The election and the polls

In the current edition of the Spectator, Stephen Glover attacks MORI in particular and the opinion polls in general on the basis of their performance in last year's general election. His argument depends upon a number of inaccuracies and misunderstandings.

Mr Glover attacks the accuracy of the polls during the 2001 election by comparing average leads over the whole election campaign period with the final result, which ignores all the changes of mind which are the whole point of fighting an election campaign. A quarter of those who voted in the 2001 election said they made up their minds who to vote for after the campaign began. The Liberal Democrat share of the vote rose from around 13% in the polls in the opening week to 19% by the end of the campaign. So it is hardly surprising that the average of the campaign polls didn't match the final result, and it doesn't prove the campaign polls were wrong because they didn't.

But probably more important was the low turnout by Labour supporters which significantly depressed the party's lead on election day. Millions of adults whose views were represented in the campaign polls didn't turn out on election day. In the final eve-of-election polls we, and the other pollsters, attempted to correct for this, though none of us allowed for a turnout quite as shockingly low as eventually transpired; I believe the real lead in the country through the campaign, including those who didn't vote but might have voted and would have supported Labour if they had, was always much larger than the eventual lead in votes actually cast.

All this seems to have passed Mr Glover by. He apparently thinks that polls, during the campaign as well as on the final day, are predictions - he states at one point that "many of them had predicted a Tory wipe-out". Polls don't predict, they measure the present. None of the final polls had the Tories below 30%, which hardly equates to a wipe-out. (But, of course, since he talks purely in terms of the lead, and not party shares, he misses the point that a higher Labour lead can sometimes be better for the Tories anyway, since it means a low Lib Dem share and saves the Tories seats in the Con-LD marginals.)

In the same way, Mr Glover's disbelief of the leads which the polls are currently finding depends on a misunderstanding of what opinion polls at this point in a parliament are measuring. Of course Labour would not beat the Tories by 51% to 27% if there were a general election tomorrow. But if there were a general election tomorrow, Labour would probably not have a 51% to 27% lead in the polls. When there is no election in the offing, voting intention is very much a hypothetical question, a measurement of national mood rather than a considered prediction of future behaviour (which is what, as far as we can, we are trying to measure during an election campaign).

The figures reflect a continuing failure of the Tories to register on the political landscape as a credible opposition; but of course many of those who currently say their voting intention is Labour will grudgingly return to the Tories when the election comes, and many others will not vote at all. The importance of the figures is not in relation to a non-existent election but by comparison with the corresponding figures measured in the same way in previous parliaments, and in the month-on-month changes (or lack of changes), and taken in conjunction with all the other monthly political indicators.

But Mr Glover is not very reliable in reporting the poll figures, either. He states that after the change in methodology in MORI's polls for The Times during the election (using a show card with a list of candidates in the respondent's constituency, which we introduced as soon as candidate names were available after nominations closed) "the leads remained almost as vast". In the three polls before the methodological switch, Labour's lead was 24, 26 and 25 points; in the two after it was 18 and 15 ? a pretty substantial difference.

When he comes to compare the "eve-of-election" polls with the final result, he is no less misleading. It is simply untrue that "none of the pollsters came close": ICM's average error on party share was barely half a percentage point, and every one of the final polls had an average error within the 3% that is usually (though over-simplistically) quoted as the "margin of error".