Identifying target races

This project begins with the identification of the 23 U.S. Senate seats currently held by Republicans that are up for re-election in 2020. All 23 races were assessed against a set of standard criteria that included PVI, vote history, and whether Clinton won in that district, among others. This initial analysis generated a list of four priority races with another eight to watch.

Simulating turnout

Next we simulated total turnout in each of the 23 prospective races. Whereas aggregate-level indicators are often used to estimate total turnout in a race (e.g., total turnout in the last comparable election, total number of registered voters in the last comparable election), we used individual-level voter files for our analyses.[1]

Analyses of individual-level voter files

While aggregate-level indicators can be used to estimate total turnout, it is only through analyses of individual-level voter files that one can answer important questions about who is and is not likely to vote. We used the partisan score, race, and vote history variables found on the Catalist Voter File for these analyses.

Predicting a high-turnout election in 2020

We assume that 2020 will be a high-turnout election given the historic turnout of voters for the 2018 midterm cycle, sustained enthusiasm, and the high turnout for presidential election years in general. Accordingly, our models adjust for those anticipated changes in turnout rates among the Democratic and Republican voters in 2020. Similarly, we used the 2012 presidential general election, 2014 midterm general election, 2016 presidential general election, and 2018 midterm election to construct our 2020 turnout estimates.

The resulting turnout estimates for Democratic and Republican voters in each state enable us to assess the likely election outcomes for 2020. In some races Republican candidates are estimated to receive more votes than the Democratic candidates, while in others the opposite outcome is estimated.

We also determined the number of eligible voters who could be registered by calculating the total citizen voting age population less the total number of registered voters. We then examined the racial composition of eligible voters and applied a percentage of those who voted for Clinton the 2016 presidential election in each state by race/ethnicity to estimate the number of likely voters of color and progressive whites who would likely vote for Democratic candidates if they registered and cast ballots..

The analysis resulted in the identification of Republican-held U.S. Senate seats where Democrats are most able to elect their candidates.


[1] Voter files were obtained from Catalist.