Show Summary Details

Page of

PRINTED FROM the OXFORD RESEARCH ENCYCLOPEDIA, ECONOMICS AND FINANCE (economics.oxfordre.com). (c) Oxford University Press USA, 2016. All Rights Reserved. Personal use only; commercial use is strictly prohibited (for details see Privacy Policy and Legal Notice).

date: 15 October 2018

The Business Cycle and Health

Summary and Keywords

The impact of macroeconomic fluctuations on health and mortality rates has been a highly studied topic in the field of economics. Many studies, using fixed-effects models, find that mortality is procyclical in many countries, such as the United States, Germany, Spain, France, Pacific-Asian nations, Mexico, and Canada. On the other hand, a small number of studies find that mortality decreases during economic expansion. Differences in the social insurance systems and labor market institutions across countries may explain some of the disparities found in the literature. Studies examining the effects of more recent recessions are less conclusive, finding mortality to be less procyclical, or even countercyclical. This new finding could be explained by changes over time in the mechanisms behind the association between business cycle conditions and mortality.

A related strand of the literature has focused on understanding the effect of economic fluctuations on infant health at birth and/or child mortality. While infant mortality is found to be procyclical in countries like the United States and Spain, the opposite is found in developing countries.

Even though the association between business cycle conditions and mortality has been extensively documented, a much stronger effort is needed to understand the mechanisms behind the relationship between business cycle conditions and health. Many studies have examined the association between macroeconomic fluctuations and smoking, drinking, weight disorders, eating habits, and physical activity, although results are rather mixed. The only well-established finding is that mental health deteriorates during economic slowdowns.

An important challenge is the fact that the comparison of the main results across studies proves to be complicated due to the variety of empirical methods and time spans used. Furthermore, estimates have been found to be sensitive to the use of different levels of geographic aggregation, model specifications, and proxies of macroeconomic fluctuations.

Keywords: business cycle conditions, mortality rates, infant health, mental health, weight disorders, addictions

Introduction

In the last several decades, there has been much interest in the impact of macroeconomic fluctuations on health and mortality rates. Ruhm (2000), using for the first time fixed-effects models, found that health improves during temporary economic slowdowns. Other studies such as Neumayer (2004; Germany), Granados (2005; Spain), Buchmueller, Grignon, and Jusot (2007; France), Lin (2009; Asia-Pacific nations), Gonzalez and Quast (2011; Mexico), and Ariizumi and Schirle (2012; Canada) also found that total mortality is procyclical. However, there are some exceptions; a smaller number of studies find that mortality is countercyclical, for example, in Sweden (Gerdtham & Johannesson, 2005; Svensson, 2007) and 13 countries of the European Union (Economou, Nikolaou, & Theodossiou, 2008).

Yet studies examining the effects of more recent recessions are less conclusive, finding mortality to be less procyclical (Stevens, Miller, Page, & Filipski, 2015), countercyclical (McInerney & Mellor, 2012; Svensson, 2007), or even unrelated to macroeconomic conditions (Ruhm, 2015). Ruhm (2015) suggests that changes over time in the relationship between business cycle conditions and mortality could be driven by the fact that the instability over time is poorly measured when using short periods of analysis. Another possible reason could be that the mechanisms behind the associations between macroeconomic conditions and mortality are not stable over time due to the changes in institutions or changes in health behaviors (or in any other determinant of health outcomes). Similarly, the impact of mild recessions on mortality might be different than the impact of strong economic crisis (Ruhm, 2016).

Part of the literature on business cycle conditions and health has focused on understanding the effect of economic fluctuations on infant health at birth and/or infant and child mortality. It is important to note that for infant health at birth, an additional mechanism can play an important role: selection. The type of mother that conceives during recessions might be different from those that get pregnant during economic expansions. Then, differences in health outcomes at birth through the business cycle may arise as a consequence of an endogenous composition of births. The seminal paper by Dehejia and Lleras-Muney (2004) finds that babies’ health at birth improves during recessions in the United States (and infant mortality is procyclical). This finding is consistent with the positive health outcomes found for adults. The same conclusion has been reached for Spain (Aparicio-Fenoll & Gonzalez, 2014). However, infant mortality is found to be countercyclical in some developing countries (Baird, Friedman, & Schady, 2011; Bhalotra, 2010; Bozzoli & Quintana-Domeque, 2014; Paxson & Schady, 2005).

