Abstract

This study uses aggregate panel data on French départements to investigate the relationship between macroeconomic conditions and mortality from 1982 to 2014. We find no consistent relationship between macroeconomic conditions and all-cause mortality in France. The results are robust across different specifications, over time, and across different geographic levels. However, we find that heterogeneity across age groups and mortality causes matters. Furthermore, in areas with a low average educational level, a large population, and a high share of migrants, mortality is significantly countercyclical. Similar to the case in the United States, the relationship between the unemployment rate and mortality seems to have moved from slightly procyclical to slightly countercyclical over the period of analysis.

Introduction

Most macroeconomic studies have found that good economic times are bad for population health. The literature on the topic dates back as far as Ogburn and Thomas (1922) who documented that mortality seems to decline when unemployment levels rise, a notion confirmed by a series of more contemporary studies by Ruhm (2000, 2003, 2005).1 The procyclicality of mortality has subsequently been documented in countries other than the United States, at least for motor vehicle accidents and cardiovascular diseases (Ariizumi and Schirle 2012; Gerdtham and Ruhm 2006; Lin 2009; Neumayer 2004; Regidor et al. 2016; Tapia Granados 2008; Tapia Granados and Ionides 2011; Tapia Granados and Rodriguez 2015).2

The main contributions of this study are threefold. First, we update previous results from the only unpublished but highly quoted French study in this field of research (Buchmueller et al. 2007). In contrast to Buchmueller et al. (2007), we do not find that increases in the local unemployment rates are associated with reductions in all-cause mortality rates over the same period.3 Moreover, we also examine the relationship over a longer time frame and add département-specific linear trends to the econometric specification. By including data that incorporates the Great Recession (from mid-2007 to the beginning of 2010), we are able to exploit greater exogenous variation in the unemployment rate across départements and years, as compared with Buchmueller et al. (2007). In addition, Ruhm (2015) found that over a long period (1976–2010), total mortality shifted from strongly procyclical to being weakly related to macroeconomic conditions in the United States. We also examine this issue by using different sample windows for the French case. We find that the relationship is not procyclical in France.

Second, we examine heterogeneity in the relationship for several age groups and specific causes of death. Accidents and cardiovascular diseases should exhibit more procyclical patterns, for instance. We find evidence that this is the case for traffic accidents but not for cardiovascular diseases. For those whose unemployment status does not change—particularly the elderly between ages 65 and 74—the relationship is countercyclical, meaning that a higher unemployment rate increases general mortality. Furthermore, the relationship seems to be particularly countercyclical for men. Therefore, the aggregate fluctuations in mortality are not concentrated among only those of working age. Put differently, the estimated impact of individuals’ own job loss cannot be directly reconciled with the aggregate patterns. To explore this aspect further, we also analyze how own-group unemployment rates are related to own-group mortality. We find that they are not systematically related.

Third, as an exploratory analysis, we focus on other factors that may influence this relationship. Heterogeneity indeed matters. For instance, in areas with a low average educational level, a large population, and a high share of migrants, mortality is significantly countercyclical. The French case might also help us to understand why mortality rates were procyclical elsewhere, particularly in the United States. We thus test a number of alternative variables (such as equipment rates, nursing services for the elderly, or nurse density) that are supposed to proxy the quality of health care or a certain level of budgetary and human resources constraint.4 Last, we provide a discussion on the ability of the local unemployment rate to represent the business cycle in France.

Our results add an interesting perspective to the recent U.S. literature that investigates the robustness of the result of procyclical mortality. For instance, Ruhm (2015) analyzed the stability of the relationship over time and found that the procyclicality of mortality has decreased in recent years. Lam and Piérard (2017) found that mortality has even become countercyclical for some age groups. Furthermore, Lindo (2015) analyzed the impact of the level of aggregation in the United States and found that the use of smaller geographical levels results in smaller coefficients because of greater spillover effects between smaller geographical units. Ionides et al. (2013) tested the robustness of results across various methods of detrending the data and found procyclical mortality in all age groups. Our analysis suggests that France looks a lot like the United States, with less procyclicality occurring in recent years and a better social insurance system probably leading to less procyclicality of mortality.

The aforementioned literature and our study focus on the unemployment rate rather than individual unemployment. Individual-level studies and mechanisms may be very different from what is found at the aggregate level. At the individual level, greater economic prosperity is generally found to be positively related to health, whereas the opposite is often true at the aggregate level. Until recently, evidence on the channels through which the unemployment rate affects mortality had been rather scarce. The exception is mortality due to motor vehicle accidents, which is heavily procyclical given that traffic, and thus accidents, increase during economic expansions. For cardiovascular diseases, evidence shows that increased pollution during economic upturns is responsible for the procyclical relationship (Deryugina et al. 2016; Heutel and Ruhm 2016). These explanations are similar in spirit in that they proclaim that it is not primarily own employment status that explains the macro relationship between unemployment levels and cause-specific mortality. Stevens et al. (2015) emphasized this point, attributing higher mortality for older people during economic upswings to lower health care quality as the health care sector struggles to find adequate staff during economic upswings.5 There is a priori no reason that own-group employment should matter more in France than elsewhere, and our analysis confirms this point. Explanations that focus on externalities of unemployment on health are probably more attractive because they can explain why macroeconomic studies find unemployment to be inversely linked to mortality but microeconomic evidence points to the contrary.

