Is Tony Blair losing it?

Has the Harman affair inflicted lasting damage? Peter Kellner explores conflicting polls

Adlai Stevenson, intellectual, wit and failed US presidential candidate, once remarked that opinion polls should be taken but not inhaled. The trouble is, they are addictive, and politicians are the greatest addicts of all. Claims to the contrary should be regarded as the routine protestations of people in denial.

As addicts, the politicianswill be frantically absorbing this week's three offerings: the ICM poll in Wednesday's Guardian, the Gallup Poll in today's Daily Telegraph and the Opinion Research Business panel of former Tory voters conducted for the stockbrokers James Capel. They send apparently contradictory messages. Tory support is up, say ICM and Gallup; no sign of deserters yet returning to the fold, says ORB. Labour support is down, says Gallup; oh, no it isn't, says ICM. Then there is the continuing, huge gulf between Labour's lead as reported by ICM and the much larger, if declining, lead observed by Gallup. So what is really going on?

First, the gap between the parties is closing. Labour's lead, which peaked just over a year ago, was unsustainable. It was always likely to contract.

Second, the long, slow underlying pace of the Tory recovery has been concealed by one-off events that have disrupted the trend. Last July, John Major's leadership election campaign gave his party a brief extra boost; then Tory support slipped back. Last autumn's party conference season helped Labour; but this, too, did not last. Emma Nicholson's defection to the Liberal Democrats and the Harriet Harman schools row have caused turbulence in some polling figures since the new year.

Third, media headlines areliable to be affected by "pollsters' wobble". Labour's underlying lead has probably been declining by an average of one point a month since early last year. But this is too small to be detected reliably on a month-by-month basis. Individual polls, taken a month apart by the same company, can easily show a variation of up to six points in the gap between the parties simply as a result of sampling fluctuations.

Some of the reported decline in Labour's lead in today's Daily Telegraph flows from the fact that Gallup's snapshot poll a month ago - which put it as high as 39 points - was probably a rogue.

Fourth, the problem of interpreting the polls is made harder by the different methods used by each polling company. Gallup's samples are more "downmarket" than its rivals'. They contain (mainly Labour) council tenants and fewer (mainly Tory) voters in two-car households. At the other extreme, ICM adjusts its raw data more aggressively than any other pollster to take account of the "spiral of silence" - Conservative supporters who are reluctant to admit their allegiance to the pollsters.

My own judgement is that Labour's underlying lead peaked a year ago at 32-35 points; it is now down to 20-23 points. This is still high by historical standards. In the last parliament, Labour's underlying mid-term lead never exceeded 10 points: the much higher leads reported in the spring of 1990 were the result of (a) short-term reaction to the introduction of the poll tax, and (b) errors in polling methods that finally came to light on the night of the general election two years later.

Labour's lead is likely to decline still further over the months ahead, as tax cuts and mortgage rate reductions feed into voters' pockets, and the "feel-good" factor improves. This alone, however, is more likely to save the Tories from humiliation than from defeat. For the Conservatives to secure a fifth term, they need either to secure an additional boost to their image, or to dent new Labour's, or both. On its own, the Harman affair is unlikely to be much help. Equally, the Scott report next week is unlikely to do much lasting harm.

In both cases, the task facing the two main parties is to translate essentially ephemeral controversies into ingredients of a more permanent change in their rival's image. It will take more than the blurred snapshots taken by three polls in a single week to tell whether the latest dip in Labour's lead is a short-term kink in an unchanged trend - or the start of a new trend that portends a really close election.