How historically accurate are election projections based on polls?

[d000hg]

The BBC's neat thingie lets you predict actual seats based on polls. It's still looking pretty good for Labour, dead-even at worst, which makes me wonder why everyone is so optimistic about Labour not getting in?
Historically, how much out have such projections been? Very close, OK, or fairly useless?

[Fred Bloggs]

The tendency is to underestimate the eventual Tory vote. Apparently many folk who vote Tory don't say they will beforehand to pollsters. I don't know why. Classic example was for John Major when he won the election that he should really have lost for the good of the country. If Kinnock had won, the Tories would have bounced back in 97 instead of letting NL into power. Sadly, I really think Cameroon will win with a small majority, but I think he'll be booted back out after 1 term.

[Green Mango]

The tendency is to underestimate the eventual Tory vote. Apparently many folk who vote Tory don't say they will beforehand to pollsters. I don't know why. Classic example was for John Major when he won the election that he should really have lost for the good of the country. If Kinnock had won, the Tories would have bounced back in 97 instead of letting NL into power. Sadly, I really think Cameroon will win with a small majority, but I think he'll be booted back out after 1 term.

That used to be the case, but apparently this has now changed to LAbour.
Many Labour voters are now ashamed to say that they will vote for Labour.

That is not difficult to understand given Labour have wrecked the country's finances
for a generation. Also the thousands of deaths that Labour have caused through their perfidious
pursuit of the Iraq war has made honest Labour voter ashamed that they ever voted for them.

[VectraMan]

As well I think the LibDem support is over-estimated, because people are more likely to say they'll vote Lib Dem, perhaps because it appears to be the alternate, or even trendy vote, but won't do so in the privacy of a voting booth.

It's interesting that the media continue talking about the percentages in the polls, and I don't think have made enough of this issue of the system being fixed in favour of Labour. Really the LibDems shouldn't be getting the coverage they are, and shouldn't have been part of the debates, because when it comes to counting the seats, they're not in it at all. And also when it comes to the seats, the Conservatives and Labour are neck and neck, which is not the impression you get from the TV coverage.

[BolshieBastard]

Not certain about the 'election predictor' as that possibly measures people prejudices and favours?

Anyone doing an opinion poll will tell you that a representative sample of about 2000 people in a number of constituencies across the country is needed to form a trend.

These polls tend to be correct + / - about 5%. So, in a constituency where the final voting difference between the return MP and other main challenger, the poll for that constituency could be totally incorrect.

[Gonzo]

From their own description:

Party seat totals are calculated by applying a uniform national swing. This assumes that for every seat in the country, each party's vote share changes by the same amount.
So this is a crude model - in reality every seat is unique.

The reality is that the UK is diverse. There are some seats that are so rock-solid Conservative that applying the national swing is pointless. Likewise to the seats that are rock-solid Labour.

The make-up of the next government is decided by the voters in a much smaller area than the UK as a whole, which makes the whole election prediction business so interesting.

[NickFitz]

...which makes me wonder why everyone is so optimistic about Labour not getting in?

pessimistic

[NickFitz]

There are some seats that are so rock-solid Conservative that applying the national swing is pointless. Likewise to the seats that are rock-solid Labour.

Yet even they can fall. In 1992 Cheltenham, that bastion of retired Colonels and GCHQ spies, went from Tory to LibDem - the first time Cheltenham had been represented by a non-Conservative (allowing for some Independent Conservatives) since 1906, as far as I can determine. (A Liberal Baronet won in December 1910, but was "unseated by petition" - not sure how that works...)

[Gonzo]

Yet even they can fall. In 1992 Cheltenham, that bastion of retired Colonels and GCHQ spies, went from Tory to LibDem - the first time Cheltenham had been represented by a non-Conservative (allowing for some Independent Conservatives) since 1906, as far as I can determine. (A Liberal Baronet won in December 1910, but was "unseated by petition" - not sure how that works...)

There will always be exceptions. I think in that election the retired Colonels and GCHQ spies demonstrated that they were more conservative than Conservative.

[NotAllThere]

New Scientist published an article that predicts "...a hung parliament, with 290 seats going to the Conservatives, 247 to Labour, 70 to the Liberal Democrats, and 25 to other parties"

We'll see.