Research Design

Our regression specification mostly follows the prior literature (Ruhm 2000, 2015). The data comprise 3,135 observations corresponding to 95 départements in metropolitan France and 33 years (1982–2014).6 For a specific source of death, in département j, at year t, we estimate the following equation:

lnMortj,t=λUj,t+βXj,t+δj+δt+θjt+εj,t,
1
where U is the area-specific unemployment rate, X is a vector of covariates, δj represents area fixed effects, δt represents year fixed effects, and θjt represents département-specific time trends. Estimations are weighted by the population in each area, and standard errors are clustered at the area level. The coefficient of interest is λ, which measures the effect of unemployment on the natural log of the mortality rate (number of deaths per 100,000). X includes demographic control variables. In our preferred specification, these controls include the proportion of the population that is female and the proportion aged 0–4, 5–24, 45–54, 55–64, 65–74, 75–84, and 85 and over. Moreover, we control for education level and share of migrants in an area.7 In most regressions, the dependent variable is the natural log of the mortality rate in département i in year t. As a robustness check, we report models in which the dependent variable is the number of deaths per 100,000 persons. We also investigate mortality by cause. For these estimations, we use the nontransformed mortality rate as the dependent variable and estimate the model with Poisson pseudo-maximum likelihood (PPML). The main reason for this is the presence of zeros in the dependent variable when we analyze specific mortality causes.8

Note that the residual variation in state-year unemployment rates after controls are included (1 minus the R squared from regressing unemployment rates on state and year dummy variables, area-specific time trends, and the preferred set of controls) is relatively low (at about .06).9 This raises the possibility that unemployment rates are not the best proxy for macroeconomic conditions in France. Nonetheless, unemployment has been extensively used in most studies and is thus used in our study to enable comparisons. We further discuss this point in a later section.

Data and Graphical Exploration

Crude mortality rates are obtained from the Centre d’épidémiologie sur les causes médicales de décès (CépiDc).10 The CépiDc maintains a database with currently more than 20 million records (deaths since 1979). Statistics on deaths are based on information gathered from two documents: the medical certificate and the bulletin of civil status of death. Total deaths are available by département, year, gender, and age (in five-year groups). We also use an extensive set of demographic controls to account for any shifts in the age distribution of the population as, for example, Ruhm (2015) did. The use of age-adjusted mortality rates over crude mortality rates has only a moderate impact on the results in the United States.11

Bridges have been established between ICD-9 and ICD-10 coding systems in the United States (Anderson et al. 2001). These issues are typically minor when looking at broad causes of death (e.g., those from malignant neoplasms) but may be important for many specific sources of mortality (Ruhm 2015). Table A1 in the online appendix details the ICD codes used to classify causes of death in France. CIM9 and CIM10 equivalence are provided by the CépiDc, a WHO collaborating center for the family of international classifications in French.12 Issues of data comparability are likely to be minor and well-captured by the inclusion of year fixed effects in the regressions (Ruhm 2015). Figures A1A4 in the online appendix provide mortality rates (per 100,000 population) by gender, age groups, and selected causes, and show how the sources of death changed over the period of analysis.

Population data and unemployment data were collected from the Institut National de la Statistique et des Etudes Economiques (INSEE).13 Because unemployment is the core issue of our paper, we rely on official statistics provided by INSEE.14 Unemployment rates are available at the département level for 1982 onward. INSEE provides population data by gender in five-year age brackets. Furthermore, we also obtained data on the share of migrants and on the education level for the census years 1982, 1990, 1999, 2009, and 2014.

In the administrative divisions of France, the département is one of the three levels of government below the national level, between the 13 administrative regions in metropolitan France and the municipalities (communes). In 2014, the last year of our analysis, the average département had a population of 676,000 (median 539,000).15 We also report some results on the level of regions for comparison. Before January 2016, the French metropolitan area was divided into 22 regions, including Corsica. Ideally, we would have liked to run these tests on the level of old regions, but we could obtain the unemployment rate from only 1982 onward for the 13 new regions. However, these new regions do not split old regions but instead aggregate several old regions into one new region. As a result, we can still run additional robustness tests based on former regions.

We begin our analysis with a graphical exploration of the relationship between unemployment and mortality. Figure 1 displays national mortality and unemployment rates in each year. The variables are detrended, using a linear trend, and standardized by subtracting the mean of the detrended variable and dividing by its standard deviation (Ruhm 2000). A comparison with Buchmueller et al. (2007) is provided in the online appendix (section B).

