On October 23rd, 2016, ABC News inaugurated its tracking poll with a Hillary Clinton advantage  of 12 points. It was a lead, ABC News noted, signifying the “broad disapproval” of Donald Trump. Around the same time, the Associated/GfK poll was giving Clinton +13 percent, USA Today/Suffolk +10 and CNBC +10. Of course, there were some other polls at the time, most often ranging from +2% to +7%  for Clinton—and two outliers, IBD/TIPP tracking giving Trump +2 percent and the LA Times/USC Tracking giving him +4.

The RCP average would register Clinton ahead until the end, with the final lead at 3.2 percent. The most notable numbers in 2016 came from state polls which gave Clinton a decisive advantage in the battleground states. Models incorporating various data gave Hillary Clinton a 71 to 93% chance to win the White House.

 Circulating the conventional narrative to explain what happened with the polls in ’16, Tom McCarthy wrote in The Guardian last month that “While some key state polls were off in 2016, the national polls in aggregate were right on target, showing Clinton three points ahead at the end; she won the popular vote by two points but lost in the electoral college.” That is a misleading picture about the polls at the national level. If we are to exclude the perennial outlier (LA Times/USC Tracking) that gave Trump a consistent and incorrect lead, and if we also exclude the nearly 900,000 extra votes that came from California for Clinton—an outcome which no one polled, predicted or even speculated about—then the national polls were off to a comparable degree to the battleground polls. Namely, the national polls seemed to be less erroneous for all the wrong reasons. A glaring misunderstanding like this invites the question whether the mainstream pollsters and poll readers have learned anything since 2016. 

Two pollsters who got 2016 right think that the mainstream polls are wrong again, and although they grant that the election is very close, at this point they predict a Trump electoral college victory. Patrick Basham of the Democracy Institute predicted a Trump win in ’16 and also got the Brexit referendum right as well. Basham, in his latest poll for 2020 predicts an easy electoral college victory for Trump with all battleground states ending up in Trump’s column. Robert Cahaly of Trafalgar Group in his 2016 polls predicted the exact number of electors awarded to Trump. Now Cahaly predicts that most battleground states will go for Trump with an electoral college victory in the mid-270s.

What both these pollsters are aiming to tackle is what is called social desirability bias in the polls. Social desirability bias is when a poll interviewee gives an answer to a question based on what he considers socially acceptable, rather than his true opinion on the subject. It has been observed that voters were more likely to choose Trump in a poll that felt more anonymous, such as a poll that used an automated, interactive voice response system instead of a live caller.

Cahaly thinks that the social desirability bias is even more prevalent now than 2016. The people who were called deplorables in 2016 are now called racists and white supremacists. Groups like white women in the suburbs are particularly sensitive to those kind of charges. A percentage of them will give pollsters the socially acceptable answer, even if that is not their actual choice. This attitude might not change the winner of a particular demographic, such as white suburban women or working class blacks, but it could prove crucial in a close election.

Robert Barnes is a trial lawyer with a number of notable clients, from Wesley Snipes to the Covington Catholic students, Ralph Nader and Alex Jones. But he is also the most formidable and comprehensive critic of the polls. He has trained for that role by putting some “skin in the game”—namely, betting against the polls when he thinks they are wrong. In 2016 The Times of London reported that Mr. Barnes “won a combined total of €470,000 by betting on Mr Trump in London and Dublin.”

Mr. Barnes claims that there are numerous problems with the polls. A very important issue is how difficult it is for the polls to reach working class whites, blacks and Hispanics. The difficulty is even more pronounced when pollsters try to get representative samples in rural areas. In an email exchange, Barnes gave me the example of Greene county in Pennsylvania, where a recent New York Times poll had it “going 2 to 1 for Biden (a county Biden is likely to lose 3-1)”. Greene county went 62%-36% for Trump in 2016. The county is heavily unionized and is dependent on its steel mills. One has to assume, in order to believe the numbers of the poll, that the white rural steel workers have turned massively against the president, even though he has imposed a 25% steel tariff and seems to embody in his positions their values and concerns.

Mr. Barnes argues that “polling is now more art than science, and all the artists are talentless artists.” You are not a good pollster if you don’t know the demographics, culture, and political history of the place you are trying to poll. “Few pollsters could tell you the difference between the Polish vote in Pittsburgh, the Italian vote in south Philly, the mainline Episcopalian in Montgomery, the Welsh son of coal miners in Wilkes Barre, and the large rural working class spread throughout the state, with their differences between coal mining country, old industry, and pastoral economics/society.” It’s a nowheres trying to poll the somewheres problem.

Apart from the issue of social desirability bias, Mr. Barnes believes that this will be an election sharply divided by class. And when that happens, polls go very wrong. The most dramatic example of an election of that kind was 1948, when President Truman defeated Thomas Dewey decisively despite all polls predicting the opposite outcome. When the political preferences of an upper class female, who is much more likely to answer the phone call and complete a poll questionnaire, differ from those of a male laborer, then you are in trouble. One of the biggest problems with polls is that 98% of the people refuse to take them. It’s hard to get a representative sample of a group when so few people answer pollsters’ calls. 

For Barnes, a more accurate indicator of the direction of an incumbent election like the one we are having now is to check for “division within the incumbent party, as measured by early primary vote share.” Or how charismatic the opposition candidate is. Has the incumbent President started any new wars? What is the “primary participation rate in historically low-propensity voter groups?” What is the answer to the question “are you better off than you were four years ago? What are the “between-presidential-election 4-year voter registration trends in key states?” What is the “partisan differential in voter registration trends in the 6 months prior to the election in swing states?” The answers to these questions, and a few others that he considers, point to a Trump win this year.

Then there is the Primary Model of professor Helmut Norpoth of Stony Brook University. It has predicted the winner of 25 out of 27 elections, missing 1960, when the walking dead were voting for Kennedy in Illinois, and 2000, where the election ended up in the Supreme Court’s lap. “It is a statistical model that relies on presidential primaries and, in addition, on an election cycle as predictors of the vote in the general election”. In the five cases where Norpoth’s model gave a different result than what the polls predicted, Norpoth’s model won 5-0. This year Norpoth’s model predicts a decisive electoral college victory for Trump in November. 

Thomas Dewey, certain of his victory in 1948, limited himself to inane platitudes of the type: “You know that your future is still ahead of you.” Now Joe Biden, with a similar attitude, has his own don’t-rock-the-boat strategy for the last few weeks of the election. He hopes to “run out the clock,” as the Hill reported: 

“They don’t want to break anything in the last three weeks,” said one Democratic strategist. “If the election were held today, they would win. They know it, and more importantly, Trump knows it.” “To put it simply, they’re winning,” the strategist added. “And they don’t need to do much. They need to firm up their support and call it a game.”

Even the few pollsters who see President Trump as the favorite to win admit that this is a close election. It could definitely go Biden’s way. But perhaps it is telling that the Democratic strategist quoted in The Hill viewed his statements as risky enough to require anonymity. What if we are witnessing one of the biggest political blunders in history?

We’ll see. The future is still ahead of us.

Napoleon Linarthatos is a writer based in New York.