If you've been to a Rainbow Warrior home game - which I personally recommend - you know the one. There's a picture of a racetrack and three colored dots that race around while gambling-deprived fans try to pick the winner. Some actually cheer more loudly for their dot than for the 'Bows. In the end, a dot that's trailing usually zooms ahead to win.
Doing a poll during a political campaign is like stopping that scoreboard partway through the race. It's interesting and newsworthy to see who's ahead and to analyze where their support lies, but a poll really can't predict infallibly who's going to win come election day.
For a poll to make an accurate prediction weeks or even days before an election would mean that nothing could happen in the last days and hours of the campaign to change voters' minds. Of course the candidates, who study polls of their own, work very hard to make sure lots of things happen to bring more votes their way and to peak at the end.
A poll is a snapshot of public opinion frozen in time. As soon as voters see another ad, watch another debate or read another story about a candidate, the dynamics have changed. Support becomes more or less firm. Some people make up their minds; others change horses.
Just ask Harry Truman or Pat Saiki.
