Literature review

Do unexpected, large-scale events that threaten the security and livelihoods of individuals and communities—such as phenomena associated with natural hazards—affect electoral behavior? If so, what is the role of the government’s response to the crisis in this relationship? Scholars have found solid evidence that voters are responsive to good performance in areas which politicians can more or less control, such as the management of the economy (Fiorina Reference Fiorina1978; Duch and Stevenson Reference Duch and Stevenson2008; Powell and Whitten Reference Powell and Whitten1993; Singer and Carlin Reference Singer and Carlin2013) or public safety (Carlin, Love and Martinez-Gallardo Reference Carlin, Love and Martinez-Gallardo2011).

Nonetheless, other authors point out that voters are also sensitive to events that their representatives cannot either anticipate or influence. Some examples of these include fluctuations in the international commodity markets (Campello and Zucco Reference Campello and Zucco2016), the occurrence of pandemics (Gutierrez, Meriläinen, and Rubli Reference Gutiérrez, Meriläinen and Rubli2022; Piazza and Schneider Reference Piazza and Schneider2021), exchange rate crises (Murillo and Visconti Reference Victoria and Visconti2017), the timing of natural hazards (Healy and Malhotra Reference Healy and Malhotra2009), and even shark attacks (Achen and Bartels Reference Achen and Bartels2004; Fowler and Hall Reference Fowler and Hall2018). In this article, we focus on natural hazards as a type of large-scale and unpredictable events.

Political scientists have explored both the effects of exposure to natural events on electoral behavior and the variables that explain politicians’ distribution of postdisaster policies. However, despite the growing attention to this issue, there are still a few gaps in our understanding of the politics of natural hazards. We focus on two of them: whether voters reward or punish their representatives for damages associated with these events and the role of disaster-related policies as a mechanism in this relationship.

Regarding the first gap, the extent to which citizens punish or reward their leaders after a disaster is still a matter of debate. Whereas some studies show evidence of a negative relationship between exposure to a natural hazard and incumbent vote shares (Abney and Hill Reference Glenn and Hill1966; Arceneaux and Stein Reference Arceneaux and Stein2006; Gasper and Reeves Reference Gasper and Reeves2011; Heersink, Peterson, and Jenkins Reference Heersink, Peterson and Jenkins2017; Healy and Malhotra Reference Healy and Malhotra2010; Flores and Smith Reference Flores and Smith2013), others present results in the opposite direction or even null findings (Ramos and Sanz Reference Ramos and Sanz2020; Bodet, Thomas, and Tessier Reference Bodet, Thomas and Tessier2016; Bovan, Banai, and Banai Reference Bovan, Banai and Pavela Banai2018). These disparate findings seem to suggest that the political consequences of disasters are highly dependent on the context.

What is the political role of disaster-relief policies?

In terms of the second gap, the existing literature is not yet conclusive about three dimensions of the politics of disaster-relief policies: the role of electoral calculations in the distribution of aid, the drivers of citizens’ attitudes toward specific aid strategies, and the electoral effectiveness of different relief policies. Regarding the first element, some research claims that electoral variables play an important role in governments’ allocation of postdisaster spending. For example, Sainz-Santamaria and Anderson (Reference Sainz-Santamaria and Anderson2013) argue that US politicians distribute more disaster preparation funds to electorally competitive counties. Moreover, Gasper and Reeves (Reference Gasper and Reeves2011) show that American presidents are more likely to issue disaster declarations in competitive states. In addition, Chen (Reference Chen2008) presents evidence of an electoral bias in the distribution of Federal Emergency Management funds.

From a comparative perspective, Cooperman (Reference Cooperman2022) shows that local politicians in Brazil are also strategic in their drought declarations, as they tend to issue them more often when elections are close. Similarly, Gallego (Reference Gallego2018) finds that Colombian political elites took advantage of severe floods and landslides to exchange immediate relief for votes. Nonetheless, not all studies point out to such electorally oriented strategies of postdisaster-aid distribution, as Kumar (Reference Kumar2016) posits for the case of India.

Regardless of politicians’ motivations, there is still some disagreement about voters’ preferences and responsiveness to different types of postdisaster policies. For example, Healy and Malhotra (Reference Healy and Malhotra2009) find that voters in the United States prioritize short-term relief policies over long-term strategies to reduce risk to future disasters. Bechtel and Mannino (Reference Bechtel and Mannino2022) posit that citizens prefer a need-based approach to distributing aid after a natural disaster. Visconti (Reference Visconti2022) argues that these events increase support for left-leaning politicians, who are more likely to implement social policies that support the reconstruction process. Visconti (Reference Visconti2021) also shows that natural hazards have an impact on the policy priorities of citizens.