Results

All-Cause Mortality

Table 1 explores the robustness of the regression of total mortality on unemployment across different econometric specifications. Panel A shows results for our main level of analysis, the départements. Panel B adds the same regressions on the regional level as an additional robustness check. Column 1 starts with a simple specification that includes year and area fixed effects but not control variables or area-specific trends. This is similar to the specifications Buchmueller et al. (2007) used. As they found, the coefficient is negative and statistically significant. However, the other columns show that the coefficient turns positive once area-specific time trends or demographic controls are included.16 Column 2 shows the full set of control variables that include the share of migrants; the share low-educated; the share high-educated; and the proportion of the département population that is female, aged 0–4, 5–24, 45–54, 55–64, 65–74, 75–84, and 85 or over. Column 3 drops these controls but adds a département-specific linear trend. Column 4 adds basic population controls to column 3. These include only the proportion of the population aged 0–4 and 65 and older. Column 5 adds the full set of control variables again together with the département-specific trend, and column 6 presents the preferred specification that includes weighting by the population in each département at time t.

Weighting is commonly done in analyses of mortality and brings the regression interpretation closer to the microdata analysis of individual unemployment and mortality. The weighted least squares (WLS) estimate appears to be slightly greater.17 OLS with fixed effects weights départements according to the within-département variance of the treatment and sample frequency (which is identical in our case).18 The difference in coefficient estimates due to the weighting already hints at the fact that the effect might be heterogeneous, an issue that we investigate in more detail in the next section. Estimates on the regional level tend to mirror results on the département level but are not overly informative because of the increased standard errors. As an additional robustness check, column 7 of Table 1 uses the preferred specification, but the outcome is now the mortality rate—not the natural logarithm of it. Substituting the log of mortality for the untransformed mortality rate does not have a qualitative impact on our results. The importance of the area-specific trends also points to a high correlation between unemployment rates and other local factors.19 If we use the age-adjusted mortality rate as the dependent variable, the coefficient is positive in all specifications on the département level (see online appendix, Table D1).

Heterogeneity by Age, Gender, and Cause

Having confirmed that mortality is not procyclical from 1982 to 2014 in France, the question we explore is why the probability of dying is not affected—or increases weakly—when economic times are bad in France. This finding is more in line with Ruhm’s (2015) finding that total mortality shifted from being strongly procyclical to being weakly or unrelated to macroeconomic conditions from 1976 to 2010. He observed, however, that the mechanisms behind (pro)cyclical variation in mortality remain poorly understood. In our case, even in the 1980s or 1990s, the relationship between macroeconomic conditions and total mortality is not procyclical.

Two broad mechanisms have been proposed in the literature for the procyclical relationship between mortality and economic conditions that has been observed in other countries: (1) changes in individuals’ own work hours or opportunity cost of time, and (2) the impact of external factors that fluctuate with the economy and geographic mobility. The underlying mechanism behind the relationship between unemployment and mortality has also long been investigated in studies of individuals (Burgard et al. 2009; Sullivan and von Wachter 2009), with general agreement that being unemployed raises the risk of death, although reverse causality seems to be also present (i.e., poor health increasing the risk of being jobless).

In our next step, we analyze whether the association between unemployment and mortality varies by age and gender. For this purpose, Table 2 provides the effects of unemployment on mortality by gender and age (at the département level). Again, the results show that mortality is not consistently procyclical in France. Males will be more affected than females by crises, particularly men aged 25–44 and 65–74.

Table 3 also shows regressions of cause-specific mortality by gender. Because some cells contain zeros in the mortality rates, all regressions are estimated by PPML.20 We also report additional deaths from a 1 percentage point increase in the unemployment rate.

Regarding the results for cause-specific mortality across all ages, all-cause mortality and mortality due to alcohol abuse are significant and positively related to unemployment. In contrast, accidents are negatively and significantly linked with unemployment. The other mortality rates show no significant relationship with unemployment.

Some estimates are similar to those reported elsewhere. Ruhm (1995) found that traffic deaths vary procyclically because of reductions in income rather than unemployment (U.S. data, 1975–1988). The procyclicality of mortality has been documented in many countries for motor vehicle accidents as well. Table 3 shows that this relationship is confirmed in France for traffic accidents and all accidents.21 Currie et al. (2018) compared the United States and France in an analysis of leading causes of death in the two countries. They argued that no one cause of death is a “smoking gun” that would explain the large differences in mortality across all age groups. However, they found that for younger groups, differences in deaths due to accidents account for more than one-third of the gap in mortality between the two countries, the United States being more impacted by accidents and experiencing a slower decline in accident-related mortality. In both countries, accidents seem to vary procyclically. Stevens et al. (2015) showed that motor vehicle accidents account for just 18 % of all cyclically induced deaths in the United States, and cyclical variation in non-elderly mortality due to other causes explains a very small fraction of the overall variation. Variation of mortality in France over the business cycle is certainly also not driven by motor vehicle accidents. In addition to Table 3, we split the sample in several periods to explore whether mortality is procyclical when the average risk of dying on the road is higher and becomes noncyclical or even countercyclical when this risk becomes smaller. Indeed, traffic-related deaths plummeted in France between around 2002 (when the Buchmueller et al. (2007) study ended) and 2014. However, we do not find strong evidence that this is the case. As a last test, we subtracted traffic accidents from all-cause mortality and ran a regression of our preferred specification on this variable. The resulting coefficient is 0.0019, with a standard error of 0.0009 (significant at the 5 % level), which is slightly higher than the coefficient on all-cause mortality (0.0016). This increase was to be expected given that the number of people killed in traffic accidents as a share of all deaths is relatively low but varies strongly with the economic cycle.

