Our online polls prove a winner for accuracy

Last Thursday, as the election-day edition of the Evening Standard started rolling off the presses, I told colleagues and journalists: "Tomorrow night, either Ken Livingstone or I will be out of a job."

I was not joking. YouGov gave a final forecast that Boris Johnson would win by between 47 and 53 per cent. It was our sixth mayoral campaign poll and the sixth to show Boris ahead by a comfortable margin. Six polls by other companies had all disagreed, either showing the candidates neck-and-neck, or Livingstone ahead.

Had Livingstone won, the clamour to condemn YouGov's panel-based internet polling methods would have been deafening. I would have no excuses. I would have to go.

In the event, the result was exactly as we predicted. It is our rivals who must now rethink their methods and, to their credit, our friends at Mori have already announced a review of the pro-Livingstone bias in their figures.

Here is my advice to the traditional polling companies:

Recognise you have a long-standing problem in London. Most telephone polls in 2000 and 2004 significantly overstated Livingstone's support.

Rethink your sampling methods. Response rates for telephone polling companies in London are worse than in the rest of Britain. Well under 20 per cent of calls result in an interview. And, because telephone pollsters dial only landline numbers, they question too many people who stay at home, and too few of the many Londoners who lead socially active lives. In contrast, YouGov's panel members can respond to our surveys at any time from any computer within a 48-hour window.

Make sure your samples are politically representative. It is not enough to have the right proportions of men and women, young and old, rich and poor, black and white. To achieve a sample that is fit for purpose, the aim of the survey must be considered. For a political survey, this means ensuring that the sample contains the right mix of voters by party allegiance. We at YouGov use sophisticated techniques to ensure this. Last week's performance demonstrates their effectiveness.

Take interviewer effects into account. One reason why national polls underestimated Tory support in the Nineties was that they failed to allow for "shy Tories" - people reluctant to admit to a stranger on the phone that they would vote for John Major. This time phone pollsters may have encountered "bashful Borisians". YouGov does not have this problem. By entering answers anonymously on a computer, our respondents are more willing to come clean on a variety of topics: sex, drugs, tax - and even voting Tory.