The American Election Part III: Polling

Opinion polls tally voter preferences. They ae almost as old as America itself. The Raleigh Star and the North Carolina State Gazette, the Wilmington American Watchman and the Delaware Advertiser all published surveys in 1824 predicting that Andrew Jackson would defeat John Quincy Adams. He did – 335 to 169.

Gallup is the longest reputable pollster in the United States. Since 1948, it has conducted eighteen presidential polls. What is not recognized is that it has significantly erred in five of them, three since the turn of the century. If your batting average had dropped from being in the eighty percentile to fifty percent, there might be some expectation that you would be benched.

The first major error took place in 1948. Gallup erroneously predicted John Dewey’s victory over Harry Truman. The predicted breakdown of the vote was 50% for Dewey, 44% for Truman. It was a consensus view and was the most embarrassing moment for polling in U.S. history. Harry Truman, of course, won.

Gallup erred again in 1976, 2000, 2012, 2016 and 2020. In the twenty-first century, it was wrong 3 out of 6 times and in the last three elections, there was a hundred percent error record. In 2000, on 27 October, Gallup predicted George W. Bush would garner 52% of the vote while Al Gore only obtained 39%. In a 51.2% turnout, the Bush/Chaney team actually obtained 50,462,412 votes or 47.87% of the total. The Gore/Lieberman team obtained 51,009,810 votes or 48.38% of the total.

In 2004, the turnout increased to 56.7%. Bush/Cheney received 50.7% of the popular vote and won by 35 electoral college votes. The Kerry/Edwards team won 59,028,444 votes or 48.3%. Polls generally predicted a close race with Bush the likely winner. It was the only election in the twenty-first century that Republicans won more of the popular vote than Democrats. Pollsters appeared to recover their reputations.

The turnout increased again in 2008 when 58.2% of the electorate cast ballots. Barack Obama received an amazing 52.9% of the votes cast while John McCain, a true hero in the American firmament, but one who tethered himself mistakenly to Sarah Palin, won only 45.7% of the vote. In the final Gallup pre-election poll, it was estimated that Barack Obama would win with 52% of the vote while McCain would get only 42%, pretty accurate for Obama, but beyond the room for error for McCain. This was a clear indicator that pollsters were underestimating Republican votes.

In Gallup’s final election survey in 2012, Romney was supposed to get 49% to Obama’s 48% out of a vote total predicted to remain constant. Obama actually won by four percentage points. Gallup issued a report explaining its gross error. The organization blamed:

  • Misidentification of those likely to vote
  • Under-representation of regions which were too broad
  • A system of racial representation that gave extra weight to white voters
  • A new sampling system based on a random selection of numbers listed in the phone book rather than based simply on random numbers.

Who would use Gallup again?

Nothing, however, prepared pollsters for the disaster of 2016. Bunching the states most strongly or very strongly in one camp or another, these were the estimates of the distribution of electoral college votes and a selection of swing states:

   SwingStates – Cl.SwingStates– Tr.
   WiscMi    PennOh.   Nev.   Az.
Crystal Ball322216 xxxXx
Associated Press274190xxx  x
FiveThirtyEight272214 xxx x
CNN Electoral268204xxxx  
Cook Political278214xxxx x
Rothenberg/Gonzalez323197xxxx x
NPR274214xxxx x
The Fix275215xxxx x
Louis Jacobson274186xxx   
ABC News274188xxxx  

The swing states either leaned towards Clinton or towards Trump. Generally, the results of the polls were remarkably consistent. The results of the election, however, were not consistent with the polls.

The problem of such inaccuracy is made much worse because people believe polls, believe they have a record for accuracy. In this century, and most notably in the last two elections, it has not been accurate in predicting the margin of victory for the victor and in 2016 even the candidate that won.

While sweeping judgments on the overall performance of polls in 2020 are premature before all the ballots are counted, while, as Joe Biden said, patience is required, nevertheless a number of conclusions can be drawn without fear that they will be subsequently contradicted.

In 2020, polls badly missed the actual results. Most predicted an easy win for former Vice President Joe Biden. Most predicted that the Democrats would make net gains in the Senate, enough to wrest control from the Republicans. They even predicted the strong possibility of gains in the House of Representatives where the Democrats actually lost seats. The mistakes were generally outside the margins of error. Trust in sophisticated prognostication imploded. The ability of expert elites to divine the behaviour of everyday citizens was put into question.

George Gallup, like most pollsters, had a deep faith in numbers and what they represented. True to the Platonic tradition, numbers provided a different order of truth than direct observation. Aggregates matter. Aggregates offer insights that an observer would not otherwise have. George Gallup originally asked who was reading the Daily Iowan that he edited. And which parts did they prefer? In 1936, after mastering statistics and refining his methods, and noting that the Literary Digest had got it right in the previous five presidential elections, predicted, based on its postcard surveys, that the Literary Digest would give Republican Alf Landon 56% of the votes and would, therefore, win.

Gallup was out by only 1% about how the Literary Digest would predict the election. But his own predictions and the magazine’s on the outcome, as well as all the other polls, were also wrong beyond any margin of error. Gallup had not only seriously underestimated the amount by which Franklin Delano Roosevelt would win the election, he expected him to lose to Alf Landon. It was predicted that Landon would win in a landslide, 57.1% to 42.9% taking 370 of the electoral college seats. The actual result: Roosevelt whipped Landon 60.8% to 36.5%.

Gallup spent the next eight years perfecting his methods and breaking the American voting population into demographic groups, a method that remains the mainstay of the polling profession to this day. But what if geography, what if education, what if gender, what if race were all better predictors? What if a pollster had to get the right mixture? What if ideology counted so much that it deformed who would answer questions and how?

