One more poll: L.A. Times. And it's tied: 48%-48% (likely voters, 3-way race), 47%-47% (registered voters, 3-way race).

Gallup? Feh. Over at MyDD, ProfAlan calculates that if you take the percentage of Democrats, Republicans, and independents in Gallup's polling sample and rejigger the numbers based on the actual breakdown of voters by party in the 2000 election, Kerry's 5-point deficit turns into a lead. (Albeit a tiny one.)

The article on John Zogby in the October 18 New Yorker talks about Gallup's thinking:

Figuring out who is likely to vote is difficult.... Gallup, for instance, asks respondents seven questions, such as: "Do you happen to know where people in your neighborhood go to vote?" and "In the election for President in November of 2000, did things come up that kept you from voting, or did you happen to vote?" ... Gallup then assumes that only fifty or fifty-five per cent of the population will vote in the election, since that is historically the case, and counts only the fifty or fifty-five per cent of its sample that scored the highest on the likely-voter questionnaire. This method of calculation means that Gallup is throwing out half of its sample (rather than, say, weighting the whole sample), and that is risky ...: it means that Gallup is relying on a key variable in the election -- in this case turnout -- remaining unchanged.

So Gallup thinks every year the same people who always vote will vote, and no bloc of voters -- blacks angry about 2000, say, or young people worried about a draft -- will improve their rate of participation. Does that make any sense?