Exclusive Prospect polling: Do we want Trident?
Polling isn't dead. That was the conclusion (unsurprisingly) reached by the end of Ipsos MORI’s The Death of Polling? debate in Westminster last night, where five experts in the field from around the world discussed the challenges pollsters face in their respective countries.
Many in Britain have been pessimistic about political polling since the 2015 General Election. In the run up to the vote, the prediction across the board was a hung parliament with no party set to win enough seats to secure an overall majority. This might well have had an impact on the public’s voting intentions: as it was predicted that The Labour Party would not win outright, the Conservatives began hammering home the prospect of a Labour-SNP coalition in a bid to scare would-be Labour voters.
The polls turned out to be utterly misleading. The Tories went on to win a 330 seats, a majority of 12, when a week before the election YouGov was predicting that they would win 272 (this figure was 18 per cent out). The pollsters were confounded; the public incensed.
But where did the experts go wrong? At Ipsos’ event last night Ben Page, Chief Executive of Ipsos MORI UK, offered an answer, the first of many myth-busting interventions made throughout the evening.
It has become widely accepted that “Shy Tories” were what confounded the pollsters last year. That is, lots of people misled the people asking them about their voting intentions, saying they would not vote Tory but secretly planning on doing exactly that.
Page explained that contrary to popular belief this was not the case. Shy Tories did damage the accuracy of last year’s polls, he explained. However, this phenomenon was tiny when compared with a bigger problem: the failure to accurately predict “likelihood of voting.” Many people who were expected to vote for the Labour Party did not go out to vote. While the Conservatives did do better than expected last year, the more important fact is that Labour did worse.
Later in the evening Darrell Bricker, Canada's leading pollster, explained to the audience how “likelihood of voting” is measured in Canada. He said that they take account of three factors and combine the results: whether someone says they intend to vote, whether that person has a record of voting, and whether they display signs of being interested in the election.
This seems like a solution—or at the very least, part of a solution—to our own polling dilemma. Bricker’s explanation was typical of the evening: all THE speakers were open about their various failings, and by the time you’d heard all the speeches, it became clear that the solution to one country’s problems may be found through careful analysis of another country’s successes. In light of this, knee-jerk pessimism about polling seems unjustified.
Julia Clark, polling expert for Ipsos US, talked about the accuracy of opinion polls for different Presidential elections in the United States. She explained that in 1936, the average candidate error in major opinion polls was 6 per cent—pollsters really did miss the mark. A margin of error this substantial renders a poll all but useless. For the 2012 Presidential Election, however, the average candidate error was 1.5 per cent—the lowest it has ever been. (It’s also worth pointing out just how huge the American polling operation was: in the two months leading up to that election, Ipsos conducted 350 interviews a day and 17,000 polls were carried out across the board.)
The 1.5 per cent statistic was representative. US pollsters’ accuracy in 2012 wasn’t a blip; rather, it falls neatly into a pattern of near-constant improvement in the US over the past 80 years. While replicating this success in Britain may prove difficult, Clark’s figures do seem a cause for optimism.
David Ahlin, Public opinion expert for Ipsos Sweden, discussed the rise of the “Sweden Democrats” (A far-right party we have written about in Prospect) in his country’s General Election in 2014. He explained that polls in Sweden are skewed by the fact that those who were planning to vote for the Sweden Democrats were much less likely to tell that to ethnic minority pollsters than they were to white pollsters. This was an issue to which a solution was not proposed—though perhaps a move away from face-to-face interviewing might be the solution.
At the end of the evening, someone pointed out that the record of pollsters looks poor when you consider that in 14 of the last 18 British elections the colour of the shirt of the FA Cup-winning team the previous year predicted the winner (if the team that won played in blue or white, the Conservatives would win; if they played in red or yellow, Labour would win). This was an amusing statistic, but it would be very odd if no random sequence corresponded to General Election results.
Pollsters across the world face challenges, and sometimes get it wrong. At our last General Election they got it very wrong. But last night made clear that improvement is possible if pollsters from different countries collaborate.