Finally, in terms of the electoral effectiveness of different types of aid, Cooperman (Reference Cooperman2022) posits that voters in Brazil reward politicians who issue drought declarations close to elections. Similarly, there is evidence that distributive spending is electorally beneficial among core voters in the United States (Chen Reference Chen2008). Nonetheless, other scholars point to different conclusions. Heersink, Peterson, and Jenkins (Reference Heersink, Peterson and Jenkins2017) argue that President Herbert Hoover suffered important electoral losses in places affected by the 1927 floods despite the allocation of aid by the federal government.

In summary, there are still a few important gaps in our understanding of the electoral consequences of natural hazards. First, we need more evidence on whether voters reward or punish their representatives for these events and under what conditions, particularly in non-US contexts. Second, the existing research has not fully explored the mechanisms behind this relationship, especially the role of postdisaster social policies; moreover, with a few exceptions (Healy and Malhotra Reference Healy and Malhotra2009; Visconti Reference Visconti2021, Reference Visconti2022), scholars focus on only one type of government actions in the aftermath of a disaster, either relief or prevention policies. Our article aims to contribute to address some of these theoretical gaps.

Natural disasters, social policies, and electoral outcomes

We advance two arguments. First, we posit that the ability to distribute government resources to the affected populations offers an electoral advantage to candidates from the incumbent party in charge of these resources. Hence, exposure to a natural disaster could benefit certain candidates in the elections following the event, conditional on the implementation of social policies in response to the crisis. Second, we expect that different characteristics of postdisaster policy (whether they offer private or public goods or whether they address the short-term effects of a disaster versus the long-term causes of vulnerability) affect their electoral performance.

The first element of our theory centers on the ability of the incumbent party to allocate public funds to the affected populations in the aftermath of a disaster. We argue that having control over the initial response to an emergency benefits candidates from the party in government under two scenarios. First, state bureaucracies may distribute relief and reconstruction funds on the basis of need, such as levels of damage in a particular neighborhood. Voters observe the performance of their representatives and vote accordingly. Second, incumbents may allocate these resources strategically, on the basis of electoral criteria and clientelistic networks. Party brokers influence the vote choice of the affected citizens.

Regardless of the specific mechanism, the observable implication is the same: exposure to a natural hazard could benefit the candidates from the incumbent party, conditional on the implementation of postdisaster social policies. Indeed, the literature on the politics of natural disasters presents examples of both frameworks. In any case, this distinction between programmatic responses and clientelistic networks lies at the core of the literature on distributive politics (Golden and Min Reference Golden and Min2013).

The second component of our theory disentangles the role of social policy in this relationship. We argue that different postdisaster policies may have heterogeneous electoral impacts. These are a function of two features of a given government response: whether it provides public versus private goods and how effective it is to provide relief and address the consequences of the crisis. In terms of the first one of these characteristics, a large body of literature in political science, including case studies in Mexico, shows that citizens are responsive to particularistic goods that provide an immediate economic benefit, for example, cash handouts (Díaz-Cayeros, Estévez, and Magaloni Reference Díaz-Cayeros, Estévez and Magaloni2016; Hicken and Nathan Reference Hicken and Nathan2020).

In the context of natural disasters, Visconti (Reference Visconti2021) shows that exposure to these events makes individuals more likely to prioritize investments in housing (an individual good). Nonetheless, another equally relevant strand of research suggest that public goods represent an electorally effective strategy because they benefit a larger subset of the population (Flores and Smith Reference Flores and Smith2013).

It is important to mention that we do not have a strong position one way or another with respect to the electoral impacts of policies at either side of this range—that is, on whether more universalistic interventions are more or less effective from a political perspective than individualized transfers are.

The second feature of a disaster response that should affect its electoral impact is the extent to which the response effectively addresses the crisis. In the literature, the classical distinction is between policies that provide immediate relief and strategies to reduce vulnerability to future disasters. Although both are relevant, there is an increasing consensus that long-term adaptation measures are cost effective and should be prioritized as general prevention strategies. For example, Aldrich (Reference Aldrich2012) and Aldrich and Meyer (Reference Aldrich and Meyer2015) suggest that postdisaster actions that strengthen social capital and are implemented in close interaction with citizens are particularly beneficial to populations affected by a disaster.

Nonetheless, as mentioned before, some scholars show that voters prefer short-term postdisaster policies (Healy and Malhotra Reference Healy and Malhotra2009; Gailmard and Patty Reference Gailmard and Patty2019; Fox and Van Weelden Reference Fox and Van Weelden2015). As with the previous dimension, we do not have a strong position one way or another about the relative electoral impact of these two policy types for the case of Mexico, although we recognize that existing research tends to find that short-term responses are more electorally advantageous.

