How COVID-19 affects voting for incumbents: Evidence from local elections in France

Morisi, Davide; Cloléry, Héloïse; Kon Kam King, Guillaume; Schaub, Max Article — Published Version How COVID-19 affects voting for incumbents: Evidence from local elections in France PLOS ONE Provided in Cooperation with: WZB Berlin Social Science Center Suggested Citation: Morisi, Davide; Cloléry, Héloïse; Kon Kam King, Guillaume; Schaub, Max (2024) : How COVID-19 affects voting for incumbents: Evidence from local elections in France, PLOS ONE, ISSN 1932-6203, Public Library of Science (PLoS), San Francisco, CA, Vol. 19, Iss. 3, pp. 1-19, https://doi.org/10.1371/journal.pone.0297432 This Version is available at: https://hdl.handle.net/10419/312942 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by/4.0/ RESEARCH ARTICLE How COVID-19 affects voting for incumbents: Evidence from local elections in France Davide MorisiID 1,2 , He ´loïse Clole ´ryID 3 , Guillaume Kon Kam King 3,4 , Max SchaubID 5,6 * 1Department of Political Science and Public Management, University of Southern Denmark, Odense, Denmark, 2Collegio Carlo Alberto, Turin, Italy, 3CREST, E ´cole Polytechnique, IP Paris, Palaiseau, France, 4INRAE, Universite ´Paris-Saclay, Jouy-en-Josas, France, 5Department of Political Science, University of Hamburg, Hamburg, Germany, 6WZB Berlin Social Science Center, Berlin, Germany *[email protected] Abstract How do voters react to an ongoing natural threat? Do voters sanction or reward incumbents even when incumbents cannot be held accountable because an unforeseeable natural disaster is unfolding? We address this question by investigating voters’ reactions to the early spread of COVID-19 in the 2020 French municipal elections. Using a novel, finegrained measure of the circulation of the virus based on excess-mortality data, we find that support for incumbents increased in areas that were particularly hard hit by the virus. Incumbents from both left and right gained votes in areas more strongly affected by COVID-19. We provide suggestive evidence for two mechanisms that can explain our findings: an emotional channel related to feelings of fear and anxiety, and a prospective-voting channel, related to the ability of incumbents to act more swiftly against the diffusion of the virus than challengers. 1. Introduction Natural disasters influence voting. After catastrophic events—such as earthquakes, floods, and hurricanes—voters often reward political candidates and incumbents [1–5] or punish them electorally [6–11], depending on how they responded to such events. Virtually all available evidence focuses on how voters retrospectively evaluate politicians’ performance after natural disasters have occurred [12], and whether these evaluations influence voting. However, we do not know how voters respond to a threatening event while it is unfolding. How do voters react to an ongoing natural threat? Do voters sanction or reward incumbents even when incumbents cannot be held accountable because an unforeseeable natural disaster is still unfolding? We address this question by investigating voters’ reactions to the early spread of COVID-19 in France. COVID-19 differs from other natural disasters since it is a prolonged threat that does not result in physical destruction but spreads invisibly, potentially affecting all human beings. Like other infectious diseases [13], COVID-19 generates fear, which, we argue, increases voters’ tendency to support incumbent candidates as a means to obtain safety and security. This dynamic applied especially at the onset of the pandemic when there was uncertainty regarding the best strategy for containing the virus and guaranteeing citizens’ safety. PLOS ONE PLOS ONE | https://doi.org/10.1371/journal.pone.0297432 March 19, 2024 1 / 19 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Morisi D, Clole ´ry H, Kon Kam King G, Schaub M (2024) How COVID-19 affects voting for incumbents: Evidence from local elections in France. PLoS ONE 19(3): e0297432. https://doi. org/10.1371/journal.pone.0297432 Editor: Eugenio Proto, University of Glasgow, UNITED KINGDOM Received: June 17, 2023 Accepted: January 4, 2024 Published: March 19, 2024 Copyright: ©2024 Morisi et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All replication material and code have been published in a public repository, here: https://doi.org/10.7910/DVN/ EFDGLY. Funding: DM received a special COVID-19 grant by Collegio Carlo Alberto (Turin, Italy) for conducting the research included in this manuscript. The sponsor did not play any role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The publication of this article was supported by the Open Access Publication Fund of the University of Hamburg. We focus on the municipal elections that took place in France on 15 March 2020, when SARSCov-2, the virus causing COVID-19, was already circulating in the country. To estimate the “threat” of COVID-19, we developed a novel, fine-grained measure of the spread of the virus (the “prevalence” of COVID-19) at the municipality level, using excess-mortality data weighed by the age and sex structure of the population. This measure accounts for the Infection Fatality Ratio [14] and allows us to address the problem of underreporting with regard to the circulation of the virus at the beginning of the pandemic [15]. Even if the total number of confirmed cases of COVID-19 in France on 15 March 2020 was relatively low—a total of 4532 [16]—our estimates indicate that a substantial share of French voters already knew of other people who suffered from COVID-19 in their local area. Specifically, we estimate that there was a small but sizeable group of municipalities (7 percent of the total), in which a voter had at least a 30 percent probability of knowing at least another person with COVID-19 symptoms in their inner circle of friends and acquaintances. Even if the proportion of people with COVID19 symptoms in a given municipality is low, the probability that a random voter knows at least one person with COVID-19 symptoms can be large (as further explained below). We exploit the uneven diffusion of the virus in the country to test whether support for local incumbents increased in municipalities that were particularly hard hit by COVID-19. In particular, we estimate the effect of being severely affected using a) regression models that control for the lagged dependent variable and condition on baseline mortality rates, b) a difference-indifferences approach, and c) propensity score matching. In line with evidence for