While the association between business cycle conditions and mortality has been extensively documented, the concrete mechanisms behind these associations remain poorly understood. A possible mechanism is that during bad economic conditions health-producing activities are less costly due to a decrease in the opportunity cost of time. As a consequence, many studies have examined the association between macroeconomic fluctuations and smoking, drinking, weight disorders, eating habits, and physical activity. However, results are rather mixed (Arkes, 2009; Böckerman et al., 2007; Charles & DeCicca, 2008; Latif, 2014; Ruhm, 2000, 2005).

The reduction in income that can experience some individuals during downturns could be another important mechanisms through which business cycle conditions affect health. Even if the individuals manage to remain employed, they may still suffer from stress, depression from unemployment risk, and economic insecurity because of the recessionary environment. Other studies have explored how these associations differ depending on the individuals’ labor force status to evaluate the importance of the income and stress mechanisms. While it is well established that mental health decreases during economic slowdowns (Charles & DeCicca, 2008; McInerney & Mellor, 2012; Ruhm, 2000, 2003), the impact of the individual’s employment status is less conclusive. Thus, a much stronger effort is needed to understand the mechanisms behind the relationship between business cycle conditions and health.

An additional important drawback of the literature is that comparisons of the main results across studies are problematic given the variety of empirical methods used. For one, the level of geographic aggregation chosen will influence not only the results but also the interpretation of those findings (Lindo, 2015). While more disaggregated measures of economic conditions tend to give more precise estimates, there could be spillover effects across these small regional units. Also, many studies include regional-specific linear time trends in their model specifications as an attempt to control for unobserved factors that vary within regions over time. However, one potential issue is that these trends might absorb most of the within-region variation in unemployment rates. Finally, there is no consensus on the best proxy for macroeconomic fluctuations (unemployment rates, employment-to-population ratios, changes in regional real GDP, among others), which indicate the lack of robustness of the results when several of these measures are used.

Business Cycle and Mortality

In recent decades there has been much interest in estimating the effect of short-term economic fluctuations on health and mortality rates. Ruhm (2000)’s path-breaking work, using panel data and controlling for time-invariant state-specific effects, was able to overcome some of the omitted variable biases from which previous time-series analysis suffered (Brenner, 1973, 1975, 1979, 1987, 1995). Using a 20-year period (1972–1991) panel dataset for the 50 US states and the district of Colombia, Ruhm (2000) estimates that the total death rate behaves procyclically. Furthermore, 8 of the 10 specific causes of fatalities that he distinguishes are also procyclical. In particular, he finds that a one percentage point decrease in the state unemployment rate is associated with a 0.5% increase in the total mortality rate and with a 3, 1.9, 1.6, 0.7, and 0.5% increase in the predicted mortality from motor vehicle crashes, homicides, other accidents, influenza/pneumonia, and cardiovascular diseases, respectively. As expected, he finds no significant variation in cancer fatalities, which are unlikely to respond to short-term changes in health-enhancing habits. On the other hand, a one percentage point increase in the unemployment rate is predicted to increase by 1.3% the incidence of suicides. The countercyclicality of suicides indicates that recessions, while improving physical well-being, could have negative effects on mental health. Finally, the procyclicality of overall mortality seems to be larger among young adults (20–44 years). Miller, Page, Stevens, and Filipski (2009) use the same methodology as Ruhm but extend the analysis until 2004 and find a strong procyclical pattern among young adults (20–34 years) and children, precisely the age groups that are less likely to be working. Very similarly, Stevens et al. (2015) find that the own-group unemployment rate is not significantly positively related to that group’s mortality. These results suggest that there might be other mechanisms besides individual work and leisure that may be playing an important role in procyclicality. In addition, Stevens et al. (2015) document that the cyclical variation among the elderly living in nursing homes, a group with very low labor force attachment, accounts for a large part of the total cyclical variation in mortality.

Granados (2005) finds similar results pooling data from the 50 Spanish provinces over the period 1980–1997. However, the influence of business cycle conditions on total mortality seems to be less pronounced for the case of Spain. A drop of 0.11% in mortality is estimated from an increase in one percentage point in the province unemployment rate. Using very similar methods, Neumayer (2004) confirms the procyclical fluctuation in total mortality rates, using aggregate panel data for 16 German states from 1980 to 2000. He documents that a 1% decline in unemployment rate is associated with an increase in mortality by 0.7–1.1%, a slightly larger effect than found by to Ruhm (2000). Moreover, he finds not only deaths from cardiovascular diseases, influenza/pneumonia (only for males), and motor vehicle accidents (only for females) to be procyclical, but also suicides (contradicting, in this last association, Ruhm’s results). Using aggregate panel data from 1982 to 2002, the procyclicality of total mortality has been also confirmed for France (Buchmueller et al., 2007). Ariizumi and Schirle (2012) also find a procyclical relationship with total mortality for middle-aged individuals (30–39 years) in Canada, while they do not find any significant relationship for infant and seniors.

