On Friday, Gallup published this report discussing the differences between registered and likely voters. Perhaps unintentionally, Frank Newport, the head of Gallup, made a huge admission that bears directly on the September 8th likely voter poll:
Second, we are at this point reporting likely voter estimates on only an occasional basis. We feel that the trends among registered voters give us the best way to track election preferences in our daily poll, in part because many voters are not yet in a position to accurately estimate their chances of voting on Election Day. But from time to time, we do estimate (and report) likely voter results to give us a feel for the potential difference turnout could make in November. So far this summer, there have been occasions when -- as was the case this past weekend after the GOP convention -- likely voters were decidedly more Republican. But there have also been occasions when there was little difference between the vote patterns of likely voters and those of registered voters.
In other words, Gallup is admitting the following:
1. At the time it released the September 8th poll (showing McCain up by 10), it believed institutionally that likely voter results were less accurate than registered voter results.
2. Likely voter results have only occasionally diverged from the registered voter results.
3. Despite these facts, Gallup deliberately chose to release, to the widest fanfare possible, a poll using an admittedly less accurate method (the likely voter method) at the time of McCain's maximum convention bounce, knowing that it would show a large divergence (+10 for McCain vs. only +4 with registered voters) based on the likely voter method, even though such a divergence is not often present.
4. In short, they combined all possible factors in McCain's favor to make his lead seem as big as possible -- and the media went wild with it.
more at:
http://www.dailykos.com/storyonly/2008/9/15/74531/5933/788/599244