Other estimates are also similar to those found in previous studies. For instance, Stevens et al. (2015) found that deaths due to cancer and suicide decline when the economy strengthens. We find similar results, although additional deaths from a 1 percentage point increase in the unemployment rate are much lower in France than in the United States in both cases because of a size effect (the United States being much larger than France). Focusing on coefficients and not predicted number of deaths, the effect of unemployment on mortality is actually higher in France than in the United States.

However, some of our estimates by age and cause differ from those found by Miller et al. (2009) and Stevens et al. (2015), who found that cyclical deaths are concentrated among the elderly and are driven by cardiovascular deaths, respiratory deaths, degenerative brain diseases, and deaths due to infections. Ruhm (2007) showed that coronary heart disease (CHD), the most important component of heart disease, increases rapidly when the economy strengthens but returns to its baseline level within five years (U.S. data, 1979–1998). In France, cardiovascular deaths increase only for men when the economy strengthens.

Ruhm (1995) also found that alcohol consumption varies procyclically. By contrast, we find that mortality due to alcohol abuse in France is positively related to unemployment, probably indicating that stress-induced increases in alcohol consumption in France are not offset by reductions in income. The coefficient on alcohol abuse is strongest among women, a somewhat surprising finding given that prior studies have found that it is mostly males who experience deterioration in mental health from unemployment.

Heterogeneity Across Time and Subsamples

Recent developments in the literature highlight the possibility that the coefficient changes over time (Lam and Piérard 2017; Ruhm 2015). Thus, we investigate this possibility in Fig. 2, which replicates part of Ruhm’s (2015) analysis using the preferred specification from Table 1, column 6 (including the full set of control variables, département-specific linear trends, département and year fixed effects, and weighting by population).22 Figure 2 confirms that the association is sensitive to the period of analysis, with relatively shorter periods showing more instability. The coefficient is significantly positive only over the full period. In fact, the point estimate is negative (even if not statistically significant) for the beginning of the period and then turns positive. As in the United States, we thus find more countercyclicality of mortality occurring in recent years.

Table 4 provides heterogeneous effects by subsamples of départements using population, population growth, average educational level, unemployment rates, and share of migrants to segregate départements. For low average educational level, large population, high population growth, and high share of migrants, the relationship between unemployment rates and mortality is significantly countercyclical over the period of analysis. This is an important finding from a political perspective. Although our analysis is not done at such a fine level, these characteristics—densely populated, less educated on average, with more immigration and population growth—could be those of some underprivileged suburbs, which probably deserve more attention.

We also split the sample by the share who are religious.23 In his seminal work, Durkheim (1897) explained that religion, family, and certain political situations protect from death and particularly from suicide. High suicide rates are often cited as evidence of social failure (Case and Deaton 2015; Hamermesh and Soss 1974). Religion and the family are instances of integration of individuals who protect them from suicide by morally forbidding them to commit suicide. Here we start by splitting general mortality by a low or high proportion of Catholics, Muslims, and nonreligious people. For départements with a high proportion of nonreligious people and Muslims but a low share of Catholics, the relationship between unemployment rates and mortality is significantly countercyclical over the period of analysis. For suicides in particular (results available upon request), the relationship is significantly countercyclical only for those départements with a low proportion of nonreligious people. This means that a higher rate of unemployment (bad economic times) will increase suicides only in those départements with a low proportion of seculars, in accordance with Durkheim (1897) and surprisingly despite the evolution of the French society since the 1890s. Unemployment might thus be a better proxy for social anomie than for economic health.

A Tentative Exploration of Mechanisms

Does the Unemployment Rate Represent the Economic Cycle?

In most specifications, we use the unemployment rate as a proxy for local economic conditions. Because we include département-specific trends in our analysis and use the local unemployment rate, this does not necessarily proxy the “position in the business cycle” of the overall economy.24

Furthermore, other proxies of local economic conditions might yield different results. Section C in the online appendix examines this issue. Table C1 (online appendix) shows that using alternative indicators, such as the employment-to-population ratio, does not change our conclusions. Table C2 (online appendix) compares GDP and unemployment as regressors in different econometric specifications (1990–2014).25 Although the results with GDP per capita are not very stable, our preferred specification (column 6) shows that GDP per capita has a positive—and almost significant—effect on crude mortality. This result indicates that unemployment might not be the best proxy for economic cycles in France. Future research could examine quarterly data (instead of yearly mortality and unemployment) in order to examine the effect of other proxies of economic activity, such as strikes or school holidays (Adda 2016).