The procyclicality of mortality is also confirmed in less developed countries. Lin (2009), using panel data from eight Asia-Pacific countries for the years 1976–2003, documents mortality decreases during recessions as well as some cause-specific mortality rates such as those for cardiovascular diseases, motor vehicle accidents, and infant mortality. Gonzalez and Quast (2011) find that for Mexico, total mortality rates are procyclical, with the largest impact on those aged 20–49 years.

On the other hand, Gerdtham and Johannesson (2005) find evidence of a countercyclical variation of mortality in Sweden. Using individual level data from 1980 to 1996 on the annual mortality risk, they find that mortality risk increases during recessions but only for males and when using alternative business cycle indicators, such as the 135 rate, the industrial capacity utilization rate, a confidence indicator, and the change in GDP. They find no effect for women or when using other measures of business cycle conditions such as the unemployment rate or deviations from the GDP trend. Regarding cardiovascular mortality, they also find a countercyclical relationship for men when using three proxies of the business cycle conditions. However, some of the differences in their findings with respect to the previous literature could be explained by their use of individual data (instead of aggregated regional data) and the different time period analyzed. Nevertheless, Svensson (2007) confirms these last results for Sweden using similar methods as those used in the previous studies from the United States, Germany, or Spain (Gerdtham & Ruhm, 2006; Granados, 2005; Neumayer, 2004; Ruhm, 2000). Using aggregate panel data from the 21 Swedish regions during the period 1987–2003, he finds a countercyclical relationship for incidence, mortality, and lethality in acute myocardial infarction only for those in prime working ages (20–49 years).

Using a panel data of 13 European Union countries from 1977 to 1996, Economou et al. (2008) also find a positive relationship between recessions and total mortality. Besides using a fixed-effect model, they also control for potential confounders, such as lifestyle risk factors (smoking, alcohol consumption, and caloric intake), urbanization (population density and carbon dioxide emissions), and medial intervention indicators (inpatient care admissions).

Why do we find a countercyclical relationship of total mortality in these countries? One potential explanation is that these countries have very different social insurance systems and labor market institutions. The macroeconomic effects could be mitigated by the availability of more comprehensive health insurance (in contrast to a employment-based health system), income-replacement policies, or labor protection laws, such as maximum allowable hours/days of work, more stringent job safety regulations, or regulated worker termination policies. In these countries with more social protection, individuals might feel less pressure to work harder during booms, as they don’t have to compensate so much for the decrease in income during recessions. Thus, institutional factors may help to reduce or reverse the procyclical variation in mortality. In fact, this is precisely what Gerdtham and Ruhm (2006) state in their study. Gerdtham and Ruhm (2006), using data from 23 OECD countries between 1960 and 1997, find similar results to Ruhm (2000). A 1% point decrease in the national unemployment rate is associated with a rise of 0.4, 0.4, 1.1, 1.1, and 0.8% in total mortality and deaths from cardiovascular diseases, influenza/pneumonia, liver diseases, motor vehicle deaths, and other accidents, respectively. Yet they show that the procyclical fluctuation is stronger in countries with a higher average public social expenditure as a percentage of their GDP (proxy for a strong social insurance system), which might explain some of the disparities in the effects found for different countries and time spans.

