-Reply-To: <1176306820.331452.262680@d57g2000hsg.googlegroups.com>
Content-Type: text/plain; charset="us-ascii"
Catherine:
I think that we are getting the idea that you don't actually have much
of a sample design! Assuming that a random sample from your original
data sources gave almost every specialist a equal chance of selection
into the sample of specialists, I'd say you have an very simple but
adequate sample design at that level. Unfortunately the two-stage sample
design that you described earlier makes things more complex. Specialists
have likely selected convenience samples of ten patients.
You really have two choices: finesse the sample design issues and
attempt to specify a covariance structure that takes into account
correlations among observations and specialists, or hire a good
statistician as a consultant on both sample design and estimation
issues. In other words, you may be able to develop a useful explanatory
or predictive model that adjusts for biases due to sample selection and
for covariances across patients of the same specialist, but you can't
then claim to have representative estimates for a population.
Many of the real sampling and statistical modelling experts on the 'L
are making their way to Florida for the SGF. In any event I'd expect
that you will need to explain what you hope to do with your sample
before sample design experts on the 'L could offer advise. You might
start with what outcomes you would expect to see related to the ranges
and distributions of covariates by specialist.
S
-----Original Message-----
From: owner-sas-l@listserv.uga.edu [mailto:owner-sas-l@listserv.uga.edu]
On Behalf Of cat..
Sent: Wednesday, April 11, 2007 11:54 AM
To: sas-l@uga.edu
Subject: Re: Identifying a sample design
Does anyone have an idea ?
Thanks.
Catherine.
| 
