By David Knowles/Editor Yahoo News
October 24: Thanks to the hidden support from voters who are embarrassed to admit they will vote for Donald Trump, the President will be narrowly reelected on November 3, says one of the few pollsters who correctly predicted his 2016 victory.
Pollster Robert Cahaly, the head of the Georgia-based Trafalgar Group, saw interest in his company skyrocket in 2016 after he bucked the consensus of other pollsters and forecast that Trump would beat Democrat Hillary Clinton in Michigan and Pennsylvania, two states that were crucial to his victory. He credits a proprietary model of calculating a polling sample that takes into account the social pressure that leads many people to hide their support of a controversial, polarizing candidate like Trump, even from anonymous pollsters.
It’s that same demographic that Trump often refers to as the “silent majority,” whom the president says will help him “win this election big.” Unlike the rally goers in MAGA hats who have memorized the lyrics and sing along to Lee Greenwood’s “God Bless the USA,” these “shy Trump voters,” Cahaly told Yahoo News in an interview Friday, will show that the methods of most polling outfits are outdated and unreliable.
Unlike those bigger, more traditional pollsters, Trafalgar eschews long, in-person interviews with voters, providing them with other options to record their voting preferences — including online and by text — that critics say are less scientific but that Cahaly said in the following interview (which has been lightly edited for clarity) more accurately reflect human psychology.
Can you explain the theory behind that so-called “social desirability bias” that you’ve said caused some state polls to be off about Trump in 2016?
The social desirability bias is a theory that’s been around for a long time. It is that, especially when dealing with a live caller, the person being asked the questions will craft their answer in a way that puts them in the best light to the person asking the question. Kind of in the sense that if somebody hideously dressed walks up to you and says, “How do I look?” You say, “You look fine.” I think every guy’s been in a position where some lady asks how they look, and you don’t say “bad.” There’s politeness. I grew up in the South, where people never really tell you what they think. This was nothing new to me. I’m used to this. Some of my first experiences in politics… I saw every Jesse Helms race, all the ads, and I remember hearing Jesse Helms was only losing by 5, so he was going to win by 2. People won’t admit they’re going to vote for him. I saw this in practice, grew up with it.
What specifically about the question ‘Who do you plan to vote for, Hillary Clinton or Donald Trump?’ made people go out of their way to hide their intention to vote for Trump?
For the first time in a long time we had not only a candidate being attacked, but their supporters being attacked. Remember, [Clinton] called his supporters “deplorable.” It was beyond what we’d seen in the past. If you supported him, you were everything they were accusing him of being. There was a shaming of his supporters. He was a rejection of political correctness, so if you were for him, you were not politically correct. There’s a penalty to be paid for not being politically correct, so we saw that and we used a tactic in 2016 that I probably shouldn’t have explained to the whole world — the neighbor question [asking people who they think they’re neighbors will vote for] — seeing as how many other polling groups now use it and never give us credit for it.
In 1982, Los Angeles’s African American mayor Tom Bradley lost the governor’s race in California despite leading in polls. The theory is that the surveys were skewed by white voters hiding their intention to vote against Bradley out of a reluctance to be perceived as racist. How does one accurately go about quantifying such hidden motivations?
You really don’t. You have to figure out a way to minimize it, and there is a way to minimize it. You poll people in a way that makes them feel very, very anonymous. The more anonymous they believe a poll is, the more the social desirability effect doesn’t play. In the same way that you’ve got a person that has their public Facebook or their Twitter account where they put pictures of their family, and then some of these people also have a little troll account that they go on to cause mayhem. Well, their genuine emotion, their true feelings are in that troll account. They don’t want to own those feelings, but those are what go in the ballot box. I don’t want the person that’s behind your public Facebook account, I want the person that’s behind your troll account. I want that person’s opinion, because that’s how you’re going to vote.
You can minimize hidden motivations [in polling responses], but you can’t eliminate it. Our goal is to minimize it. When I look at Pennsylvania, for example, I’ve got Biden up by one point, but I don’t think Biden is going to win Pennsylvania. I think Trump is probably going to win it. Even in my poll I see people in there who are saying they are for Biden, they are undecided, they are for Jorgensen, who [based on] the other questions we ask, they’re not really for Biden. I see that they’re not expressing opinions that are consistent with a Biden voter. I’m going to report that they’re for Biden, because it’s what they said. I think there’s even social desirability in our polls. We’ve minimized it, but it still exists. I think Trump will out-perform our polls by a point or two.
In an interview with Sean Hannity this week, you said that most polls are not accounting for “shy Trump voters,” do you think that there are any “shy” Biden voters, and if not, why?