Some recent studies suggest that mortality has become less procyclical or even countercyclical when including more recent recessions. McInerney and Mellor (2012) using both aggregate and micro-level data from the United States, find that a one percentage point increase in the state unemployment rate lowered the mortality rate of elderly individuals by 0.27% during 1976–1991 but raised it by 0.49% from 1994 to 2008. Moreover, they estimate that elderly persons’ mental health deteriorates during recessions and they are not more likely to engage in healthier behaviors. These findings suggest that either seniors’ health responds differently to business cycle fluctuations than prime-age working adults or the relationship between economic conditions and health has changed in the last several decades. Ruhm (2015) confirms the latter suspicion raised by McInerney and Mellor. Using panel data covering 1976 to 2010, he documents that overall mortality has shifted from being strongly procyclical in 1976–1993 to being unrelated to economic conditions in recent years (1991–2010). Moreover, he finds that, while fatalities due to motor vehicle accidents and cardiovascular diseases continue to be procyclical (although weaker), cancer and other external sources of death have become countercyclical. Similarly, Tekin, McClellan, and Minyard (2013), using data from the Behavioral Risk Factor Surveillance System (BRFSS) from 2005 and 2011, argue that during this period the relationship between self-reported general and mental health and the economy was essentially zero (thought estimates are small and imprecisely estimated). They also report some differences among various subpopulation and age groups. Lam and Piérard (2017) confirm these changes in the relationship between mortality and the economy over time using tests for structural breaks. They show that a structural change occurred around 1995, after which mortality no longer seems procyclical. They also suggest some of the potential factors that contributed to this change in the relationship: advances in medical technology and changes in the effect that working hours have on health.

Finally, it is important to note that the majority of previous work has mainly focused on the effect of recessions on objective health measures, such as mortality rates. Yet one common result is that fatalities due to suicides are countercyclical or unrelated to macroeconomic conditions. This suggests that recessions have different effects on mental health than on physical health. Charles and DeCicca (2008), using micro-data from the National Health Interview Survey in the 58 largest metropolitan statistical areas in the United States from 1997 to 2001, find that mental health is procyclical. Moreover, this effect is more pronounced for individuals with low employment probabilities, African Americans, and individuals with low education. Bradford and Lastrapes (2014) also show that mental health drug prescriptions in the US Northeast increase during bad economic conditions. Similarly, doctor visits related to mental health diagnoses also increase during recessions (though the effect is weaker). These findings are consistent with economic stress mechanisms being more important than the reduction of the opportunity cost of time for mental health issues, as opposed to what is reported for physical health.

To sum up, while there is strong evidence of procyclical fluctuations of total mortality when examining data from the 1970s to the 2000s for different countries, the relationship has changed in recent years. The mechanisms that explain this relationship as well as the reasons for the recent change remain poorly understood. Thus, these are important questions for future research.

Business Cycle and Infant Mortality

Some studies have focused their attention on understanding the effect of business cycle conditions over infant health at birth and/or infant and child mortality. There are three main mechanisms through which economic fluctuations might be impacting infant health. First, recessions might have a direct income effect for the household, reducing their consumption of nutritious foods and lowering expenditures on other expensive inputs that can contribute positively to children’s health outcomes. Thus, lower incomes at the household level could imply worse health outcomes for children. However, if income is reduced, the number of times that the family can afford to eat in bars and restaurants might decrease, leading to an improvement in child nutrition (typically home-cooked food is healthier). Second, economic downturns could lower the opportunity cost of parent’s time. During recessions, parents that lose their jobs could do more time-consuming activities that directly benefit the health of their children, such as breastfeeding, cooking healthy meals, and taking the children to the doctor, among other health-enhancing activities. Finally, differences in health outcomes at birth throughout the business cycle may arise as a consequence of an endogenous composition of births. Families could defer fertility or have a higher incidence of fetal deaths during recessions. The selection effect is stronger for women with poor economic backgrounds, which have a higher probability of experiencing job insecurity and negative labor market prospects during recessions. These women have also an ex ante higher risk of experiencing infant mortality or having children with deteriorated health status. Then, during recessions women with low economic status might decide to deter or postpone conception, resulting in a reduction of infant mortality and healthier live births. This selection effect should be weaker for women with high economic status.

The seminal study by Dehejia and Lleras-Muney (2004) using panel data for the United States from 1975 to 1999 finds that babies’ health at birth improves during recessions. A higher unemployment rate at the time of conception is associated with a reduction in the incidence of low- and very-low-birthweight babies as well as lower infant mortality. Dehejia and Lleras-Muney (2004) show that this countercyclicality of infant health could be attributed to both selection (smaller proportion of babies from low-educated black mothers born during recessions) as well as to changes in maternal health behaviors during pregnancy (improvements in prenatal care and reductions in smoking and drinking during pregnancy in recessionary periods). For Spain, Aparicio-Fenoll and Gonzalez (2014) also find infant health to be countercyclical, even though these two countries have very different healthcare systems and labor markets features. Aparicio-Fenoll and Gonzalez (2014) also show changes in the composition of parents during the cycle; recessions lead to lower fertility among low-skilled parents. To closely examine if selection could be driven the results, they match multiple births from the same parents that take place at different points of the cycle and include parents’ fixed effects (note that Dehejia & Lleras-Muney, 2004, only controlled for mother fixed effects for California and a very short time span). Given that some of the results survive to the inclusion of parental fixed effects, they conclude that selection plays a role but does not account for the entire countercyclical relationship between infant health and business cycle conditions. They further examine some potential behavioral channels and show that while maternal employment is not the main mediating channel, women seem to engage in healthier behaviors during recessions (they smoke, drink, and weigh less and sleep more).