a “rallyaround-the-flag” effect at the beginning of the pandemic [17–22], our findings show that vote shares for local incumbents increased as the threat of COVID-19 increased. In municipalities with a high circulation of the virus (tenth decile), the average vote share for incumbents was 2.5 percentage points higher than in places with a low circulation (first decile). The effects are insensitive to the political affiliation of the incumbents, meaning that support for incumbents increases regardless of whether they are left-wing, right-wing, or from other parties. We also show that the effect of COVID-19 on voting is more pronounced in areas a) where people feel particularly anxious about the pandemic, and b) where economic welfare improved during the outgoing administration. This evidence supports our argument that, when faced with the circulation of the virus, voters, on the one hand, turned to incumbents as a means of reducing fear and anxiety–an “emotional channel” in line with evidence from the United States [23,24]. In addition, following the logic of prospective voting, we argue that voters support incumbents to maintain continuity with the current administration as the most efficient means of tackling the virus, especially in areas where the incumbent mayors had performed well. In the next section, we review existing research on the effects of natural disasters on voting, and advance our arguments for why the early spread of COVID-19 should increase support for local incumbents. We then provide some brief background information on the 2020 municipal elections in France, describe our data and methods, and present the results. In the conclusion, we summarize our findings and discuss the implications and the limitations of our study. 2 Natural threats and support for incumbents Research shows that voters punish or credit political candidates depending on how these candidates respond to severe weather phenomena. For example, studies indicate that incumbent politicians lose support if their response to natural disasters is perceived as inadequate [6,8, 11,25]. Voters also punish incumbents in the aftermath of severe weather events that cause economic damage, such as tornadoes, floods and extreme rainfall [4,7,9,10], and even in the PLOS ONE How COVID-19 affects voting for incumbents PLOS ONE | https://doi.org/10.1371/journal.pone.0297432 March 19, 2024 2 / 19 Competing interests: NO authors have competing interests. case of events that are clearly out of politicians’ control, such as shark attacks [26]. On the other hand, there is evidence that voters reward governments and incumbent politicians who are seen as having responded well to natural disasters [1–3,5,12]. The primary mechanism describing voters’ reaction after a disaster has occurred is retrospective voting [27]: on election day, voters sanction or reward politicians depending on how well they responded to the damage caused by the natural catastrophe. At the beginning of the COVID-19 pandemic, however, it was unclear how to best respond to the circulation of the virus. As an invisible threat that does not lead to physical destruction and persists for an indefinite period, COVID-19 cannot be tackled with the same means used after a natural catastrophe. Relief-spending and the types of reconstruction undertaken after natural disasters were not viable options for incumbent politicians at the beginning of the pandemic. This is especially true for local mayors who lacked the legal authority to introduce strong containment measures, such as lockdowns. Thus, the initial spread of COVID-19 should not have affected voting for local incumbents through retrospective evaluations since it was unclear how politicians ought to have responded to the pandemic. In the absence of retrospective evaluations, threatening events can still induce people to express greater support for incumbent political leaders via emotional reactions, the so-called ‘rally-aroundthe-flag’ effect [28–30]. This effect has been identified in relation to terrorist attacks (see for example [31–35]) and military crises [28], both of which have been found to lead to a boost in presidential popularity and government support. Recent studies have argued that the threat of COVID-19 should have similar rally effects [21], at least at the beginning of the pandemic. Indeed, evidence indicates that support for world leaders [17–19,22] and trust in governments [20,36–38] increased at the beginning of the COVID-19 pandemic. We do not know, however, whether approval ratings actually translated into voting behaviour (but see [39]), and it is unclear what mechanisms could connect the spread of the virus to voting behaviour. We argue that the early spread of COVID-19 should lead voters to increase support for local incumbents mainly through two channels. Similar to rally effects, the first channel is emotional, although involving a different type of emotion: anxiety (instead of anger or patriotic arousal). It has been argued that rally effects are activated mostly by a sentiment of patriotism and by anger towards a clearly identifiable external aggressor, such as a terrorist group [29,40]. The threat of COVID-19, however, should not activate the same feelings of anger and patriotism, since there is no deliberate attack behind the circulation of the virus (i.e., the virus did not “intend” to attack a population, as terrorists do), and there is no clearly identifiable human enemy against whom the population should mobilize. Instead, we argue that COVID-19 triggers fear and anxiety, as indicated by psychological studies [41,42] and by recent evidence in the U.S. [23] (see also [21]). As with the Ebola virus when it reached the US in 2014 [13], at the beginning of the pandemic, COVID-19 represented an invisible threat against which there was no clear remedy, thus potentially inducing fear in the population. Different strands of theories indicate that sentiments of fear and anxiety lead to risk-aversion (for a review, see [43]), which, in turn, should translate into increased support for incumbent candidates, as they represent a safer, status quo option compared to the risk of electing an as-yet-untested challenger. Furthermore, according to terror-management [44] and motivated-social-cognition theory [45,46], people want to see the world as a secure place. Therefore, in the presence of a fearinducing threat such as COVID-19, voters should turn their support towards incumbent politicians who offer an actual or symbolic sense of safety and security because