It is important to note that, although the two dimensions are related, they are conceptually distinct. Hence, governments can implement strategies along a two-dimensional space, defined by the extent to which a policy offers private versus public goods and addresses the short-term versus the long-term impacts of a disaster. In this article, we explore the role of two specific policies that Mexico City’s government implemented after the earthquake: actions to mitigate future risk for a small subset of the affected population and cash transfers that offer immediate housing relief to all affected citizens. Although these represent only a subset of all official responses to the crisis, we focus on these two for reasons of data availability. We do not have strong positions about which is more electorally advantageous.

A final consideration in our theory is the relevance of party labels. We argue that candidates—all down the ballot—who are from the party in charge of distributing risk reduction and housing-relief policies also benefit electorally from this policy implementation advantage. In the case of Mexico City, the office of the governor (jefatura de gobierno) controls the vast majority of public spending. We posit that candidates for mayor (alcaldes) and deputies from the same political party of the governor are able to receive some benefits from the actions implemented at higher levels of government. Regardless of whether the allocation of these resources followed a need-based or a clientelistic logic, the observable implication is the same: exposure to a natural hazard benefits copartisans of the governor, conditional on the implementation of postdisaster policies.

In summary, our theory claims that large-scale and unexpected events associated with natural hazards, such as earthquakes, can affect electoral outcomes. The implementation of policies in response to the crisis is a key mechanism behind this relationship. From the policy supply side, candidates from the party in charge of distributing resources have an electoral advantage over challengers from other parties in the election following the emergency. From the policy demand dimension, citizens update their evaluations of political parties and candidates on the basis of the responses they implement; however, as mentioned earlier, we have no strong position as to whether any of these characteristics is more electorally advantageous; therefore, the analysis of the relative effectiveness of different postdisaster policies is mostly exploratory.

Finally, we emphasize that, in this article, we focus on the political dimension of the earthquake at the aggregate (electoral precinct) level. As mentioned earlier, some researchers have studied the individual-level consequences of exposure to natural disasters (Carlin, Love, and Zechmeister Reference Carlin, Love and Zechmeister2014; Visconti Reference Visconti2021, Reference Visconti2022). The following are the key observable implications of our theory:

The 19-S earthquake and the 2018 elections in Mexico City: Natural disasters and local politics

Mexico, especially the country’s capital, is severely vulnerable to different types of natural hazards. Seismic events, in particular, have affected the metropolis socially and politically throughout its history (Monsiváis Reference Monsiváis2005). The 1985 earthquake, for example, contributed to the transformation of the disaster prevention policy in Mexico (Estrada Díaz Reference Estrada Díaz2014), catalyzed the democratization process of the city, and exposed the incapacity of the autocratic Partido Revolucionario Institucional (PRI) to respond to a major emergency (Leal Martínez Reference Leal Martínez2014; Davis Reference Davis1990; Massolo Reference Massolo1986).

When the 19-S event hit, the city—as the rest of the country—was in the middle of one of the most consequential electoral periods since democratization. At both the federal and the state levels, unpopular incumbents (President Peña Nieto from the PRI and Mexico City’s Governor Mancera from the PRD) faced a serious challenge from MORENA, a political party created in 2014 by then presidential candidate—and former governor of Mexico City—Andrés Manuel López Obrador (AMLO). In its first election (2015), MORENA made substantial gains in the capital, seizing five boroughs from the PRD.

During most of the 2018 campaign, AMLO’s political party was the front runner and clear favorite to win the presidency, nine governorships (including Mexico City), and majorities in both chambers of the federal congress. After the election, MORENA became the predominant party in the city, whereas the PRD, which had been the main political party in the metropolis for two decades, practically disappeared from the political arena. The majority of the former’s victories in the 2018 election came from the latter’s losses.

Claudia Sheinbaum (MORENA), who was mayor of the Tlalpan borough from 2015 to 2018, won the governorship of Mexico City in the 2018 election by a landslide. At the municipal level, MORENA kept the five boroughs it won in 2015 and seized six more—eleven of sixteen. The alliance between the PRD and the Partido Acción Nacional (PAN) and the PRI retained four and one municipalities, respectively. In summary, the 2018 local election was a contest between a widely unpopular city incumbent (PRD) and a powerful rising challenger (MORENA), which resulted in a massive defeat of the former (see Figures S1 and S2 in the Online Appendix).

What was the role of the 19-S earthquake in the public debate during the electoral campaigns? In the aftermath of the crisis, the affected populations received support from the federal and the local governments. The National Fund for Natural Disasters provided resources for the reconstruction of critical infrastructure. ^{Footnote 2} However, most of its funding went to other states affected by this and a previous earthquake (the Online Appendix provides a detailed description of this response).