Own-Group Unemployment and Mortality

Following our analysis by causes, there is a priori no reason that own-group unemployment should matter more in France than elsewhere. Following Stevens et al. (2015), we look at how mortality rates for different subgroups respond to variation in that subgroup’s unemployment rate relative to variation in other groups’ unemployment rates. If most of the mortality effect is driven by changes in own behaviors, then a group’s own unemployment rate should have the strongest impact on that group’s mortality. We then reestimate Eq. (1) separately for men and women but also for different age-specific unemployment rates and mortality. Last, we examine the question of gender-specific mortality and its relationship with same- and opposite-gender unemployment rates. In this last test, we thus estimate cross-gender effects on gender-specific diseases within a single specification.

Table C3 (online appendix) provides the response of mortality by age and gender to gender- and age-specific unemployment. The results show no real support for own-group unemployment effects, consistent with findings of Stevens et al. (2015). Table C4 (online appendix) provides results for gender-specific causes of death. Two variables are used separately for this analysis. We look at the effect of gender-specific unemployment on gender-specific mortality using the same gender and the opposite gender in the same regression model. Again, we find no evidence of cross-gender effects or gender-specific effects.

Unemployment and Health Care Quality

We devote the rest of this discussion to exploring other possible mechanisms involved in whether health care quality could mediate the influence of unemployment on mortality. On the one hand, if government resources allocated to the health care sector are more constrained during economic crises, we should expect a negative link between increased unemployment and quality of care. Efforts such as those implemented to reduce the social security deficit might be more constraining during economic crises. Many strikes in the medical sector in France over the period are evidence of budgetary and human resources constraints. On the other hand, when unemployment is high, it becomes much easier to hire providers (perhaps not doctors but certainly orderlies, as Stevens et al. (2015) found). However, in France, there might be no relationship between provider density and unemployment because most health care professions are heavily regulated. For the United States, Stevens et al. (2015) looked at the effect of the unemployment rate on the employed in health occupations and found that staffing levels in nursing homes rise during periods of high unemployment but move procyclically for high-skilled health occupations.

As an exploratory attempt, Table C5 in the online appendix shows a number of variables (such as equipment rates, nursing services for elderly, or nurse density) that are thought to proxy the quality of health care or a certain pressure on budget or human resources at the département level.26 Our results confirm that in a context of high regulation, the relationship between provider density or equipment rates and unemployment is weak. We also find a negative coefficient—although not significant—for self-employed nurse density. The fact that only self-employed (private) activities seem to be affected more negatively by unemployment makes sense given that officials in the public sector can hardly become unemployed, which provides a kind of falsification test. Furthermore, we also test the effect of unemployment on the health insurance coverage of people (see online appendix, Table C6). We do not find any support for a cyclicality of universal insurance coverage or CMU-C in France. Last, including these variables directly in the mortality regressions does not change our conclusions (Table C7, online appendix).

Conclusions

This study explores the relationship between business cycles and mortality, using time series of official aggregate data. We use the case of France, a country with universal health care coverage and a highly regulated health care sector. Although this case had already been studied, we add more years, use novel estimation strategies, and explore pathways for our findings. The literature on business cycles and health is ever growing and has received considerable impetus with the Great Recession. Furthermore, the literature suggests that results may be quite context-specific and more complex than previously thought. Our work also contributes to demonstrating the importance of the econometric specification for previous findings and providing more evidence on the robustness of the results for different subsamples and demographic groups.

Are bad economic times bad for your health? The answer to this question is not obvious. In France, economic downturns seem to have little impact on aggregate mortality. France is indeed a useful case study regarding mortality and unemployment rates from 1982 to 2014. In summary, mortality rates in France are roughly acyclical, although some specifications reveal a countercyclical variation. We find some differences over time, with some evidence suggesting that the variations have moved from slightly procyclical to slightly countercyclical and then, for the most recent 15-year period studied, back to neutral. We find some differences by age and sex, but they are generally hard to characterize and show no clear patterns. We also find some differences across causes, but for the most part, they do not fall into clear patterns. France also looks a lot like the United States, with less procyclicality or more countercyclicality of mortality occurring in recent years. For mortality due to specific causes, such as traffic accidents, the findings are in line with prior research showing that mortality due to traffic accidents decreases during economic downturns. However, Stevens et al. (2015) showed that motor vehicle accidents account for less than 20 % of all cyclically induced deaths in the United States, and cyclical variation in non-elderly mortality due to other causes explains a very small fraction of the overall variation. Moreover, accident-specific mortality rates are lower in France than in the United States.

This convergence between our results and recent results from Ruhm (2015) for the United States could reflect the universality of insurance coverage in France, as the United States experienced a dramatic expansion of public health insurance over the past decades (Currie and Gruber 1996; Currie et al. 2018). Because health insurance coverage is mostly universal in France and not impacted by economic cycles, it might be logical that no effect or a weak effect is found when one looks at the impact of economic activity on mortality and health in general. Gerdtham and Ruhm (2006) showed that countries with stronger social insurance systems have less procyclicality of mortality. Recent improvements in social security in France (e.g., with the introduction of the Carte Vitale in 1998, Couverture Maladie Universelle-CMU in 2000, and Protection Universelle Maladie-PUMa in 2016) could thus explain part of this puzzle and partly and partly drive the differences between France and the United States. Moreover, heterogeneity matters in France, and this has not been discussed extensively elsewhere. For départements with low average education levels, high population density, high population growth, and a high share of migrants, the relationship between unemployment rates and mortality is significantly countercyclical over the period of analysis. This might imply that other less-visible conditional mechanisms could be at play, such as the quality of education, inequality, or demographic trends.

