You can see there’s definitely a correlation. Six of the 12 eventual nominees were leading at this point in the national polls. But they were all polling better than Trump is now. Not only that, but 52 percent of the variation in the eventual results go unexplained. That’s a mathy way of saying that a lot tends to happen from this point on. There have been collapses: Giuliani in 2008 and John Glenn in 1984. There have been surges: Barack Obama in 2008 and John Kerry in 2004. Glenn, for example, consistently polled as well or better than Trump is nationally and in the early primary states, and Glenn fell off the map completely once voting began.
Put another way, past campaigns suggest that 95 percent of the time, Trump’s eventual percentage of the national primary vote will be between 8 percent and 64 percent. And there’s reason to think Trump will end up on the lower end of that range. He doesn’t have a single endorsement from a governor or member of Congress, and those endorsements have historically been predictive of the eventual winner. [NW: But Jeb had all of those and did terribly, right?]
From the late nineteen-nineties to 2012, twelve hundred polling organizations conducted nearly thirty-seven thousand polls by making more than three billion phone calls. Most Americans refused to speak to them. This skewed results. Mitt Romney’s pollsters believed, even on the morning of the election, that Romney would win. A 2013 study—a poll—found that three out of four Americans suspect polls of bias. Presumably, there was far greater distrust among the people who refused to take the survey.
The modern public-opinion poll has been around since the Great Depression, when the response rate—the number of people who take a survey as a percentage of those who were asked—was more than ninety. (...) A typical response rate is now in the single digits.
(...)
Using methods designed for knocking on doors to measure public opinion on the Internet is like trying to shoe a horse with your operating system. Internet pollsters can’t call you; they have to wait for you to come to them. Not everyone uses the Internet, and, at the moment, the people who do, and who complete online surveys, are younger and leftier than people who don’t, while people who have landlines, and who answer the phone, are older and more conservative than people who don’t. Some pollsters, both here and around the world, rely on a combination of telephone and Internet polling; the trick is to figure out just the right mix. So far, it isn’t working. In Israel this March, polls failed to predict Benjamin Netanyahu’s victory. In May in the U.K., every major national poll failed to forecast the Conservative Party’s win.
(...)
“It’s a little crazy to me that people are still using the same tools that were used in the nineteen-thirties,” Dan Wagner told me when I asked him about the future of polling. Wagner was the chief analytics officer on the 2012 Obama campaign and is the C.E.O. of Civis Analytics, a data-science technology and advisory firm. Companies like Civis have been collecting information about you and people like you in order to measure public opinion and, among other things, forecast elections by building predictive models and running simulations to determine what issues you and people like you care about, what kind of candidate you’d give money to, and, if you’re likely to turn out on Election Day, how you’ll vote. They might call you, but they don’t need to.
(...)
With the turn to the secret ballot, beginning in the eighteen-eighties, the government began supplying the ballots, but newspapers kept printing them; they’d use them to conduct their own polls, called “straw polls.” Before the election, you’d cut out your ballot and mail it to the newspaper, which would make a prediction. Political parties conducted straw polls, too. That’s one of the ways the political machine worked.
Straw polls were usually conducted a few days or weeks before an election. This August, to cull the field for the first G.O.P. debate, Fox News used polls conducted more than four hundred and sixty days before the general election. (These early polls have become so unreliable that neither Gallup nor Pew conducts them.)
A century ago, newspapers that wanted to predict the outcome of a Presidential election had to join forces. In 1908, the New York Herald, the Cincinnati Enquirer, the Chicago Record-Herald, and the St. Louis Republic tallied their straws together. William Randolph Hearst’s newspapers did the same thing. But the best predictions were made by a national magazine, the Literary Digest, beginning in 1916. It regularly miscalculated the popular vote, but for a long time it got the Electoral College winner right. In 1920, the Digest mailed out eleven million ballots. By 1932, its mailing list had swelled to twenty million. Most of those names were taken from telephone directories and automobile-registration files. George Gallup was one of the few people who understood that the Digest risked underestimating Democratic votes, especially as the Depression deepened, because its sample, while very big, was not very representative: people who supported F.D.R. were much less likely than the rest of the population to own a telephone or a car.
(...)
When Gallup started out, he was skeptical about using a survey to forecast an election: “Such a test is by no means perfect, because a preelection survey must not only measure public opinion in respect to candidates but must also predict just what groups of people will actually take the trouble to cast their ballots.” Also, he didn’t think that predicting elections constituted a public good: “While such forecasts provide an interesting and legitimate activity, they probably serve no great social purpose.” Then why do it? Gallup conducted polls only to prove the accuracy of his surveys, there being no other way to demonstrate it. The polls themselves, he thought, were pointless.
(...)