At the local level, the office of Mexico City’s governor implemented several policies to address the crisis. In this article, we focus on two of them for reasons of data availability. First, Mexico City’s Housing Institute (Instituto de Vivienda de la Ciudad de México) gave immediate rent relief—equivalent to US$200 per month—and housing reconstruction credits (US$700). Second, the Social Attorney’s Office (Procuraduría Social de la Ciudad de México) offered some services to address long-term risk of future seismic events: support for neighborhood organizations, inspection of buildings, and institutional capacity building. These policies required the strong participation of citizens (for a detailed description of these policies, see the Online Appendix). In both cases, the state-level incumbent (the governor, or jefe de gobierno) had discretion over the amount and allocation of resources.

Some of these actions, in particular the distribution of the reconstruction budget, were quite controversial. Candidates for governor from MORENA and PRI attacked the candidate from the PRD-PAN alliance for the overall response of Governor Mancera and the local assembly to the crisis. The opponents to the incumbent governor (along with civic organizations) criticized the disbursement of hundreds of millions of pesos in reconstruction efforts. According to the media, deputies from the PRD and the PAN were involved in serious malfeasance as leaders of the Commission for Reconstruction. ^{Footnote 3} The discretionary ability of Mexico City’s governor to spend emergency funds was at the core of the controversy.

At the municipal level, local candidates from the PRD-PAN alliance accused Sheinbaum of negligence in the enforcement of building codes, which led to the collapse of a private school in southern Mexico City. ^{Footnote 4} In summary, the 19-S earthquake happened during one of the most relevant political changes in Mexico since democratization. The main theme of the election was the threat that MORENA posed to the party system dominated by the PRI, PAN, and PRD. The earthquake exacerbated the divisions between the establishment and the challenger parties and became one of the central issues in the local and national campaigns.

Research design

To analyze the relationship between the damages associated with the 19-S earthquake and the electoral outcomes of the 2018 elections in Mexico City, we employ a quantitative approach at the electoral precinct level. Estimating the causal effects of natural disasters on political variables is challenging for a variety of reasons, in particular the nonrandom assignment of the treatment (Visconti Reference Visconti2021, Reference Visconti2022; Carlin, Love, and Zechmeister Reference Carlin, Love and Zechmeister2014). We do not claim to identify a causal effect; nonetheless, our empirical design accounts for some key geological and socioeconomic drivers of the treatment assignment process, thus allaying some of these concerns. Our approach has two main building blocks. Figure 1 provides an illustration of our argument.

Figure 1. Structure of the argument: direct and indirect effects of treatment (natural disaster) on outcome variable (electoral outcomes). Panel A describes the basic setup. The damage from a natural disaster affects electoral outcomes. Panel B considers the confounding role attributable to the relative risk associated to seismic zones. Panel C adds the mediating role of government relief efforts.

First, we do not assume that the impacts of events associated with natural hazards are randomly distributed. As the existing research shows, political decisions and institutional quality define the extent to which a hazard becomes a disaster (Flores and Smith Reference Flores and Smith2013; Anbarci, Escaleras, and Register Reference Anbarci, Escaleras and Register2005; Raschky Reference Raschky2008; Cohen and Werker Reference Cohen and Werker2008). In the case of earthquakes and other geological phenomena, although they are practically impossible to predict, it is much more straightforward to identify what regions will be more affected when they hit.

The second dimension of our empirical strategy consists in identifying the set of variables that account for most of the variation in the treatment assignment. The literature on disaster management and adaptation to other natural hazards (e.g., climate change) offers good theoretical guidance. According to the Intergovernmental Panel on Climate Change, the vulnerability of a household, a community, or a polity to natural hazards depends crucially on their exposure and sensitivity to that event. The first term refers to the extent to which an actor or system is located in the area of influence of a hazard, for example along the path of a hurricane or near the epicenter of an earthquake. The second variable denotes how resilient is that actor or system to the effects of said event (Cardona et al. Reference Cardona, van Aalst, Birkmann, Fordham, McGregor, Pérez, Pulwarty, Schipper and Tan Sinh2012, 72). Hence, households or communities with equal exposure to a hazard but different levels of sensitivity have varying degrees of vulnerability.

We argue that, for the case of Mexico City, besides socioeconomic factors, two other characteristics that have an impact on the vulnerability of neighborhoods to earthquakes are the geological exposure of a precinct to that type of disaster and the average year of building construction for said neighborhood. Whereas the former approximates the geophysical exposure of an actor or system to a seismic event, the latter is an appropriate measure of how sensitive that actor or system is to an earthquake.

The first one of these variables is the geological location of an electoral precinct. Much of contemporary Mexico City was built on the former Lake Texcoco bed, which explains in part the high vulnerability of the metropolis to earthquakes. Softer soils, such as those on the former lake, are associated with a faster expansion of seismic waves and, therefore, more vulnerability to disasters. In contrast, more solid and rocky soils are correlated with a slower expansion of seismic waves. There is a clear correspondence between being located on the former lake and having greater seismic intensity. Figure 2 shows the geographic distribution of this variable across Mexico City. Darker shades of gray indicate more seismic intensity and lighter shades of gray denote less. Areas shaded in white have the lowest levels of intensity.

