2020 isn't 2016, but Trump could still win

Screenshot of the New York Times’s election needle, taken by Vox’s Libby Nelson at 11:15pm Eastern on the night of the 2016 presidential election.

Donald Trump winning the presidential election is unlikely. But, it's a possibility. If you look at elections forecasters like 538, the Economist, or Plural Vote, a Donald Trump win is 4-35% likely—non-zero odds. Even a 4% win chance is slightly more likely than the probability of there being a full moon tonight. When folks hear numbers like those, though, it is hard not to remember 2016 when several forecasters gave Hillary Clinton ridiculously high win chances—like the HuffPost, who assigned Clinton a 98.2% chance of winning. However, there are several factors in this race that make it fundamentally different from 2016 in ways that bode poorly for Trump.

First, let's revisit 2016. Eleven days before Election Day, FBI Director James Comey reopened an investigation into Clinton's emails. While the polls were generally turbulent throughout the cycle, the letter likely contributed to a last-minute narrowing of the race from about a 6 point Clinton lead to a 3 point lead. Even then, the race still had large numbers of undecided and third-party voters, probably owed in part to the fact that Clinton and Trump were historically unpopular nominees. These undecided voters broke heavily for Trump, often by 20+ point margins in battleground states, when it came time to vote. This dynamic on its own made polling less predictive—what can a pollster do about surveying someone who has not made up their mind yet? Then, there was also a systemic polling error that disproportionately impacted the accuracy of polls in northern battleground states. In 2016, we learned educational attainment strongly correlates not only with a respondent's willingness to talk to a pollster on the phone, but also their voting intention. Thus, because Trump performed especially well with the non-college educated white voters who were less likely to complete election surveys, his support was famously understated in states like Wisconsin, Michigan, and Pennsylvania. In Wisconsin, for example, RealClearPolitics's polling average overestimated Clinton's margin by a whopping 7.2 points. On the other hand, national popular vote polls withstood the many quirks of 2016 and were actually quite accurate, leading to the Electoral College-popular vote mismatch that handed Trump the presidency.

Thinking about 2020 again, we can see a completely different race. Joe Biden has held a steady lead in averages of the national popular vote the entire cycle, unlike the volatility seen in 2016. There are also far fewer undecided and third-party voters going into Election Day. Importantly, Biden is polling at 50%+ in the Trump-won Rust Belt states of Pennsylvania, Wisconsin, and Michigan, meaning he doesn't need to bank any voters who are not already firmly in his column to win. These states are notable because they are among the likeliest "tipping point" states, i.e. the states most likely to deliver the 270th electoral vote and decide the election. Moreover, Biden's favorability numbers are much stronger than Trump's (and Clinton's in 2016, for that matter). Biden has actually grown his net favorability from about +0 at the beginning of the race to +6 now, according to Morning Consult. This could be due in part to Biden's success at presenting himself as a moderate. Trump's net favorability, on the other hand, has stayed around -10 to -15 for most of the cycle. Trump's path to victory becomes that much more complicated, as he needs to do even better with undecided voters this time around, and also convert a few Biden voters—all the while running against a much better-liked competitor. Moreover, Trump is unlikely to be able to count on the same polling error that saved him in 2020. Many pollsters revised their polling methods for the 2018 midterm elections to weight for education in their samples, and it was one of the strongest years for the polls in awhile.

However, it is not all sunshine and roses for Team Biden. While he has an impressive 8.4 point lead in an average of national polls, there is no national election—rather, there are 50 simultaneous state elections, and the winner is decided by the Electoral College. And Biden's lead in the likeliest tipping point state, Pennsylvania, is only 4.8 points. If a 2016-size polling error occurred in 2020, Biden's lead would shrink to 0.4 points. For the reasons mentioned above, it is unlikely that the polls will be off this year in the same way they were in 2016—not only that, they have to be even more wrong than in 2016 because of Biden's impressive lead. They would also have to be biased against Republicans across a variety of geographies and demographics, because of Biden's leads in North Carolina, Florida, Georgia, and Arizona. It is possible that such a polling error exists, it is just very unlikely—hence Trump's roughly 10% chance of winning.

At the end of the day, there is a lot more on the ballot tomorrow than just the presidential race! The House seems to be all but a lock for Democrats. The race for control of the Senate is shaping up to be quite competitive, and the precise margin of control will be important in determining what either party will be able to accomplish. I wrote a piece back in July in which I discussed the competitive Senate races this cycle—the only caveat I would add is that South Carolina's Senate race has become more competitive since, but otherwise, it is a good overview. There are dozens of ballot initiatives across the country, ranging from raising the minimum wage to $15 an hour in Florida, to changing the voting system in Massachusetts (check your state here!). Control of state government is up for grabs in a number of states, which will have major implications for not only policy setting, but also creating Congressional districts for the next decade.

In the meantime, do not discount any possibility within the realm of reason from occurring tomorrow. As easy as it is, and as much as we would like to, we cannot change probabilities into a binary by rounding them up or down. 10% is not 0%. 90% is not 100%. Also, do not expect the race to be decided on election night–results could take until the end of November to be fully settled as mail ballots are counted. Finally, I did not even touch upon the possibility of drawn-out legal battles leaving the race undecided for days or even weeks. Trump has signaled that he intends to contest any election result that shows him losing, and has spent months attacking the integrity of U.S. elections. For now, just make sure you keep on breathing, and make your vote count!