they represent the status quo. Indeed, evidence from the United States indicates that the spread of COVID-19 PLOS ONE How COVID-19 affects voting for incumbents PLOS ONE | https://doi.org/10.1371/journal.pone.0297432 March 19, 2024 3 / 19 and lockdown measures at the beginning of the pandemic increased support for candidates representing the “status quo” [24]. Although these different theoretical backgrounds speak in favour of a positive association between the circulation of COVID-19 and support for incumbent candidates, a negative association can also be conceivable for two different reasons. First, given accumulated evidence that “fear, threat, and anxiety often contribute to right-wing extremism” [47], it is possible that the spread of COVID-19 shifts voters away from incumbents, if incumbents are predominantly not extremists. Although this theoretical possibility is plausible, it seems highly unlikely to occur in the context of local elections in France, where the vote share for extreme rightwing candidates is negligible. Second, it is conceivable that where the virus circulates more, it is mostly risk takers who go to vote, because risk-averse voters decide to stay home in the first place. Thus, in these areas we should observe a decline in both turnout and vote for incumbents. Although this theoretical channel is also plausible, it rests on the assumption that risk takers are particularly likely to choose challengers over incumbents. However, it is also possible that risk takers prefer incumbents over challengers for purely instrumental reasons related to the prospective channel described below, in line with evidence that instrumental risk taking is linked to more rational decisions [48] and that risk takers “tend to decide more strictly on the basis of cost-benefit considerations” [49]. The discussion of these alternative scenarios leads to the second channel related to a simple version of prospective voting, that is, voting on the basis of how candidates will perform in the upcoming administration. This channel is based on the simple consideration that during a crisis “it is not a good time for a change.” In line with the idea of prospective voting [39,50,51], voters compare how challengers versus incumbents will perform in relation to containing the virus. Anticipating that the threat of COVID-19 is going to last, and given that virtually no candidate has a track record in dealing with a pandemic, voters might consider that incumbents will perform better simply because they have had several years of experience with the local administration and can swiftly mobilize all the available local resources. For instance, French mayors had to cooperate with existing local actors (schools, local NGOs, and volunteers) during the pandemic to share information, distribute masks, or coordinate with other administrations. Voters could interpret the politicians’ ability to work with local actors as a signal of better future management of the crisis. Because they already knew these local actors, the incumbents probably had an advantage over the other candidates. For these reasons, in a scenario of a looming threat, voters might prefer to maintain continuity with the current administration as the most efficient means of tackling the virus and restoring safety. On the other hand, choosing a challenger over the incumbent might delay the necessary response to the pandemic due to the period of transition to the new administration. Furthermore, voters might simply consider that in a situation of high uncertainty produced by the circulation of the virus, electing a challenger would add further uncertainty regarding how the local administration will deal with the pandemic. Thus, they might prefer incumbents as a safer way of preserving public safety. Both the emotional channel and the prospective-voting channel lead us to the expectation that support for incumbents should increase in areas where the spread of COVID-19 was higher at the early stage of the pandemic. According to these theoretical channels, support for incumbents should increase regardless of their political affiliation, as voters threatened by COVID-19 turn to incumbents mainly because they represent the status quo. This said, previous studies indicate that it is mostly right-wing incumbents that gain from military threats because their aggressive defense strategies are seen as more in line with citizens’ need for protection compared to other parties’ strategies [52,53]. Although the spread of COVID-19 differs PLOS ONE How COVID-19 affects voting for incumbents PLOS ONE | https://doi.org/10.1371/journal.pone.0297432 March 19, 2024 4 / 19 from military threats, if right-wing incumbents are better suited to reducing anxiety and addressing citizens’ need for safety and security, we might expect the effect to be strongest for right-wing incumbents. This expectation is also in line with cited evidence that fear and anxiety contribute to a shift towards right-wing political attitudes (for a review, see [47]). 3 Municipal elections in France On March 15, 2020 French voters went to the polls to choose the mayors of all of the nearly 35,000 municipalities in the country. Municipal elections in France are held every six years, and, depending on the size of the municipality, consist of one or up to two rounds. In small municipalities with under 1,000 inhabitants, voters choose lists of candidates and are allowed to cast a vote for more than a single candidate in a majoritarian system. In addition, the candidates in towns with less than 1000 inhabitants are not required to declare any political affiliation. In larger towns and cities with more than 1,000 inhabitants, elections consist of one or two rounds. In the first round, all lists of candidates standing for the election compete against each other. In case no list obtains more than 50 percent of the vote, a run-off election is held in which all the lists that obtained more than 10 percent of the votes compete again for election. This second round is usually held two weeks after the first round. In 2020 the first round of the municipal elections was held as scheduled, despite a partial lockdown in some places. However, as the number of cases of COVID-19 continued to increase, the day after the first round of the elections, President Macron announced that the second round would be postponed to an unspecified date, which was later set to June 28, 2020. Compared to the previous local elections in 2014, turnout dropped substantially. While in 2014, 63.5 percent of eligible voters went to cast their ballot in the first round of the elections, on March 15, 2020 less than half of eligible voters (44.7 percent) went to vote. This drop in participation was likely due to the circulation of the virus, as confirmed by recent studies [54–56]. 