And in a political election we know what the speeches are, how the candidates react to current events, their stated positions on issues. Like election cycles, football games are very unpredictable. Other than injury, there are many "unexpected outside events likely to throw a wrench in things" including kick/punt returns, interceptions, fumbles, penalites, player mistakes, etc.

Okay, now you're making a completely different analogy than what I was using. This analogy doesn't work because it's not the audience's job to determine the winner, as it is in an election. A better analogy would be polling of viewers over a season of American Idol to try to predict who the audience will vote to win at the end of the season.

I was saying that interpreting polling data is like watching a football game. While it is impossible to predict with 100% certaintly who will win the game up until the last few second, you often have a good idea based on past performance and how well the teams have been playing. Likewise, the polling "score" may go up and down with time, but you can certainly get a feel for who has the momentum, what "plays" are working, etc.

I'm not going to say polls are completely accurate -- that's why there's a margin of error. Many polls (e.g. Zogby) correct for your problem of polling too many fans of one team over the other by weighting the scores based on actual proportions observed in the previous election. The polls that don't do this are the ones that had Bush leading by double digits, meaning their sample either had a higher proportion of Republicans than actually exists, or that there are more Republicans now than there were in 2000.

I still don't see why. They surely aren't perfect, but they're really the only way to assess how well a candidate is doing at any given time.