What Do the Polls Know?

The pollsters have had some high-profile misses in 2016 and did not always nail the margin, but surveys during the primaries have done a decent job of predicting the winners.

This is especially true for states with the most polls and states that have held primaries. Low-turnout caucuses are notoriously difficult to poll.

“They did a good job in most of the states,” said Andrew Smith, director of the University of New Hampshire Survey Center.

Smith said a poll that is accurate at the time it was taken can appear to miss the mark on primary day if there are a large number of late-deciding voters who break heavily toward one candidate. That frequently is the case in early-voting states like New Hampshire, where as many as 40 percent of the voters make up their minds in the last three days.

“We tend to put over-reliance on polls, when people really haven’t made up their minds,” he said.

Smith attributed some of the misses to methodological errors. He said the media kept referring to his poll as an “outlier” because it showed a wider lead for Sen. Bernie Sanders over former Secretary of State Hillary Clinton. The end result surprised many pollsters. The senator’s winning margin was 9.1 percentage points higher than the final RealClearPolitics polling average.

“Most of the polls … overestimated Clinton’s support,” Smith said. “Most of them rely on samples from previous primaries or on lists of registered voters.”

[lz_table title=”2016 Polling Misses” source=”RealClearPolitics”]Republican Races
|State,Predicted,Actual
Iowa,Trump,Cruz
Oklahoma,Trump,Cruz
Arkansas,Cruz,Trump
Kansas,Trump,Cruz
|Democratic Races
|State,Predicted,Actual
Oklahoma,Clinton,Sanders
Kansas,Clinton,Sanders
Michigan,Clinton,Sanders
Indiana,Clinton,Sanders
[/lz_table]

Smith said his poll instead uses random-digit dialing, which does a better job of capturing new voters. He said 15 percent of voters, for instance, registered on the day of the primary — and they skewed young, a demographic dominated by Sanders.

The RealClearPolitics average called five Republican contests and six Democratic races wrong. But most of those were either low-turnout caucuses or contests with only one poll. And in some of those cases, the single poll was conducted a good bit before the voting.

Sam Wang, who analyzes polls at the Princeton Election Consortium, offered a positive review of the performance of pollsters during the 2016 cycle.

“Polls are fine, and have done well so far — in D and R primaries, for instance,” he wrote in an emailed response to questions from LifeZette. “They’re the main thing we have left in such a crazy year.”

If past is prologue, it probably means that Clinton’s lead over Sanders in the upcoming California primary is small but real. The current RCP average is 6 points, but the margin has been just 2 points in three of the last four surveys. Sanders has no realistic path to the Democratic nomination regardless of what happens Tuesday, but a victory in California would be a major embarrassment to Clinton.

And a victory for Sanders is more plausible than some other wins he has scored. Clinton had a 6.8-point lead in the RCP average in Indiana, for instance, but lost the primary by 5 points. More stunningly, she lost the Michigan primary by 1.5 points even though her lead had been 21.4 points in the final RCP average.

Such volatility has some experts questioning whether polls in the general election will live up to the accuracy of the recent past, particularly since both Clinton and presumptive GOP nominee Donald Trump both have historically high negative ratings.

[lz_related_box id=”147911″]

“We are looking at … the most disliked candidates in a single election,” pollster Jennifer Dineen told The Hill.

Karlyn Bowman, a polling expert with the conservative American Enterprise Institute, told the paper that Trump’s divisiveness could cause some of his supporters to lie to pollsters.

“It’s really hard to gauge a closet vote,” she said.

Smith said pollsters have a pretty good record in the general election and can draw on proven models in building their surveys.

“The past is usually a pretty good predictor,” he said.

The wrinkle is turnout, Smith said. If pollsters overestimate or underestimate turnout of particularly groups of people — young voters or minority voters, for instance — the could get the result wrong.

“That is the one thing that might make it harder to predict,” he said.