4 Data and methods 4.1 Voting and census data We retrieve electoral data for the municipal elections in 2020 and 2014 from the Ministry of the Interior [57–59]. We consider only the municipalities that in both election rounds had more than 1000 inhabitants due to the differences in electoral rules explained above. We also exclude overseas territories and the cities of Paris, Marseille and Lyon, since the voting system also differs in these areas. After excluding these observations, we are left with a sample of 8,193 municipalities. To determine which candidates running in 2020 were incumbents, we matched the names of those who were elected mayor in 2014 with the names of the 2020 candidates. This operation gives us our effective sample of 4,952 municipalities (around 56%) in which an incumbent was running in 2020. Balance statistics indicate that municipalities with and without incumbents are statistically indistinguishable across all our indicators, apart from population density and turnout in 2014, as municipalities with an incumbent are more densely populated and have a higher turnout in 2014 than municipalities without an incumbent (see S1 Table in S1 File). Our outcome of interest is the share of votes for the incumbents in the first round of the municipal elections in 2020. Crucially, in all regression models we include the vote share for the same candidates in 2014, which allows us to estimate whether the spread of COVID-19 influences the change in vote shares for incumbents in 2020 compared to 2014. We also retrieve information about the political affiliation of the incumbent candidates to test for heterogeneous effects depending on the incumbents’ political leaning. In 2014, all candidates competing in municipalities above 1000 inhabitants had to indicate their political PLOS ONE How COVID-19 affects voting for incumbents PLOS ONE | https://doi.org/10.1371/journal.pone.0297432 March 19, 2024 5 / 19 affiliation, which was then classified by the Ministry of the Interior. We use this information to distinguish between incumbents related to left-wing parties (22%), right-wing parties (47%), and a residual category including mostly candidates with no clear left-right affiliation and a few centrist candidates (31%). In S3 Table (S1 File) we provide detailed information about each political group. Furthermore, we collect census data at the municipality level from the French national institute for statistics for the year 2017 to control for key socio-economic indicators that might confound the relationship between the spread of COVID-19 and voting for incumbents [60]. Specifically, in each model we include measures of population density, male over female ratio, the share of people aged above 65, the share of immigrants, the share of blue-collar workers, the share of unemployed inhabitants, the median income of households in the municipality, and the share of people with a junior high school degree (“CAP” or “BEP” degree) and with a bachelor’s degree. We use the logged transformation for each of these variables (apart from the median income) in order to account for the presence of clear outliers. Furthermore, we control for the level of turnout in 2014 to account for imbalances in voters’ electoral participation at the municipality level, and for the number of candidates running in the first round of the elections. The latter variable is particularly important, since the incumbent vote share is partially determined by the level of electoral competition in a given municipality (with higher vote share where fewer candidates compete). A directed acyclic graph (DAG) justifying the choice of control variables is included in the supplementary S2 Fig in S1 File. For summary statistics, see S2 Table in S1 File. 4.2 Estimating the spread of COVID-19 Our core independent variable is a municipality’s degree of affectedness by COVID-19. As no such measure was readily available, we estimated it ourselves using standard epidemiological methods. The aim of this section is to provide an overview of how this measure is constructed (additional technical details are available in Appendix C (S1 File)). Specifically, we sought to construct a fine-grained measure of the spread of Covid-19 at the municipality level at the time of the election, which we denote Cov m,20 . This quantity is then used as an independent variable to predict the vote share received by incumbents, with the aim to measure the impact of COVID-19 on voting behaviour. Estimating the spread of COVID-19 at the municipality level is challenging because of scattered data availability and the low reliability of common measures. For instance, reported cases of COVID19 at the beginning of the pandemic were sensitive to differences in testing frequencies across the country and underestimated the circulation of the virus [15], while data about hospitalisations are not available at the municipality level since hospitals are located only in a few cities. To estimate the spread of COVID-19 at the municipality level, we therefore utilized exhaustive mortality records at the municipality level to estimate COVID-19 prevalence in each municipality at the time of the election. We took into account the differential probability of dying depending on age and sex (the Infection Fatality Ratio (IFR, [14])), which essentially weights the excess mortality data by the age and sex structure of the population at the municipality level. More precisely, we first computed the excess mortality (or excess hazard) at the municipality level by collecting all mortality records between 2015 and 2020 [61]. We estimated excess mortality around the time of the first round of elections by comparing mortality figures to those of the same period 2015–2019, thus avoiding issues of seasonal variability. Excess mortality has been presented as a reliable method to estimate the number of COVID-19 cases because mortality appears fairly stable over a period of a few years, making extreme mortality events PLOS ONE How COVID-19 affects voting for incumbents PLOS ONE | https://doi.org/10.1371/journal.pone.0297432 March 19, 2024 6 / 19 clearly visible in the form of peaks [62,63]. Yearly influenza outbreaks, for instance, are clearly identifiable in country-wide weekly mortality records (see S4 Fig in S1 File), while aggregated mortality rates seem reasonably stable over the last half-decade for the period considered (see S5 Fig in S1 File). Given that there is a delay between infection and death, we consider all deaths occurring between election day (March 15, 2020) to six weeks later, which should provide information about the number of infected at the time of the election. There is a risk of simultaneity bias: (i) the spread of the disease can affect the vote share for incumbents, but (ii) turnout—and thus vote shares for incumbents—itself can affect the spread of COVID-19 [64]. In a supplementary analysis, we consider a shorter time window of four weeks, less likely to be affected by the simultaneity issue. We obtain substantially similar results with this four-week window (see S7 Table in S1 File). Scholars have argued that excess death may be related indirectly to COVID-19 through the saturation of hospital facilities, interruption of preventative programmes or the impact of social restrictions on physical and mental health [62]. However, it seems unlikely that these factors played a strong role at the very beginning of the pandemic. Instead, researchers have argued that the probability of dying from COVID-19 once infected by the virus (Infection Fatality Ratio, IFR) is heavily determined by age and sex [14] and that excess mortality should be weighed by these factors if we intend to use it as a proxy for the circulation of the virus. In this sense, a 5% excess mortality in a relatively young city indicates that SARS-Cov-2, the virus causing COVID-19, is circulating more than in a relatively old city with the same level of excess death since young people are less likely to die from COVID-19 than elderly people. Therefore, we build a model to estimate COVID-19 prevalence (the proportion of the population infected by COVID-19) at the municipality level from mortality data and use it as a novel measure describing the spread of COVID-19, which weights excess death by age and sex-specific infection fatality ratios to estimate the number of cases more precisely. Specifically, we use a type of Poisson mortality model where the municipality-level baseline hazard over the 6 weeks period is considered constant between 2015 and 2019 and death counts in 2020 result from the baseline hazard plus an excess hazard due to COVID-19. Excess hazard is related to prevalence modulated by the age and sex-specific Infection Fatality Ratio from [14]. We obtain a municipality-level measure of prevalence, along with age and sex-specific baseline hazards for each municipality. Population data is obtained from Census data from 2010 to 2017 [65] and then either linearly or log-linearly extrapolated for 2018–2020 (see details in Appendix C.2 (S1 File)). To deal with the presence of outliers among the municipalities, in the regression models we use either a decile version (treated as a continuous variable) or a quartile version of our measure (to test for non-linear effects). In supplementary analyses, we show that substantially similar results are obtained using a logged transformation or a quintile version of our measure (see S4 and S5 Tables in S1 File). Although in the majority of French municipalities the SARS-Cov-2 virus was either not present or barely spreading on election day on 15 March 2020, we estimate that in a small but sizeable number of cities voters had a relatively high probability of knowing other people with COVID-19 symptoms. Assuming that the average voter knows around 15 people, and assuming homogeneity among voters, based on our prevalence measure we estimate that on 15 March 2020 there were 654 cities (7 percent of the total) and 255 cities (3 percent of the total) in which a person had a probability of at least 30 percent or 50 percent respectively of knowing at least one person with COVID-19 symptoms in their circle of friends and acquaintances. This is a very conservative estimate, considering a small social network of 15 people instead of the network of 150 people calculated by Dunbar [66], to account for the fact that not all COVID-19 infected people may experience symptoms or share that information in their social network. This stems from the fact that even if only few people suffer from COVID-19 in a PLOS ONE How COVID-19 affects voting for incumbents PLOS ONE | https://doi.org/10.1371/journal.pone.0297432 March 19, 2024 7 / 19 municipality, the probability that a random voter knows at least one person with COVID-19 symptoms can be large—a counter-intuitive finding in line with the so-called “birthday paradox” [67,68]. These basic estimates suggest that on election day there was already a substantial number of voters who were aware that the virus causing COVID-19 was circulating in their local area of residence. 4.3 Estimating the effect on incumbent voting In order to estimate the effect of COVID-19 on incumbent vote shares, we estimate OLS regression models in the form Im;d;20 ¼aþbCovm;20 þgXc;t1þZIm;d;14 þldþ�m;d;tð1Þ where I m,d,20 is the vote share received by the incumbent in a given municipality min 2020; Cov m,20 measures the spread of COVID-19 in a given municipality in 2020; X c,t−1 is a vector including the municipality-level census variables presented above (measured in 2017), in addition to the level of turnout in 2014 and a measure of baseline mortality. The baseline mortality is the municipality-specific probability of dying during the period prior to the onset of the pandemic. The construction of both Cov m,20 and the baseline mortality was discussed in the previous section. Controlling for the baseline mortality is especially important since it captures various contributions to mortality—both observed and unobserved—thereby adjusting the otherwise quite diverse set of municipalities in terms of expected future mortality. I m,d,14 is the incumbent’s vote share in the election 2014, and λ d are county (Département) fixed effects that remove time-invariant factors at this level from our estimations. Our parameter of interest is β. Since our models include the lagged dependent variable (i.e., vote for incumbents in 2014), βmeasures the change in incumbent support between the two elections related to the spread of COVID-19, all else being equal. We opted for this specification because it allows for more flexibility in the way the independent variable can be specified. Below we show that our results also hold in a difference-in-differences framework, when matching municipalities using propensity scores and a punishing