Abstract
On October 17, 1989, the Loma Prieta earthquake struck the central coast of California. Three weeks later, on November 7, 1989, San Francisco held a municipal election. A dozen polling places had been destroyed or were otherwise inaccessible to the public due to the earthquake, so new polling places were selected for the affected precincts in the days leading up to the election. This case represents a credible natural experiment examining how changes in the costs of voting affect political participation, with the “as-if” random assignment of voters to the treatment group determined by earthquake damage to individual buildings rather than election administration decisions which could conceivably be related to turnout, such as precinct consolidation or the location of precinct boundaries. We use a difference-in-differences design, with the difference in turnout between the 1987 and 1989 municipal elections as the outcome variable. We find that voter turnout was 2.9 [5.1, 0.6] percentage points lower in precincts in which the polling place was relocated due to earthquake damage as compared to precincts that kept their original polling place.
Introduction
On October 17, 1989, the Loma Prieta earthquake struck the central coast of California, causing 63 deaths, thousands of injuries, and extensive property damage. In the City of San Francisco, hundreds of buildings were destroyed or damaged to the point that they posed a danger to the public (Seekins and Boatwright, 1994).
Three weeks later, on November 7, 1989, San Francisco held a municipal election. A dozen polling places had been destroyed or were otherwise inaccessible to the public due to the earthquake. The San Francisco Department of Elections selected new polling places for the affected precincts in the days leading up to the election (Sandalow, 1989; San Francisco Department of Elections, 2009).
This case represents a credible natural experiment examining how changes in the costs of voting affect political participation, with the “as-if” random assignment of voters to the treatment group determined by earthquake damage to individual buildings rather than election administration decisions which could conceivably be related to turnout, such as precinct consolidation (Brady and McNulty, 2011; McNulty et al., 2009) or the location of precinct boundaries (Cantoni, 2020). We expect that changing the location of a polling place in the days leading up to an election will increase the transportation and search costs of voting, and thus decrease voter turnout (Brady and McNulty, 2011; McNulty et al., 2009).
Data
Treatment and outcome variables
Our treatment variable measures whether a precinct’s polling place was relocated due to earthquake damage. Twelve precincts were forced to relocate their polling places due to earthquake damage (Sandalow, 1989; San Francisco Department of Elections, 2009). The distance each polling place was moved ranged from 0.1 to 0.6 miles, with an average distance of a little over 0.25 miles.
Our outcome variable is the precinct-level difference in registered voter turnout between the 1989 municipal election and the November 3, 1987 municipal election. There were a total of 711 precincts in San Francisco in 1987, which were consolidated into 621 precincts in 1989. The difference in turnout for each precinct was calculated as the percentage point difference between polling place turnout in the 1989 precinct and the equivalent precinct or set of precincts in 1987.
Voters who were unable to vote at their precinct’s polling place were allowed to request a mail ballot (San Francisco Office of the Registrar of Voters, 1989). Mail ballot voters were not assigned to a specific geographic precinct in the statements of vote, and are thus not included in the analysis. However, mail ballots needed to be requested by mailing a postcard to the Registrar of Voters, and given the disruption to the postal service in the three weeks between the earthquake and the election, as well as earthquake damage to City Hall (which housed the Office of the Registrar of Voters), it is unlikely that many voters who had intended to vote at their polling place before the earthquake would have been able to obtain mail ballots in time.
Information on voter turnout and precinct boundaries was obtained from the archives of the Government Information Center at the San Francisco Public Library. These archives had paper records of the 1987 and 1989 precinct-level statements of vote, as well as a paper precinct map that applied to both the 1987 and 1989 elections. All of this information was digitized for the following analysis. No individual-level information on voter registration or turnout was found in the archives despite an extensive search.
Earthquake Disruption
Precincts with an earthquake-damaged polling place were also likely to have experienced other earthquake-related disruptions that could affect voter turnout. The amount by which each precinct was disrupted by the earthquake was assessed through heat emissions captured by satellite imaging. All else equal, higher levels of human activity should lead to higher heat emissions in an urban environment as people use energy in their daily activities (e.g., driving vehicles, using electrical appliances). A drop in heat emissions relative to the historic baseline is an indication that human activity in that area has declined. A growing number of studies have measured human activity through remote sensing data, most notably by measuring economic activity through nighttime light emissions (Donaldson and Storeygard, 2016). Our measure of earthquake disruption is the percentage difference between actual heat emissions captured the day after the election (November 8, 1989) and predicted heat emissions based on the historic baseline captured by earlier satellite images, with higher numbers indicating a greater reduction in heat emissions relative to baseline levels.
The measure of earthquake disruption across all precincts is presented in Figure 1, along with the locations of the precincts with earthquake-damaged polling places. Our measure of earthquake disruption produces sensible results, with the largest estimated reductions in human activity in areas known to be the most affected by the earthquake (such as the heavily damaged Marina District on the north coast of the city).
Disruption in San Francisco due to the 1989 Loma Prieta earthquake.
