Poll Watch: How accurate were election opinion polls?

By David Cowling
Editor, BBC Political Research Unit

With the national votes for the three main parties in Britain standing at 37% for the Conservatives, 30% for Labour and 23% for the Liberal Democrats, how did the final eve of Election Day opinion polls perform?

THE FINAL POLLS

There were seven polls published on 6 May with fieldwork mostly carried out two days before.

Here's a graphic of how they break down. The lines down the middle of the graphic show the actual share of the vote achieved by each of the parties. The lines running across, by the name of each pollster, show how close their predictions have come to the actual result.

If your computer does not display the graphic, the figures are written out in full at the bottom of this story.

THE DIFFERENCES

What are the percentage point differences between each poll and the actual result?

ICM Guardian: Con -1; Lab -2: LD +3

ComRes/ITV/Independent: Con 0; Lab -2: LD +5

Angus Reid/PB: Con -1; Lab -6; LD +6

Populus/Times: Con 0; Lab -2; LD +4

YouGov/Sun: Con -2; Lab -2; LD +5

Harris/Daily Mail: Con -2; Lab -1; LD +4

MORI/London Evening Standard: Con -1; Lab -1; LD +4

The sharpest difference with the final result is seen in the ratings for the Liberal Democrats, with every pollster over-estimating them by between 3-6%.

The discrepancy with the Conservative vote was the smallest of all three parties. Two of of the polling companies - ComRes and Populus - were bang on the button.

But Labour's ratings offer the second biggest surprise of election night.

For at least two decades the polls have tended to over-estimate the Labour share at general elections. But this time, in every case, they have under-estimated it.

THE AVERAGE ERROR

What is the average error for each of the polling companies? Here we add all the three columns of difference for each company and divide the total by three to reveal the average polling error:

ICM: 2.0

MORI: 2.0

Populus: 2.0

ComRes: 2.3

Harris: 2.3

YouGov: 3.0

Angus Reid: 4.3

Unlike the 2005 election where one company, NOP, correctly predicted the outcome for all three main parties, this time nobody came anywhere near taking the crown.

The heart of the error for the pollsters was the extraordinary outcome for the Lib Dems.

The televised debates and 'Cleggmania' seemed to shape the entire election campaign.

And yet the Liberal Democrats have ended up with share of the national vote up less than 1% on 2005 and fewer seats than five years ago.

Were the polls wrong throughout the campaign?

I doubt it, but then how do I explain the stark contrast between them and the hard reality of 6 May?

At the moment I cannot. For me and for many others I imagine, there is considerable work in progress as we try to reconcile the past four weeks of campaigning with the actual outcome.

---

Here, again, are the full figures for each pollster's final predictions in a form that is more easily readable by some computers. The number in brackets at the end of the row is each pollster's prediction of the Tory lead.

ICM Guardian: Con 36%; Lab 28%; LD 26%; (8%)

ComRes/ITV/Independent: Con 37%; Lab 28%; LD 28%; (9%)

Angus Reid/PB: Con 36%; Lab 24%; LD 29%; (12%)

Populus/Times: Con 37%; Lab 28%; LD 27%; (9%)

YouGov/Sun: Con 35%; Lab 28%; LD 28%; (7%)

Harris/Daily Mail: Con 35%; Lab 29%; LD 27%; (6%)

MORI/Lon Eve Standard: Con 36%; Lab 29%; LD 27%; (7%)