Figure 2. Different seismic zones in Mexico City (darker shades of gray indicate more vulnerability to earthquakes).

The second variable is the average year of construction of the electoral precinct. As we mentioned earlier, the 1985 earthquake motivated crucial reforms in the building codes in Mexico City. Over time, these regulations have become more strict, at least on paper. Our assumption is that, on average, older buildings, particularly those built before the 1980s, are more vulnerable to severe damages during an earthquake than are more recently built ones. In contrast, newer housing units and commercial construction should be more resilient.

Although we cannot claim that these two variables alone make the treatment—the damages resulting from the earthquake—assigned as if random, we argue that, when controlling for them in our analysis, they reduce some of the endogeneity concerns associated with the treatment assignment process. See, for example, the clear correlation between geological zones and the distribution of damaged housing units in Figures 2–5. That is, we assume that for any two electoral precincts in the same seismic zone and with the same average building age, the distribution of damage after the earthquake should not be related to the main outcome variable (electoral performance), as they share both the same exposure (geology) and sensitivity (urban structure) to this type of natural hazard.

Figure 3. Geographic distribution of all damaged housing units after the September 2017 earthquake.

Figure 4. Geographic distribution of severely damaged housing units after the September 2017 earthquake.

Figure 5. Geographic distribution of damaged multifamily housing units after the September 2017 earthquake.

Data sources

In our empirical analysis, the main dependent variables are the electoral performance of the city incumbent (PRD) and of the main challenger (MORENA), at the state, municipal, and local legislative levels. We measure these variables as the change in the vote share for the candidates of each party for governor, mayor, and local deputies in 2018 compared to the immediate previous race. For the gubernatorial race, we analyze the change from 2012 to 2018 (elections are every six years); for the other elections, we calculate the change from 2015 to 2018 (elections are every three years).

Although the 2018 elections in Mexico City, as in the rest of the country, had multiple political parties competing, we focus on the performance of PRD and MORENA candidates at the electoral precinct level for two reasons. First, we are interested in explaining how the ability to distribute disaster relief affected local elections; since the PRD governor was in charge of the reconstruction process from September 2017 to July 2018 (when the election happened), we analyze the performance of candidates running with this party. Second, as mentioned, the rise of AMLO’s political party against unpopular incumbents at the federal and local levels was the defining feature of the 2018 elections. Hence, our main goal is to understand how the 19-S earthquake affected the electoral competition between a weak incumbent and a powerful new challenger. The data come from the Electoral Institute of Mexico City (IECM 2021).

The key independent variable in the article is exposure to damages caused by the 19-S earthquake at the precinct level in Mexico City, which we measure in two ways. First, we calculate number of damaged housing units, number of severely damaged housing units, and number of damaged multifamily housing units (known in Mexico City as multifamiliares) per capita in a given electoral precinct. Second, we compute the distance from the centroid of an electoral precinct to the closest damaged housing unit, the closest severely damaged housing unit, and the closest multifamily housing unit. For the latter measures, we follow an empirical approach similar to that of Enos (Reference Enos2016). These different variables allow us to take full advantage of the geospatial granularity of our data and estimate how different intensities of and geographic exposure to the earthquake damage affected electoral results.

We make three assumptions about the different damage variables. First, regarding the distance measures, we assume that the shorter the distance between a damaged housing unit and the centroid of a neighborhood, the greater is the exposure of said neighborhood to the effects of a natural disaster. Therefore, this measures how much damage was concentrated in a particular precinct and how visible it was for citizens there. Second, in terms of per capita variables, we assume that the larger the normalized number of damaged housing units, the more intense is the impact of the earthquake. Third, for both types of independent variables, we assume that the impacts of a damaged large multifamily unit are larger than the effects of severely damaged housing, which are, in turn, more intense than exposure to any damaged housing unit.

The damage data comes from Gobierno CDMX (2021b), which created a specific repository of data related to the earthquake. The unit of analysis of these data are the geocoded addresses of more than 11,500 damaged housing units, which we aggregated to the precinct level (for the geographic distribution of the 19-S earthquake, see Figures 2, 3, and 4).

Our empirical strategy conditions exposure to damage on the geological characteristics of the precinct and its average year of construction, which together form the seismic profile of a given neighborhood. For the first variable, we gathered georeferenced data from the National Center for Disaster Prevention (CENAPRED 2021). For each electoral precinct, we calculated the percentage of its area overlapping each seismic intensity zone (see Figure 2). When the precinct extends over two or more such zones, we calculated the weighted average of both. For the second variable, we employed data at the building level from the Urban Land Registry, or Catastro Público de la Propiedad (Gobierno CDMX 2021a); we summarized this data at the precinct level to obtain an average year of construction, which we use to calculate the average building age.

