Why polls and pre-election predictions can be meaningless..

[J.D.]

Case Study: Canadian Federal Election 2004.

The Liberals are in the middle of a massive sponsorship scandal, and have been in power for 12 years already (3 terms of Jean Chretien). The resurging Conservatives have a new leader and a strong following. The stage was set for the Conservatives to give the incumbent Liberals the smack-down.

Polls conducted just prior to the election (Popular Vote):
Liberals 32%
Conservatives 31%
NDP 17%
Bloc 12%
Green Party 8%

Experts were predicting a Conservative minority government, with a possibility they could dominate Ontario (the province with the most seats up for grabs) and might even land a majority (155 seats). Ontario was ripe for the picking, after the newly elected provincial Liberals went back on almost every promise they made and engraged the voting public.

In short, just in case you're getting bored, the Liberals were in big trouble.

Here are the election results:
Liberals 36.7% (135 seats)
Conservatives 29.6% (99 seats)
BQ 12.4% (54 seats)
NDP 15.6% (19 seats)

Although no party achieved a majority, this was seen as a huge collapse by the conservatives, who had all but been predicted to win the day before.

Oh, and how did Ontario (the province the Conservatives were expected to roll right through) do?
Liberals (75 seats) 44.6%
Conservatives (24 seats) 31.4%
NDP (7 seats) 18.0%

Oops! I guess those polls weren't so helpful after all. This was a mark on all polling organizations in Canada, as they weren't even CLOSE to getting the results right.

So what happened? Well it appears that a lot of the undecided voters chose to vote for the incumbent party rather than the challenger. But all we're hearing on the news now is that the challenger usually gets most of the undecided votes. Not true.

Have fun tomorrow, I know I'll be watching with great interest!

EDITED because the thread title was too much of a generalization.