Polls Were Way Off on Donald Trump. Here’s What It Means

I'm not being treated fairly here (hint, hint)

(New York Times) - Donald Trump was at the top of each of the last 10 polls in Iowa, but his lead failed to hold up on caucus night Monday. In the end, his seven-point lead in polling averages amounted to a three-point loss to Ted Cruz.

That 10-point swing was enough to make Mr. Trump’s defeat the biggest polling error in an early primary since Hillary Clinton defeated Barack Obama in New Hampshire in 2008. But even that measure understates the extent that the polls misjudged Mr. Trump’s strength.

Mr. Trump was at 31 percent in the final polls, but finished with just 24 percent. In our data set of early primary polls from New Hampshire and Iowa since 2004, no candidate underperformed the final surveys by as much as Mr. Trump. Mrs. Clinton, for instance, mainly beat Mr. Obama by outperforming her polling, not because Mr. Obama fell short.

It’s probably not a coincidence that the candidate who underperformed the polls by the most is also the one who had a mediocre turnout operation and enjoyed seemingly nonstop media coverage.

It’s always hard to figure out why polls are wrong, but this time the stakes are higher. Republican strategists have hoped for months that Mr. Trump’s lead was an illusion. The results in Iowa at least raise the possibility that they’re right — which would call into question Mr. Trump’s advantage elsewhere.

This time there is evidence to support one of two possibilities for why polls overestimated Mr. Trump: Voters broke strongly against Mr. Trump in the final days, or the electorate was more conservative and more religious than polls anticipated.

In general, there are three basic ways polls go wrong:

■ an unrepresentative sample that doesn’t accurately reflect the population it’s trying to measure.

■ a flawed likely-voter model that misjudges the composition of the electorate.

■ late events or changes in the race after the poll was conducted that moves voters.

This year, there is extremely strong evidence to support the “late movement” scenario, some evidence to support the likely-voter problem and little evidence to support the sampling problem — even if it can’t be ruled out. Read more.