Donald Trump doesn’t have a campaign pollster, but, while he was leading them, his campaign loved polls. Polls admitted Trump into the first G.O.P. debate and polls handed him a victory. “Donald J. Trump Dominates Time Poll,” the Trump campaign posted on its Web site following the August debate, linking to a story in which Time reported that forty-seven per cent of respondents said that Trump had won. Time’s “poll” was conducted by PlayBuzz, a viral-content provider that embeds quizzes, polls, lists, and other “playful content” items onto Web sites to attract traffic. PlayBuzz collected more than seventy-seven thousand “votes” from visitors to Time’s Web site in its instant opt-in Internet poll. Time posted a warning: “The results of this poll are not scientific.”
(...)
The statistician Nate Silver began explaining polls to readers in 2008; the Times ran his blog, FiveThirtyEight, for four years. Silver makes his own predictions by aggregating polls, giving greater weight to those which are more reliable. This is helpful, but it’s a patch, not a fix. The distinction between one kind of poll and another is important, but it is also often exaggerated. Polls drive polls. Good polls drive polls and bad polls drive polls, and when bad polls drive good polls they’re not so good anymore.
(...)
Douglas Rivers is a professor of political science at Stanford who is also the chief scientist at YouGov. He started trying to conduct public-opinion surveys via the Internet in the nineties, and has done much of the best and most careful work in the field. When he co-founded Knowledge Networks and conducted polls through Web TV, he used probability sampling as an alternative to quota sampling. The initial response rate was something like fifty per cent, but over time the rate fell into the single digits. Then came the Internet crash. “We slimmed down,” Rivers told me when I visited him in Palo Alto. “I went back to teaching.”Rivers then started a company called Polimetrix, which he sold to YouGov for an estimated thirty-five million dollars. There he developed a method called “matched sampling”: he uses the U.S. Census Bureau’s American Community Survey, which surveys a million people a year, to generate a random sample according to “fifteen variables of representativeness” and to determine who will participate in polls. “You get a million people to take the poll, but you only need a thousand, so you pick the thousand that match your target population,” he explained to me.
(...)
The best and most responsible pollsters, whether Democratic, Republican, or nonpartisan, want nothing so much as reliable results. Today, with a response rate in the single digits, they defend their work by pointing out that the people who do answer the phone are the people who are most likely to vote. Bill McInturff, of Public Opinion Strategies, told me, “The people we have trouble getting are less likely to vote.” But the difficulty remains. Surveying only likely voters might make for a better election prediction, but it means that the reason for measuring public opinion, the entire justification for the endeavor, has been abandoned.
(...)
Donald Trump is a creature of the polls. He is his numbers. But he is only a sign of the times. Turning the press into pollsters has made American political culture Trumpian: frantic, volatile, shortsighted, sales-driven, and anti-democratic.
A cruise through the betting websites (PaddyPower, Bovada, Intrade, the University of Iowa’s futures contracts and so on) paints a much less volatile — though not static — picture, in which Mrs Clinton has strengthened her standing. (...) She is trailed by Mr Rubio at 400, Mr Trump at 600 and Mr Cruz at 1000*. It may be noted that the oddsmakers rank Mr Trump only second favourite to win the GOP nomination, in sharp contrast to the opinion polls that have him ahead by streets. They must have some faith that the party establishment — which loathes both Messrs Trump and Cruz — will find a way of asserting itself.
Of course stuff does happen and change the odds, as when people actually cast a vote rather than express opinions. Before the Iowa caucuses in 2008 Mrs Clinton was given a 70 per chance of winning the Democratic nomination against Mr Obama with 25 per cent. Even after the caucuses, which he won, she was given a 52-44 per cent edge. By April, however, with several big state primaries remaining in her strongholds, he was north of 80 per cent while she was logging in at barely 12 per cent.
The analysis, by Morning Consult, a polling and media company, looked at an odd occurrence that has cropped up repeatedly this year: Trump generally has done better in online polls than in surveys done by phone.
The firm conducted an experiment aimed at understanding why that happens and which polls are more accurate -- online surveys that have tended to show Trump with support of nearly four-in-10 GOP voters or the telephone surveys that have typically shown him with the backing of one-third or fewer.
Their results suggest that the higher figure probably provides the more accurate measure. Some significant number of Trump supporters, especially those with college educations, are "less likely to say that they support him when they’re talking to a live human” than when they are in the “anonymous environment” of an online survey, said the firm's polling director, Kyle Dropp.
(...)
The most telling part of the experiment, however, was that not all types of people responded the same way. Among blue-collar Republicans, who have formed the core of Trump's support, the polls were about the same regardless of method. But among college-educated Republicans, a significant difference appeared, with Trump scoring 9 points better in the online poll.The most likely explanation for that education gap, Dropp and his colleagues believe, is a well-known problem known as social-desirability bias -- the tendency of people to not want to confess unpopular views to a pollster.
Blue-collar voters don't feel embarrassed about supporting Trump, who is very popular in their communities, the pollsters suggested. But many college-educated Republicans may hesitate to admit their attraction to Trump, the experiment indicates.