Date: Wed, 18 Dec 2002 14:14:10 -0800
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: "David L. Cassell" <Cassell.David@EPAMAIL.EPA.GOV>
Subject: Re: help on power analysis
Content-type: text/plain; charset=us-ascii
"Lai L." <laihere@YAHOO.COM> wrote:
> Hi, I am working on power analysis for a study sample design with
correlated observations and need your
> recommendations on useful referece books, websites or formula.
> Briefly, here is what our study design is like: we will recruit 20
health workers, each of whom will find
> "n" smokers in his community. Half of them will be randomly assigned
to receive certain treatment, half
> of them don't. At the end of study period, we will check if smokers
quit smoking in both groups. So the
> outcome will be binomial. And these "n" smokers will be correlated to
some degree. Our power analysis has
> to address this correlation problem.
You have a sampling theory problem rather than an
experimental design problem.
Rather than a 'power analysis', you need to be looking at
optimum allocation issues in sampling theory. You appear
to be looking at what would be some sort of multi-stage
sampling design, but you have a lot of potential problems
as you have stated the study. Please consult with a
statistician who has a sampling theory background, so you
can get some help on the sorts of response biases you may
introduce with this arrangement (I hesitate to call it a
'design'). You clearly have seen that there is a huge
risk of correlation, but you haven't addressed it as yet.
What if some health workers choose 'n' people who all work
together and live together, while some choose 'n' people
who are as widely separated as they can manage?
David Cassell, CSC
Senior computing specialist