There have been substantial differences between poll projections and election results in many nations in recent months.
Before last week’s British election, the consensus of all reputable pollsters was that the governing Conservative Party and the opposition Labour Party would each receive around 33 percent of the vote and between 260 and 290 seats in the House of Commons. The Conservatives ended up with 37 percent of the vote and 331 seats, compared with Labour’s 30 percent of the vote and 232 seats.
The pollsters also did not project the Scottish National Party’s overwhelming sweep of parliamentary seats in Scotland. While most polls showed that the SNP would be the largest party in Scotland, few polls projected the SNP winning 56 of the 59 Scottish seats it actually obtained.
The British election was only one of several recent votes in which pre-election polling diverged considerably from actual election results. Also last week, the New Democratic Party won a substantial majority of seats in the provincial parliament in Alberta — the first time a left-wing party has ever formed the government in that Canadian province. Pre-election polls showed the NDP with, at best, a likelihood of forming a minority government in a parliament with substantial representation from two other parties.
In Israel in March, polls indicated a very close race between Prime Minister Benjamin Netanyahu’s Likud Party and the opposition Zionist Union coalition. On Election Day, Likud ended up with leads of 5 percent of the vote and 6 parliamentary seats over Zionist Union, substantial gaps in Israel’s multi-party political system.
Finally, here in Vermont last November, no publicly available poll indicated the collapse of Gov. Shumlin’s support in the last weeks of the campaign and his near-defeat by Republican Scott Milne.
Three factors appear to explain the gaps between poll projections and election results. First, pollsters have not adequately adjusted their sampling methods to take account of the worldwide increase in cell phone usage. Many poll samples over-represent landline telephone users.
The Marketing Research Association estimates that close to 45 percent of American households currently use only mobile phones, with no landlines. That percentage will increase in the months and years ahead. Also, polling mobile phone users in the U.S. is more difficult than polling landline users, because federal regulations prohibit the use of automatic dialing systems to call cell phone numbers.
Second, even if mobile phone users are properly represented in poll samples, pollsters are having increasing difficulty projecting which people surveyed will actually vote on Election Day. In Scotland, turnout was considerably higher than pollsters’ estimates, explaining at least some of the SNP’s surge at the end of the campaign.
In Vermont last November, polls over-estimated turnout, a particular problem for the Shumlin campaign in what turned out to be the lowest-turnout election in modern Vermont history. Many voters inclined to support Shumlin were not enthusiastic enough about him, or thought he would win re-election without them, that they simply did not cast a ballot.
Finally, parties and candidates are becoming more effective at using last-minute public, advertising, and social media appeals to voters. In Israel, Netanyahu presented stark warnings about the dangers of Zionist Union’s policies to nationalist voters on Election Day itself. In Britain, Prime Minister David Cameron successfully used fears of SNP influence over a Labour government to win votes in England in the final days of the campaign. These appeals came too late to be reflected in pre-election polls.
I am going to be very cautious about relying on poll projections in Vermont’s 2016 election cycle. The only poll that counts is the one that is taken on Election Day.
Eric L. Davis is professor emeritus of political science at Middlebury College.