Date:         Tue, 8 Apr 2003 15:02:34 -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: Statisticians (was S.O.S. (Save Our SAS))
Content-Type: text/plain; charset=US-ASCII

The problem here, of course, is that there are huge areas of research
(including almost all the work we do where I work)  which simply cannot
be done with random samples, because there is no sampling frame.  e.g.,
work with drug users, or sex workers, or men who have sex with men,
etc.

So, when we use convenience and/or snowball and/or targeted sampling,
what have we got?  It's more than nothing, but it's hard to say exactly
how much more than nothing it is.  Generally, when research like this
gets written up (our work and others), no EXPLICIT inference to a
general population is made, and the limitations section includes a bit
about the problems of the sample.

OTOH, the research DOES add something to the knowledge pool.  But it
makes the usual things like p-values even murkier than usual. But if you
try to submit something WITHOUT p-values, the journal editor says "where
are the p-values?"  ah well.....

Peter

Dale  wrote
<<<
Let's back up a bit further.  What is the population for which
you wish to make inference?  What is your sampling frame?  Does
your sampling frame represent the population on which you wish
to make inference?  How were samples drawn?  Oh, you used
convenience sampling?  Sorry, that does not represent any
specifiable population.  Come back with an experiment which
employs proper sampling methods!
>>>