Pre-election polls

MORI's final poll of the 2005 election campaign, published in the London Evening Standard on election day, predicted the Conservative and Liberal Democrat shares of the vote spot-on, and missed Labour by two percentage points - an average error of 0.67 points on the three major parties, well within the normally accepted margins of error.

Nor was this a one-off fluke. Three of MORI's final four published polls¹ were this close to the final result, clear evidence of the effectiveness of our procedures for reaching a representative section of the British public.

Apart from the necessary fine-tuning to the demographic balance to correct the vagaries of sampling, and the exclusion of those unlikely to vote, we were able to directly report the unmanipulated answers of our sample of members of the public. As the table below shows, all these adjustments made only a very small difference to the figures in our final prediction poll.

*Percentages exclude undecided, would not vote and refused to answer+Published in the London Evening Standard, 5 May 2005: MORI interviewed a representative quota sample of 1,628 GB residents aged 18+ by telephone on 3-4 May 2005. Data are weighted to match the profile of the population.

Unlike the polls of some of our competitors, we did not need to subject our figures to radical weightings or other major adjustments to reach an accurate prediction. Although all the final polls were creditably close to the election result, MORI's raw figures needed less adjustment to produce this accurate prediction than those of any of the other four main pollsters.

This is vindication of our belief that it is still perfectly possible to achieve a representative sample of the British public through sound and straightforward sampling methods, using quota sampling and either telephone or face-to-face interviewing. Furthermore, because the production of an accurate voting prediction depended on the basic representativeness of the sample rather than radical adjustment to the data, we can be confident that the methodology should be equally applicable to the measurement of any other characteristics of the British public.

Exit poll

In 2005, for the first time, BBC and ITV combined to commission a single exit poll, which was conducted jointly by MORI and NOP.

It predicted the Labour majority in seats exactly, and the shares of the vote within one percentage point. The poll was conducted in 120 polling stations across the country.

*In 1997 the precise NOP projection was not published, but was broadcast as indicating an overall majority of "about 200"

Full credit must be given to the academic team, led by John Curtice, which was responsible for the statistical model used to convert the polling data into a prediction, surpassing the already impressive record of accuracy of the British exit polls in recent elections. This was despite the complications caused by the huge increase in the number of postal votes, which naturally cannot be included in the exit poll and must be measured indirectly to calculate a compensating factor.

¹The exception being the poll conducted over the Bank Holiday weekend, which in retrospect looks like a rogue poll. Interestingly, the Populus poll conducted for The Times over the same period found a very similar result. However it seems likely that this reflects disruption to the normal sampling process by the public's different behaviour patterns over the holiday weekend than any real short-lived blip in voting intentions. Under normal circumstances we would not attempt to conduct polls over a Bank Holiday.