How did you vote?

MORI's political polls frequently include a question asking respondents how they voted at the last general. However, although the responses are useful to us in a number of ways, we do not expect them to be an entirely accurate reflection of how the respondents did, in fact, vote. Consequently the responses of a representative sample will NOT normally match the actual result of the last election, and the fact that a sample's recalled vote differs from the election result is not evidence that the sample is unrepresentative.

The phenomenon of inaccurate recall of past vote has been known in Britain for many years. It was noted in the Sixties that recall of Liberal vote tended to be lower than reality in polls between elections, and there has also been a general tendency for more respondents to claim to have voted than actually did so. MORI was aware of this back in the 1970s, and Peter Kellner, who was then our 'link' at the Sunday Times, wrote it up for that newspaper. Himmelweit, Baberian and Stockdale produced a seminal paper on the subject in the British Journal of Political Science in 1978.

This phenomenon is not confined to Britain: in the USA, for example, where John F Kennedy won the Presidency by a tiny margin, the margin on recalled vote steadily increased during his term, and, after he was killed, some two-thirds recalled that they had voted for him.

That the failure of the recalled vote on surveys to match the actual result is caused by inaccuracy of recall, rather than by unrepresentativeness of the samples, is relatively easy to demonstrate. One obvious point is that the polls have a generally good record of final election prediction, which would not be the case if past voters for a particular party were systematically unrepresentative. In the case of the Liberals/Alliance/Lib Dems, whose share of the vote has been most consistently undercounted by the vote recall question, the average finding of the final election day polls has not been out by more than one percentage point in 'predicting' the share in any of the last six elections. The unreliability of recall has also been demonstrated experimentally on a number of occasions using a panel survey, where the same respondents are questioned about their past vote on more than one occasion, and can be shown to have given inconsistent answers.

The pattern of answers to the vote recall question since 1997 has been generally consistent: the proportion of people who say that they voted Conservative in 1997 is usually quite accurate; however people who did not actually vote in 1997 seem liable to say that they voted Labour even if they did not, and the number remembering having voted Liberal Democrat, though initially accurate, is now consistently lower than the number who actually did.

The table below shows MORI's findings from the vote recall question since 1997, which should be compared with the election result (expressed as percentages of the adult population rather than of those who voted) shown in the top line.

Q Which party did you vote for at the last General Election, in May 1997? If you are not sure, or did not vote, please say so.

Source: MORI

For these reasons, we regard recall of past vote as being unsuitable for weighting purposes.