With the conventions over, the presidential race enters the fall home stretch. A few things are clear. Joe Biden enters the fall as the leader over Donald Trump in national and state-by-state polls. That lead has eroded markedly since the beginning of August. And this election will be unlike any in American history. By any reasonable estimate, Biden remains the favorite to win. There are as many things we don’t know, however, as things we do know. We must take stock of the known unknowns.
Begin with the polls. In the RealClearPolitics national poll average, Biden has led Trump head-to-head since entering the race. His largest lead since clinching the nomination came on June 24: 51 percent–41 percent. That’s bigger than Hillary Clinton’s widest lead of the general-election campaign, which was 48 percent–40 percent on August 9, 2016. Moreover, Biden has repeatedly nudged above the magic 50 percent mark, whereas Hillary was never above 50 percent after March. Biden closed out August up 50 percent–43 percent. One of the strongest arguments that Trump will have a harder time surprising the pollsters in 2020 is that current polls show fewer voters as undecided and available to swing his way at the end. That is, if you believe pollsters are accurately projecting the electorate.
Consider what happened in 2016. The final RCP average in 2016 had Hillary winning the national popular vote by 3.2 points, and she won it by 2.1 points. Polling is, even at its best, not that precise a science, so this is not a large underestimate of Trump. Presidential elections are, however, state-by-state races. At the end of August, the RCP state-by-state map shows Biden leading Trump 337–201 in the Electoral College, claiming Florida, Pennsylvania, Ohio, Michigan, Wisconsin, and Arizona from Trump’s 2016 coalition. At the same time in 2016, however, Hillary led 362–176, a wider lead than Biden’s at any time in the race. That closed to 272–266 by Election Day; the final poll averages got 46 states right but missed Trump’s trifecta in Pennsylvania, Wisconsin, and Michigan, while incorrectly showing Trump ahead in Nevada.
How much of that shift was due to the polls’ accurately measuring movement in Trump’s direction, and how much to pollsters’ being wrong all along? Consider eleven competitive states where Trump ended up outperforming his poll standing as of August 31, 2016: Arizona, Florida, Georgia, Iowa, Maine, Michigan, New Hampshire, North Carolina, Ohio, Pennsylvania, and Wisconsin, and Trump won nine of the eleven. Weighted by the number of votes cast in each state, Hillary’s poll lead across those states shifted from 43.2 percent–39.6 percent on August 31 to a 45 percent–45 percent tie on Election Day. Voters in those states went for Trump 48.9 percent–46.1 percent. Both candidates’ percentages rose from late August to November, reflecting undecided voters’ making up their minds, but the final polls underestimated Trump by almost four points, Hillary by only a point.
At the end of August this year, Biden was above 50 percent in only one of those states (New Hampshire), but he was between 47 percent and 49 percent in eight others. Using the same weighted average, Biden ended August up 47.8 percent to 45.2 percent, a smaller lead than Hillary’s at the same point, but with far fewer undecideds. If the race followed precisely the same trajectory, Trump would be up 51.4 percent to 49.8 percent across these eleven states in the final polling, 56 percent to 51 percent in the actual voting. That is, obviously, mathematically impossible, which illustrates the difficulty of projecting a carbon copy of 2016. But if every state moves proportionately the same from its current poll average as it did from August 31, 2016, to Election Day, 2016, then the only 2016 Trump state that would flip to Biden is Arizona.
Consider as well the FiveThirtyEight projection model, which is mostly just a sophisticated way of systematically interpreting the polls. At the end of August, the model gave Biden a 67 percent chance of winning, his lowest projection in the model since it was rolled out in June; it had gone as high as 79 percent. He ended August lower than either Hillary’s 74 percent at the end of August 2016 or her 71 percent on Election Day (although Hillary’s odds were much more volatile, dropping to nearly even with Trump’s at the end of July and the end of September). The model also projected 313 electoral votes for Biden, down from 342 in late June; it projected 321 for Hillary at the end of August 2016, and 302 on Election Day. FiveThirtyEight’s model recognizes uncertainty; Nate Silver took a lot of heat in 2016 for giving Trump nearly a one-in-three chance of winning, as other models were more skeptical. The erosion in Biden’s standing in August and the fact that he is now behind where Hillary stood in the same model at the same juncture offer Trump some cause for hope of a repeat.
