home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!paladin.american.edu!auvm!IRLEARN.UCD.IE!RCONROY
- Message-ID: <STAT-L%92090204593206@VM1.MCGILL.CA>
- Newsgroups: bit.listserv.stat-l
- Date: Wed, 2 Sep 1992 09:42:46 GMT
- Sender: "STATISTICAL CONSULTING" <STAT-L@MCGILL1.BITNET>
- From: Ronan M Conroy <RCONROY@IRLEARN.UCD.IE>
- Subject: polls, confidence, point estimates
- Lines: 24
-
- Geoff Selig is right when he points out that there is a region around
- a survey result representing out imprecision. The true value that we
- are trying to measure has an X% (usually calculated as 5%) chance of
- being contained in that region. But we sometimes act as if the
- true value had a uniform chance of being anywhere in the region.
- In fact, the observed value is our single best guess as to the
- true value, and the values nearest to it are more likely than the
- values far away. In the case he mentions, we should calculate the
- difference between the two proportions and IT'S confidence interval.
-
- To get a 4% error margin (+/-) 95 times out of a hundred, sampling from
- a large population, you need a sample of 600. If the samples were actually
- 600 in each poll, the difference between the two percentages is 8% with
- a confidence interval of 2.4% to 13.6%. So the change in Clinton's
- popularity is likely to be somewhere between 2.4% (trivial) and
- 13.6% (uh-uh time).
-
- Gee - sometimes I think I'll do ANYTHING in the mornings except work.
-
- ronsn
-
- (eep! can't even get the name right...)
-
- ronan
-