These first two studies focused on developed countries, but negative economic shocks might affect infant health in developing countries differently. In these countries, cyclical fluctuations tend to be larger and more severe. Furthermore, there are no (or fewer) state or market insurance systems against macroeconomic shocks, so that individuals might have more difficulties in smoothing these fluctuations through borrowing. The majority of the studies examining the relationship between business cycle conditions and infant mortality in developing countries find a countercyclical relationship. For instance, Baird et al. (2011), using micro-level survey data for 59 developing countries, find that a 1% decrease in per capita GDP results in an increase in infant mortality of between 0.24 and 0.40 per 1,000 children born. An important result is that girls’ mortality is more sensitive to economic fluctuations than boys’ mortality, and this is particularly true for negative economic shocks. Finding very similar results, Paxson and Schady (2005) examine the effect on infant mortality of a severe crisis that took place in the late 1980s in Peru. They show that, during the crisis, infant mortality increased by 2.5 percentage points. They suggest that the collapse in public and private expenditures on health could have been one of the mechanisms to explain the increase in infant mortality. However, they find no evidence that changes in the consumption of food or in the composition of mothers giving birth played a significant role in the countercyclical relationship. Similarly, other infectious diseases or terrorism attacks were not important determinants in explaining infant health outcomes and business cycle fluctuations. Using siblings born at different phases of the business cycle to control for the potential endogenous heterogeneity in the composition of live births, Bhalotra (2010) finds that rural infant mortality in India is countercyclical. A decrease of one standard deviation in the low aggregate income is estimated to increase infant mortality risk by 1.2 percentage points. This effect is greater for children of uneducated or teenage mothers and for girls. This result mimics the one by Baird et al. (2011) confirming that in developing countries girls are less protected during recessions than boys. These results are very important, especially considering the fact that births of uneducated parents and women of scheduled tribes are underrepresented in recessions, which indicates that high-risk women are either timing fertility or suffering from fetal loses. This, in turn, lowers death risk. On the other hand, mothers do not seem to be able to time their labor supply to maximize infant health. Balotra (2010) finds that maternal labor supply is countercyclical, especially for mothers that work in agriculture. Moreover, recessions also lower antenatal and postnatal health-seeking behaviors. Thus, the increase in maternal labor supply during recessions could increase the relative price of child health, which reinforces the direct income effect of recessions, leading to an increase in rural infant mortality. Finally, Bozzoli and Quintana-Domeque (2014) study the impact of the 2001–2002 economic crisis in Argentina on birthweight. They find that birthweight is procyclical (thus the prevalence of low-birthweight countercyclical). Previous literature has pointed out that nutritional deficits, as a consequence of an economic downturn, affect birthweight more if they are experienced during the third trimester of pregnancy (Almond, Hoynes, & Schanzenbach, 2011; Stein & Lumey, 2000), while maternal psychosocial stress affects birthweight more if it occurs during the first trimester of pregnancy (Camacho, 2008; Torche, 2011). Following this previous literature, Bozzoli and Quintana-Domeneque (2014) document that economic fluctuations during the first trimester matter for both low- and high-educated mothers, suggesting that maternal stress affects birthweight independently of the mother’s educational status. However, fluctuations during the third trimester only matter for low-educated mothers, which is consistent with nutritional deficiencies affecting only more credit-constrained households.

