• The Staff

A Global Perspective on Predicting the 2020 Election

By Stefan Wojcik, Ryan Kennedy, and David Lazer


The US has never had an election like this before--but other countries have. Comparative analysis provides leverage for forecasting 2020 US elections.



As Vice President Biden has widened his lead in polls leading up to the 2020 election and President Trump has signaled a willingness to take actions that might undermine the results (see here, here and here for example), forecasters find themselves in somewhat new territory. Models based solely on presidential elections in the U.S. do not have many past elections on which to base their probability estimates (17 since the famous Truman/Dewey election changed polling methods). Moreover, they do not generally take into account the potential willingness of a leader to take actions that might undermine the outcome of the election.

There is, however, another way to try doing this.


In a paper published in Science in 2017, we assessed more than 146 rounds of direct presidential elections from both democratic and non-democratic countries around the world, collecting the results of about 4,331 polls. This allowed us to develop a more comprehensive view on both the reliability of polls internationally and to figure out how successful we could be in predicting election outcomes, even in regimes that were less-than-fully democratic. The results were encouraging, with 90% prediction accuracy in the final model.


So what would the data say about this election? How likely is it that President Trump would be re-elected to the presidency? And, if he should take some non-democratic steps to try and maintain his office, what is the likelihood that they will be successful?

How Likely is President Trump to Remain in Office?


Polls have shown a widening national lead for Biden in the past weeks, up to 8 points for the week ending October 24. In democracies globally, the average polling error tends to creep downward as the election draws near. While in nondemocracies the average error tends to vary more widely over the course of the election, they generally give less information about the outcome.


But this does not mean that incumbent party candidates poll this badly even in nondemocracies. If we plot the distribution of polling margins at various points in the election cycle for democracies and nondemocracies, we see that winning incumbents rarely poll this low in the weeks or months leading up to the election.


The results suggest that, even in non-democracies, incumbents are not likely to maintain office when the polls suggest this large of a skew towards their opposition. Table 2 shows the average percent of polls in each election giving the incumbent an 8 point deficit at various points in the campaign - the same deficit where President Trump was polling nationally in mid-October.


These results suggest that President Trump would be among a relatively small group to poll this badly and retain his office, even if he were willing to take actions that might undermine the level of democratic fairness of the election.


But we rarely look at a single poll. According to simple tabulations based on this data, only 1.5% of incumbent party candidates in democracies show a polling average this low in the last 30 days of an election and ultimately win. 4.8% of incumbent party candidates in nondemocracies do the same. Examples of this tend to be under extreme circumstances that deviate substantially from the U.S. case.


The first round of Ecuador’s 2006 elections -- where Alvaro Noboa won a surprising plurality over Rafael Correa in the first round of the election, but ultimately went on to lose in the second round -- is a case of an election in a democracy, but where votes were split between 13 candidates, resulting in last minute strategic switches by voters.


One example from an autocracy is Algeria in 1995, where, in the midst of a civil war and with armed groups threatening violence, Liamine Zéroual won a victory over Mahfoud Nahnah despite one poll in the run-up indicating that he trailed by about 10 points.


While the set of non-democracies likely includes both countries in the process of transition and countries exercising far more extensive electoral manipulation than that threatened by President Trump, this low success rate of incumbent party success is still striking. The model from our original paper is more conservative than the average among democracies and more confident than the average of nondemocracies. In Figure 2, we plot 1,000 simulations from our model. It gives Trump a 2.5% chance of winning nationally. This is approximately in line with the latest FiveThirtyEight estimates which give the President a 4 in 100 chance of winning the popular vote.


Put simply, there is simply not much President Trump can do to avoid losing the popular vote in this election. Even in non-democratic countries, only very few incumbent parties have their candidate win an election when trailing by this much.


The “Yuge!” Caveat


Of course, we have to caveat this by pointing out that there is one very big difference between the U.S. system of electing a president and the system of just about every other country in the world. In the U.S., presidents are not elected based on the proportion of the popular vote they receive, but by the number of electors who vote for them. The number of electors is proportioned to the states based on their population numbers in the last census, and, with the exception of Maine and Nebraska, the number of electors received by a candidate is winner take all. Our model is based on the national popular vote, and might be mistaken in the U.S. case.


This raises the possibility that Vice President Biden could win the popular vote by a substantial margin, but still lose in the electoral college. Prior to 2000, only two presidents won the electoral college outright while losing the popular vote – Rutherford B. Hayes in 1876 and Benjamin Harrison in 1888. Since 2000, two of the last five elections have produced such an outcome.


Generally speaking, the electoral college has become much more favorable to President Trump and other Republicans, as Democratic voters are increasingly located in a few highly populated states. The reality is that only a few states really matter to the outcome, and the size of victory in a state means nothing for the distribution of electoral votes.


One could imagine a scenario where Vice President Biden wins by large margins in large states like California and New York, while President Trump ekes out very narrow victories in places like Michigan, Pennsylvania, and (yes) Texas. This could produce a popular vote landslide, at least in contemporary terms, for Vice President Biden while still returning President Trump to office.


In sum, while expanding our analysis to incorporate executive elections from around the world gives us more variety on which to base predictions, it is difficult to find a case that looks quite like the U.S. this year -- a democratic election taking place in the middle of a health crisis, with a skewed electoral college, an incumbent signaling potential anti-democratic actions, and Republican-appointed majority Supreme Court that may be asked to intervene on key issues of vote counting.

How Likely is This?

While the electoral college is tilted in Trump’s favor, the size of Vice President Biden’s lead has now reached a point where an electoral college inversion is relatively unlikely. A number of polls in swing states suggest that Biden is comfortably leading in several important battleground states, and forecasts that include the state battlegrounds similarly suggest a high probability of his winning the electoral college vote (see e.g., here, here and here). Current forecasts tend to give President Trump about a 1 in 6 to a 1 in 8 chance of winning. But, as others, have pointed out, these chances are not outside the realm of possibility. The former is about the same chances as a person has of getting the bullet in Russian Roulette, while the latter is about the probability of a couple having three female children in a row.

As always, polls are not a substitute for people showing up to vote.


Stefan Wojcik is a computational social scientist and data analyst at Twitter; Ryan Kennedy is associate professor of political science and founding director of the Center for International and Comparative studies at the University of Houston; David Lazer is university distinguished professor of political science and computer science and co-director of NULab for Texts, Maps, and Networks at Northeastern University.

©2019 by Mischiefs of Faction.