By Jon Cohen and Mark Blumenthal
Every time Donald Trump has won another contest, or risen in the polls, or survived another round of criticism, a chorus of voices has risen up to protest: This can’t possibly be real.
It happened again this past week. In an op-ed column in the New York Times, academics Norman Ornstein and Alan Abramowitz announced that recent polling showing Trump within striking distance of Hillary Clinton in the general election race can’t possibly be right. Relying on the most superficial evidence, they dismissed recent surveys as “wrong,” “cringeworthy,” built on “distortions” and reflecting not reality but “artifacts of sample design and polling flaws” produced by “mediocre or bad artists.”
But as we should all realize by now, it’s not enough for Trump’s opponents to wish him away. It’s important for political professionals to actually explore what is buoying Trump — even if they find his rise unfathomable.
Let us be clear: We believe it’s always dangerous to make predictions about the election more than six months out, especially at a moment when one party’s nomination contest is resolved, and another remains contested. Polls are a snapshot in time, not rarified glimpses into a crystal ball.
Our latest NBC News/SurveyMonkey online election tracking poll, which found Clinton leading Trump narrowly among registered voters (48 to 45 percent), is just such a topline glance. Any individual poll — certainly including ours — merits close scrutiny.
Ornstein and Abramowitz fail this test of seriousness.
First, they argue that polls have been on a “wild ride” recently, showing “wild fluctuations.” That’s partly true — polls can vary at this stage of a campaign for a lot of reasons. But one way to smooth out the vagaries of varying methodologies and smaller samples, which Ornstein and Abramowitz actually recommend in their write-up, is to turn to polling averages. And this year, the HuffPost Pollster and RealClearPolitics averages show a fairly clear trend: The Clinton-Trump margin has narrowed significantly, following Donald Trump’s elevation to presumptive Republican nominee following his victory in the Indiana primary on May 3.
Second, Ornstein and Abramowitz refuse to believe recent polls, including ours, that show Trump favored by 28 percent of Latino voters. That result “wouldn’t raise eyebrows” if it were about Bush or Romney, they write, but “most other surveys have shown Mr. Trump eking out 10 to 12 percent among Latino voters.”
“Most other surveys?” Six other recent national polls, all conducted since Trump attained presumptive nominee status, found his support among Hispanic voters varying between 15 and 31 percent. SurveyMonkey’s 28 percent result was certainly on the high side of that range, but our estimate of Hillary Clinton’s Latino support (65 percent) was the third highest of the six polls. Why? Because our questions are structured to encourage respondents to make a choice, one of the many design decisions polls make at this stage that have a real effect on the results. For example, when asked to choose between Trump and Clinton, our latest poll had only 7 percent of respondents undecided; the latest Morning Consult poll had that number at 17 percent.
We asked Abramowitz, via email, for more specifics on the polls he relied on for this conclusion. The exchange yielded only two conducted at the national level showing Trump’s support among Latinos in the 10–12 percent range, and both were conducted in April, when the Republican nomination remained contested. We have no beef with the methods used by these polls, but the claim that “most” other surveys show Trump at just 10–12 percent of Hispanics is simply false.
A bigger and more critical issue raised by their column is the advice to “use gold-standard surveys, like Pew.” We certainly share their respect for the work of the Pew Research Center, but like Gallup, the Center has not done any presidential horserace polling in 2016.
The late Andrew Kohut, founding director of Pew Research, once summed up his polling philosophy with these words: “I’m not a handicapper, I’m a measurer. There’s a difference.” He meant that he relied, not on “ad-hoc judgments” about the likely outcome of an election, but rather on trusted survey measures, honed for decades, to produce his findings.
That credo remains essential. Now more than ever, at this moment of reinvention for public opinion polling, we need many independent estimates of voter preferences, not a herd of handicappers issuing their best guesses about the eventual outcome.
The methodology behind SurveyMonkey’s Election Tracking is relatively new, and will continue to evolve over the course of the 2016 election and beyond. We are committed to transparency about our approach, and open to honest reflection about its strengths and weaknesses, but not dismissiveness based on cherry picking.
Sadly, the telephone methodologies long considered “gold standard” have their own challenges, as Ornstein and Abramowitz acknowledge. “Getting reliable samples of voters is increasingly expensive and difficult,” they write, as Americans drop their landline phone and response rates plunge. That’s certainly true.
The consequence, unfortunately, is that while telephone surveys continue to gather quality data, completed samples of a thousand or more interviews have become prohibitively expensive for most news organizations.
Case in point, the latest New York Times survey, published on the same day as the Ornstein and Abramowitz column, which reported results for white voters and African Americans but nothing at all about preferences of Hispanics.
Perhaps the definition of “gold standard” polling needs an update.
Jon Cohen is Chief Research Officer of SurveyMonkey, and formerly ran polling for The Washington Post. Mark Blumenthal is SurveyMonkey’s Head of Election Polling and formerly Senior Polling Editor for The Huffington Post. More information on SurveyMonkey’s Election Tracking is available here.