Our results include some limitations. Mortality rates measured at a small local area may be affected by migration. For instance, retired people may have migrated south and west to the band of départements known as the Sun Belt a few years before dying. However, this does not appear to be an issue except in two regions (Baccäıni 2001). Even to the extent that this type of migration occurs, there is little reason to expect it to be correlated with short-term changes in local unemployment rates because it likely concerns retired individuals (Buchmueller et al. 2007). Our regression coefficients are also particular to the period of analysis (1982–2014).

Acknowledgments

We thank the CépiDc for the mortality data used in this analysis. We thank Thomas Buchmueller for kindly sharing part of his data set with us for the replication of Buchmueller et al. (2007) provided in section B of the online appendix. We also thank Janet Currie for helpful guidance.

Notes

1

Brenner (1971) and Brenner and Mooney (1983) conducted similar research, but their empirical methods have subsequently been heavily criticized.

2

An exception to the rule that higher unemployment is associated with fewer deaths due to cardiovascular diseases might be Sweden (Svensson 2010), even though Tapia Granados and Ionides (2011) cast doubt on this conclusion.

3

Personal communication with Buchmueller (February 28, 2017) confirmed that the project was abandoned because this research group discovered that the results were not robust.

4

See Stevens et al. (2015) for a similar analysis in the United States.

5

In contrast to these more recent theories, earlier research tended to explain procyclical mortality chiefly with people’s own employment status. In this light, the direction of the effect of unemployment on health is not evident a priori. On the one hand, unemployment can be a psycho-social stressor that reduces quality of life and subjective well-being, possibly resulting in higher morbidity and mortality. Downward mobility could also be associated with increased morbidity and mortality. On the other hand, economic downturns could reduce stress and overtime hours, which should particularly reduce deaths due to cardiovascular diseases. Rising opportunity cost of time that accompanies better labor market opportunities might lead to higher mortality (Miller et al. 2009) because it makes it more costly for individuals to undertake time-intensive, health-producing activities. Finally, income growth may increase risky activities, such as drinking and driving (Ruhm and Black 2002). Stress-induced increases in alcohol consumption were found to be more than offset by reductions in incomes during economic crises.

6

We treat Corsica as one département because of data constraints.

7

These latter controls are obtained by interpolating census data from 1982, 1990, 1999, 2009, and 2014.

8

Coefficients in this specification continue to be interpreted as semi-elasticities. For a discussion of the advantages of PPML over a log-linear model estimated by ordinary least squares (OLS), see Manning and Mullahy (2001) and Santos Silva and Tenreyro (2006, 2011).

9

When we split our sample in two parts, the value was .03 from 1982 to 1998 and was stable from 1998 to 2014 (i.e., .03). Ruhm (2015) found this value to be .09 for the period 1999–2010 in the United States.

11

As an additional robustness check, we present results with the age-adjusted mortality rate as dependent variable in the online appendix (section D).

12

Anderson et al. (2001) provided comparability ratios for the United States. For the specific sources of mortality we are considering, most of the estimated comparability ratios are close to 1, suggesting that a similar number of deaths are reported using either ICD system.

14

For details on how unemployment statistics are calculated in France, see Fougère et al. (2009) or http://www.insee.fr/en/methodes/default.asp?page=sources/ope-taux-chomage-localises.htm.

15

In comparison, the average state in the United States had a population of 6,280,000; the average county, 100,000. The size of geographic unit is expected to have a moderate influence on the estimated coefficients. The channels through which the unemployment rate influences mortality might vary with the level of aggregation. Smaller geographical units will have higher spillover effects (Lindo 2015).

16

Section B of the online appendix shows that the inclusion of département-specific time trends similarly changes most of the results of Buchmueller et al. (2007).

17

See also Ruhm (2015) for a discussion of the influence of weights on the unemployment coefficient for the case of the United States.

18

Another reason for weighting that is sometimes given is to achieve more efficient estimation given heteroskedasticity. However, as shown in Table 1, the standard errors between WLS and OLS estimates hardly differ, indicating that weighting by area population does not do a better job of dealing with heteroskedasticity. See Solon et al. (2015) for a discussion of the relative advantages of WLS and OLS.

19

As an additional check, we ran the département-level regression with the full set of controls, weighting, year fixed effects, and département fixed effects, but we replaced the département-specific time trends by region-specific time trends. This yielded a coefficient estimate of 0.0014 (standard error = 0.0011). This should make our analysis more robust to the claim that the département-specific time trends take too much variation out of the analysis. Unlike in panel B of Table 1, the old regions were used to create the region-specific time trends because these were the administrative levels in place during the analysis period. Using the new regions for the time trends in an otherwise similar specification yields an estimate of –0.0001 (standard error = 0.0015).

20

In unreported regressions, we also used the nontransformed mortality rate as a dependent variable and estimated by OLS. However, because mortality rates can be viewed as count data, we prefer to use a count model and thus keep the preferred functional form and the interpretation of coefficients as semi-elasticities.

