Probably the best piece I've read on the Forum Polling cock-up in Manitoba:
Instead, what we’re dealing with appears to be flawed methodology. Bozinoff has admitted that some respondents may have been called for 3 consecutive polls, and that likely wouldn’t have happened unless the response rate was in the neighbourhood of 1% (typical for robo-polls, when you don’t do callbacks). Heck, Sunday being the Grey Cup, it may have been even lower. Sampling methodology only works if you assume survey respondents are similar to the public at large – otherwise, these polls are no more accurate than the “self selecting” click polls you see on websites, asking what you think of Miley Cyrus’ antics.
Read the whole thing, and next time you see a poll from Forum Research, just say to your self "Oh, that's just a poll from Forum Research!"