The Rattler’s Tail and Snakebit Pollsters
By Robert Graboyes and Charles Blahous
There are countless, conflicting interpretations of 2020’s election results, but one fact is indisputable: across the board, polls failed miserably. Several reasons have been offered for this failure, but one explanation is especially compelling and worthy of further attention, namely, the fraying of social trust in America. It is not possible to produce accurate public opinion surveys where there isn’t tolerance of opposing viewpoints—even those we detest.
In 2020, poll after poll foretold a “blue wave,” with Democrats easily recapturing the White House, toppling several Republican incumbents to retake the Senate and swelling their majority in the House of Representatives. Instead, Republicans may retain their Senate majority, holding at least 50 and possibly as many as 52 seats. Republicans may also end up only a few seats short of a majority in the House.
Meanwhile, the final presidential popular vote count remains incomplete, but the number of votes cast for President Trump will undoubtedly far exceed what polls throughout 2020 had indicated. Indeed, it appears that Donald Trump will become only the second president (after John Quincy Adams) to lose reelection while increasing his percentage of the popular vote from the previous election—and the first since Grover Cleveland in 1888 to lose while increasing his total number of popular votes.
So, in fact, there was no blue wave—merely the illusion of such a trend, based on the opinions of a politically skewed sample of voters willing to share their thoughts with pollsters. But why were those who were willing to speak with pollsters so unrepresentative of the electorate as a whole?
Many observers have attributed the polls’ failures to “shy Trump voters,” unwilling to reveal their presidential preference because of fears of opprobrium and repercussions over supporting a particularly polarizing president. That’s undoubtedly true for some, but it doesn’t explain the magnitude of the problem. It doesn’t explain, for example, why Maine polls would also be distorted by a pool of “shy Susan Collins voters.” Yet virtually every pre-election poll showed Maine voters poised to oust the moderate, mild-mannered Collins, who instead won a comfortable victory.
This underestimation of GOP support suggests that segments of the political left may have cowed many conservative voters into public silence, while stoking resentment among those same voters, to be freely expressed only in the privacy of the voting booth or on a mail-in ballot. To throw the polls off in a systematic way merely requires that this phenomenon occur with a small but nevertheless real slice of the American electorate. If this is in fact happening, it means that the American left is systematically blinding itself (and pollsters) to the numbers and viewpoints of their conservative opponents—a situation that operates to the detriment of the left as much as to anyone. Would down-ballot Democrats have perhaps performed better had their party not been lulled into overconfidence by months of unrelentingly sunny polls?
This hypothesis is not intended to absolve Trump or Republicans of any of their own failures of policy or breaches of decorum over the past four years. Nor does it suggest that ideological intimidation is limited to the left; “McCarthyism,” after all, was named for a Republican. However, conditions in 2020 (specifically, the strong leftward tilt in the messages of cultural arbiters, including leaders in universities, news organizations, entertainment, sports and the tech sector) have conveyed persistent approval of opinions on the left, combined with episodic public shaming or other negative repercussions for those deviating from those opinions. The resulting behavioral differences between voters on the left and right today may simply reflect their respective pragmatic reactions to the signals they receive concerning how public they should be about their views.
Both of us are economists—focused on policy, not politics. But we are concerned that this loss of social trust is spilling over into the economic sphere, as well. A business executive’s out-of-fashion viewpoint (even one expressed years ago) can prompt boycotts, firings and organizational turmoil. This can and does act as a damper on free economic activity. Markets depend as heavily upon social trust as do elections. Our hope is that tolerance for opposing viewpoints will increase—for the benefit of all Americans from the left to the right, and in economic as well as political spheres.
But before examining the decay of social trust in more detail, let’s look at the numbers—both pre-election polling and election results.
2020: Polls versus Results
The three tables below examine the disparities between pre-election polls and actual election results in the races for the presidency, the Senate and the House. In each case, the analysis is limited to the 13 most hotly contested jurisdictions, as forecast by the FiveThirtyEight website.
Table 1 shows the 13 most closely contested battleground states, as forecast before Election Day, in the race for the presidency. Trump so outperformed his pre-election projections that we as a nation have nearly forgotten what those projections were. Analysis after Election Day has focused on the close outcomes in Pennsylvania, Georgia, Arizona, Wisconsin, Nevada and North Carolina. Pre-election, however, states such as Ohio, Iowa and Texas were also supposed to be close. Yet Trump won each of those states by comfortable margins of 5 points or more.
The projection errors shown on table 1 are not random fluctuations. The probability that random projection errors would undercount Trump’s performance in 13 out of 13 states is less than 1 in 8,000. The likelihood of individual results being off by as much as they are—more than 4 points in the majority of instances, and always in the same direction—is even less. Pre-election forecasts consistently and systematically underestimated Trump’s actual support.