Demographics and precinct consolidation
Demographic data for San Francisco was obtained from the 1990 U.S. Census (U.S. Census Bureau, 1992). We estimated the demographics of each precinct by overlaying a map of the 1990 U.S. Census block groups on the precinct map, and assigning the Census population to each precinct based on the percentage of each block group that fell within the precinct boundaries. We then calculated the percentages of the precinct population that were over 65 years of age, college educated, black, Asian, and female. The mean income for each precinct was calculated by assigning each individual the median income for their Census block group and averaging across individuals within the precinct.
Some of the 1987 voting precincts were consolidated for the 1989 election. As with earthquake damage, precinct consolidation also changes the location of the polling place for at least some voters in the consolidated precinct. Precincts might also be consolidated in areas where election officials anticipate lower turnout. We created a dummy variable for each precinct that was consolidated in 1989.
Methods and results
We use a difference-in-differences design, with the difference in turnout between the 1987 and 1989 municipal elections as the outcome variable. Due to the disruption from the earthquake, and the fact that 1987 was a mayoral election year while 1989 was not, overall voter turnout was expected to be lower in 1989 than in 1987. Our natural experiment tests whether this decline in voter turnout was steeper for those precincts in which the polling place was relocated due to earthquake damage.
We used entropy balancing to reweight our data, creating covariate-balanced treatment and control groups (Hain-mueller, 2012). We then estimated an entropy-weighted least squares linear regression on all 621 precincts.
Table 1 presents our estimate of the average treatment effect on the treated (ATT). Our estimate indicate that precincts in which the polling place had been relocated due to earthquake damage showed a statistically significant decline in voter turnout of approximately 2.9 [5.1, 0.6] percentage points.
The effect of earthquake-damaged polling places on voter turnout in the 1989 San Francisco municipal election.
Note: Standard errors are heteroskedasticity-corrected (HC1).
Our difference-in-differences design controls for time invariant confounders within each precinct. If the parallel trends assumption holds, our estimate of the difference between the treatment and control groups is an unbiased estimate of the ATT.
Unobserved time-varying confounders that could lead to non-parallel trends for the treatment and control groups, such as large changes in the population composition of the treated precincts relative to the control precincts, are unlikely to have occurred in the two years between the 1987 and 1989 elections. To support this assumption we undertook a placebo test utilizing the same methods described above, with the placebo group defined as the 43 precincts immediately adjacent to the treated precincts (two large, sparsely populated precincts were excluded from the placebo group), and excluding the treated precincts from the analysis. Since the treated precincts are small geographic areas (see Figure 1), any unobserved time-varying confounders influencing the treated precincts would also be likely to affect neighboring precincts. The 95% CI on the estimated placebo effect [-2.274, 0.793] both included zero and excluded the estimated ATT, providing greater confidence that the parallel trends assumption is reasonable.
Discussion
Our results are consistent with those from other natural experiments examining voter turnout and the location of polling places. For example, Brady and McNulty (2011) examined changes in polling place location due to the consolidation of precincts in Los Angeles County during California’s 2003 gubernatorial recall election, and found that turnout at the polling place was reduced by about 3 percentage points among individuals for whom the polling place had changed. Similarly, Cantoni (2020) examined geographic discontinuities in travel distance to polling places based on precinct boundaries, and found that increasing travel distance by about 0.25 miles decreased voter turnout by 2 to 5 percentage points – in our case the relocated precincts moved by an average of about 0.25 miles (although the distance to the polling place did not necessarily increase for all voters in the treated precincts).
Our estimated decline in voter turnout is smaller than that estimated by McNulty et al. (2009), who examined changes in polling place location due to the consolidation of precincts in a school district election in New York State, and found a 7 percentage point reduction in turnout among individuals for whom the polling place had changed. McNulty et al. (2009) suggest that the decline in voter turnout in the Los Angeles case studied by Brady and McNulty (2011) may have been mitigated by the high profile nature of the recall election. The 1989 San Francisco municipal election, although not a mayoral election, did have two high profile ballot initiatives (on domestic partnerships for same-sex couples and a new baseball stadium) that may have had a similar effect. Another mitigating factor may have been induced by the earthquake itself, as natural disasters may inspire pro-social behavior that subsequently increases voter turnout (e.g., Fair et al., 2017).
More broadly, the relatively modest decline in voter turnout we find in response to a significant disruption to the electoral process is consistent with research that finds that for most individuals the decision to vote does not depend heavily on the cost of voting (e.g., Blais and Daoust, 2020).
Supplemental Material
sj-pdf-1-rap-10.1177_20531680211004254 – Supplemental material for The Loma Prieta earthquake and the cost of voting in the 1989 San Francisco municipal election: Evidence from a natural experiment
Supplemental material, sj-pdf-1-rap-10.1177_20531680211004254 for The Loma Prieta earthquake and the cost of voting in the 1989 San Francisco municipal election: Evidence from a natural experiment by Garrett Glasgow, Pavel Oleinikov and Rhoanne Esteban in Research & Politics
Footnotes
Correction (June 2025):
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
Supplementary Material
The replication files are available at: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/K86FNS&version=DRAFT
Carnegie Corporation of New York Grant
This publication was made possible (in part) by a grant from the Carnegie Corporation of New York. The statements made and views expressed are solely the responsibility of the author.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
