home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: soc.motss
- Path: sparky!uunet!think.com!paperboy.osf.org!gr.osf.org!emcmanus
- From: emcmanus@gr.osf.org (Eamonn McManus)
- Subject: Re: Telephone surveys
- Message-ID: <hasardeux@kaa.gr.osf.org>
- Sender: news@osf.org (USENET News System)
- Organization: Open Software Foundation Research Institute, Grenoble
- References: <1992Nov16.183905.25643@macc.wisc.edu> <dialSforSurvey@kaa.gr.osf.org> <67@bbx.basis.com>
- Date: Thu, 19 Nov 1992 10:50:04 GMT
- Lines: 28
-
- scott@bbx.basis.com (Scott Amspoker) writes:
- [emcmanus:]
- >>First of all, how can you
- >>tell that your results are accurate to "within about three
- >>percent"? I see no way to check apart from other surveys with
- >>the same assumptions about accuracy.
- >One can never tell that a statistical sample is accurate within 3%.
- >One can mathematically determine that, given the population size and a
- >(truely random) sample, that there is a X% chance that
- >the survey is within Y% accuracy.
-
- I think what you are saying is that you can mathematically determine
- how likely your results are to have a certain accuracy relative to the
- results you would get if you surveyed the population as a whole. But
- my point is that there is no way of knowing how this compares with the
- true proportions. In some surveys this is not too important because
- you have some independent means of confirmation. For instance if you
- ask people how they are going to vote in an election, you can later
- compare your results with the actual outcome of the election.
-
- But this is particularly a problem when asking about homosexuality
- because we know lots of people (like a former you and indeed a former
- me) will lie; we don't know and can't know how many such people there
- are, short of taking readings on the hypothetical Kinsometer.
-
- ,
- Eamonn
- name no one man
-