There are reasons to doubt, however, that the trend will be the same. Undecideds tend to break against the incumbent party; even if that doesn’t work against Trump now, it is unlikely to work as well in his favor as it did in 2016, when Democrats had held the White House for eight years. Turnout in 2018 was historically high for a midterm election, suggesting that the Democratic coalition will not be as complacent as it was in 2016. Pollsters have also strained mightily to avoid 2016’s underestimation of the share of white working-class voters in the electorate, so Trump may be picking up fewer voters who are off the pollsters’ radar. Then again, in 2018 there were still states where the final polls significantly underestimated Republican support. In states with both a Senate and a governor’s race, Republicans exceeded the final 2018 polls in Florida, Michigan, Ohio, Pennsylvania, Tennessee, and Wisconsin. Trafalgar — the one pollster who saw Trump’s surge in the Midwest coming in 2016 — is once again more bullish on pro-Trump turnout.
Besides polls, there are two other places we look for evidence of what is likely to happen in November: hard fundamental factors such as incumbency and economic news, and unquantifiable soft factors such as narratives, enthusiasm, and momentum.
On the fundamental side, the economic factors are a known unknown: The situation is too unique to be reliably measured against prior-year metrics, but an elephant you can’t see is still an elephant. No previous election has featured an incumbent running amid an abrupt election-year economic collapse, stemming from noneconomic factors, along with a strong stock market. That’s before we factor in a possible sharp recovery (or reversal) in GDP or unemployment. An October announcement of a coronavirus vaccine (all but promised by Trump in his convention speech) would be a wild card, while polls showing weakening Trump support among senior citizens amid the pandemic are a grave warning light. Voter turnout in a general election with unprecedented levels of mail-in voting is also impossible to project with more than guesswork.
Biden is also a historically unique challenger. At 77, he’s four years older than Ronald Reagan in 1984, to date the oldest candidate ever elected (Trump is 74). Every incumbent president to lose has been defeated either by a fresh face on the national stage or in a rematch with a prior opponent; a challenger like Biden has never won. He’s also the first major-party nominee from a state so small that it has only three electoral votes.
Betting markets are a soft factor — they represent the perceptions of a few rather than the preferences of the many — but they drew Trump nearly even by the end of August. Enthusiasm is hard to measure, given the asymmetry between the two parties’ potential voter pools and the social stigma around publicly supporting Trump. The campaign narratives seem to suggest that Trump had a major rebound in August. Trump built his campaign around a strong economy and an opposing party in thrall to identity politics and enamored of socialism; it wan’t until the waves of rioting that Republicans settled on a message other than mocking Biden’s age. And the Republican convention fired up its partisans more than the Democrats’ did, whether or not that lasts.
Even if Biden wins, a Trump revival in the fall could have major consequences. Parties that see the bottom drop out down the stretch, such as the Republicans in 2008 and the Democrats in 1980, tend to get brutalized further down the ticket. The opposite happened when Republicans rallied late in losing causes in 1976 and 1996. Even if the GOP is unlikely to retake the House, and although there are few marquee governor’s races, the fall elections will decide control of the Senate and of the state legislatures that will draw a decade’s worth of congressional and state legislative districts. That means that Republicans who don’t think much of Donald Trump’s chances should still be hoping for the best from the unknown.
This article appears as “Known Unknowns in 2020’s Stretch Run” in the September 21, 2020, print edition of National Review.
Something to Consider
If you enjoyed this article, we have a proposition for you: Join NRPLUS. Members get all of our content (including the magazine), no paywalls or content meters, an advertising-minimal experience, and unique access to our writers and editors (conference calls, social-media groups, etc.). And importantly, NRPLUS members help keep NR going. Consider it?
If you enjoyed this article, and were stimulated by its contents, we have a proposition for you: Join NRPLUS.