caliper, and are robust to a placebo test. 5 Results Fig 1 presents the results of our regression models. Our model explains a large part of the variation in votes for incumbents in 2020 (R-squared = 0.59). Support for incumbents in 2020 is higher in municipalities with higher population density, higher share of people with a bachelor’s degree, lower share of people with a junior high school degree, lower share of people aged above 65, and lower median income of households. As expected, vote share for incumbents increases as electoral competition decreases, as captured by the number of candidates running in each municipality (for complete results, see S4 Table in S1 File). Despite the high explanatory power of the regression model, our measure of the spread of COVID-19 has a positive effect: vote shares for incumbents are substantially higher where COVID-19 is more widespread. The vertical bars in the left-hand plot of Fig 1 show the predicted vote share for incumbents in 2020 at each decile of COVID-19 prevalence, treated as a categorical predictor, holding all other factors constant. Despite variation, we observe a roughly linear increase moving from places with low to high circulation of COVID-19. If we treat COVID-19 prevalence as a linear predictor (the dotted line in the left-hand plot), we find that support for local incumbents is 2.5 percentage points higher in the municipalities with high circulation of the virus (tenth decile) than in the municipalities with low circulation (first decile)–an increase that is statistically PLOS ONE How COVID-19 affects voting for incumbents PLOS ONE | https://doi.org/10.1371/journal.pone.0297432 March 19, 2024 8 / 19 we lack individual-level data that can address these mechanisms directly. We also call for further research into the nature of rally-around-the-flag effects. While our findings are in line with evidence of rally effects at the beginning of the pandemic [17–20,22,36–38], we argue that the threat of COVID-19 differs from the type of threats that, according to previous studies, lead people to rally around the flag, such as terrorist attacks and military crises. In particular, COVID-19 should not activate the same feelings of anger and revenge that have been theorized to explain rally effects [29], since the virus lacks the intention to attack a population. Further research should test whether facing a natural threat (such as COVID-19) or a human threat (such as a terrorist attack) is associated to different emotional states of fear or anger, which, in turn, can lead to different behavioural outcomes (for a review, see [43]). Lastly, we should not overstate the “power” of the incumbency status in influencing voting behaviour. Although the effects we detect can change the outcome of elections, they are likely to be short-lived, in line with research on the effects of terrorist attacks [31] and COVID-19 as well [76]. As the pandemic unfolds, a number of considerations probably influence voters’ decisions, including the evaluation of how incumbents and political representatives responded to the spread of the virus [77]. Indeed, evidence indicates that support for governments in Western democracies has decreased during the course of the pandemic [78]. In this sense, voters are not blind to the performance of their political representatives. After the initial emotional reaction to the spread of COVID-19, it is likely that other considerations shape voting decisions. Supporting information S1 File. Supporting information including additional analyses. (PDF) Acknowledgments We thank our three reviewers and conference participants at the Collegio Carlo Alberto and the European Political Science Association’s Annual Meeting 2021 for comments on earlier versions of this research. Author Contributions Conceptualization: Davide Morisi, He ´loïse Clole ´ry, Guillaume Kon Kam King, Max Schaub. Data curation: Davide Morisi, He ´loïse Clole ´ry, Guillaume Kon Kam King, Max Schaub. Formal analysis: Davide Morisi, He ´loïse Clole ´ry, Guillaume Kon Kam King, Max Schaub. Funding acquisition: Davide Morisi, Guillaume Kon Kam King, Max Schaub. Investigation: Davide Morisi. Methodology: Davide Morisi, He ´loïse Clole ´ry, Guillaume Kon Kam King, Max Schaub. Project administration: Davide Morisi. Resources: Davide Morisi. Writing – original draft: Davide Morisi, He ´loïse Clole ´ry, Guillaume Kon Kam King, Max Schaub. Writing – review & editing: Davide Morisi, He ´loïse Clole ´ry, Guillaume Kon Kam King, Max Schaub. PLOS ONE How COVID-19 affects voting for incumbents PLOS ONE | https://doi.org/10.1371/journal.pone.0297432 March 19, 2024 15 / 19 References 1. Bechtel MM, Hainmueller J. How Lasting Is Voter Gratitude? An Analysis of the Shortand Long-Term Electoral Returns to Beneficial Policy. American Journal of Political Science. 2011; 55(4):852–868. 2. Belloc M, Drago F, Galbiati R. Earthquakes, Religion, and Transition to Self-Government in Italian Cities. The Quarterly Journal of Economics. 2016; 131(4):1875–1926. 3. Healy A, Malhotra N. Myopic Voters and Natural Disaster Policy. American Political Science Review. 2009; 103(3):387–406. 4. Heersink B, Peterson BD, Jenkins JA. Disasters and Elections: Estimating the Net Effect of Damage and Relief in Historical Perspective. Political Analysis. 2017; 25(2):260–268. 5. Masiero G, Santarossa M. Natural Disasters and Electoral Outcomes. European Journal of Political Economy. 2021; 67(January):101983. 6. Akarca AT, Tansel A. Social and Economic Determinants of Turkish Voter Choice in the 1995 Parliamentary Election. Electoral Studies. 2007; 26(3):633–647. 7. Cole S, Healy A, Werker E. Do Voters Demand Responsive Governments? Evidence from Indian Disaster Relief. Journal of Development Economics. 2012; 97(2):167–181. 8. Eriksson LM. Winds of Change: Voter Blame and Storm Gudrun in the 2006 Swedish Parliamentary Election. Electoral Studies. 2016; 41:129–142. 9. Gasper JT, Reeves A. Make It Rain? Retrospection and the Attentive Electorate in the Context of Natural Disasters. American Journal of Political Science. 2011; 55(2):340–355. 10. Healy AJ, Malhotra N, Mo CH. Irrelevant Events Affect Voters’ Evaluations of Government Performance. Proceedings of the National Academy of Sciences of the United States of America. 2010; 107 (29):12804–12809. https://doi.org/10.1073/pnas.1007420107 PMID: 20615955 11. Lay JC. Race, Retrospective Voting, and Disasters: The Re-Election of C. Ray Nagin after Hurricane Katrina. Urban Affairs Review. 2009; 44(5):645–662. 12. Bechtel MM, Mannino M. Retrospection, Fairness, and Economic Shocks: How Do Voters Judge Policy Responses to Natural Disasters? Political Science Research and Methods. 2022; 10(2):260–278. 13. Campante FR, Depetris-Chauvin E, Durante R. The Virus of Fear: The Political Impact of Ebola in the U.S. National Bureau of Economic Research; 2020. 26897. 14. O’Driscoll M, Ribeiro Dos Santos G, Wang L, Cummings DAT, Azman AS, Paireau J, et al. Age-Specific Mortality and Immunity Patterns of SARS-CoV-2. Nature. 2021; 590(7844):140–145. https://doi.org/10. 1038/s41586-020-2918-0 PMID: 33137809 15. Russell TW, Hellewell J, Jarvis CI, van Zandvoort K, Abbott S, Ratnayake R, et al. Estimating the Infection and Case Fatality Ratio for Coronavirus Disease (COVID-19) Using Age-Adjusted Data from the Outbreak on the Diamond Princess Cruise Ship, February 2020. Eurosurveillance. 2020; 25(2000256). https://doi.org/10.2807/1560-7917.ES.2020.25.12.2000256 PMID: 32234121 16. Center for Systems Science and Engineering (CSSE) at Johns Hopkins University. COVID-19 Data Repository by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University; 2021. https://github.com/CSSEGISandData/COVID-19. 17. Bol D, Giani M, Blais A, Loewen PJ. The Effect of COVID-19 Lockdowns on Political Support: Some Good News for Democracy? European Journal of Political Research. 2021; 60(2):497–505. 18. De Vries CE, Bakker BN, Hobolt SB, Arceneaux K. Crisis Signaling: How Italy’s Coronavirus Lockdown Affected Incumbent Support in Other European Countries. Political Science Research and Methods. 2021; 9(3):451–467. 19. Dietz M, Roßteutscher S, Scherer P, Sto ¨vsand LC. Rally Effect in the Covid-19 Pandemic: The Role of Affectedness, Fear, and Partisanship. German Politics. 2023; 32(4):643–663. 20. Kritzinger S, Foucault M, Lachat R, Partheymu ller J, Plescia C, Brouard S. ‘Rally Round the Flag’: The COVID-19 Crisis and Trust in the National Government. West European Politics. 2021; 44(5–6):1205– 1231. 21. Schraff D. Political Trust during the Covid-19 Pandemic: Rally around the Flag or Lockdown Effects? European Journal of Political Research. 2021; 60:1007–1017. https://doi.org/10.1111/1475-6765. 12425 PMID: 33362332 22. Yam KC, Jackson JC, Barnes CM, Lau J, Qin X, Lee HY. The Rise of COVID-19 Cases Is Associated with Support for World Leaders. Proceedings of the National Academy of Sciences of the United States of America. 2020; 117(41):25429–25433. https://doi.org/10.1073/pnas.2009252117 PMID: 32973100 23. Bechtel MM, Obrochta W, Tavits M. Can Policy Responses to Pandemics Reduce Mass Fear? Journal of Experimental Political Science. 2022;: 1–11. https://doi.org/10.1017/XPS.2022.7 PLOS ONE How COVID-19 affects voting for incumbents PLOS ONE | https://doi.org/10.1371/journal.pone.0297432 March 19, 2024 16 / 19 24. Bisbee J, Honig D. Flight to Safety: COVID-Induced Changes in the Intensity of Status Quo Preference and Voting Behavior. American Political Science Review. 2021;: 1–17. 25. Akarca AT, Tansel A. Voter Reaction to Government Incompetence and Corruption Related to the 1999 Earthquakes in Turkey. Journal of Economic Studies. 2016; 43(2):309–335. 26. Achen CH, Bartels LM. Democracy for Realists: Why Elections Do Not Produce Responsive Government. Princeton, NJ: Princeton University Press; 2016. 27. Fiorina MP. Retrospective Voting in American National Elections. New Haven: Yale University Press; 1981. 28. Baker WD, Oneal JR. Patriotism or Opinion Leadership? Journal of Conflict Resolution. 2001; 45 (5):661–687. 29. Lambert AJ, Schott JP, Scherer L. Threat, Politics, and Attitudes: Toward a Greater Understanding of Rally-’round-the-Flag Effects. Current Directions in Psychological Science. 2011; 20(6):343–348. 30. Mueller JE. Presidential Popularity from Truman to Johnson. American Political Science Review. 1970; 64(1):18–34. 31. Dinesen PT, Jæger MM. The Effect of Terror on Institutional Trust: New Evidence from the 3/11 Madrid Terrorist Attack. Political Psychology. 2013; 34(6):917–926. 32. Gaines BJ. Where’s the Rally? Approval and Trust of the President, Cabinet, Congress, and Government since September 11. PS: Political Science & Politics. 2002; 35(3):531–536. 33. Hetherington MJ, Nelson M. Anatomy of a Rally Effect: George W. Bush and the War on Terrorism. PS: Political Science & Politics. 2003; 36(1):37–42. 34. Hintson J, Vaishnav M. Who Rallies around the Flag? Nationalist Parties, National Security, and the 2019 Indian Election. American Journal of Political Science. 2023; 67:342–357. 35. Perrin AJ, Smolek SJ. Who Trusts? Race, Gender, and the September 11 Rally Effect among Young Adults. Social Science Research. 2009; 38(1):134–145. https://doi.org/10.1016/j.ssresearch.2008.09. 001 PMID: 19569296 36. Baekgaard M, Christensen J, Madsen JK, Mikkelsen KS. Rallying around the Flag in Times of COVID19: Societal Lockdown and Trust in Democratic Institutions. Journal of Behavioral Public Administration. 2020; 3(2):1–12. 37. Esaiasson P, Sohlberg J, Ghersetti M, Johansson B. How the Coronavirus Crisis Affects Citizen Trust in Institutions and in Unknown Others: Evidence from ‘the Swedish Experiment’. European Journal of Political Research. 2020;1–13. 38. Nielsen JH, Lindvall J. Trust in Government in Sweden and Denmark during the COVID-19 Epidemic. West European Politics. 2021; 44(5–6):1180–1204. 39. Leininger A, Schaub M. Strategic Alignment in Times of Crisis: Voting at the Dawn of a Global Pandemic. Political Behavior. 2023. 40. Vasilopoulos P, Marcus GE, Valentino NA, Foucault M. Fear, Anger, and Voting for the Far Right: Evidence from the November 13, 2015 Paris Terror Attacks. Political Psychology. 2019; 40(4):679–704. 41. Hyland P, Shevlin M, McBride O, Murphy J, Karatzias T, Bentall RP, et al. Anxiety and Depression in the Republic of Ireland during the COVID-19 Pandemic. Acta Psychiatrica Scandinavica. 2020; 142 (3):249–256. https://doi.org/10.1111/acps.13219 PMID: 32716520 42. Petzold MB, Bendau A, Plag J, Pyrkosch L, Mascarell Maricic L, Betzler F, et al. Risk, Resilience, Psychological Distress, and Anxiety at the Beginning of the COVID-19 Pandemic in Germany. Brain and Behavior. 2020; 10(9):e01745. https://doi.org/10.1002/brb3.1745 PMID: 32633464 43. Wagner M, Morisi D. Anxiety, Fear, and Political Decision Making. Oxford Research Encyclopedia of Politics. 2019;. 44. Greenberg J, Solomon S, Pyszczynski T. Terror Management Theory of Self-Esteem and Cultural Worldviews: Empirical Assessments and Conceptual Refinements. In: Zanna MP, editor. Advances in Experimental Social Psychology. vol. 29. Academic Press; 1997.: 61–139. 