On Fri, 22 Dec 2000 14:36:03 GMT, [EMAIL PROTECTED] (J. Williams) wrote: > The following article appeared on CNN.com outlining the VNS exit > polling errors. However, VNS believes the networks jumped the gun in > predicting Florida at first for Gore, then over to Bush. The future > of exit polling in close heats may be questioned more closely prior to > network "predictions." My local paper, the Pittsburgh Post Gazette, has an article by Howard Kurtz, with credit to the Wash. Post, which seems to be the reference document for that CNN report. The article mentions that neither VNS (who did the polling) nor the AP made the mistake at 2:16 a.m. I know from elsewhere that the first bad-call was by Fox, followed within 5 minutes by four other networks. At Fox < 'the Republican Network'>, the person in charge was John Ellis, George Bush's cousin who ran the 'call' desk . Has everyone already blamed Ellis too many times before, or was it a cowardly act of GOP-political-correctness, not to name him? > > December 22, 2000 > Web posted at: 5:35 a.m. EST (1035 GMT) > (CNN) -- An internal investigation by the polling organization that > incorrectly said Al Gore won the state of Florida on election night > concluded that its projections were plagued by errors all night long. [ ... ] > The networks and the Associated Press created VNS in 1990 as a > cost-cutting measure. CBS, NBC, ABC, CNN and Fox all relied heavily on > VNS data when they first projected Gore, then George W. Bush, the > winner in Florida. The networks were forced to retract both > projections after it became clear the candidates were separated by > only a razor-thin margin of votes. [ snip, concerning the problems at 7:50, call for Gore.] Here is some information that I asked about when I posted to one .stat. group a week ago. How bad was the 'finite sampling estimator' and how bad was the data? >From the article, Wash. Post as in the P.G. -- "At 2:20, with 97 percent of the state's precincts reporting, VMS estimated that there were 179,713 votes outstanding. In fact, more than 359,000 votes came in after 2:10. In Palm Beach County alone, VMS projected that there were 40,000 votes outstanding, but 129,000 votes came in." Well, 3% of the 6-million vote is about 180,000. It seems as if they were not taking into account that detail, that one county might have 500,000 votes while another has 5000. I read that some precincts came in with 50% extra voters. But that number, above, is 100% excess for the whole thing, and the other is over 200% for the Palm Beach numbers. Not a good job. Since votes are reported by precinct, the 'projections' were *apparently* based on that simple count of "97% of precincts" and nothing more precise. That is, "3%= 180,000" is a state-wide projection: based on state-wide total-vote, and the bare count of precincts. I am pretty sure that someone has done much better than that, in certain times and certain places -- I have witnessed much better jobs done in predictions. But now I don't know whether any really 'good job' was ever due to statisticians, or if the statisticians have, up to now, been rescued by the local experts. I have heard the analyses on election night for my own state of Pennsylvania, and they have been quite clever, for two different sets of races. They often use the current numbers for votes, but I have gotten the impression that the advice is fairly direct from some long-time pols. FOR PENNSYLVANIA (in particular). For the President: A Democrat has to build up a huge lead in the two cities at either end of the state, if he hopes to beat the huge Republican majority in all the central counties. (Pittsburgh is near the western border, Philadelphia is at the eastern.) For the Governor (and statewide): Republican/Democrat and East/West are both important. A successful candidate needs to have a specially big margin from his 'home' half of the state. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html ================================================================= Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =================================================================
