Published: 29-Apr-2021
In early 2020, the Covid-19 pandemic has forced most of the planet to stop any non-essential activity involving large groups of people until disease transmission is eventually controlled. Whether one should consider elections an essential activity in this context has become a matter of intense debates leading to dramatically different decisions across the globe. Some governments have postponed elections (local elections in Austria, Italy & UK, Ohio primaries), some have switched to mail-in voting (Poland, Alaska primaries), and others have decided to hold the elections as planned with further sanitary precautions (Israel, South Korea, Wisconsin primaries). The empirical evidence to guide public policy is very scarce, owing to how sudden the Covid-19 outbreak has been and to the small number of elections held since it began.
This paper seeks to bridge this gap: we measure the impact of holding the first round of French town hall elections on the mortality of the individuals it exposed the most to Covid-19, i.e. politicians. These elections provide an unusual setting as they were held at the apex of the epidemic in France – on Sunday the 15th of March 2020 – one day after the nationwide closure of non-essential retail businesses was ordered and two days before a shelter-at-home order.
Our analysis is based on individual death records from the French institute of statistics (Insee) matched with a public registry of candidates in town hall elections. The resulting dataset enables us to compare the mortality rates of male candidates in the 2014 and 2020 municipal elections relative to the general population as well as depending on the degree of participation into the 2020 election.
The correlations we can estimate between political activity and mortality using this new dataset do not necessarily reflect a causal relationship. In particular, citizens might choose to become candidates only after careful consideration of their health status and election participants may adapt their contact behavior to their current knowledge about the epidemic. To deal with these threats to identification, we implement two distinct strategies to estimate the causal effect hall seats, French voters choose between competing lists of candidates, and turnout and campaign intensity are much lower when only one candidate list is running for election. Therefore, one identification strategy is to compare the mortality rates of candidates when there are either one or two lists competing for entrance in the city council. Second, we examine whether, among the full population of male candidates in the previous election held in 2014, being either an incumbent councilor or a candidate in 2020 caused an increase in the probability of death. We implement a regression discontinuity design in the spirit of Lee (2008), since electoral success in the prior election has a large impact on active participation in the 2020 election. This strategy identifies the impact of running in the 2020 election at the individual level.
Three main sets of findings emerge from our empirical exercise. First, we compute the mortality rates in March and April 2020 among candidates in the March 15th election. This is a group of around 900,000 individuals, representing 1.8% of the adult French population. They were among the most likely to contract the disease as they actively participated in the campaign and closely monitored the voting process on election day. Many of them also belong to demographic groups with a high risk of dying from Covid-19: 160,000 candidates were men aged between 60 and 79 years old, representing 2.6% of their category in the general population. We compare their mortality rates with overall mortality rates among individuals of same gender and same age. We find that, across March and April 2020, candidates died significantly less than members of the general population from the same age bracket. The weekly mortality rate among men aged between 60 and 79 years old from March to April was 4 for 10,000 in the general population against 1 for 10,000 among town hall candidates. This could reflect selection into candidacy based on good health. However, we do not find that this mortality “advantage” was smaller in areas where Covid-19 had the greatest incidence, i.e. the Ile-de-France and Grand-Est regions. The difference in mortality rates between the two groups was also of similar size around the onset of the epidemic and during the second wave of the Covid-19 pandemic in the fall of 2020, well after the election. This further suggests the excess mortality of candidates due to Covid-19 was in fact similar to the general population.
Second, we compare the mortality of candidates in cities where two candidate lists compete with the mortality of candidates in cities where there is only one candidate list. In the data, the number of competing candidate lists is a very strong determinant of turnout. Yet, we do not find any difference in mortality among candidates between the two types of cities, even though they otherwise do not differ much in observable characteristics. The lack of effect of electoral competition on mortality during the pandemic holds not only in the pool of candidates but also among city residents in general, which further suggests the lack of a large effect of holding the elections on epidemic diffusion.
Third, in order to go beyond estimating citywide effects of the election on mortality, we investigate the mortality patterns of the people who were candidates in the previous municipal elections, held in 2014. We find that, among them, those who decided to run again in 2020 were not more likely to die from March to April 2020. Using a regression discontinuity design, we observe that candidates in 2014 are far more likely to participate again in 2020 if their candidate list had won the election in 2014. However, we do not detect any significant effect of winning the election in 2014 on mortality in March-April 2020. This further confirms, using an entirely different identification strategy, that intense participation in the 2020 elections did not generate any specific mortality risk.
This is to our knowledge the first paper to document the mortality patterns of politicians in the middle of a Covid-19 pandemic. It is of interest in its own right as politicians are key decision-makers whose good health is of paramount value in times of crisis. By the nature of their occupation, politicians may also be more likely to contract the disease. Examples of how this has hampered policy-making abound: at certain times in the epidemic, the UK and the US governments were leaderless while the French National Assembly had to stop its proceedings once an internal cluster developed.1 The main research hurdle is to obtain a large enough sample of politicians to analyze. In this respect, our setting, which includes more than 160,000 politicians with significant exposure to mortality from Covid-19, provides a unique opportunity.
Our research also innovates by investigating the effects of elections on the diffusion of Covid-19 using individual data. Several studies have by now investigated this issue at higher levels of aggregation and for the general population. Some of this research (Zeitoun et al., 2020, Cassan and Sangnier, 2020, Cotti et al., 2020) has focused on Covid-19 outcomes measured at the level of each county and related them to turnout at the same level of aggregation, with conflicting results.2 Those analyses seek to explain and predict the full development of the epidemic. This comes at the cost of making strong assumptions on the behavior of the disease and its diffusion patterns. This is why we focus instead on a narrower goal of identifying individual transmissions of the disease that were directly caused by the election.