In The Hill today, Jonathan Easley, does an excellent job in getting to the root of the polling problem in the United States. How wrong were most of the pollsters in 2016! The reporter highlights current measurements, and discusses with representatives of the polling agencies what may be missing in their assessments. Here is the report:
Most pollsters show Democratic nominee Joe Biden with a sturdy and stable lead over President Trump at a time when tens of millions of people have already voted and there is almost no time to change the course of the race.
But a handful of contrarian pollsters believe Trump’s support is underrepresented and that election analysts could be headed for another embarrassing miss on Election Day.
The battles have spilled on to social media, where some well-known political analysts have dismissed polls that show Trump leading Biden.
The Trafalgar Group, which was the only nonpartisan outlet in 2016 to find Trump leading in Michigan and Pennsylvania on Election Day, show Trump with small leads in both states, which would be keys to another Trump win in the Electoral College. Nearly every other pollster shows Biden with a comfortable lead.
Trafalgar’s Robert Cahaly says there is a hidden Trump vote that is not being accounted for in polls that show Biden on a glide path to the White House.
“There are more [shy Trump voters] than last time and it’s not even a contest,” Cahaly said, adding that it’s “quite possible” that the polling industry is headed for a catastrophic miss in 2020.
FiveThirtyEight’s Nate Silver and Cook Political Report editor Dave Wasserman are among those deeply skeptical of Cahaly’s polling.
Both have dug into the crosstabs of Trafalgar polls and pointed to questionable breakdowns as evidence Trafalgar doesn’t know what it’s doing. For instance, the crosstabs in a Michigan poll, which are no longer online, appeared to show Trump leading Biden by 8 points among young voters, a Democratic stronghold.
“[Trafalgar] doesn’t disclose their ‘proprietary digital methods’ so I can’t really evaluate what they’re doing,” said Jon McHenry, a Republican pollster with North Star Opinion Research. “They’re far enough out on a limb that a year from now, we’ll all remember if they were very right or very wrong.”
FiveThirtyEight’s model gives Trump about an 11 percent chance of winning – roughly equal to pulling an inside straight in poker – after giving him about a 30 percent chance on Election Day in 2016.
Those who think the majority of polls showing Biden with a more comfortable lead than Hillary Clinton had at this point in 2016 note a couple of factors. They say Trump is underperforming – in some cases dramatically – among the key coalitions that powered his 2016 victory. Biden also is a more popular candidate than Clinton.
McHenry said he does not think there are many “shy” Trump supporters who would lie about their intentions.
Rather, there is concern about a “skewed response rate pattern,” whereby Trump voters would be less likely to participate in a survey or answer the phone when a pollster calls.
Still, McHenry noted that this wouldn’t be an automatic benefit for Trump. In Pennsylvania, for instance, he found Democrats were less likely to answer the phone than their registration would suggest.
“I can’t definitively say there is no response bias, but I’m skeptical of it, and it certainly wouldn’t be enough to explain the national deficits we’re seeing,” he said.
That said, Trafalgar is not the only contrarian voice in polling. Several other pollsters have joined it in arguing that other pollsters are missing pro-Trump voters.
Jim Lee of Susquehanna Polling and Research has been another proponent of the “submerged” Trump voter theory.
A recent Susquehanna survey of Wisconsin found Trump and Biden tied, making it the only poll to not show Biden in the lead in the Badger State since August, when the Trafalgar Group found Trump ahead by 1 point. In Florida, Susquehanna shows Trump leading by 4 points, while the FiveThirtyEight average gives Biden a 2-point advantage.
“There are a lot of voters out there that don’t want to admit they are voting for a guy that has been called a racist, that submerged Trump factor is very real,” Lee said this week on WFMZ’s Business Matters. “We have been able to capture it and I’m really disappointed others have not.”
The University of Southern California’s Dornsife Center is publishing results from its regular national poll but is also asking parallel “experimental” questions asking people who they think their social contacts are voting for and who they think will win their home state.
In 2016, USC-Dornsife made headlines for being one of the few polls to show Trump leading nationally. Hillary Clinton ended up winning the national vote and USC later adjusted its methodology, saying it oversampled rural voters in the last election.
This time around, the USC-Dornsife poll shows Biden leading by 11 points nationally.
However, the race tightens to 5 points when voters are asked about their social circles and to 1 point when voters are asked who they expect others in their state will vote for. That survey suggests Trump would once again win the Electoral College in 2020.
USC-Dornsife notes that the social circle question was a better indicator than the “own intention” question in five recent elections, including the 2016 U.S. presidential election and the 2018 battle for the House.
Still, the Santa Fe Institute’s Mirta Galesic, one of the researchers studying the added polling questions for USC, noted that the poll’s state-level findings suggesting Trump will win the Electoral College should be viewed with skepticism.
The USC-Dornsife poll of 5,000 national participants has very small samples in some of the battleground states and could be less accurate than public state polls.
“We anticipate that with such small samples, the social-circle question will produce more accurate state-level predictions than the own-intention question, because the social-circle question may provide more information and smooth out some of the bias of the small state sample,” Galesic said. “But this does not mean that predictions based on the social-circle question will be more accurate than large state polls.”
In addition, Galesic says the social terrain is extremely volatile due to the coronavirus pandemic, making it more difficult to gather reliable data about voter’s social circles, which have shrunk dramatically in recent months.
And Galesic said the specter of 2016 still colors what many voters think they know about how their friends and family will vote in 2020, even if the dynamics have changed dramatically in that time.
“This incites a lot of pessimism among Democrats about Biden’s chances and optimism among Republicans about Trump’s chances,” Galesic said. “It also contributes to the belief that there are some Trump voters that are not accounted for in polls. Taken together, these beliefs could bias social-circle expectations towards a more narrow margin between the two candidates.”
- Jonathan Easley