To build a baseline model of the 2020 presidential election we started with the results of the 2016 election, measuring the support for President Donald Trump and Hillary Clinton by race, college education and age in each state. Specifically, we used exit polls and data from the 2016 Cooperative Congressional Election Study to estimate each demographic group’s partisan preference, and data from the Census Bureau’s 2016 report on voting and registration to estimate each group’s turnout level. We then applied these estimates at the county level and used a statistical technique (called iterative proportional fitting) to “fit” the estimates to actual vote results before reconstructing them at the state level. Finally, we adjusted the size of each demographic group by doubling census-estimated population change for voting-age citizens from 2016 to 2018.
