Health as an independent predictor of the 2017 French presidential voting behaviour: a cross-sectional analysis

Two great democracies have recently encountered outcomes that were considered as unexpected and that have been interpreted as reflecting a rejection of established political parties. Namely, vote for Brexit in United-Kingdom (UK) and Donald Trump’s election had not been anticipated by most polling institutes [1, 2]. Even though multiple explanatory factors were thought to be involved for each, it has been argued that health and healthcare had a significant influence on voters [3]. In the UK, self-reported health has been measured as declining since 2010 and death rate in most age categories rose over the last years [4]. Moreover, public healthcare funding was a key issue in political debates, with proponents pretending that Brexit would bring substantial savings for the UK that would be subsequently reallocated toward the National Health System, a claim that has been rapidly retracted just after the vote [5]. In the US, it has been shown that the strongest factor associated with an increase in Republican voting for Donald Trump, as compared to prior elections, was a decrease in health status when measured by several indicators [6]. In the US also, the healthcare system and its funding were paramount in the campaign debates although Donald Trump’s intentions to repeal Affordable Care Act have not succeeded yet [7]. Even though those elements cannot be considered as definitive evidence, they strongly suggest that health and healthcare topics have been indeed important factors of citizen choices in the UK and US votes. This would be consistent with prior peer-reviewed work showing that health may influence attitude toward democracy and turnout to elections [8, 9]. However, in-depth studies regarding true votes and psychological factors that could have driven behaviours are lacking, despite rapidly accumulated evidence that those two western countries may be failing to maintain life expectancy [10, 11].

France also has issues with people’s health and its healthcare system. According to official estimates, there has been an unexpected peak of deaths in 2015, with no clear explanation [12, 13]. Even though France’s healthcare system is frequently considered as equitable, budgets are highly pressured and concerns over sustainability of the current model are growing [12, 14]. For the sake of comparison between the three countries, a graph has been constructed, showing health expenditures over the last years of the UK, the US, and France, according to the data established by the Organisation for Economic Cooperation and Development (Fig. 1). Soon after the UK and US votes, France entered into a major political campaign, namely the 2017 presidential election. After the first round, two candidates were selected for a second round in accordance with the established rules of the country’s 5th Republic. Emmanuel Macron (EM), a novel politician claiming a “nor left-nor right wing” position and Marine Le Pen (MLP), the far-right leader competed for the second and final round, as foreseen by most polling institutes. Even though Emmanuel Macron won by far the election, Marine Le Pen’s votes both at the first and second round reached unprecedented levels for a far-right party in France, with about 11 million people voting for her at the second round. While social factors, like unemployment, are strongly linked with turnout and voting behaviour [15], we are unaware of any investigation regarding possible relationships between health and voting patterns in France whereas, like in many high-income democracies, our healthcare system experiences difficulties, both regarding funding and inequalities [12]. Moreover, the presidential campaign has emphasized people’s concerns regarding funding of the combined set of public and private health insurances that characterizes the French healthcare system [14]. Since the 2017 presidential election, health and economic issues have been placed at the top of the political agenda [16] but we still lack data regarding a possible association between health status and voting patterns. Such measures would help to determine whether recent trends observed in the UK and the US reflect a more general phenomenon, to address concerns of people living in areas affected by poorer health indicators, and to inform the overall public health debate.

Therefore, we sought to analyse possible associations between voting patterns at the first round of the 2017 French presidential election and relevant health indicators. Since there is a well-known relationship between wealth and health, [17, 18] we aimed to adjust our analyses on several socioeconomic indicators.

The 95 departments were included in the study, with a median population of 536,694 per department [76,607-2,595,536]. Departments’ characteristics with respect to socioeconomic and health indicators are presented in Table 1. The outcome of the first round of the election for both leading candidates is represented on the French map in Fig. 2. Abstention rate was 22.23%.

In this analysis of the first round of the 2017 French presidential election, we showed that, even after adjustment on wealth and social confounders, most health-related indicators were strongly associated with voting patterns both for Emmanuel Macron, the “nor right-nor left wing” candidate, and Marine Le Pen, the far-right candidate. Diabetes mellitus, being affected by a severe chronic condition, and mortality were among the strongest predictors in both models. Whereas other researchers have raised the possibility that health recently influenced votes for Brexit and Donald Trump respectively in the United-Kingdom and the US, our findings come to support health as an independent political marker in France also.

Our study has several strengths. First, it is to our knowledge, the first assessment in the medical literature of the relationship between health and voting pattern at a French presidential election. Second, our analysis relies upon official, updated and very reliable data sources. All collected variables came from public registries and databases, which are constructed through rigorous and transparent methods, and revised almost every year. Except for one of them (the Gini coefficient), we retrieved very recent data for each selected indicator. Third, since health is a highly complex and multifaceted concept, we chose an extensive set of relevant indicators, with so-called “hard” endpoints such as mortality and severe chronic conditions. Fourth, because there is a well-established relationship between wealth and health status, [18, 26] we adjusted our analysis on another set of income indicators through sophisticated statistical methods. Fifth, we used a sophisticated and established method for conducting the search of associations between variables and our outcome. Principal component analysis has extensively been used, including in landmark papers [27].

