The direct cost to voters of polling site closures and consolidation
The direct cost to voters of polling site closures and consolidation (DOI: https://doi.org/10.1017/psrm.2025.10034) was written by Scott Abramson and Sharece Thrower in 2025. It was published in Political Science Research and Methods.
The authors make use of a primary election day tornado event in Davidson County, TN, that forced polling sites to close. (Note that TN has open primaries, and in 2020 the Republican presidential nominee was unchallenged.) The authors can then isolate:
- the effect of moving one's polling location on their propensity to vote
- the effect of overcrowding one's polling location on their propensity to vote
Importantly, because this was an unexpected event hours before polling began, there likely were no mobilization or informational campaigns that could have provided a countervailing effect.
14% (21) of polling locations in the county were closed due to power loss. 20% of registered voters were affected and then reassigned to 1 of 6 consolidated 'Super Sites'. 2 'Mega Sites' were also established at the board of elections' main office and satellite office, and any registered voter was able to vote at these regardless of precinct. Public voter files indicate whether and how a person voted, which precinct they are (ordinarily) assigned to, etc.
To control for direct effects of the tornado, data from Nashville Electric Service (NES) is leveraged. Addresses that experienced property damage or power loss are directly identified; the authors broaden the definition to any address within 500m from a damaged or affected address.
The authors model propensity to vote, first in-person voring and second by early/mail-in voting. (The expectation is that early and mail-in voting should not see any significant effects, as those decisions predated the storm event.) For difference in differences reporting, they regress on first differenced measures. Standard errors are clustered by precinct.
In a third set of models, they leverage early/mail-in voting propensity as a uniformly untreated measure for people's instrumental preferences (i.e., their preference for an outcome by the expected probability their participation will be pivotal). They use the first differenced measure as a predictor.
The 2018 general election is used as the untreated baseline.
- Alternative baselines in the August 2018 and March 2016 primaries are tested. Findings are robust to these.
- Also robust to subsetting the set of voters to those who voted in the Democratic primary in the baseline election.
- In any case, observations are subset to those who were registered to vote for both the baseline election and the March 2020 primary election.
The treatment vector Tie includes fixed and variable cost terms. The estimated coefficients are denoted by piie.
- Fixed cost of reassignment is a binary indicator for having the ordinary polling location closed; estimated coefficient is the average effect.
- Variable cost of reassigned is the difference between logged distances (residential address to normal polling location, and to reassigned polling location).
- Fixed cost of consolidation is a binary indicator for voting at a consolidated site; estimated coefficient is the average effect.
- Variable cost of consolidation is the difference between logged voter counts (number of registered voters in ordinary precinct, and in consolidated precinct).
Before introducing the variable treatment terms, the coefficients are consistent with and without controls. Reassignment has a significant negative effect, lowering probability of turning out by about 5.7%. Consolidation is not significant.
After decomposing fixed and variable costs, for reassignment, the former becomes insignificant while the latter is both significant and negative. Interpretation is that variable costs of reassignment drive reductions in turnout. Furthermore, the variable cost of consolidation is now significant and negative.
Then the propensity to early/mail-in vote is controlled for, and for the most part the model's interpretation does not change.
- Fixed consolidation costs now have a statistically significant positive coefficient, suggesting a non-linear effect, and also possibly explaining why the first models did not detect a significant effect.
- "At the largest increase in precinct size, about 340%, turnout is reduced by 14.1 pp."
- Note that no treatment had a statistically significant effect on propensity to early/mail-in vote, as expected.
To demonstrate parallel trends, the authors run this model on the first differences of every pair of elections since Nov. 2012. Almost all parameters are statistically insignificant.
- "That is, for the 14 estimated coefficients, only one is indistinguishable from zero—which we would expect by chance." Is that valid?
- The authors also plot the marginal propensity to vote among the control and 'treated' groups across all of these elections.