We test the mechanisms of our theory with two previously untapped public-spending data sets. The first one includes geocoded information of the distribution of a program that provided immediate housing relief in the affected areas. This policy, administered by Mexico City’s Housing Institute (Instituto de Vivienda de la Ciudad de México), consisted of a credit for housing reconstruction (INVI 2019). The second data set has geocoded information of a program administered by the Social Attorney’s Office of Mexico City (Procuraduría Social de la Ciudad de México). Although this policy also was meant to provide relief to affected neighborhoods, its components were quite different, as they focused on fostering the organization of neighbors, providing technical assessments of buildings, forming neighbors’ assemblies, and training specialists in risk reduction and civil protection, among other issues (PROSOC 2019). In both cases, we divide the number of benefits per one thousand inhabitants to account for different population sizes across electoral precincts. We obtained the data through an official information request.

Finally, our models also include a wide array of control variables. First, one of the most relevant features of the 2018 election was the success of MORENA at the federal, state, and municipal levels. It has been argued that the coattails of the presidential election explain much of the gains of MORENA candidates for other types of office (Johnson and Cantú Reference Johnson and Cantú2020; Garrido and Freidenberg Reference Garrido and Freidenberg2020). Therefore, we control for the levels of support for MORENA’s candidate in the presidential elections, as this could be an important confounder in the relationship between damage and local electoral returns. The data come from the National Electoral Institute. Second, socioeconomic variables are closely associated with political behavior and exposure to natural disasters in Mexico. To account for the potential effects of these variables, our models include illiteracy rates, percentage unemployment, and primary health-care coverage; the data comes from the national 2010 census—the most up to date at the time of the 19-S earthquake (INEGI 2010).

Empirical findings

Exposure to earthquake damage and electoral performance

We present two sets of empirical results. First, we explore the relationship between exposure to the earthquake damage and the electoral performance of candidates from two political parties (the city-level incumbent, PRD, and the rising challenger, MORENA), which addresses Hypothesis 1. Second, we study the statistical association between the distribution of two types of disaster relief and the vote, conditional on the levels of damage, the seismic profile, and the socioeconomic characteristics of the precinct, which addresses Hypotheses 2 and 3. Figures 6–8 show the statistical relationship between the different damage variables and the performance of the PRD and MORENA candidates in the gubernatorial (Figure 6), mayoral (Figure 7), and legislative (Figure 8) elections in the city. For mayors and deputies, we compare the two parties. For the gubernatorial race we include only the PRD, as MORENA did not compete in 2012.

Figure 6. Relationship between damage associated with the earthquake and the electoral performance of the PRD candidate for governor. The outcome variable is the change in the vote from 2012 to 2018. The coefficients in the plot correspond to the distance-based and per capita measures of damage associated with the earthquake. Major building refers to large multifamily housing, high-risk house refers to houses with severe damage, and any damaged housing includes all damaged units. All models include seismic, socioeconomic, and political covariates; municipality fixed effects; and heteroskedasticity-robust standard errors.

Figure 7. Relationship between damage associated with the earthquake and electoral performance of the PRD and MORENA candidates for mayor. The outcome variable is the change in the vote from 2015 to 2018. The coefficients in the plot correspond to the distance-based and per capita measures of damage associated with the earthquake. Major building refers to large multifamily housing, high-risk house refers to houses with severe damage, and any damaged housing includes all damaged units. All the models include seismic, socioeconomic, and political covariates; municipality fixed effects; and heteroskedasticity-robust standard errors.

Figure 8. Relationship between damage associated with the earthquake and the electoral performance of the PRD and MORENA candidates for deputy. The outcome variable is the change in the vote from 2015 to 2018. The coefficients in the plot correspond to the distance-based and per capita measures of damage associated with the earthquake. Major building refers to large multifamily housing, high-risk house refers to houses with severe damage, and any damaged housing includes all damaged units. All models include seismic, socioeconomic, and political covariates; municipality fixed effects; and heteroskedasticity-robust standard errors.

We highlight three findings. First, there is a strong and positive association between exposure (distance) to damage and damage per capita and the electoral performance of the city-level incumbent PRD. Starting with distance-based measures, the top-left panels in Figures 7 and 8 and the top panel in Figure 6 show that as the distance from the precinct to a damaged housing unit increases, the vote for the PRD decreases, suggesting that proximity to the damage was beneficial for the PRD candidates. In terms of the different levels of damage, we find that proximity to any damage housing unit had the largest coefficient, whereas proximity to a large multifamily home was the smallest, though statistically significant.