Our findings raise several issues worth considering about the relationship between health and voting behaviour. According to most political analysts, Emmanuel Macron and Marine Le Pen had very different projects with respect to almost all possible aspects. Emmanuel Macron comes from the left wing of the political spectrum, yet he repeatedly claimed from the beginning of his campaign a “nor left-nor right wing” position. His program was liberal, both from an economic and societal standpoint. Conversely and even if she rejects this categorization, Marine Le Pen is considered as supporting a far-right doctrine, expressing among others strong criticisms against liberalism, the European Union project and immigrants. Therefore, it comes with no surprise that social and economic indicators were evenly distributed among both candidates, with departments associated with greater wealth more likely to vote for Emmanuel Macron. However, we also found that most health-related indicators were associated with voting patterns in an independent and very significant way. In our view, this was less expected since France has a healthcare system considered as highly protective and universal [12, 14]. Therefore, and despite significant challenges regarding its sustainability, [12, 14, 28] one might have hypothesized that health would not substantially influence voting behaviour. However, we measured statistical associations that strongly suggest such influence. For instance, we found that when the rate of diabetes mellitus in a given department is around a higher range, people tend to vote less for EM but vote more for MLP. It should be emphasized for non-French readers that the first round of the presidential election did not only include EM and MLP but also 9 other candidates, and that MLP and EM together only attracted 45.31% of all votes. We chose to analyse the results of the first round as it is more likely to reflect the actual voting behaviour for a candidate rather than against one of them, as it is commonplace at the second round when there are only two options. The ecological nature of our study precludes definitive interpretation of our findings. The most immediate explanation would be that sicker people are more likely to vote for a far-right candidate. Another hypothesis is that people living in areas with deprived health indicators are more likely to choose a candidate perceived as representing a greater potential for change, as compared to the one that already had political responsibilities at the national level. Those assumptions deserve further investigation. Also, further work could search for similar statistical associations between health indicators and results from the second round of the 2017 French presidential election and the first round of the legislative election. Our study also found that socioeconomic factors were related to turnouts and blank votes. However, turnout and blank votes were not associated with the same variables than MLP and EM, suggesting that some factors may influence the decision to vote or not and others may influence the choice of the candidate. This last point was clearly visible after projection on the first factorial plane where MLP and EM projected in the same line but in opposite direction whereas turnout and blank votes combined projected orthogonally to this line.

Our research has limitations, the main one probably being ecological fallacy. Estimated effects are conditional-on-observables and we cannot exclude unobserved confounders. Indeed, this study is based on aggregated data at the department level, not upon individual data. Substantial heterogeneity exists within a department and thus aggregating data may engender a loss of information and statistical power. Additionally, both the socio-economic variables and health variables were calculated on the whole population of the department, but due to French electoral law, not all the population living in a department are voting. There might be a bias as patients who vote may be different from those who do not. Consequently, our findings need to be interpreted with caution. In particular, they do not necessarily mean that people with poorer health had significantly different votes. It can for instance be hypothesized that people unaffected by substantial medical condition but who live in degraded areas were influenced by their environment. Yet, individual patients’ data would have been almost impossible to obtain due to patient doctor privilege and the secrecy of the vote. Instead, high quality aggregates (as measured by French Public Institutes and the Ministry of the Internal Affairs) are reliable and free of the secrecy problem. The French department is the smallest administrative unit for which both voting patterns and indicators were available. Regarding the choice of the statistical model, one can argue that another model (multinomial) or a more complex representation of the relation between the variable could have performed better. We believe that this model was the most adequate choice as it is easier to understand and less prone to over fitting. Moreover, the elastic-net procedures confirmed that the variables we selected in this approach are reliable even in presence of multiple, highly correlated predictors. Last, we selected several health indicators to be incorporated in the model, some of them being reference indicators such as mortality, life expectancy, and being affected by a chronic health condition for instance. Other indicators could have been added to the analysis, namely cancer prevalence, mental health, and cardiovascular diseases, and may deserve future dedicated research. Similarly, other economic variables could have been included in the current analysis, such as economic growth, inflation rate, or even subjective variables.

Our analysis of the first round of the 2017 French presidential election found strong and independent associations between almost all relevant health indicators and voting behaviours. It emerged that areas where health indicators exhibited worse results were also those which were more likely to favour the far-right candidate, even after adjustment on all wealth and social variables. Conversely, areas with better scores in health variables significantly favoured the more moderate and clearly liberal candidate, who eventually was elected two weeks after. Our findings are consistent with recent reports about Brexit and 2016 US presidential election and highlight the considerable attention that health issues deserve, even in such a protective country like France.