On Nov. 9, 2016, as much of the world reacted with shock and dismay to Donald Trump’s victory, people across the spectrum began to repeat phrases that continue to dominate the narrative of today, saying things like “the polls were wrong” or “don’t trust the polls.” Even intellectual people on the left, who champion numbers and data as they lambast the right’s inaction on COVID-19, all too often still buy into this misconception about polls four years later. They argue that Trump has a good chance of winning and the polls are not an indicator that Democratic party candidate Joe Biden is a strong candidate. While anything is possible in the next two weeks, it is important that we emphasize that the polls are an accurate source of information, for they provide valuable insight into the state of the race.
On Election Day 2016, noted polling aggregator FiveThirtyEight gave Donald Trump a 28.6 percent chance of winning the election. Somehow, numerous people spun this into evidence that FiveThirtyEight was wrong and that its polling aggregation was clearly incorrect. But 28.6 percent is only slightly less than the chance of rolling any two numbers on a six-sided die. If I roll a 1 or a 2, no one is going to tell me that my probabilities were wrong. Yet for some reason, Trump’s victory was treated by those on the right as something completely unexpected by the polls and evidence that we shouldn’t trust them — and by extension, trust math or science itself. In addition, Hillary Clinton led Trump by an average of 3.2 points in polls leading up to the election. She won the popular vote by 2.1 points, a 1.1 point margin that is well within the margin of error of most polls. In short, Trump’s win was entirely possible based on this polling: He was an underdog who had a small but not insignificant chance of winning. And his actual path to victory mirrored that; he lost by 3 million votes while sneaking away a combined 80,000 votes in three Rust Belt states to win the electoral vote.
In the aftermath of the election, a majority of those on the right pointed to a singular reason why the polls had been wrong: the silent majority. The idea of the silent majority, or the Shy Tory Effect, as it was coined in England, is that it is controversial to be a Trump supporter (true), and thus scores of his supporters do not publicly declare their support for him (false), thus polls and other measures of support for Trump underrate how popular he is. If this were true, a wave of shy supporters of Trump’s Republican coalition would only materialize on election day at the ballot box. However, it’s a rather ridiculous claim given the fact that Clinton won the popular vote by 3 million votes in 2016, and Democrats won the House popular vote in the 2018 midterms by over 9.7 million votes. In addition, anecdotal evidence suggests that Trump supporters are loud and proud about their support.
A far more compelling case can be made that in the days leading up to the election, the polls quite accurately predicted support for the presidential candidates; what they were unable to predict was which way the remaining undecided voters would swing. The number of undecided voters was extraordinarily high in the weeks leading up to the election, with 15 percent of the electorate still unsure two weeks before. This was due to a number of factors, including two historically unfavorable candidates and a number of October surprises. And while undecided voters generally split rather evenly between the late polls and the election, in 2016 they overwhelmingly broke for Trump across the country, and by over 10 points in Pennsylvania, Wisconsin and Michigan. Part of this was because Trump also won voters who disliked both candidates by a near two to one margin.
While these are all good reasons to believe that the polls were actually largely accurate in 2016, the lessons learned from analyzing voting behavior since then have also motivated pollsters to improve their methodology this time around. For example, pollsters have recognized a greater need to weight by education, given the large divide between college educated and non-college educated white voters.
Yet many still disregard these explanations for the legitimacy of political polling and simply see this election season as a replica of 2016. They could not be more wrong. Undecided voters have halved from 2016, and not only does Biden have a larger lead than Clinton did in 2016, but perhaps more importantly, he has already eclipsed the magic 50 percent number in many national and state polls. In short, even if every single undecided voter broke for Trump, the polls predict Biden would win the election. That was far from true in 2016. But among undecided voters, Biden does have an advantage; undecided voters are overwhelmingly hispanic and under 40, and voters who dislike both candidates actually are tilting toward Biden. In addition, the greater use of mail-in and early voting this election could mute the impact of a late October surprise; people will have voted long before a last-minute event hits the newswaves and influences their vote. Early voting turnout is at 4,000,000 already, more than 50 times what it was in 2016 at this point. As a result, a late announcement into an investigation against the Democratic candidate, for example, which caused Clinton to lose support in 2016, might have less of an impact than it did four years ago.
Trump has roughly one path to victory. He needs to win at least one of the three Rust Belt states he flipped in 2016. In addition, he needs to defend a number of toss-up states including Arizona, Florida, North Carolina and increasingly Ohio (as Trump recently pulled all midwest ad buys to focus on the Sun Belt). Trump’s standing in those states, compared to four years ago, is far worse — specifically because of Biden’s success in capturing undecided voters over him. As of Oct. 18, the RCP polling averages in Michigan, Pennsylvania and Wisconsin show Biden leading Trump 50 to 42.7, 48.8 to 45, and 49.7 to 43.5, respectively. By contrast, in 2016, over the same period Clinton-Trump was 47-43.4, 46.8-44.7, and 46.8-40.3, respectively. While the margins between the candidates are roughly the same, what matters is that Biden has actually broken 50 percent in these swing states, thus nullifying Trump’s ability to tap into undecided voters to win the election again.
Even in polls that predicted a Trump victory in 2016, Biden is in good shape. Trafalgar Group, considered a hack Republican polling shop for its inclusion of “social desirability bias” (re: the assumption of an existence of a silent majority), which projected Trump wins in 2016, now has Biden winning in Pennsylvania. That is not to say that the poll should be trusted, but within Trafalgar’s polling, Biden has remained ahead by two points since a month ago, suggesting that even the pollsters whose methods most favor Republicans cannot see a likely Trump win at the moment.
Over the past four years, Trump and his supporters have pushed misinformation on a multitude of scientific issues. To push for people-centered policy, we need accurate measures of public opinion on issues and politicians. We cannot afford to delegitimize polls, for they are the best window into what the populace wants. In this time when science itself is under attack, and historic endorsements of Joe Biden in the name of science are being made, we must be the bulwark against anti-intellectualism. Polling — and the science that backs it — should be no exception to this rule.
Caleb Apple ’21 can be reached at email@example.com. Please send responses to this opinion to firstname.lastname@example.org and op-eds to email@example.com.