The situation is somewhat different when we measure damage in terms of magnitude: although the number of damaged large multifamily buildings per thousand inhabitants is positively associated with vote for the PRD (first models in top-right panel of Figures 7 and 8), damage per capita measured as both total number of housing units and high-risk housing units is not. Moreover, the signs and magnitudes of the coefficients are remarkably similar for the gubernatorial, mayoral, and legislative candidates, suggesting that Mexico City inhabitants applied similar assessments to politicians from the same party across different levels of office. The one exception is the model using the number of damaged multifamily units per capita for the PRD gubernatorial candidate. This is the only instance in our analysis in which the damage caused by the earthquake is associated with a lower vote share for the PRD candidate in 2018 than in 2012.

Regardless of the statistical significance and sign, the coefficients are relatively small from a substantive point of view. For example, for the mayoral race, a one standard deviation change in the distance to any damaged housing unit is associated with 0.9 percent better performance for the PRD candidate. Although in competitive elections this could make the difference between winning and losing a race, in the context of the decisive victory of MORENA, it was relatively minor. The same occurs when we measure damage in per capita terms. A one standard deviation change in the levels of large-scale damage per capita (measured as number of damaged multifamily housing units) is associated with approximately 0.26 percent higher vote share for the PRD mayoral candidates; the magnitudes are similar for legislative candidates (0.2 percent).

The second finding we highlight is that the impact of the earthquake seemed to have been mostly concentrated among PRD candidates. Although there is some evidence that places with more exposure to it punished the candidates from the rising challenger (MORENA) at the polls, the findings are less consistent across different races and ways to measure damage. Hence, we cannot conclude that the 19-S earthquake damaged the electoral performance of MORENA across Mexico City.

In general, the signs of the damage coefficients for the models explaining the electoral performance of MORENA candidates are the opposite from the PRD. For example, the models using a distance-based measure of damage (bottom-left plots in Figures 7 and 8) show that MORENA underperformed in precincts closer to damaged housing units. In contrast to the findings for the PRD, only the broadest distance measure of damage (to any housing unit) is associated with the MORENA vote for mayors and deputies. Regarding the per capita measures of damage, there is some evidence that places with more per capita exposure to damaged multifamily housing units punished MORENA candidates for mayor (bottom-right plot in Figure 7). This finding is consistent with the conventional wisdom claiming that some MORENA mayors, in particular then soon-to-be governor Claudia Sheinbaum, had somewhat controversial responses to the emergency.

As in the case of the PRD, the coefficients corresponding to the different damage measures are relatively minor given the enormous electoral success of the party. For example, for the mayoral race, a one standard deviation increase in the number of high-profile damaged buildings per capita (multifamily units) is associated with only a 0.20 percent decrease in vote share for MORENA candidates. Similarly, a one standard deviation increase in distance from any damaged housing unit is associated with a 0.48 percent decrease in vote share for the party. In many competitive elections, similar effects would be enough to flip one district from one party to another; however, as mentioned, the political landscape of Mexico City by the time of the 2018 election meant that the party was the clearly favorite to win the gubernatorial, mayoral, and legislative races across the country.

To summarize, the empirical analysis of our article provides some strong support for our first hypothesis. We find that exposure to the damage from the 19-S earthquake had a mostly pro-incumbent effect during the 2018 local elections. Across the different elections (mostly for mayors and deputies) and different ways to measure damage, places more affected by the earthquake showed a stronger support for the PRD candidates in 2018 compared to the previous election—with the notable exception of the candidate for governor.

In contrast, although there is some evidence that voters punished MORENA candidates in more affected places, particularly among mayoral candidates, the findings are somewhat inconsistent across different races and ways to measure damage. Regardless of the specific party, office, or damage variable, the overall effects of the earthquake were surprisingly small, given the salience of the topic in the media and among the electorate (Tables S1–S10 in the Online Appendix present the full results of the analysis for Figures 6–8).

The role of post-disaster social policy

Regarding our second hypothesis (about the role of different disaster-relief policies), we highlight two main findings. First, we show that risk reduction and rent-relief policies had slightly different effects on electoral performance of the PRD and MORENA. Figure 9 presents the results associated with Equations 3 and 4; the main independent variable is distribution of disaster relief per capita, conditional on the levels of damage, the seismic profile and the socioeconomic characteristics of the precinct. Both types of social policy had a pro-incumbent effect: PRD candidates for office at the state, local, and legislative levels had better electoral performances in precincts with more aid per capita.

Figure 9. Relationship between damage associated with the earthquake, the electoral performance of the PRD and MORENA candidates, and the distribution of disaster-relief policies. The outcome variable is the change in the vote from 2018 to the previous election. The coefficients in the plot correspond to two different social policies implemented by the local government to address the crisis. Risk reduction refers to the actions implemented by the Social Attorney’s Office to reduce risk to future natural disasters. Reconstruction credit refers to the cash transfer from the local Institute for Housing to rebuild damaged housing units. All models include seismic, socioeconomic, and political covariates; municipality fixed effects; and heteroskedasticity-robust standard errors.