No. If Bernie [Sanders] had won the nomination and you would have been approaching moderate Democrats in a poll, then I would have seen shy on both sides — moderate Democrats who did not want to admit they were voting for a socialist. But because of Biden’s persona being so moderate and in the middle, there’s just no stigma to voting for him. No one looks at you and says you are so-and-so for supporting Biden. No one calls you a racist, sexist, homophobe — nobody says that.
Polling guru Nate Silver has been critical of Trafalgar, giving you a C-minus rating. “They know what result they want, then engineer their sample, etc., to get there,” he has written. How do you respond to that?
Nate Silver makes money off of clicks. God bless him for doing so, that’s his business. He’s got to be provocative. We make most of our money in the private sector only because we have a reputation for getting it right.
Do you consider yourself a pro-Trump pollster?
No. Don’t get me wrong — as a Republican, of course I would like to see him win, but you know what I’d like more? I’d like to be right.
The general consensus about why state polls of the Rust Belt were wrong in 2016 is that education was underweighted as a factor, specifically that less educated voters were much more likely to vote for Trump and were underrepresented in the samples used by pollsters. Some pollsters say they’ve since adjusted their models to fix that mistake and some now include the neighbor question. Do you think they’ll still be wrong?
They’re not using [the neighbor question] properly. They factor it in, but they don’t understand it. They think that’s what they got wrong. They are not going to let go of this [practice] of calling a live person and asking a crazy amount of questions, 20, 30, 40 questions. That is the hill they’re going to die on. If they admit that model is now flawed and outdated, that in this modern world average people don’t participate in crazy long polls and that people lie to pollsters, if they admit that, they’ve lowered the threshold to entry into the polling game to digital, and when they do that, they can’t justify charging the crazy numbers they charge for polls. They are the Pony Express trashing the telegraph lines. They don’t believe that people are hiding their votes for Trump. Tell me this, why were they all wrong in Florida in predicting [Ron} DeSantis vs. [Andrew] Gillum, where the social desirability factor was in play, yet it wasn’t in the [Bill] Nelson race. We got both of them right. If they had fixed it, why did they all get Florida wrong? What they don’t understand is: In an age when people will lie to their doctor, lie to their priest, lie to their accountant, suddenly they turn into honest Abe when they answer a polling call? That is silly.
But in 2018 you were also pretty far off, more than 10 points, in your prediction for the governor’s race in Georgia [between Democrat Stacy Abrams and Republican Brian Kemp]—
I’ve owned it. That’s one of the only ones where we were way off. Unlike them, I admit my mistakes. We still have 96.2 percent of the time we call it right and I’ll put that against any of them.
Did the social desirability bias play into that error?
No. The reason we were wrong is we did not pull a new voter list. We did not go get new data. Stacy Abrams did a tremendous job registering new people to vote. We are still a small operation. We did not spend the money. It was a mistake.
In 2016 you were right about Michigan. You were the only pollster showing Trump winning that state. Your latest poll of Michigan shows Trump with a 1.8 percentage lead, but the Real Clear Politics average shows Biden with a 7.8 percent lead. That’s a greater discrepancy between you and the other polls than in 2016. What explains it?
I think we’ve got a few factors going on: One, the [Republican] U.S. Senate candidate John James is a significant benefit to Trump. Trump is also doing extremely well with the Black vote in Michigan, and we’re also seeing a number of people who are saying they’re undecided in the Black community that I think are going to break toward Trump. With everything that’s happened with [Gov.] Whitmer, she might as well be Biden’s running mate. She’s dragging him down in Michigan. It is a state where people are being polled to death. It’s the state that takes us the longest time to get a good sample, and Republicans are very unlikely to participate in polls, if you’re not trying extra-hard to get them. … You’ve got to try to get your sample right, and I think [other pollsters] are not making the effort to get a fair representation of Republicans. … We’re going to know whether we’re right or wrong pretty soon, but I feel Michigan is going to break Trump’s way, I really do.
Frank Luntz said last week that “If Donald Trump surprises people,’ and polls are shown to have been wrong again, ‘my profession is done.” I take it you think there would be a future for Trafalgar if you end up getting it right again.
I think so. Polling, like everything else, must evolve. We’ve seen what happens when industries don’t evolve. This one has not evolved. Forget all of the technical mumbo-jumbo and just think about your own life. Can you imagine the phone ringing at 6:30 or 7 and stopping what you are doing to spend 7 to 15 minutes answering a poll? Who is the person who does that? They’re extremely on the right, they’re extremely on the left or, worst of all, they’re bored. Average people do not have that kind of time. If you are not giving them a short survey that they can take at the moment or at their convenience where some of emails and texts and other online [options] can let them receive it in one moment and participate later, how in the world are you supposed to get average people participate? And we all know at least one person who is for Trump and has people in their lives they don’t want to know it. Explain to me how that person, who won’t put a Trump sign in their yard, is going to tell a stranger on the phone they’re for Donald Trump.