21

In unreported results, we found that the coefficients steadily decline with age. Younger people who are inexperienced drivers are involved in more accidents. It also makes sense that the effect would be weakest among people above the retirement age given that this demographic group is least likely to participate in auto travel, partly because they no longer commute to work.

22

As a robustness check, we created the same figures excluding the weighting by population. Results were qualitatively the same but coefficients were closer to 0, as can be expected from Table 1.

23

The data were provided by IFOP based on available data up to 2009 (Institut Français d’Opinion Publique: Eléments d’analyse géographique de l’implantation des religions en France, unpublished).

24

Interestingly, in his reanalysis of his 2000 paper, Ruhm (2015) actually added such a state-specific time trend.

25

We can compare unemployment with GDP by region only from 1990 to 2014.

26

Importantly, we did not have access to the full time series for all variables, but this is the most complete data set we can find for France.

References

Adda, J. (
2016
).
Economic activity and the spread of viral diseases: Evidence from high frequency data
.
Quarterly Journal of Economics
,
131
,
891
941
. 10.1093/qje/qjw005
Anderson, R. N., Mininõ, A. M., Hoyert, D. L., & Rosenberg, H. M. (
2001
).
Comparability of cause of death between ICD-9 and ICD-10: Preliminary estimates
(National Vital Statistics Reports, No. 49/2).
Hyattsville, MD
:
National Center for Health Statistics
.
Ariizumi, H., & Schirle, T. (
2012
).
Are recessions really good for your health? Evidence from Canada
.
Social Science & Medicine
,
74
,
1224
1231
. 10.1016/j.socscimed.2011.12.038
Baccäıni, B. (
2001
).
Les migrations internes en France de 1990 à 1999: L’appel de l’Ouest [Internal migration in France from 1990 to 1999: The call of the West]
.
Economie et Statistique
,
344
,
39
79
.
Brenner, M. H. (
1971
).
Economic changes and heart disease mortality
.
American Journal of Public Health
,
61
,
606
611
. 10.2105/AJPH.61.3.606
Brenner, M. H., & Mooney, A. (
1983
).
Unemployment and health in the context of economic change
.
Social Science & Medicine
,
17
,
1125
1138
. 10.1016/0277-9536(83)90005-9
Buchmueller, T., Grignon, M., & Jusot, F. (
2007
).
Unemployment and mortality in France, 1982–2002
(
Working Paper Series, 2007-04
).
Hamilton, Canada
:
Centre for Health Economics and Policy Analysis, McMaster University
.
Burgard, S. A., Brand, J. E., & House, J. S. (
2009
).
Perceived job insecurity and worker health in the United States
.
Social Science & Medicine
,
69
,
777
785
. 10.1016/j.socscimed.2009.06.029
Case, A., & Deaton, A. (
2015
).
Suicide, age, and wellbeing: An empirical investigation
(NBER Working Paper No. 21279).
Cambridge, MA
:
National Bureau of Economic Research
.
Currie, J., & Gruber, J. (
1996
).
Saving babies: The efficacy and cost of recent changes in the Medicaid eligibility of pregnant women
.
Journal of Political Economy
,
104
,
1263
1296
. 10.1086/262059
Currie, J., Schwandt, H., & Thuilliez, J. (
2018
).
Pauvreté, égalité, mortalité: Mortality (in)equality in France and the United States
(NBER Working Paper No. 24623).
Cambridge, MA
:
National Bureau of Economic Research
.
Deryugina, T., Heutel, G., Miller, N., Molitor, D., & Reif, J. (
2016
).
The mortality and medical costs of air pollution: Evidence from changes in wind direction
(NBER Working Paper No. 22796).
Cambridge, MA
:
National Bureau of Economic Research
.
Durkheim, E. (
1897
).
Le suicide: Étude de sociologie
[Suicide: Sociology study].
Paris, France
:
F. Alcan
.
Fougère, D., Kramarz, F., & Pouget, J. (
2009
).
Youth unemployment and crime in France
.
Journal of the European Economic Association
,
7
,
909
938
. 10.1162/JEEA.2009.7.5.909
Gerdtham, U-G, & Ruhm, C. J. (
2006
).
Deaths rise in good economic times: Evidence from the OECD
.
Economics & Human Biology
,
4
,
298
316
. 10.1016/j.ehb.2006.04.001
Hamermesh, D. S., & Soss, N. M. (
1974
).
An economic theory of suicide
.
Journal of Political Economy
,
82
,
83
98
. 10.1086/260171
Heutel, G., & Ruhm, C. J. (
2016
).
Air pollution and procyclical mortality
.
Journal of the Association of Environmental and Resource Economists
,
3
,
667
706
. 10.1086/686251
Ionides, E. L., Wang, Z., & Tapia Granados, J. A. (
