I earlier linked to a Washington Post piece on the poor showing of the exit polls in the Wisconsin recall election. There was a section of the article that explains the conventional thinking perfectly:
"One way to avoid Election Day confusion is to focus on what exit polls are good for — the tally of how different groups voted in an election, and their relative size in the overall electorate — not what they’re not: predicting results."See? Exit polls are allegedly "good" for showing the turnout percentages of various different groups - but these percentages are based upon a Democratic leaning sample! When you weigh a poll today so that it reflects a sample which has been proven to be skewed Democratic, you are necessarily skewing your current poll towards the Democrats.
This skewing is in addition to the problem of weighing current polls to reflect exit polls taken from a Democratic wave election.