Enlarge /. A socially distant vote may not be required if a vote is accepted by mail.
In the face of an uncontrolled pandemic, many states have considered switching to an email voting system. This turned out to be politically controversial as President Trump made evidence-free claims that it will lead to massive fraud or give the Democrats some other unspecified advantage. Meanwhile, absentee voters sometimes present the solution to the generally low turnout of US citizens in elections. They claim that a postal vote will significantly increase voter turnout.
Two researchers decided that this was something they could answer. Michael Barber of BYU and John Holbein of the University of Virginia used states where voting was carried out by county-to-county mail. Their results suggest that voting via email did not give any party an advantage, but neither did it do much to increase participation.
A story of two states
The work relies on voter records and census data from two states who have voted by mail but have not implemented it consistently across the state. Instead, both Utah and Washington had different counties that implemented the system at different times, and provided a number of changing test and control conditions. The two states balance each other out well because they can be viewed as mirror images. Eastern Washington is largely rural and conservative, but state politics is dominated by the urbanized coastal region. In Utah, conservative politics dominate the state, while the Salt Lake City area is somewhat liberal.
The challenge to work is that each election has a unique dynamic, so in addition to rolling out the voting in the mail, you expect a specific backdrop for change. To handle this, Barber and Holbein do what is known as a difference-in-difference analysis. This form of analysis tracks changes in each county during the elections and compares these differences with differences in other counties. Ideally, this will largely explain the differences between the elections, with the only thing that will stand out being the differences caused by each district's switch to mail when voting.
However, there are several ways to implement difference-for-difference analysis, and the researchers tried a number of different mathematical models on their data. They also broke things up at the county level using individual voting papers and census data.
When it comes to voter turnout, the impact of absentee voting is small but positive. Depending on the model used, voter turnout increases by 1.8 to 2.9 percentage points, an effect that is statistically significant.
A story of two parties
However, there was no indication that either party benefited from this increased turnout. Here you can think of the authors' approach as the opposite of P-hacking: they tried different models and didn't get a statistically significant effect in any of them, which gives them confidence in the negative outcome. While the average of the models represents a 0.7 percentage point increase in the proportion of Democratic voters, the confidence intervals extend to negative numbers – which means that this could just as easily increase the proportion of Republicans.
"For us it seems possible, maybe even likely," suggested Barber and Holbein, "that VBM will attract people on the fence about voting, many of whom are individuals in both political parties."
With this study there are definitely nits to be picked. While Utah and Washington are great for balancing liberal / conservative issues, they are definitely not representative of US demographics as a whole. And it's possible that there are some subtle demographic differences associated with how each county in these states will accept votes. However, we are unlikely to do much better than this analysis, and we certainly will not do it until critical decisions are made about this year's election. When states debate what to do, they should do so with the knowledge that the best evidence available suggests that voting by post does not provide any party political advantage.
Science Advances, 2020. DOI: 10.1126 / sciadv.abc7685 (About DOIs).