2013
).
Macroeconomic effects on mortality revealed by panel analysis with nonlinear trends
.
Annals of Applied Statistics
,
7
,
1362
1385
. 10.1214/12-AOAS624
Lam, J-P, & Piérard, E. (
2017
).
The time-varying relationship between mortality and business cycles in the USA
.
Health Economics
,
26
,
164
183
. 10.1002/hec.3285
Lin, S-J (
2009
).
Economic fluctuations and health outcome: A panel analysis of Asia-Pacific countries
.
Applied Economics
,
41
,
519
530
. 10.1080/00036840701720754
Lindo, J. M. (
2015
).
Aggregation and the estimated effects of economic conditions on health
.
Journal of Health Economics
,
40
,
83
96
. 10.1016/j.jhealeco.2014.11.009
Manning, W. G., & Mullahy, J. (
2001
).
Estimating log models: To transform or not to transform?
.
Journal of Health Economics
,
20
,
461
494
. 10.1016/S0167-6296(01)00086-8
Miller, D. L., Page, M. E., Stevens, A. H., & Filipski, M. (
2009
).
Why are recessions good for your health?
.
American Economic Review: Papers & Proceedings
,
99
,
122
127
. 10.1257/aer.99.2.122
Neumayer, E. (
2004
).
Recessions lower (some) mortality rates: Evidence from Germany
.
Social Science & Medicine
,
58
,
1037
1047
. 10.1016/S0277-9536(03)00276-4
Ogburn, W. F., & Thomas, D. S. (
1922
).
The influence of the business cycle on certain social conditions
.
Journal of the American Statistical Association
,
18
,
324
340
. 10.1080/01621459.1922.10502475
Regidor, E., Vallejo, F., Tapia Granados, J. A., Viciana-Fernández, F. J., de la Fuente, L., & Barrio, G. (
2016
).
Mortality decrease according to socioeconomic groups during the economic crisis in Spain: A cohort study of 36 million people
.
Lancet
,
388
,
2642
2652
. 10.1016/S0140-6736(16)30446-9
Ruhm, C. J. (
1995
).
Economic conditions and alcohol problems
.
Journal of Health Economics
,
14
,
583
603
. 10.1016/0167-6296(95)00024-0
Ruhm, C. J. (
2000
).
Are recessions good for your health?
.
Quarterly Journal of Economics
,
115
,
617
650
. 10.1162/003355300554872
Ruhm, C. J. (
2003
).
Good times make you sick
.
Journal of Health Economics
,
22
,
637
658
. 10.1016/S0167-6296(03)00041-9
Ruhm, C. J. (
2005
).
Healthy living in hard times
.
Journal of Health Economics
,
24
,
341
363
. 10.1016/j.jhealeco.2004.09.007
Ruhm, C. J. (
2007
).
A healthy economy can break your heart
.
Demography
,
44
,
829
848
. 10.1007/BF03208384
Ruhm, C. J. (
2015
).
Recessions, healthy no more?
.
Journal of Health Economics
,
42
,
17
28
. 10.1016/j.jhealeco.2015.03.004
Ruhm, C. J., & Black, W. E. (
2002
).
Does drinking really decrease in bad times?
.
Journal of Health Economics
,
21
,
659
678
. 10.1016/S0167-6296(02)00033-4
Santos Silva, J. M. C., & Tenreyro, S. (
2006
).
The log of gravity
.
Review of Economics and Statistics
,
88
,
641
658
. 10.1162/rest.88.4.641
Santos Silva, J. M. C., & Tenreyro, S. (
2011
).
Further simulation evidence on the performance of the Poisson pseudo-maximum likelihood estimator
.
Economics Letters
,
112
,
220
222
. 10.1016/j.econlet.2011.05.008
Solon, G., Haider, S. J., & Wooldridge, J. M. (
2015
).
What are we weighting for?
.
Journal of Human Resources
,
50
,
301
316
. 10.3368/jhr.50.2.301
Stevens, A. H., Miller, D. L., Page, M. E., & Filipski, M. (
2015
).
The best of times, the worst of times: Understanding pro-cyclical mortality
.
American Economic Journal: Economic Policy
,
7
(
4
),
279
311
.
Sullivan, D., & von Wachter, T. (
2009
).
Job displacement and mortality: An analysis using administrative data
.
Quarterly Journal of Economics
,
124
,
1265
1306
. 10.1162/qjec.2009.124.3.1265
Svensson, M. (
2010
).
Economic upturns are good for your heart but watch out for accidents: A study on Swedish regional data 1976–2005
.
Applied Economics
,
42
,
615
625
. 10.1080/00036840701704519
Tapia Granados, J. A. (
2008
).
Macroeconomic fluctuations and mortality in postwar Japan
.
Demography
,
45
,
323
343
. 10.1353/dem.0.0008
Tapia Granados, J. A., & Ionides, E. L. (
2011
).
Mortality and macroeconomic fluctuations in contemporary Sweden
.
European Journal of Population / Revue Européenne de Démographie
,
27
,
157
184
. 10.1007/s10680-011-9231-4
Tapia Granados, J. A., & Rodriguez, J. M. (
2015
).
Health, economic crisis, and austerity: A comparison of Greece, Finland and Iceland
.
Health Policy
,
119
,
941
953
. 10.1016/j.healthpol.2015.04.009

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary data