Distance traveled to polling locations: Are travel costs imposed equally on party members?

ABSTRACT

Previous studies show that geographical distance to polling locations represents a genuine cost of political participation and influences voter turnout. Here we examine whether travel costs are equally distributed across party members and do not manifest themselves randomly.

Methods: Using 2016 data obtained from a County Clerk’s Office (N = 81,323), and a scripted algorithm developed in Python to interact with Google Maps, we were able to identity the road distances between registered voters’ addresses and their polling locations. We then regressed this distance against several available measures to determine whether a relationship existed between voter’s political affiliation and distance to the polls.

Results: The multivariate analyses show that travel distances to polling locations were not equally distributed across partisan groups. Rather the dominant party traveled significantly less distance. Additionally, we used the same methodology to demonstrate the suitability of alternative polling locations. We introduced polling locations that significantly reduce the overall travel distances and shrink existing disparities between party members.

Conclusions: It is important to recognize travel distances as a cost to individual voters and determine whether present polling locations represent greater costs for some groups than others.

KEYWORDS:

• Travel distances
• Polling locations
• Party
• Cost of voting
• Local election officials

Notes

¹ See Gimpel and Schuknecht (2003), Dyck and Gimpel 2005 for studies that utilize a city-block method – which applies longitude and latitude coordinates – for calculating distance to polling locations.

² Distances may vary depending on time and date. Google constantly up-dates travel distances based on traffic, road construction, and other potential hazards. Our data were generated on October 2, 2017.

³ Zero distances were recorded when voters’ address matched a polling location. A polling location at a retirement community for example resulted in zeros for voters living in that community. This occurred for 199 voters, representing .24% of registered voters.

⁴ For our analyses, Libertarians were not used. There were 947 Libertarians, representing 1.16% of registered voters. Including them did not alter results.

⁵ We tested the model for heteroscedasticity using the Breusch–Pagan/Cook–Weisberg test. Chi-square values indicated heteroscedasticity present. As a correction of potentially biased standard errors, we used OLS with robust standard errors. Standard errors changed slightly, though not enough to alter the significance of estimates.

⁶ This measure closely approximates identification of precincts by city and rural categories. That is, in rural precincts the Trump vote share is much higher than in city precincts. Further we utilized the party competition measure, compared to the simple percentages of Republicans in a precinct, because it reveals actual voting behavior.

⁷ The specific finding for partisan differences in precinct 1 reflected well the general changes across the 17 substituted locations.

⁸ While these are possible polling locations, we recognize there are other factors that may make such alternatives inappropriate.