|Date: ||Sat, 24 Jan 2009 18:55:25 -0500|
|Reply-To: ||Peter Flom <firstname.lastname@example.org>|
|Sender: ||"SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>|
|From: ||Peter Flom <peterflomconsulting@MINDSPRING.COM>|
|Subject: ||Re: Analysing data from observational studies|
|Content-Type: ||text/plain; charset=UTF-8|
"cat.." <cat.b41@GMAIL.COM> wrote
>I'd like to get feedback from those of you who have experience in
>analyzing data from observational studies.
>Do you take into account the sampling design or do you just analyze as
>if it was a simple random sampling design ?
That depends ... often, there's no way to account for the sampling design in an observational study. Sometimes there is.
>Taking into account the sampling design requires to know already some
>statistics about the target population. E.g, in a design stratified
>let say by geographic zones, estimating properly the mean, std, ...
>requires to know the repartition of your subjects in the different
>regions on the target population.
>But how about when you have no idea about that because the target
>population is very particular, eg: Patients who are administered a
>given drug XX ?
>Is analysing these data as if they came from a simple random design
>the only way to proceed ?
OTOH, the problem with analyzing such data is not the analysis itself, but the question of generalizing. So, if you have, as I often did, a convenience sample of injection drug users from the streets of Brooklyn, one can be somewhat confident about generalizing to street IDU in Brooklyn, but a little leerier about, say 'urban drug injectors' and much leerier about e.g. all drug injectors
Peter L. Flom, PhD
www DOT peterflom DOT com