Post-Election Vote by Mail Analysis
In spring of 2020, the coronavirus was hitting — and sweeping — our nation. As Americans across the country sheltered in place and began adjusting to the “new normal” of masks and a sluggish economy, some began to wonder how this global pandemic would impact the 2020 election.
As part of our non-partisan mission, Citizen Data seeks to produce actionable data and insights in the public interest. Early into the pandemic, our team — along with others who ultimately partnered with us on this effort — recognized that the logistics of in-person voting would present safety concerns and challenges if COVID was to continue into the fall, and that voting by mail would likely be implemented in the majority of states. However, we also identified a knowledge gap: because the circumstances were unprecedented, election administrators and non-partisan non-profits were left without the critical information they needed to plan ahead.
Over the course of the spring, summer, and fall, our team partnered with Stanford-MIT Project on a Healthy Election, National Vote at Home Institute, and more in a robust effort to predict, project, analyze, and understand vote method behavior nationwide and in key states including Iowa, Georgia, Texas, Pennsylvania, North Carolina, Michigan, Florida, and Wisconsin, among others. The goal was to provide election officials and non-partisan groups the data necessary to allocate resources strategically and help ensure a smooth, secure, and representative election.
In addition to bespoke election and voter behavior research, we conducted a series of surveys nationwide and in key states to representative and weighted samples of registered voters. Our team also coded and produced predictive models anticipating voter behavior by vote method. Ahead of the election, we compared our projections to 2016 actual voter data in order to demonstrate how much turnout and vote by mail returns would increase; then, post-election, we compared our projections to 2020 actual voter data in order to assess our model accuracy and identify areas for improvement and further precision in the future. With the benefit of a few weeks having passed between the election and now, with all states having their election results certified, our team is now releasing a comprehensive analysis of our findings from the months leading up to the election and beyond.
Our key findings presented throughout the analysis include:
- Our projections for overall voter turnout numbers and vote by mail turnout were highly accurate; 95.1% and 94.5% accurate on average in key states, respectively.
- While presidential polling data was notoriously inaccurate across the board, our team’s survey data predicting vote method was much more accurate, relatively: in most key states, the final percentage of estimated vote method drawn from our survey data fell within 2 to 3 percentage points of the actual vote method numbers.
- Democrats were much more likely to plan to vote by mail, and ultimately, they did vote by mail in numbers much higher than non-partisans and Republicans. In fact, Democrats were almost exactly twice as likely to vote by mail than Republicans.
- There is strong support in USPS among voters broadly, and that support rose steadily as we moved closer to Election Day.
- Factors such as increased convenience and transparency or COVID cases increasing in the area made these voters say they were much more likely to vote by mail; the USPS experiencing delays that would affect their ability to deliver ballots made voters 41.4% less likely to plan to vote by mail.
- In the days and weeks leading up to Election, our team predicted that election counts and calls would go more smoothly than some had anticipated; it did.