USC Pollster: Our Experimental But Accurate Model Shows Trump Winning the Electoral College Again

|
|
Posted: Oct 28, 2020 10:05 AM
USC Pollster: Our Experimental But Accurate Model Shows Trump Winning the Electoral College Again

Source: Official White House Photo by Shealah Craighead

Earlier this week, we wrote about the Trafalgar Group, an outlier pollster that routinely produces rosier results for President Trump's re-election prospects than any of its competitors. If you're on the Trump Train, you love Trafalgar, and you likely already know that its data was much more predictive of some of Trump's upset victories in 2016 -- not to mention calling the 2018 Florida governor's race correctly when basically no one else did. That said, you may be less interested in the outfit's big swings and misses, like overestimating Brian Kemp's victory margin in Georgia's gubernatorial contest last cycle by ten points. True to form, Trafalgar's latest survey shows President Trump inching into the lead in Pennsylvania:


For context, the previous six Keystone State polls gave Biden an average lead of 5.5 points (though another right-leaning pollster also shows Trump surging ahead). Trafalgar's chief pollster told me that part of their "secret sauce" is asking questions that reduce the impact of "social desirability bias" in respondents' answers. The idea is that given the loud opposition to Trump across much of mass media and other taste-making institutions, some people may not want to admit to a stranger over the phone that they plan on voting for him. Part of the way Trafalgar tries to home in on someone's true intentions is to ask them more indirect questions that may reveal their actual preferences. As it turns out, USC's pollster has been doing something similar for a number of years, and here's what they've found:

From our previous research on social judgments, we learned that people seem to know their immediate social circles quite well. Their answers about the distribution of income, health status — even the relationship satisfaction of their friends, family and acquaintances — were often in the right ballpark. And when we averaged the data from their responses across a large national sample, it provided a surprisingly accurate picture of the overall population…in all five of the elections in which we tested this question, the social circle question predicted election outcomes better than traditional questions about voters’ own intentions. These five elections were the 2016 U.S. Presidential election, the 2017 French Presidential election, the 2017 Dutch Parliamentary election, the 2018 Swedish Parliamentary election, and the 2018 U.S. election for House of Representatives. In both the U.S. elections, the social-circle question predicted national and state level results better than the “own intention” question in the same polls. In fact, data from the social-circle question in 2016 accurately predicted which candidate won each state, so it predicted Trump’s electoral college victory

Five for five thus far. So what is their data showing this year? Their traditional polling model gives Biden a double-digit lead in the national horserace, pointing to an easy win and a brutal GOP night. But their "social circle" data, which "predicted outcomes better than traditional questions" is telling a different story:

When we calculate how many electoral votes each candidate could get based on state level averages of the own-intention and social-circle questions, it’s looking like an Electoral College loss for Biden. We should note that our poll was not designed for state-level predictions, and in some states we have very few participants. Even so, in 2016 it predicted that Trump would win the electoral vote.

Small sample sizes, bank-shot methodology, strange year. Every asterisk applies. But still. Allahpundit notes that asking people how they think their state will vote produces an even bigger Trump electoral college win, but also offers some compelling counterpoints:

It’s possible that the traditional polls are right and the “social circle” data is wrong, of course. Both parties this year seem convinced that Trump will do better than his current numbers, righties because they’re convinced that Trump fans are a “silent majority” despite the fact that he’s never had a job approval average north of 50 percent and lefties because they have hardcore electoral PTSD from 2016 that makes them fear Trump fans are perennially underestimated. But neither of those things is necessarily true. USC speculates that the pandemic might be throwing off people’s assessments of how their friends intend to vote this year for the simple reason that we’re all spending less time around our friends...

...One obvious question about “social desirability bias,” though. If it’s true that that’s distorting the traditional polls and producing more encouraging numbers for Biden, whether because people are embarrassed to tell pollsters the truth or because they fear being punished somehow if their community knew their support for Trump, why are we seeing Biden faring better than Clinton in blood-red states? People who live in Montana, say, shouldn’t have reason to feel shy about telling some anonymous polling apparatchik that they’re MAGA and proud, yet Trump’s leading Biden there right now by single digits. He won the state by 20 four years ago. If anything, you might expect social desirability bias to *overstate* Trump’s support in very red states: If you’re a closet Biden supporter surrounded by hardcore Trumpers, maybe you want to keep that information to yourself.

I've wondered about these exact same effects myself. The guessing game will be over soon enough -- but as I've been saying, either Biden wins fairly comfortably, or American political polling will be viewed with even more deep-seated suspicion. The former vice president leads comfortably nationally (although a few of the latest results on RCP's aggregator are creeping closer), and must be considered the strong favorite. And as I mentioned on Monday, the canary in the coal mine for Hillary Clinton four years ago was district-level polling that deviated dramatically from the national and even state-level numbers. And in that realm, Trump is doing quite poorly this cycle, which could be a meaningful red flag for his campaign. Yet another example of that phenomenon, out of Pennsylvania:


If Trump winds up winning or coming close, people may be talking about the Trafalgar and USC models a lot more. If he ends up losing big, ominous district-level data could end up being a harbinger yet again. I'll leave you with this -- and the numbers in places like Florida and North Carolina have moved in the incumbent's favor since this was tweeted:


I'd also caution against excessive doom or exuberance based on partisan breakdowns of early voting turnout or returned ballots. Trump opponents are starting to freak out about Florida, for instance, but there are so many unknowns at play -- and the breakdown of independents seems like a really big X factor that could very well favor Biden. Wait and see: