The Kansas City pollster who called the Trump upset
In the days leading up to the election, the political establishment largely ignored, and, in some cases, ridiculed the polling being published by Remington Research Group.
“I was definitely laughed out of a couple of rooms,” says Titus Bond, director of polling for the Kansas City-based operation. “People had me feeling like a crazy person — like I was a nut on some pro-Trump Reddit page.”
There were some reasons to be skeptical. Remington falls under the umbrella of Jeff Roe’s political consulting firm Axiom Strategies. Roe, the architect of Ted Cruz’s recent presidential campaign, is notorious for being one of the nastiest and most ruthless political operatives working in America today.
As the author of this Nov. 7 post suggests, it is not impossible to imagine a Republican-centric organization like Axiom seeking to, among other things, bump up turnout in support of down-ballot Republicans. One way to do that would be to publish polls that make the race appear closer than it was.
As of Nov. 3, Remington had Trump surging ahead in the battleground states of Ohio, North Carolina, Nevada, and Florida. Meanwhile, all the leading vote forecasters — FiveThirtyEight, the Upshot, the Princeton Election Consortium — put Clinton’s chances of winning anywhere between 70 percent and 99 percent.
“We’re looking at a margin of error race in seven of the eight battleground states we surveyed,” Bond said in a release announcing Remington’s final poll of the campaign. “Either one of these candidates could realistically run the table in these seven states. The big question is if Donald Trump can mobilize his voters and get them to the polls. If he can’t, then he will lose.”
Trump’s voters came out. Clinton’s didn’t. Trump won.
I called up Bond on Wednesday evening, less than 24 hours after the race had drawn to a close. He didn’t gloat too much, but he spoke with the air of a vindicated contrarian, and rightly so. Our conversation has been lightly edited for clarity.
How exactly does Remington Research Group fit into the world of Axiom Strategies and Jeff Roe?
We were founded by [Roe] in 2014. I was an employee of Axiom at the time, working on internal polling for our clients. There were two Kansas races we were interested in that year: [Pat] Roberts vs. [Greg] Orman for the U.S. Senate, and [Sam] Brownback vs. [Paul] Davis for governor. A lot of the national surveys had the Democrats winning both. I wanted to see where Kansas was, so I did an internal poll and got numbers back that were significantly different than the numbers Fox News and Public Policy Polling had. My numbers showed that both Republicans were poised to win.
We were confident in the data and thought the other firms were wrong, so we created a brand that could release public surveys if we felt something was being missed. And so that’s Remington. We pay for those polls on our own dime. But most of our polling is internal polling for governors, members of Congress, large corporations. So the public polling is sort of a marketing thing. It’s a way to get our company in the news. When everybody else is wrong, and we are right, that’s obviously good for our business.
As far as this cycle, earlier this year, we were doing some Congressional public polling but nobody really seemed to care what we were finding on a House race in Wyoming or whatever. So I pitched the idea of doing public polling on Hillary and Trump in battleground states.
Why was Remington so much more right than everybody else about Trump?
So, one thing that I don’t think is spoken about often enough in polling is the idea of “herding.” Let’s say three polls arrive on the same day. SurveyUSA has Hillary up two points. Monmouth has Hillary up two points. Public Policy Polling has Hillary up three points. So say you’re Emerson Polling, and you get your polling results back the following day, and you’ve got Trump up one point. You look at the other guys and think, I’ll be laughed at and scrutinized and not taken seriously if I publish this. So you manipulate the data a little to make it seem a little closer to what your competitors have.
There are all kinds of ways to do this. It’s about how you decide to interpret the information. The Upshot recently had a post where they gave the same raw data to four different pollsters and got drastically different results. Part of these variations happen on the front end, with the sample you’re getting. For example, in Kansas, you can look at historical turnout by party affiliation and see that it’s 58 percent Republican, 28 percent Democrat. So if your poll sample is only 40 percent Republican, then you gotta go in and weight the data in other ways to make up for that. If you wanted to weight it more lightly, you might play those numbers against all registered voters in the state, as opposed to the ones that historically turn out — which is more accurate, and what we always do.
The short answer is, I think pollsters are afraid when they see weird numbers. So they will screw with the weighting and analysis until it’s close enough to everybody else that they feel like they’ve covered their ass and not wasted ten grand on a poll. We’d rather be right. What we care about is having an accurate snapshot of what’s actually going on.
Some forecasters had Clinton’s odds at over 90 percent. How is that possible?
Of all the major poll forecasters, Nate [Silver] at FiveThirtyEight [which had Clinton at 71 percent on Tuesday morning] was the closest, and the reason why is because he had begun to include our polls in his forecasting. I believe we accounted for something like a 28 percent bump in Trump’s odds in the FiveThirtyEight forecast. Silver uses anything he believes is a legitimate survey in his forecasting model, and when he started including our surveys in his model, he took flack for it — from Democrats and even Republicans. But he knew us, and frankly I think he knew about our history of success with internal polling. So he felt comfortable including us.
What specifically were you seeing in the Trump numbers?
We polled in the battleground states, but we also underlayed that by polling specific battleground counties in those states. So, Luzerne County, in Pennsylvania. It’s in the Scranton area. It’s economically depressed, 95 percent white, blue collar, and it typically goes for Democrats. [Obama won the county by five points in 2012.] We had Trump up ten there. So if that’s a good indicator, that means Pennsylvania is more likely to flip. [It did. Trump won by 20 points in Luzerne County.]
What surprised you?
Trump winning Wisconsin. We didn’t have that. It’s a state, like Missouri, where there’s no party registration, so you have to just try your best when forecasting what the percentages of Democrats and Republicans will be. Really, the same thing happened in Missouri as in Wisconsin, which is that the Democrats didn’t show up, and that threw everything off. We didn’t poll in Michigan but I really wish we would have.
Any thoughts on what happened in Missouri?
Nothing too surprising. We had Republicans winning safely by ten points or more, except for [U.S. Senator] Roy Blunt. Trump won by 20 but Roy only won by three or four, which is a significant difference between two Republicans in the same state. And that’s because there was a ton of crossover support between Trump and [Blunt opponent Jason] Kander. Trump’s populist “drain the swamp” message is essentially running against people like Blunt.
Still, we saw the biggest Republican margins ever in Missouri. I saw campaign internals throughout these races and nobody had the Republicans winning by 20 points. Among other things, there was a much lower turnout of African-Americans in the St. Louis area, which really helped Republicans. The Democrats were not excited, they didn’t care, and they didn’t show up.
Did your polling indicate that the situation with James Comey and the FBI had an impact on Clinton’s numbers?
None at all. This thing was going to be close all along.
What is your response to the argument that Remington essentially had a dog in the hunt, that it was in your interest as a Republican organization to inflate Trump’s poll numbers?
I mean, you don’t have to know much about the background of my firm to know that we’re not ginormous fans of Donald Trump. Jeff [Roe] was Ted Cruz’s campaign manager. Now, obviously Ted eventually endorsed Trump, and we are happy that Trump winning saved Republicans from a lot of Congressional losses. But again, we want to be right. We had Hillary winning in Colorado and Virginia. It’s not like we were trying to prop up Trump.
Do you think there is a correlation between the rise of populist rage across the globe and the inability of polling to get things like Trump and Brexit right?
My data doesn’t necessarily suggest that. There was always a theory out there during this election that people were too embarrassed to say they were voting for Trump because of how toxic he is — because it’d be like admitting you’re a racist, chauvinist, you hate poor people, you’re a horrible person, whatever. We largely use automated surveys as opposed to live callers, though, so it’s possible Trump supporters were more willing to tell the truth to a computer than to another human being.