Overview

Welcome! This website holds our forecasts for the 2020 U.S. presidential elections. Our work centers around using a compartmental model to forecast elections and is based on the model and methods introduced by A. Volkening, D.F. Linder, M.A. Porter, and G.A. Rempala in this article. It is part of our undergraduate research project in applied mathematics at Northwestern University.

 
We forecast the vote in thirteen swing states individually and combine the rest of the states into either safe red or safe blue groups. These groups are then forecast collectively as red and blue "superstates". Learn more about our methods and state groupings here.
 
Check back for continual updates of our forecast as we get closer to the election!

 

– Samuel Chian, William He, and Christopher Lee

U.S. Presidential Forecasts as of September 14, 2020, 20:00 CDT

Using polling data from FiveThirtyEight that we gathered on September 13, 2020, at 22:00 CDT, our model forecasts a victory for Joe Biden 86.74% of the time, and a victory for Donald Trump 12.82% of the time.


In this second forecast with Arizona considered a swing state, our projections did not change significantly from our prior forecast 10 days ago. Overall, Trump's odds of winning increased very slightly.

 

The distribution of our simulated elections that resulted in various electoral-vote outcomes, across 10,000 stochastic simulations of our model

 

Our forecast vote margins for each state and superstate (lines indicate our 80% confidence intervals)

Prior Forecasts

You can access the forecasts that we made for the 2020 presidential elections on earlier dates by clicking the links below to expand the page.

September 3, 2020, 23:00 CDT

Using polling data from FiveThirtyEight that we gathered on September 3, 2020, at 23:00 CDT, our model forecasts a victory for Joe Biden 87.26% of the time, and a victory for Donald Trump 12.42% of the time. This forecast reflects new polls that were conducted following the Democrat and Republican National Conventions.


Additionally, we made an important adjustment to our model: we now forecast Arizona individually as a swing state instead of as part of the Red superstate, and will continue doing so going forward. We believe this change enables our forecasts to better reflect the uncertain nature of Arizona this year (our model currently forecasts it going Democrat), as well as ensure that the Red superstate is properly composed of safe red states that will not skew its forecasted vote margin towards Democrat.


This recategorization of Arizona may have helped the Red superstate be more solidly Republican, and its 80% error bar no longer crosses over to Democrat. Still, it goes Democrat in over 5% of our simulations (although less often than before). Other changes in our forecasts include Iowa and Ohio moving more Republican, and our electoral vote distribution becoming more spread out across multiple possible outcomes, with less of a concentrated spike at the point representing all swing states going Democrat. Overall, the election odds for each candidate are similar to what they were on the previous forecast; Arizona being split off as a swing state and forecasted as leaning Democrat may have balanced out with Red superstate, Iowa, and Ohio moving Republican.

 

The distribution of our simulated elections that resulted in various electoral-vote outcomes, across 10,000 stochastic simulations of our model

 

Our forecast vote margins for each state and superstate (lines indicate our 80% confidence intervals)

August 14, 2020, 23:00 CDT

Using polling data from FiveThirtyEight that we gathered up until August 14, 2020, at 23:00 CDT, our model forecasts a victory for Joe Biden 87.17% of the time, and a victory for Donald Trump 12.7% of the time.


Notable changes from our prior forecast include that the red superstate now has a greater forecast margin of Republican victory; this is the main contributor to the slightly greater odds that we now give Donald Trump. Meanwhile, Ohio has flipped from slightly Republican to just barely Democrat, with a forecasted margin of victory near zero. Iowa and North Carolina did not flip to the other party, but the forecasted margins of both are now closer to zero, indicating very close races. Our forecasts for other states stayed mostly the same, without significantly impacting the overall electoral vote distribution.

 

The distribution of our simulated elections that resulted in various electoral-vote outcomes, across 10,000 stochastic simulations of our model

 

Our forecast vote margins for each state and superstate (lines indicate our 80% confidence intervals)

August 3, 2020, 19:30 CDT
 

Our forecast vote margins for each state and superstate (lines indicate our 80% confidence intervals)

 

The distribution of our simulated elections that resulted in various electoral-vote outcomes, across 10,000 stochastic simulations of our model


Here are our forecasts based on data from FiveThirtyEight that we gathered up until August 3, 2020 at 19:30 CDT.


Our model forecasts a victory for Joe Biden 88.21% of the time and a victory for Donald Trump 11.73% of the time.


One interesting point is that, in about 10% of the 10,000 elections that we simulate, the red superstate actually votes for Joe Biden. As a result, our model forecasts that there is about a 10% chance that Donald Trump will earn no electoral votes, which is clearly unrealistic. It should be noted, however, that even if we specify that the red superstate always votes Republican, those 10% of outcomes would still correspond to an overall Democrat victory (due to the swing states going Democrat), so this pheneomenon has little effect on the chances that we give each candidate of winning the election. The cause of the red superstate not being solidly Republican could be due to some states in our red superstate leaning more Democrat this year; for example, many other forecasters (such as Sabato's Crystal Ball) are currently calling Arizona a swing state.

We may want to consider reducing the number of states that we include in our superstates this year, recategorizing Arizona, for example, as a swing state. However, there are some drawbacks to this; for instance, our forecasts take longer to produce when we add more individual states into our modeling framework. Moreover, adjusting our superstate categorizations may make it more difficult to compare our 2020 forecasts to the model's performance on forecasts of past elections, which were made with the same arrangement of states in the superstates that we are using right now.

 

Our state ratings as Solid, Likely, Lean, and Toss-Up