Ever since last Monday night's totally bonkers presidential debate at Hofstra University, there's been a near consensus that Hillary Clinton got a significant bounce in the polls, and is effectively in the lead. As I write this, the Real Clear Politics average of national polls has the Secretary of State up by 4.1 points over Donald Trump, compared to the day of the debate, September 26, where she was up by only 2.3.
Actually, 4.1 percent may be something of an understatement. Every single recent national poll has Clinton beating Trump by between 5 and 7 points – that is, all except for one: the LA Times / USC Daybreak Poll, which has Trump up by 4 points.
That's right, two of L.A.'s most respected institutions think Donald fucking Trump is going to win this election (
“It is such an extreme outlier,” says Timothy Johnson, Vice President of the American Association for Public Opinion Research. “There’s a lot of head scratching about it. Their methodology is creative and unusual.”
Indeed. Most opinion polls call voters randomly and ask them who they're going to vote for. Then they weight their answers according to various considerations (race, age, gender, if they voted in the last election), coming up with a mathematical “model” of what they think the likely voter “universe” is.
The Daybreak Poll is an entirely different animal. It has randomly selected about 3,000 U.S. citizens to participate in a months-long survey. Every day, 400 of them are asked, in an online questionnaire, to rate, on the scale of 0 to 100, the percent chance that they will vote for a candidate.
It's sort of like the Nielsen Family model for polling voters – they're asking the same people, over and over again, who they're voting for. It's a method that's been used in sociological studies, but never for a high-profile political survey.
The poll's methodology was designed by a team at the RAND Corporation, headed by Arie Kapteyn, a professor of economics. In 2012, the team tried their poll out on the presidential race and did pretty well. According to their website, they predicted President Obama would win by 3.32 points. The result was more like 3.85 points, but the team's prediction came much closer than most other national tracking polls, which predicted a tighter race.
This year, the team moved to USC, which is paying for the survey.
We asked the USC team if they were nervous that their poll was such an outlier. Survey Director Jill Darling e-mailed us this response:
I am not worried that our poll results are different from other polls. There are several weeks to go and our non-traditional methods may be measuring voter mood in ways that traditional polls cannot. Our probabilistic questions allow voters to quantify their indecision rather than forcing a candidate choice. Also, our participants, who represent all walks of American life, can answer the questions when it is convenient for them, rather than having to be available and willing to answer when the phone rings. I am of course hoping that our final estimates will align with the actual vote, just as all pollsters do. The election poll that this group conducted at RAND in 2012 was considered to be an outlier with an Obama slant, and ended up with an outcome that was very close to the vote. We will see if our methods perform as well in this very unusual election year.
Others are expressing doubts.
In August, election messiah Nate Silver offered a half-hearted defense of the poll, essentially saying that the poll was probably biased towards Trump (his term for this is “house effect”), but hey, at least it was consistently biased, which means that sites like his, which aggregate data from all polls, find it useful.
But since the debate, most polls moved a couple of points in Hillary Clinton's favor. The Times' Daybreak poll has stayed pretty much the same. That means that the poll is qualitatively different from other polls. It isn't just a house effect.
L.A. Times chief political correspondent, David Lauter, tells us that the poll's outlier-ness is due to its model of who is going to vote.
“I don’t pretend to have a crystal ball,” says Lauter. “All I can do is look at the numbers and tell you that if the electorate ends up looking like the one that our poll forecasts, then Trump would win. If it looks more like the one in 2012, Trump won’t win.”
He adds: “You can look at the poll as somewhat of the best case scenario for Trump.”
But Lauter may be downplaying the strangeness of the Daybreak Poll asking the same people the same question over and over again. For example, if the pollsters just happened to choose a sample of people that are overly sympathetic to Trump, or that hate Clinton, that bias is going to linger in the poll, in the way that it wouldn't in other polls.
And the fact that these 3,000 people have been chosen to be part of an experiment – a somewhat high profile one, at that – may change their behavior in unpredictable ways.
“If they know they’re going to be asked, again and again, they might pay attention more than other people,” says Paul Mitchell, Vice President of Political Data, Inc. “If you knew people were measuring how high your hair was, you’d probably be measuring that every morning.” (Note: this reporter is known for having, at times, tall hair.)
Mitchell also floats another theory, one that says people being polled act in ways that they view as socially desirable: “If you view staying consistent with a candidate as socially desirable, you’re going to change your mind less.”
You could look at the Daybreak Poll as some grand experiment. If it's right, while everyone else is wrong, they're going to look like geniuses (that achievement mitigated, somewhat, by Donald Trump becoming president and wreaking all sorts of havoc and destruction upon the world). Or maybe the poll will fall in line sometime in the next month.
But if it doesn't, and they're wrong, USC and the Times are going to have egg on their faces.
TogetherUpdate, 9:00 am: This morning, the L.A. Times/USC Daybreak poll jumped a couple points in Hillary Clinton's favor, and has Trump leading by only 2.6 points. It's still the only current national poll that has Trump leading.