American Digest has a plausible argument that there are millions of voters who have made up their minds for Bush but aren't yet willing to say so in public. That would explain why the number who are willing to say that they think Bush is going to win is so much higher than the number who will admit to planning to vote for him. I posted a shorter version of this in his comments, but it seems worth expanding here:

The situation reminds me in some ways of 1980. I was working for a small company in San Francisco that measured air pollution. The owner and founder was a big fan of Jimmy Carter, perhaps partly because half our business was with the EPA. The gay employee was not so much pro-Carter as intensely anti-Reagan: he may have voted for Anderson, for all I know. I was the only open Reagan voter. What I found particularly interesting about that campaign is that every one of the other six or seven employees came up to me at some time or other in the last month or two before the election and said "don't tell [the boss] but I'm voting for Reagan, too". Right up to the end, the press was calling the election "too close to call", but I was confident that Reagan would win easily. I just wish I'd thought to make some bets on the outcome. I knew a U. Chicago professor who took several hundred dollars off his colleagues betting on Reagan that year. Adjusted for inflation, that would be well over a thousand dollars today.