Last time there was a mayoral election in Toronto, the polls put out their predictions and everyone scoffed. As the results came in they didn't appear to match the polls and everyone scoffed (it appeared that Mel Lastman was overwhelmingly overtaking Barb Hall). There was even a big media blitz on the night of the election, after most of the votes had been counted, where pundits made endless fun of the pollsters: they tell us nothing; meaningless; blah blah blah (the election results from the suburbs came in faster than the downtown results, and Lastman has a solid majority in the suburbs of Toronto). And everyone scoffed.
The next morning, with the final vote in: all of the major pollsters had nailed it *exactly* - something like a 1% differential which is pretty good when then admit to +/- 4% with a 1/20 rate. The polls had sampled from both the suburbs and the downtown area and the media only focused on the suburbs as they came in, ignoring the downtown core (even when the NDP won in Ontario, most of the polls were pretty accurate - and the media still attacked the victory as a complete surprise - an invasion of the socialist hordes...).
Of course, nothing was mentioned about this and the results of the polls weren't reprinted with the election results, nor was an apology given for slandering the pollsters.
>From my casual observations about elections, the polls are *rarely* wrong, and
almost never dramatically wrong. Yes, there are countless examples of polls running amok (and yes, those 'small samples' aren't all that accurate - for obvious reasons), and the questions are predetermining and all that. But sometimes I think there is an enthusiasm to bash polling which stems from an anxiety about knowing the result ahead of time. Maybe things are different with the US pollsters. And yes, it certainly depends on what kind of questions are being asked. Polls about health care are quite different from "Who are you going to vote for?"