Gallup discovers new way of making a point

Gallup's switch to random telephone interviewing in its opinion poll yesterday means all the polling companies have now changed their methods since their failure to gauge voters' intentions at the last election.

The main pollsters are now divided between new-style random polls and adjusted, old-style polls. However, the results are very similar. Gallup's 18-point lead for Labour means that all reputable companies now put Labour between 18 and 21 percentage points ahead.

Gallup was the last to change, and its findings increasingly stood out like a sore thumb; last month it put the Labour Party 37 points ahead.

Gallup was the first company to carry out opinion polls in Britain, just before the war, and it made its name heralding Winston Churchill's "unthinkable" defeat in 1945.

Until late last year it looked as if Gallup would do this year's election much as everyone did 1992, while the other companies would all try new methods.

But the weight of evidence against the way it was done last time was overwhelming and Gallup's nerve finally cracked.

On election day, 9 April 1992, the main pollsters all put the parties almost neck and neck. Labour was an average 0.8 points ahead. Only Gallup put the Conservatives ahead and only by half a percentage point. In the actual vote the Tories won by 7.6 points.

This time there will be two types of poll. NOP and MORI are quota polls while ICM - and now Gallup - are random. Quota polls use the old method of interviewing people in the street or at home. Interviewers have to fill quotas according to sex, age and class. Since the last election the quotas have been changed, as it turned out they were too working-class.

Random polls are new in this country and are carried out over the telephone, using computers to dial numbers at random. They are, in theory, a better way of getting a representative sample. The problem is with people who don't have phones or who don't like taking part in telephone surveys.

The companies also have different ways of accounting for the "shy Tories" factor. It is accepted that Tories are less likely to take part in opinion polls and less likely to declare how they intend to vote. ICM, which works for Tory Central Office and The Guardian, makes the most drastic adjustment for this, which until now produced the lowest Labour leads.

But the adjustments all seem to produce roughly the same outcome; the most recent polls from each of the four main companies were within a few percentage points of each other. It looks as though they will all be right or wrong together - again.