How Britain’s pollsters have changed their methods

BRITAIN’S FIRST pollster, Henry Durant, quipped that his was the “stupidest of professions”. For who would make a claim one day only to be contradicted on election night? Undeterred, beginning with their accurate forecast that Clement Attlee would beat Winston Churchill in 1945, polling firms gained a hard-won respectability.

Lately that reputation has been eroded. In 2015 pollsters suggested Labour and the Conservatives were neck-and-neck. In the end the Tories finished six percentage points ahead. A year later none of the final polls predicted the Brexit vote. Methods were duly tweaked ahead of the 2017 election, in which the Tories were expected to win a commanding victory. Instead they lost their majority. These misses prompted talk of a crisis. Now, December’s election promises to be especially tricky.

Errors may creep in from three main places. The first is sampling. Just as a chef who fails to stir the soup cannot judge its taste, pollsters who lack a representative sample of the electorate cannot read the national mood. A review after the 2015 flop concluded that the polls had been biased towards Labour. The shift to online, opt-in surveys undercounts the elderly and politically disengaged.

More mistakes can arise as pollsters try to adjust for sample bias by weighting people’s responses according to the...