On the other hand, Miller and Piedad Urdinola (2010) find a countercyclical relationship between business cycle conditions and infant health in Colombia. As Colombia is the second (after Brazil) producer of Arabica coffee, they exploit three external fluctuations in coffee prices as a proxy for the business cycle phase. They also use cohort size as a measure of infant and child mortality. They find that infant and child mortality is procyclical (or countercyclical in cohort size). They also show that when coffee prices rise, mothers have a higher probability of working and the number of work hours increases. These labor market changes indicate that those mothers have less time to engage in health-promoting behaviors for their children (child care when they are ill, health investment, and vaccination use). They claim that this increase in the opportunity cost of time during booms is one of the main drivers of the procyclical relationship documented for infant mortality. They emphasize that this effect could be common in developing countries where the determinants of child health might be not very expensive but are very time-consuming.

Mechanisms

From a theoretical point of view, health (and mortality) can be affected by macroeconomic conditions through different channels that can act in opposite directions. Yet very little is known about the precise mechanisms for the estimated effects.

For one, recessions lead to changes in health behaviors due to a decrease in the opportunity cost of time during recessions. During economic downturns, non-market leisure time increases, decreasing the cost of individuals to undertake time-intensive health-producing activities such as physical exercise or preparing homemade meals. This mechanism will predict health to be countercyclical. In addition, the number of work-related injuries or deaths and the number of traffic accidents are expected to decrease during recessions. On the other hand, the loss of income that some individuals experience during recessions could affect consumption in general, but particularly the consumption of health-enhancing goods, leading to a deterioration of health status.

Finally, stress and risk-taking behavior increases during economic downturns. Individuals suffering from financial hardships and/or employment uncertainty during bad economic times could use alcohol, tobacco, or drugs per se or for self-medication to ease the distress and anxiety they are experiencing. At the same time, stress experienced while working could also have negative effects, particularly if working hours are extended. Then, job-related stress could affect negatively the workers’ health during recessions if hours are extended for those individuals that managed to stayed employed or during short-lasting economic expansions until the firms adapt and hire more employees. Therefore, it is not clear if the economic stress argument would predict procyclicality or countercyclicality of health.

In this section we summarize the main findings in the literature regarding the mechanisms behind the relationship between health and business cycle conditions.

Weight Disorders, Exercise, and Diet

Some studies have examined the importance of the opportunity cost of time mechanism analyzing the effect of business cycle conditions on adults’ weight (the most direct consequence of physical exercise and diet). The pioneering work of Ruhm (2000) shows, using US microdata from the BRFSS from 1987 to 1995, that a one percentage point increase in the state unemployment rate reduces body mass index (BMI) by 0.016 and lowers the probability of being underweight, overweight, and obese by 0.06, 0.17, and 0.21 percentage points, respectively. In a follow-up paper Ruhm (2005) confirms the procyclicality of these weight disorders during a larger time span (1987–2000), although he provides evidence that these declines in body weight are concentrated among obese and severely obese individuals.

Other studies find that adults’ weight is countercyclical in Finland, the United States, and Canada. Böckerman et al. (2007) use the National Public Health Institute micro-level data from 1978 to 2002 and find that an improvement in regional economic conditions in Finland, measured by employment rates, decreases BMI. This result is in contrast with those for the United States. Similarly, using data from the US National Health Interview Survey from 1997 to 2001, Charles and DeCicca (2008) find that overweight, obesity, and underweight are countercyclical, while normal weight behaves in the same direction as macroeconomic conditions. However, they do not find any significant impact of recessions on BMI or on the incidence of severe and very severe obesity. Likewise, Latif (2014) finds that an increase in the province-level unemployment rate in Canada increases BMI and the probability of individuals being severely obese. However, he finds no significant relationship to the probability of being obese and overweight. Finally, Tekin et al. (2013) explores this relationship using the BRFSS and covering the time span of the Great Recession (1990–2014). They find that weight is independent of changes in the economic situation of the country.

The majority of studies tend to focus on the effect of macroeconomic conditions on weight changes for adults. However, these effects should not necessarily be the same for teenagers and children. In fact, Arkes (2009), using the US National Longitudinal Survey of Youth from 1997 to 2004, finds that teenage girls gain weight in weaker economic periods (having a higher probability of being overweight and a lower probability of being underweight), while boys gain weight in stronger economic periods. Moreover, Byron and Fertig (2012), when analyzing three waves of the Child Development Supplement of the Panel Study of Income Dynamics (1997–2007) and controlling for children fixed effects, show that recessions are associated with lower weight but only for children in households with debt. Finally, Bellés-Obrero, Jiménez-Martín, and Vall-Castello (2016) examine the effect of business cycle conditions on Spanish children’s (2–15 years old) health using cross-sectional data from the National Health Surveys for the period 1987 to 2012. They find that obesity is procyclical and underweight is countercyclical, and these effects are only significant for children younger than 6 and older than 12 years old. Thus, the effects of recessions on Spanish children’s weight are both positive and negative, reducing obesity but increasing underweight at the same time. As can be concluded from the studies reviewed above, it is difficult to draw a clear conclusion about the effect of business cycle conditions on weight disorders for adults (or children), as the findings are quite mixed.