Table 2 shows the 13 most hotly contested Senate races, again according to FiveThirtyEight’s Senate pre-election forecast, which also estimated that Democrats had a 3-in-4 chance of capturing the Senate. Thus far, Democrats have only defeated two particularly vulnerable Republican senators, while Republicans knocked off one vulnerable Democratic incumbent, for a net change in the Senate balance of only one seat.
Which party will ultimately control the Senate is still an open question. Georgia’s two Republican senators have been forced into runoff elections, to be held in January. But as with the presidential election, polls systematically undercounted the Republican share of the vote—specifically underestimating the Republican share in 12 of the 13 races.
Table 3 shows a similar story in the House. The median FiveThirtyEight House projection was that Democrats would pick up six seats. It now appears instead that Republicans will end up with 209–213 seats, a substantial increase over their previous total of 201. Again, the polls underestimated Republican support in 12 of the 13 most hotly anticipated races.
Note: The Republican undercount for California's 25th district does not equal the difference between the actual margin and the forecast due to rounding.
These substantial errors in poll accuracy have led to a spate of post-mortems wondering how, once again, the polls got it so wrong. This New York Times piece reviews several of the leading theories, including errors in weighting key groupings among the electorate, the “shy voter” phenomenon in which voters on one side are markedly more reluctant to truthfully share their opinions with pollsters, oversampling of the “Trump resistance” in public opinion surveys, erroneous assumptions about voter turnout, changes in survey participation during the pandemic and a failure to accurately model Hispanic voter preferences, among others.
We are unable to draw clear lines of causation between the origins and manifestations of inaccurate polling. However, certain realities are clearly observable. One is that the inaccuracies were not limited to the presidential race, and thus cannot be attributed solely to squeamishness about expressing support for Trump. It is a more general phenomenon, in which support for Republican candidacies is systematically underestimated and support for Democratic candidacies is systematically overestimated.
Many of the explanations come down to the same basic phenomenon: for whatever reason, Democratic voters have been more willing to express their views to pollsters than Republican voters. This phenomenon underlies several of the possible explanations offered in the Times article, including the “shy voter” explanation, differences in willingness to talk to pollsters, oversampling of the politically engaged (or, put less favorably, the politically aggressive) and getting the Hispanic vote wrong. We clearly inhabit an environment in which the political left more often chooses to express their political opinions to a stranger who records them than do individuals on the political right.
Whither Social Trust as Social Trust Withers
The pollster’s trade requires voters to trust pollsters. That seems not to be the case for everyone these days. Consider the essence of polling. A pollster asks a series of questions about the respondent’s opinions, with the utility of the results entirely dependent upon the respondent answering these questions truthfully. For public opinion polling to work, there must be voluntary cooperation. Nothing can compel the respondent to tell the truth, or even to respond at all—and indeed, most people do not respond.
The entire exercise relies on the respondent’s willingness to help the pollster, and in that person perceiving no adverse consequences from doing so. Given that everything depends on the respondent altruistically helping out a stranger by taking a nontrivial amount of time to share his or her opinions, the conditions that facilitate this voluntary assistance cannot be taken for granted. Nor can we take for granted the assumption that this willingness will be equally felt by individuals in all parts of the political spectrum.
There’s no reliable way for pollsters to adjust results to reflect this blindness. Conservatives who declined to share their views before the election aren’t likely to explain why to pollsters afterward. Adjusting results on the basis of this election’s discrepancy assumes the gap between pre-election responses and actual votes is stable over time. But if the distrust is growing, adjustments for past inaccuracy will be insufficient the next time around. This year’s polling failures are consistent with rapidly worsening distrust.
In 2020, public opinion surveys appear to indicate that Democrats are more comfortable airing their political opinions with strangers than are Republicans. This gap appears to have widened considerably since 2016, which wasn’t exactly a banner year for survey accuracy. Perhaps this shouldn’t be a surprise, given the political climate of the last several years. Again, we remind readers that in making this observation we are not in any way judging the relative merits of Democratic and Republican policy positions. We are instead noting the existence of a behavioral trend among voters.
The hostility of the mainstream press and other cultural arbiters toward not only Trump, but also his supporters and Republicans generally, has been a hallmark of contemporary American life. Its manifestations have ranged from one political leader referring to much of the electorate as a “basket of deplorables” to conservatives seeing their views routinely ridiculed by late-night comedians. There is even an effort underway to compile blacklists of Trump’s core supporters, for the explicit purpose of limiting their future employment opportunities—a move encouraged by at least one sitting member of Congress. Some conservatives mask their views from family and friends out of a general weariness of politics permeating Thanksgiving dinners, Facebook threads and sporting events.
