From a 2016 working paper by Enrico Cantoni.  Below are the abstracts to the August and November versions of the paper; I like how the writing was tightened up and extraneous phrases dropped. Fewer words conveying the same information more clearly. The standard deviation of distance is given in the second version, rather than making the reader search for the value in the paper. Also, “turnout” is replaced by the more precise “number of ballots cast, ” which has the extra virtue of eliminating the earlier ambiguity about whether “reduces … by approximately 2% to 5%” referred to turnout percentage points or percent (which the paper did make clear in the text).

August 2016 version abstract: In a sample of municipalities in Massachusetts and Minnesota, I use a novel, quasi-experimental design based on geographic discontinuities to study the turnout effects of voting costs. I compare parcels and census blocks located in close proximity to boundaries between adjacent voting precincts, which determine assignment to polling places. Geographic units that share (on either side) a precinct boundary also share observationally identical attributes. At the same time, the discontinuous assignment to polling places across boundary sides provides quasi-random treatment variation. I find that a 1-standard deviation increase in distance to the polling place reduces average turnout by approximately 2% to 5% in the 2012 presidential, 2013 municipal, 2014 midterm, and 2016 primary elections. I also document a negative but imprecise effect on census block voter registration, which suggests that higher voting costs reduce registration directly, by dissuading eligible voters from registering, or indirectly through the removal of inactive voters from voter rolls. During non-presidential elections, the effects of distance to the polling place concentrate disproportionately in high-minority, low-income, and low-car-availability areas, while no differential impact emerges in the higher-salience 2012 election.

November 2016 abstract: I study the effects of voting costs through a novel, quasi-experimental design based on geographic discontinuities. I compare parcels and census blocks located near borders between adjacent voting precincts. Units on opposite sides of a border are observationally identical, except for their assignment to different polling locations. The discontinuous assignment to polling places produces sharp changes in the travel distance voters face to cast their ballots. In a sample of nine municipalities in Massachusetts and Minnesota, I find that a 1-standard deviation (.245 mile) increase in distance to the polling place reduces the number of ballots cast by 2% to 5% in the 2012 presidential, 2013 municipal, 2014 midterm, and 2016 presidential primary elections.

Cantoni also develops an algorithm that redraws precinct lines; “the algorithm reduces the average parcel-to-polling-place distance by approximately .03 mile.” In case you were wondering, .03 miles is about 158 feet, or about 50 steps in you have a long stride like mine. So the cost of redrawing all lines seems likely to exceed the fairly modest increase in turnout, and one has to wonder whether other, cheaper alternatives to get out the vote, such as SMS messaging and targeted public service announcements or Facebook campaigns would not be more cost-effective (not that Cantoni is advocating redrawing, it is just an exercise). All in all, the paper confirms that relatively inexpensive changes in the implementation of elections can generate more participation, especially for already more-marginalized voters.

A couple quick remarks. I enjoyed the welcome reintroduction of the word “Alas” into mainstream economics on page 10. The last line of the draft paper,

However, the noticeable potential for higher turnout and lower turnout inequality – especially during less salient elections – should be both a memento and a goal for future research on the determinants of voter participation.

has the oddest use of the word memento I have ever encountered. Not really sure what he meant. Some kind of archaic usage?