As the referendum results flow in, the pollsters will be as nervous as the Brexit and remain campaigns. Having worked hard to scrape the egg of their faces after last year’s general election, they would hate having to do the same again. As things stand, some pollsters seem certain to be more embarrassed than others. A year ago, their final headline figures were much the same; they were all wrong together. (The experience was especially painful for me, as the then-president of YouGov. On the night, other pollsters could grieve in private. I had to sit for 10 hours in the BBC studio, pretending to stay calm.) This time there have been big variations, both between individual surveys by the same companies and, on average, between polls conducted online and those conducted by telephone. Monday night was typical – the ORB/Telegraph phone poll showed remain 7% ahead, while the YouGov/Times online poll reported a 2% leave lead. If that difference persists in the final polls, somebody is bound to have awkward questions to answer.
That said, the pollsters’ long-term record is generally outstanding – especially when one takes account of their little secret: that it’s becoming much harder to obtain representative samples. Twenty years ago, telephone polling companies would draw 7,000 random residential telephone numbers; these would yield 2,000 completed interviews. Now they must draw 28,000 numbers. Response rates have collapsed from 30% to 7%.
The problem for online companies is different; but, by definition, they can poll only those willing to join online panels. Pollsters therefore have to extrapolate from the people they can reach to the increasing numbers they can’t reach. The really astonishing thing is not that they got last year’s general election wrong, but that they get so many things spot on.
Their private clients generally know this. Pollsters were obviously nervous following their two big failures to predict Conservative victories, in 1992 and last year. They need not have worried. Business continued to grow. Both elections prompted public embarrassments, not commercial setbacks. One reason is that slight errors in a political contest matter less in market research. In the current referendum a 52-48 decision one way has vastly different consequences from a 52-48 result the other way. But private clients would be perfectly satisfied if predicted market shares for rival brands came within four percentage points of the outcome.