Since the news media called the 2020 presidential election, observers have offered a variety of explanations for Joe Biden’s victory over Donald Trump. These include Trump’s erratic behavior; Biden having “the most sophisticated digital apparatus in American political history”; Biden’s “bump” in appeal, above other Democrats; and 10 moments in which Trump lost the election.
While each of these explanations — and others you may encounter — could have some merit, campaign particulars were probably not that important. If we look at data from more than 100 days before the election, the result has been almost exactly as we’d expect. The real surprise — and concern — of 2020 is how stable Trump’s approval ratings have remained.
Was the outcome really predictable? Some may question this claim since the election was much closer than prominent forecasts expected. Additionally, some polls overestimated Biden’s support. For instance, Biden won Wisconsin with a much slimmer margin than predicted by the final polls, which suggested an 11- to 17-point Biden lead.
But even when results may seem surprising, as political scientists, we know to be cautious of after-the-fact attempts to explain the election outcome. Presidential elections typically come down to the “fundamentals,” like economic conditions and presidential approval ratings. Months before the election, political scientists use variables like the economy and popularity of the incumbent president to forecast the final vote with a high degree of accuracy.
2020 was no different
We developed a forecast of each state’s presidential vote based primarily on three fundamental factors: each state’s presidential vote in 2016; state economic conditions through June of the election year, and state presidential approval ratings based on surveys conducted in June and July of this year. Our forecast makes a simple assumption: The relationship between these variables and state presidential vote is the same as the average relationship between these variables and vote choice during the past nine elections. The closer our prediction is to the final outcome in each state, the more evidence we have that economic conditions and presidential approval influenced this year’s election in similar ways to past elections. Since 2000, our model would have correctly predicted 94 percent of states using data from months before each presidential election.
In 2020, based on data from 104 days before the election, our forecast was again extremely accurate. If Biden wins Arizona as currently projected, our model will have correctly predicted 49 out of 50 states plus Washington, D.C. Based on current votes, still being counted in some states, Georgia was the only state our model missed. In other words, in almost all cases, votes in each state fell about where we would have expected given economic conditions through June and approval ratings from last June and July. In Arizona, for example, Biden has 50.2 percent of the two-party vote (i.e., the percent of votes out of those cast for Biden or Trump). Our forecast expected Biden to win 50.6 percent of the two-party vote. We were similarly close in Wisconsin, where Biden has so far received 50.3 percent of the two-party vote; we forecast 50.7 percent would go to Biden.
Of course, not all of our predictions were this close. But they did quite well. Among the nine battleground states, our average error is just 2.5 percentage points. That amount can swing the outcome in a close election, but whether we consider the winner or the percent of votes received, there weren’t major surprises based on what we would have expected more than 100 days in advance. These results suggest fundamental factors have more influence than pundits are considering.
What about Georgia?
Assuming Biden’s lead persists, Georgia is the state we missed. Biden’s votes exceed what we would have expected based on the fundamentals by 3.7 percent. Stacey Abrams’s efforts to inspire soaring registration and historic turnout have been credited with Biden’s presumed victory. These accounts are certainly consistent with Biden exceeding our forecast’s expectations. At the same time, Biden exceeded our forecast in North Carolina by the same margin. Both Georgia and North Carolina were considered battleground states before the election and the final outcomes in both states were equally surprising based on the fundamentals.
Why are we not seeing more accounts of Biden’s unexpectedly high performance in North Carolina? The main difference appears to be that Trump is expected to win there. Because Georgia is the Southern state that unexpectedly appears to have gone for the Democrats, pundits are scrambling to explain that result while largely ignoring Biden’s equally surprising performance in North Carolina, where he earned 49.3 percent of the two-party vote. Not all states that deserve an explanation get one, and some that need no explanation get plenty of attention. Forecasts like ours based on fundamentals can help identify which state outcomes were truly a surprise.
The biggest — and most concerning — surprise
The overall story of Biden’s victory, then, is that despite $6.6 billion dollars in campaign expenditures, the fundamentals from June and July are what mattered for predicting the outcome. Trump was a relatively unpopular incumbent with an economy that was strong for much of his term, but was devastated by a global pandemic. His defeat was exactly what we would expect: decisive, but not a blowout. The relationship between the summer’s presidential approval and final presidential vote choice worked as it usually does.
The real surprise is the stability of Trump’s approval rating throughout his presidency. This stability is both puzzling and concerning. If the lowest unemployment rate in half a century did not noticeably increase approval and if more than 239,000 deaths related to the pandemic did not dramatically decrease approval, objective conditions may no longer influence approval ratings — which is very troubling for democratic accountability. The link between presidential approval and vote choice loses its meaning if approval ratings stop reflecting reality.
Peter K. Enns (@pete_enns) is an associate professor in the department of government at Cornell University and co-founder and chief data scientist at Reality Check Insights.
Julius Lagodny (@JuliusLagodny) is a PhD candidate in the department of government at Cornell University.