There is insufficient space in this article to list all the ways in which political expression on the right draws negative attention in modern American society, but here are a few. Instead of simply expressing disagreement with conservative views expressed by business leaders, it has become common to initiate boycotts of the products of those who disagree with progressives’ opinions. Powerful senators have indicated that companies who express opposing takes on climate change should be prosecuted. One half of “strong liberals” assert in surveys that Trump donors should be fired from their jobs.
Wearing a hat expressing support for the president has been equated with hate speech. Conservative speakers on college campuses have been met by organized disruptions and even violence. Even liberal celebrities such as Ellen DeGeneres have become the target of hostile social media campaigns, simply for asserting that those across the political aisle should be treated with kindness and respect. (Her beyond-the-pale offense was chatting politely with former President George W. Bush at a football game.)
Those who engage in these campaigns argue that the views of the targeted are so odious that “public shaming” is appropriate. We disagree vehemently and point to American history to offer reasons to stanch such impulses. For 60 years, mainstream political opinion has bemoaned the era of McCarthyism and the Hollywood blacklist. It is worth noting that McCarthy’s downfall came not from censoring him, but rather from allowing him to run his mouth until the nation wearied of his behavior.
Too often, the suppression of controversial speech is defended on the basis of Oliver Wendell Holmes’ words in a 1917 Supreme Court ruling: “The most stringent protection of free speech would not protect a man falsely shouting fire in a theatre and causing a panic.” A few years back, The Atlantic implored its readers to stop using that trope, and explained the dark history of the phrase. Holmes’ words were meant to justify the imprisonment of a Socialist whose offense was nothing more than distributing pamphlets opposing the military draft during World War l—pamphlets that did not urge violence or even resistance. Soon thereafter, Holmes’ words in another case indicated that he had grave reservations about the phrase he had launched. The court’s ruling itself was later overturned altogether. It is unintentionally apt that so many invoke Holmes’ quote today, given that then as well as now, the rationale has been employed to stifle the expression of differing opinions.
A tolerance of conflicting political opinions is a central component of the American story. For much of our history, Americans who despised one another’s politics could nevertheless worship together, dine together, work together and support one another in time of need. In his second inaugural address, Abraham Lincoln addressed the greatest philosophical schism in American history with his timeless phrase, “With malice toward none, with charity for all.”
One of America’s greatest immigrants, Albert Einstein, summed up the dual public-private nature of open expression: “[It] must be guaranteed by law. But laws alone cannot secure freedom of expression; in order that every man may present his views without penalty, there must be a spirit of tolerance in the entire population.”
This seems to be less so these days. Free speech and open discourse can be stymied by private actors—individuals and organizations alike—as much as by governments. Whenever political expression carries a substantial personal cost (or a perceived risk of personal cost), speech becomes less free and discourse less open. For the moment, some on the right seem to perceive greater cost for expressing their views. They therefore abstain from sharing their views with pollsters, thus skewing polls leftward and blinding observers. Progressives suffer along with conservatives in this environment because it leads to unpleasant surprises, such as voters this November delivering a firm rebuke to progressives that few saw coming.
We cannot accurately measure the perceived attractiveness of progressives’ ideas for governance in an atmosphere where those who oppose these ideas are inhibited from speaking. This year’s botched polling, however, reminds us of yet another reason to try.
Coda: On the Virtuous Rattling of Rattlesnakes
A personal recollection from Bob Graboyes
In the mid-1970s, William Shockley, inventor of the transistor and winner of the Nobel Prize in physics, began touring campuses and other venues, after reinventing himself as a eugenicist. He spouted racist theories, couched in pseudoscientific trappings, to anyone who would listen. Today, he would almost certainly be banned from most campuses, but in that era, free speech was campus dogma nationwide.
Shockley visited the University of Virginia, where I was an undergraduate. A considerable crowd gathered to hear him—polite, but, as far as I could tell, uniformly hostile to everything he said. Attendees wanted to grasp his mindset, to express their displeasure and to send him on his way. He said his piece (uninterrupted, I believe), earned little or no applause and left—with the audience more aware than before of his menacing strain of thought. He faded from view, dying as a pariah to many who had once been close to him.
Sometime afterward, someone asked me why great universities should provide a platform for scoundrels to air their odious views. My response was: “Don’t cut the tail off the rattlesnake. The silence is more dangerous for you than it is for the snake.”
Neither Chuck Blahous nor I think of either party or of their voters in such negative terms. But those who do see “the other side” that way might ponder the virtues of free speech—out of pragmatism and self-preservation, if not out of principle.