It's their sampling strategy. Rather than take discrete samples over a period of time, they adopt a "tracking poll" methadology, and combine three days worth of small samples.
Since each of the days' samples is quite small, about 400 people, and the response rate is just hideous (I think that Gallup admits that it's <40%, but it's probably much worse than that), you can see big fluctuations on a single day's results and that skews the tracking poll for three whole days.
Slate (http://slate.msn.com/q/00-10-20/q.asp) published something to that effect on the voter bias of the likely voter model, but to the best of my knowledge James Wilson got the democrat and republican bit wrong. The likely voter model tends to favour republicans by weighting republian-leaning constituencies much more heavily.
He's also taking a very selective read of the pre-election data (one marginal example: Reagan in 1980, which was within the margin of error, ferchristssake) and interpreting mid-October polls in exactly the wrong way. They're not forecasts of the coming election, but records of how people thought in the days preceding the poll release, which was more than a month prior to the election.
Finally, he misinterprets the value of the margin of error. The margin of error is there because the sample is unlikely to be representative of the sampled population, not because it is representative, and presents a quantitative estimate of how far off it can be. John Tukey died earlier this year, and I suspect he's rolling in his grave as we speak.
(I'm not a professional statistician, but I studied it in university, and it gets me steamed to see it so abused. No wonder the practise of statistics has such a bad reputation.)
As for my 0.02$ worth, it will come down to the voter turnout, just like it did for congressional elections in 1996 and 1998. Clinton seems to be helping out in that respect, but too little, too late? You decide. (But please, get out and vote. I've already received one letter from a lifetime Democrat, despairing of the stupidity of the electorate.)
Oh, yes. There's a snap campaign in Canada on right now. Don't ignore us. However, if it comes down to President Bush and Prime Minister Day, I'm picking up my EU passport and emigrating.
Marco
,--------------------------------------------------------------------------.
> | Now if you think it strange that <
> | mice should elect a government made <
> Marco Anglesio | up of cats, you just look at the <
> mpa at the-wire.com | history of Canada for the last 90 years <
> http://www.the-wire.com/~mpa | and maybe you'll see that they weren't <
> | any stupider than we are. <
> | --Tommy Douglas <
`--------------------------------------------------------------------------'