Research Note 48 1996-97

Polls Apart: The 1997 UK Election: Accuracy of Predictions

Rosemary Bell
Law and Bills Digest Group
May 1997

Introduction

Among the winners of the British General Election on 1 May 1997 were the British opinion poll companies. The five polls published on the morning of election day correctly predicted a Labour victory. They also accurately predicted the winning margin and the main parties' share of the vote.

By contrast, the 1992 British General Election had been a disaster for the polling companies. On average, the four polls published on the morning of the 1992 election put Labour 1% ahead of the Conservatives. The actual result was a Conservative victory of 8%.

This Research Note looks at the performance of the opinion polls in the 1997 election in the light of adjustments made to polling methods as a result of the 1992 experience.

The 1992 Experience

Polling organisations in the UK were considerably embarrassed by their inability to accurately predict the results of the 1992 election. The four polls published on the morning of the 1992 election showed average support for the Conservatives at 38% (compared with the actual result of 42%), Labour at 39% (34% actual) and 19% for the Liberal Democrats (18% actual).

A two year inquiry by the pollsters' own organisation, the Market Research Society of Great Britain, concluded that four factors had contributed equally to under-estimation of the Conservative vote:

• with quota sampling, the sample of people chosen for interview should accurately reflect the socio-economic composition of the population as a whole, but the 1992 samples were skewed too much towards traditional Labour voters;

• those who intended voting Conservative were more reluctant to be interviewed or to say how they would vote than Labour voters;

• there was a late swing to the Conservatives between the final interviews and the act of voting; and

• Labour voters failed to go to the polls, either because they were in safe seats or because they had kept themselves off the voting register.

Since 1992 all the main polling companies have made significant, though different, changes to the way they conduct their polls. As a result, there are now important differences between the companies in the methods used to carry out opinion polls.

• Three companies (MORI, NOP and Harris), have decided to stick with quota sampling but to increase the number of white-collar workers which their interviewers must seek to interview. In this way they hope to interview a more representative sample of the British population. NOP also looks at how its respondents say they voted in 1992. If the company finds that too many claim to have voted Labour, they weight their results to reduce the proportion appropriately.

• Two other companies (Gallup and ICM) have gone much further, partly or wholly switching from quota to random sampling. Equally significant, they now interview over the telephone rather than face-to-face. The difference that the new methods can make was dramatically illustrated in January 1997 when Gallup changed from quota to random sampling. Using quota sampling Gallup's polls had shown Labour with a lead of 37 points. Its new telephone-based, random-quota sample reduced that to 18 points.

The 1997 Experience

Given the performance of the polling organisations in accurately predicting the results of the 1997 election (see Table 1), it would appear that the changes in polling methods introduced after 1992 have been successful. While there appears to be some overestimate in the Labour vote (up to 6% in the case of the NOP poll), on average the final opinion polls predicted a level of support for Labour that is just within the 3% error margin associated with polls of this sample size. The performance of the polls in predicting support for the Conservatives was much more accurate, the average being only 0.6% from the actual result even though there was a 5% spread in the range of individual estimates.

One way to measure the accuracy of opinion polls is to calculate what can be referred to as the 'average error' of the poll. The average error is the average of the sum of the differences between the predicted result and the actual result for each party. Using this measure, the average error for the opinion polls in 1997 was 1.3%, with the ICM poll the most accurate and the NOP poll the least accurate. By way of contrast, the average error for the polls at the 1992 election was 3.15%.

Table 1 Last Poll Before 1997 Election

Consistency of the Polls over the 1997 Election Campaign

The graphs above indicate party support as shown by the opinion polls over the Election campaign. Most polling companies started with Labour's share of the vote above 50%, gradually reducing to between 47% and 50%. Only the ICM poll consistently predicted that Labour would receive a lesser vote. Its poll of 23 April, published in the Guardian newspaper, predicted that Labour (at 42%) was only 5 points ahead of the Conservatives. Support for the Conservatives was consistently predicted by the opinion polls at around 30%, rising slightly at the end of the campaign. The ICM poll of 23 April is again out of step in predicting a sudden and unsustained increase in support for the Conservatives to 37%. Support for the Liberal Democrats as predicted by most of the polls was below their final election result of 17.2%.

Conclusion

Compared with their predictions for the result of the 1992 British Election, in 1997 the opinion polls got it right. The ICM poll, despite appearing to be out of step with the other polls during the campaign, was the most accurate in predicting the final result.

Sources

Association of Professional Opinion Polling Organisations, http://www.mori.com/election/result.htm Another useful website is University of Keele, Dept of Politics, http://www.keele.ac.uk/depts/po/ge97.htm

John Curtice, 'What future for the opinion polls? the lessons of the MRS Inquiry', in British Elections and Parties Yearbook 1995, ed. Colin Rallings et al, Frank Cass, London, 1996: 139-56.

Robert M. Worcester, 'Lessons from the electorate: what the 1992 British General Election taught British pollsters about the conduct of opinion polls', International Social Science Journal, vol. 47(4), Dec. 1995: 539-52.

Comments to: web.library@aph.gov.au
Last reviewed 17 October 2001 by the Parliamentary Library Web Manager
© Commonwealth of Australia
Parliament of Australia Web Site Privacy Statement
Images courtesy of AUSPIC