看不懂標題
Predicting the result
Pollderdash
Why the opinion polls went wrong
May 16th 2015 | From the print edition
IT WAS supposed to be the closest general election for several decades. At least ten final opinion polls put the Conservative and Labour parties within a percentage point of each other. Politicians were being told firmly that some kind of coalition government was inevitable. But all that turned out to be wrong. The Tories ended seven points ahead of Labour in the popular vote and won a majority in the House of Commons. Why were the projections wrong?
In 1992 pollsters made a similar error, putting Labour slightly ahead on the eve of an election that the Tories won by eight points. The often-cited explanation for this mistake is so-called “shy Tories”—blue voters who are ashamed to admit their allegiance to pollsters. In fact that was just one of several problems: another was that the census data used to make polling samples representative was out of date.
Following an inquiry, pollsters improved. A similar review has now been launched by the British Polling Council (BPC), but its conclusions may be less clear cut. In 1992 all the pollsters went wrong doing the same thing, says Joe Twyman of YouGov. This time they went wrong doing different things. Some firms contact people via telephone, others online, and they ask different questions. Statistical methods are hotly debated.
That has led to almost as many explanations for the error as there are polling firms. The “shy Tories” might have reappeared, but this cannot explain the whole picture. Ipsos MORI, for instance, only underestimated the Tory share of the vote by one percentage point—but it overestimated support for Labour. Bobby Duffy, the firm’s head of social research, says turnout might explain the miss. Respondents seemed unusually sure they would vote: 82% said they would definitely turn out. In the event only 66% of electors did so. The large shortfall may have hurt Labour more.
Others reckon there was a late swing to the Tories. Patrick Briône of Survation claims to have picked this up in a late poll which went unpublished, for fear that it was an outlier 異常值. Polls are often conducted over several days; Mr Briône says that slicing up the final published poll by day shows movement to the Tories, too. Yet this is contradicted by evidence from YouGov, which conducted a poll on election day itself and found no evidence of a Tory surge.
One firm, GQR, claims to have known all along that Labour was in trouble. The polls it conducted privately for the party consistently showed Labour trailing. Unlike most other pollsters, GQR “warms up” respondents by asking them about issues before their voting intention. Pollsters tend to be suspicious of so-called “priming” of voters, which seems just as likely to introduce bias as to correct it.
The BPC’s inquiry will weigh up the competing theories. Given the range of methods and the universal error, a late surge seems the most plausible explanation for now. That would vindicate Lynton Crosby, the Tory strategist, who insisted voters would turn blue late on. Next time expect more scepticism about polls—and more frantic last-minute campaigning.
IT WAS supposed to be the closest general election for several decades. At least ten final opinion polls put the Conservative and Labour parties within a percentage point of each other. Politicians were being told firmly that some kind of coalition government was inevitable. But all that turned out to be wrong. The Tories ended seven points ahead of Labour in the popular vote and won a majority in the House of Commons. Why were the projections wrong?
In 1992 pollsters made a similar error, putting Labour slightly ahead on the eve of an election that the Tories won by eight points. The often-cited explanation for this mistake is so-called “shy Tories”—blue voters who are ashamed to admit their allegiance to pollsters. In fact that was just one of several problems: another was that the census data used to make polling samples representative was out of date.
Following an inquiry, pollsters improved. A similar review has now been launched by the British Polling Council (BPC), but its conclusions may be less clear cut. In 1992 all the pollsters went wrong doing the same thing, says Joe Twyman of YouGov. This time they went wrong doing different things. Some firms contact people via telephone, others online, and they ask different questions. Statistical methods are hotly debated.
That has led to almost as many explanations for the error as there are polling firms. The “shy Tories” might have reappeared, but this cannot explain the whole picture. Ipsos MORI, for instance, only underestimated the Tory share of the vote by one percentage point—but it overestimated support for Labour. Bobby Duffy, the firm’s head of social research, says turnout might explain the miss. Respondents seemed unusually sure they would vote: 82% said they would definitely turn out. In the event only 66% of electors did so. The large shortfall may have hurt Labour more.
Others reckon there was a late swing to the Tories. Patrick Briône of Survation claims to have picked this up in a late poll which went unpublished, for fear that it was an outlier 異常值. Polls are often conducted over several days; Mr Briône says that slicing up the final published poll by day shows movement to the Tories, too. Yet this is contradicted by evidence from YouGov, which conducted a poll on election day itself and found no evidence of a Tory surge.
One firm, GQR, claims to have known all along that Labour was in trouble. The polls it conducted privately for the party consistently showed Labour trailing. Unlike most other pollsters, GQR “warms up” respondents by asking them about issues before their voting intention. Pollsters tend to be suspicious of so-called “priming” of voters, which seems just as likely to introduce bias as to correct it.
The BPC’s inquiry will weigh up the competing theories. Given the range of methods and the universal error, a late surge seems the most plausible explanation for now. That would vindicate Lynton Crosby, the Tory strategist, who insisted voters would turn blue late on. Next time expect more scepticism about polls—and more frantic last-minute campaigning.
沒有留言:
張貼留言