45. Bonanno GA, Jost JT. Conservative Shift among High-Exposure Survivors of the September 11th Terrorist Attacks. Basic and Applied Social Psychology. 2006; 28(4):311–323. 46. Jost JT, Glaser J, Kruglanski AW, Sulloway FJ. Political Conservatism as Motivated Social Cognition. Psychological Bulletin. 2003; 129(3):339–375. https://doi.org/10.1037/0033-2909.129.3.339 PMID: 12784934 47. Jost JT. Anger and Authoritarianism Mediate the Effects of Fear on Support for the Far Right—What Vasilopoulos et al. (2019) Really Found. Political Psychology. 2019; 40(4):705–711. 48. Zaleskiewicz T. Beyond Risk Seeking and Risk Aversion: Personality and the Dual Nature of Economic Risk Taking. European Journal of Personality. 2001; 15(1 SUPPL.):S105–S122. https://doi.org/10. 1002/per.426 PLOS ONE How COVID-19 affects voting for incumbents PLOS ONE | https://doi.org/10.1371/journal.pone.0297432 March 19, 2024 17 / 19 49. Nadeau R, Martin P, Blais A. Attitude towards Risk-Taking and Individual Choice in the Quebec Referendum on Sovereignty. British Journal of Political Science. 1999; 29:523–539. 50. Lockerbie B. Prospective Economic Voting in U. S. House Elections, 1956–88. Legislative Studies Quarterly. 1991; 16(2):239–261. 51. Rosema M. Partisanship, Candidate Evaluations, and Prospective Voting. Electoral Studies. 2006; 25 (3):467–488. 52. Berrebi C, Klor EF. Are Voters Sensitive to Terrorism? Direct Evidence from the Israeli Electorate. American Political Science Review. 2008; 102(3):279–301. 53. Getmansky A, Zeitzoff T. Terrorism and Voting: The Effect of Rocket Threat on Voting in Israeli Elections. American Political Science Review. 2014; 108(3):588–604. 54. Leromain E, Vannoorenberghe G. Voting under Threat: Evidence from the 2020 French Local Elections. European Journal of Political Economy. 2022; 75:102204. https://doi.org/10.1016/j.ejpoleco.2022. 102204 PMID: 36440377 55. Noury A, Fran cois A, Gergaud O, Garel AHow Does COVID-19 Affect Electoral Participation? Evidence from the French Municipal Elections. PLOS ONE. 2021; 16(2):e0247026. https://doi.org/10.1371/ journal.pone.0247026 PMID: 33626074 56. Picchio M, Santolini R. The COVID-19 Pandemic’s Effects on Voter Turnout. European Journal of Political Economy. 2022; 73:102161. https://doi.org/10.1016/j.ejpoleco.2021.102161 PMID: 34975184 57. Ministère de l’Inte ´rieur et des Outre-mer. Elections Municipales 2020—Re ´sultats 1er Tour; 2020. Data. Gouv.Fr/Fr/Datasets/Elections-Municipales-2020-Resultats-1er-Tour/. 58. Ministère de l’Inte ´rieur et des Outre-mer. Municipales 2020—Re ´sultats 2nd Tour; 2020. Data.Gouv.Fr/ Fr/Datasets/Municipales-2020-Resultats-2nd-Tour/. 59. Ministère de l’Inte ´rieur et des Outre-mer. Elections Municipales 2014; 2014. Data.Gouv.Fr/F r/Datasets/?organization=534fff91a3a7292c64a77f53&q=%{{C3}}%{{A9lections}}%20 municipales%202014. 60. Institut national de la statistique et des e ´tudes e ´conomiques. Statistiques et e ´tudes; 2020. https://www. insee.fr/fr/statistiques. 61. Institut national de la statistique et des e ´tudes e ´conomiques. Fichier des personnes de ´ce ´de ´es—De ´cès, https://www.data.gouv.fr/fr/datasets/fichier-des-personnes-decedees/, Downloaded on July 6, 2020; 2020. 62. Beaney T, Clarke JM, Jain V, Golestaneh AK, Lyons G, Salman D, et al. Excess Mortality: The Gold Standard in Measuring the Impact of COVID-19 Worldwide? Journal of the Royal Society of Medicine. 2020; 113(9):329–334. https://doi.org/10.1177/0141076820956802 PMID: 32910871 63. Karlinsky A, Kobak D. Tracking Excess Mortality across Countries during the COVID-19 Pandemic with the World Mortality Dataset. eLife. 2021; 10:e69336. https://doi.org/10.7554/eLife.69336 PMID: 34190045 64. Cassan G, Sangnier M. The Impact of 2020 French Municipal Elections on the Spread of COVID-19. Journal of population economics. 2022; 35:963–988. https://doi.org/10.1007/s00148-022-00887-0 PMID: 35345551 65. Institut national de la statistique et des e ´tudes e ´conomiques. Statistiques Locales—Densite ´de Population, https://statistiques-locales.insee.fr/#view=map1&c=indicator, Downloaded on July 10, 2020; 2020. 66. Dunbar RIM. Neocortex Size as a Constraint on Group Size in Primates. Journal of Human Evolution. 1992; 22(6):469–493. https://doi.org/10.1016/0047-2484(92)90081-j 67. Von Mises R. Ueber Aufteilungsund Besetzungswahrscheinlichkeiten. Revue de la Faculte ´des sciences de l’Universite ´d’Istanbul. 1939; 4:145–163. 68. Borja MC, Haigh J. The Birthday Problem. Significance. 2007; 4(3):124–127. https://doi.org/10.1111/j. 1740-9713.2007.00246.x 69. Tricaud C. Better Alone? Evidence on the Costs of Intermunicipal Cooperation. Condorcet Center for Political Economy; 2019. 70. IPSOS. Datacovid—Baromètre COVID-19, Data.Gouv.Fr/Fr/Datasets/Datacovid-BarometreCOVID19/; 2020. 71. Jennings W. Covid-19 and the ‘Rally-Round-the-Flag’ Effect. UK in a Changing Europe. 2020;. 72. Alesina A, Passarelli F. Loss Aversion in Politics. American Journal of Political Science. 2019; 63 (4):936–947. 73. Morgenstern S, Zechmeister E. Better the Devil You Know than the Saint You Don’t? Risk Propensity and Vote Choice in Mexico. The Journal of Politics. 2001; 63(1):93–119. PLOS ONE How COVID-19 affects voting for incumbents PLOS ONE | https://doi.org/10.1371/journal.pone.0297432 March 19, 2024 18 / 19 74. Morisi D, Jost JT, Panagopoulos C, Valtonen J. Is There an Ideological Asymmetry in the Incumbency Effect? Evidence from U.S. Congressional Elections. Social Psychological and Personality Science. 2021; 0(0):1–11. 75. Quattrone GA, Tversky A. Contrasting Rational and Psychological Analyses of Political Choice. The American Political Science Review. 1988; 82(3):719–736. 76. Johansson B, Hopmann DN, Shehata A. When the Rally-around-the-Flag Effect Disappears, or: When the COVID-19 Pandemic Becomes “Normalized”. Journal of Elections, Public Opinion and Parties. 2021; 31(S1):321–334. https://doi.org/10.1080/17457289.2021.1924742 77. Becher M, Longuet Marx N, Pons V, Brouard S, Foucault M, Galasso V, et al. COVID-19, Government Performance, and Democracy: Survey Experimental Evidence from 12 Countries. National Bureau of Economic Research; 2021. 29514. 78. Jørgensen F, Bor A, Lindholt MF, Bang M, Jørgensen F, Bor A. Public Support for Government Responses against COVID-19: Assessing Levels and Predictors in Eight Western Democracies during 2020 during 2020. West European Politics. 2021; 44(5–6):1129–1158. PLOS ONE How COVID-19 affects voting for incumbents PLOS ONE | https://doi.org/10.1371/journal.pone.0297432 March 19, 2024 19 / 19