But the problem was prediction itself even more than the degree of accuracy. For in consumer products, a pollster wants to determine why a consumer prefers one product rather than another so that the product can be made or packaged in a more appealing way. Polling began to influence and to a degree even shape policy platforms. Further, as the 2020 results showed, what if the pollsters for one reason or another could not reach the reticent rural conservative or the highly mobile youth in the city? How many would vote became harder to estimate than how people would vote.

And what if the division into demographic groups was simply a poor way to classify voters. The polling organization More in Common in Britain has tackled that issue. It starts with the premise of deep fractures no longer based primarily on religion or ethnicity or age. “Societies are fracturing as the forces of division grow stronger, driving people apart. We are losing trust in each other and in the future. Feelings of frustration, powerlessness and a loss of belonging are making us vulnerable to ‘us versus them’ stories, which turn us against each other. Social media is magnifying the loudest and most extreme voices.” (

More in Common could be a front for Joe Biden message of healing, the message of overcoming the chasms that so divide our societies. “Our organization’s name reflects our vision: to build more united, inclusive and resilient societies in which people believe that what they have in common is stronger than what divides them. More in Common’s mission is to understand the forces driving us apart, to find common ground and help to bring people together to tackle our shared challenges. We draw from ground-breaking research to test and find solutions, working with partners that have the capacity to make a real difference at scale. And we help build the larger field of efforts to strengthen democratic societies against the threats of polarization and division.”

Unfortunately, whereas the COVID-19 pandemic in Europe created a greater sense of togetherness and an awareness of our common humanity, and even uplifted the people into a more caring society, the U.S. under the leadership of Donald Trump went the other way, deepening people’s anxieties and despair at government mismanagement. Mutual trust declined rather than improved. Further, the pandemic enhanced a feeling of localism and greater trust in government on the municipal level. More significantly, perhaps, the “invisibles,” the disengaged were more lonely and more isolated than ever before.

Severe disappointment in both government and other people, and, therefore, deep divisions were prevalent in the U.S. compared to other countries. Ironically, and extremely paradoxically, that increased the base of support for Trumpism even though Donald Trump was a major cause of this effect. This was also the case with rising unemployment, but again with the ironic effect of reinforcing support for Donald Trump who had sold himself as the better manager of the economy. Third, this distrust was more directed towards scientific experts and, shockingly, enhanced trust in Donald Trump and his message of scepticism about institutions. The result: much deeper divisions that ever before and an enormous challenge for those intent on bridging the chasm.

Thus, we had polling that explained the divide rather than just depicting it. Further, the trend line went in precisely the opposite direction as one would expect from rational choice. That, at least, is consistent. Rational choice theory had shown unequivocally that people do not vote based on their perceived best interests but were more often influenced by subliminal clues. Further, old fashioned polling had become obsolete. Monitoring choices by all kinds of ways enabled not only much better ways of predicting voting behaviour, but allowed the surveyors to influence that behaviour by sending out specifically designed reinforcement messages or, alternatively, and much more difficult, signals that undermined existing biases. Surveying had become a matter of manipulation and not just measurement.

However, polls, even as measuring tools as inaccurate as they became, had turned elections into horse races. Who was ahead? Who was behind? Who was catching up? And at what pace? What were the trend lines? Policy issues were swamped by a betting model.

The pollsters claimed they had corrected their methods which failed so badly in 2016. They were prepared for the 2020 election. Once again, most surveys missed the actual results. They predicted an easy win for the Biden/Harris ticket. It was a squeaker. They failed to predict Democratic losses in the House of Representatives. And they predicted that there was a very good chance that Democrats would elect enough senators to wrest control from the Republicans.

But the innovations in the voting system flummoxed them. The COVID-10 crisis encouraged states to facilitate voting my mail and offered more options to vote before election day. And each state had its own rules. How would these systems skew the voting? How would they effect counts since Democrats encouraged voting by mail while Donald Trump discouraged unsolicited voting by mail. He claimed that voting by mail facilitated fraud, even though he could not offer an iota of proof to justify yet another lie.

How much did fear on either side of the divide spur turnout? For it seemed clear that the real race would be determined by how many would vote on each side. Thus, each side exaggerated the fear of what would happen if the other side won. There was so much emotion that pollsters could not track what the controllers of new social media could. But the interest of the controllers of the algorithms was not in reporting voting trends, but in influencing them. They were playing a different – and more perfect – game.  

One of the conclusions of the More in Common polling in the United States is perhaps the most intriguing. Unlike other countries, 4 in 5 of Americans believe that citizens can change society, that democracy works, that the decisions and actions of individual voters have consequences. Americans have more confidence in the efficacy of democracy than any European country, in spite of the fact that U. S. democracy has suffered from the most and the deepest fractures among Western democracies. This is perhaps the greatest irony that needs to be probed all on its own.

Actually, that is not the greatest irony. That will be how the media and the public will follow polls in the next election in spite of the record of polling in the past. The polls were wrong again. Long live polling.


One comment on “The American Election Part III: Polling

  1. Ira Basen says:

    Hi Howard:

    You probably know this already, but in 1940, George Gallop wrote a very interesting book called The Pulse of Democracy in which he argued that public opinion polling woould be the saviour of democracy because it would allow leaders, for the first time, to truly know what ordinary citizens were thinking, as opposed to the rich and powerful who’s voices could always be heard. His co-author on that book was a young researcher at Princeton named Saul Rae, Bob’s father.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s