The distribution of risk reduction and rent-relief policies was associated with statistically significant decreases in the vote share for MORENA candidates for mayor; in contrast, the electoral performance of candidates for deputies was unrelated to the levels of aid per capita after the earthquake. These results (for PRD and MORENA) are consistent with the previous finding that the city-level incumbent outperformed in places with more earthquake damage and, to a lesser extent, that the rising challenger underperformed in the same type of precincts. Hence, our results suggest a close relationship between earthquake damage, social policy, and the vote. There are three additional points worth emphasizing. First, even after controlling for levels of aid, exposure to the earthquake damage is still statistically significant, suggesting that part, but not all, of the pro-incumbent effect we find could be attributed to the distribution of relief.

Second, as with the damage measurements, the substantive effects of the social policy variables are rather small. For example, a one standard deviation increase in the amount of risk reduction aid per capita is associated with 0.26 percent additional votes for the PRD mayoral candidates from 2015 to 2018. The relevance of the rent relief policies is even smaller. Third, in contrast to conventional wisdom, a rent-relief policy did not outperform a risk reduction strategy that did not rely on direct cash transfers. This suggests that, in the context of natural disasters, voters may also value policies that provide nonmonetary benefits (Tables S11 and S12 in the Online Appendix present the full set of results corresponding to Figure 9).

Robustness checks

We implemented a series of robustness checks that increase our confidence in these findings. First, we employ different matching algorithms to reduce some of the imbalances between treated and control units. Other scholars have employed a similar approach to study the politics of natural disasters in Latin America (Carlin, Love, and Zechmeister Reference Carlin, Love and Zechmeister2014). We modeled the probability of treatment—defined as having one or more damaged housing unit—as a function of the seismic characteristics of a precinct, socioeconomic indicators, and the availability of public services. This procedure allows us to significantly reduce any remaining imbalance between damaged and nondamaged electoral precincts (see Figures S7–S12 in the Online Appendix). We then estimated a linear regression with electoral performance variables as the outcomes and damage variables as the main independent regressors (using different versions of damage in per capita terms), with nearest neighbor and optimal matching. The conclusions of the previous analysis held: precincts with higher levels of damage had higher vote shares for the incumbent PRD, and the opposite occurs for MORENA candidates.

A main concern of our empirical strategy is the presence of omitted variables that could bias our findings. Although we control for key factors that influence electoral performance of the PRD and MORENA, it is likely that we are missing other important variables. To address this possibility, we perform a sensitivity analysis of our results to potential confounders, as developed by Cinelli and Hazlett (Reference Cinelli and Hazlett2020).

Their methodology allows for researchers to determine how sensitive an observational result is to the presence of confounders. In particular, it determines how much of the variance in the independent and dependent variables would need to be explained by a confounder to make the coefficient of interest statistically indistinguishable from 0. Another attribute of the methodology is that it enables researchers to compare the strength of hypothetical confounders with variables known to the researcher (Cinelli and Hazlett Reference Cinelli and Hazlett2020).

For example, in Table 1 we present the summary of the sensitivity analysis the model using a distance-based measure for PRD deputies; a confounder would need to explain 12.5 percent of the variance in the outcome and independent variable to reduce the statistical significance of the damage coefficient to 0. A matter of qualitative judgment is how big 12.5 percent is in explaining the main dependent and independent variables in the analysis. As Table 1 shows, the confounder would need to be almost as strong as the vote for AMLO to change our conclusions about the damage variable. Tables 2–6 present the sensitivity analysis for the rest of our main results. It is important to notice that, although the per capita measures are more sensitive to the presence of confounders than the distance-based measures, they are still relatively strong given that we are comparing to one of the most important drivers of the 2018 local elections. From both substantive and theoretical points of view, it is highly unlikely that any single confounder (or even a set of omitted variables) is as strong as the vote for AMLO to explain the performance of both the PRD and MORENA.

Table 1. Sensitivity analysis for main results (deputies/distance-based damage).

Table 2. Sensitivity analysis for main results (deputies/per capita damage).

Table 3. Sensitivity analysis for main results (mayors/distance-based damage).

Table 4. Sensitivity analysis for main results (mayors/per capita damage).

Table 5. Sensitivity analysis for main results (governor/distance-based damage).

Table 6. Sensitivity analysis for main results (governor/per capita damage).

Finally, the last robustness check addressed the geospatial structure of the data. After confirming that the standard errors are spatially autocorrelated, we run a model that corrects for this feature of our data. The results are consistent with our main analysis; however, the magnitude of the coefficients is smaller (Tables S13–S20 in the Online Appendix present the results associated with these models).