Polls are 100% and there are multiple errors some of which have to do with sample size and some which do not but the reality is we so polls in conjunction with real world events to a degree that makes it clear they are measuring real events. If polls had so little accuracy we would seem them all over the place to a much greater degree and they would not as effective as they at predicting the winner.

Now it is fair to say that if polls show a race to be within three points then the race is basically up for grabs but the sample size of the polls is only one part of that. Right now there is about a 4 point difference in likely voter polls and registered voter polls. How accurate that likely voter model is vs. the actual turn out could easily decide the race and that wouldn’t change even if they polled a million people instead of 10,000. It is really hard to predict who will actually get off their ass and vote. Even if the model is dead on things like bad weather or a bad flu season can swing a race. My personal prediction is that since most of these models are based on who voted in the last election and the enthusiasm gap has done a 180 Romney will outperform the model and Obama will underperform the polls but we will have to wait and see.