Two key determinants of weight are exercise and nutrition. Studies that have investigated the association between macroeconomic conditions and physical activity (reviewed above) have reached contradictory results. Ruhm (2000) and Ruhm (2005) report evidence that unemployment is associated with increased physical activity. The countercyclical relationship of exercise is also confirmed by the work of Xu (2013). An and Liu (2012), using BRFSS micro-data from 1990 and 2006, show that while recessions do not seem to modify the probability of exercising, they do increase the number of hours adults spend exercising. On the other hand, Charles and DeCicca (2008) find no significant link between macroeconomic conditions and exercise. Tekin et al. (2013) find the same result using the BRFSS data from 1990 to 2014.

All of these studies focus on the relationship between business cycle conditions and recreational exercise, using mainly self-reported measures of moderate or vigorous exercise available in the BRFSS and National Health Interview surveys. As recreational exercise constitutes only 4% of total physical activity for adults, Colman and Dave (2013) examine the effect of economic downturns on total daily physical activity and exertion using the American Time Use Survey from 2003 to 2010. They confirm the countercyclical nature of recreational exercise previously reported by Ruhm (2000), Ruhm (2005), Xu (2013), and An and Liu (2012). However, total physical exertion is procyclical. In other words, during a recession, the increase in exercise does not offset the reduction in work-related exertion, leading to a reduction in the total physical activity.

Very little is known about the response of eating habits to changes in business cycle conditions. Ruhm (2000) is one of the exceptions. He analyzes the association between macroeconomic conditions and the consumption of fruits and vegetables and does not find any significant relationship between them. Dave and Kelly (2012), on the other hand, use the BRFSS data from 1990 to 2009 to analyze the effect of business cycle conditions and adults’ monthly consumption of several healthy (fruit, fruit juice, carrots, green salad, vegetables) and unhealthy products (snacks, hamburgers, hot dogs, French fries, fried chicken, doughnuts). They find that the consumption of the majority of the healthy products is strongly procyclical, while the countercyclicality of the consumption of unhealthy products is less clear and is only significant for females and individuals with good health.

Addictions

Alcohol abuse or dependence generates great private and social costs in terms of health, mortality, productivity, and crime. Excessive alcohol consumption has been identified as an important determinant of more than 200 diseases and injuries, such as liver cirrhosis, cancers, and infectious diseases such as tuberculosis and HIV/AIDS (WHO, 2014). In addition, alcohol consumption is associated with 3.8% of all deaths (Rehm et al., 2009).

Similarly, the harmful consequences of tobacco consumption for health have also been extensively documented. Smoking can cause cancer, heart disease, stroke, lung diseases, diabetes, and/or chronic obstructive pulmonary diseases. As a consequence, it has been estimated that smoking is related to about one of every five deaths in the United States each year (U.S. Department of Health and Human Services, 2014).

Finally, the large prevalence of illicit drugs, especially non-medical use of prescription opioids and heroin, constitutes an important problem, particularly in the United States. The abuse of opioids has serious consequences for individuals’ health as well as social and economic welfare. It has been estimated that in 2012, 2.1 million people in the United States suffered from substance use disorders related to prescription opioids, and 467,000 were addicted to heroin (Substance Abuse and Mental Health Services Administration, 2013). During the Great Recession, the number of unintentional overdose deaths involving opioids tripled (Volkow, 2014). Moreover, there is some evidence that increases in heroin abuse can be related to the increased use of prescription pain relievers.

Alcohol Consumption

A large number of studies have analyzed the relationship between business cycle conditions and alcohol consumption. The majority of these studies have used US pre–Great Recession data. Ruhm (1995) analyzes aggregate data of 48 states over the 1975–1988 period to determine how macroeconomic conditions are related to alcohol consumption and highway vehicle fatalities. He concludes that state-level unemployment rates are negatively correlated with total alcohol consumption. These findings confirms that alcohol acts as a normal good. Freeman (1999) re-examines Ruhm’s work to correct for the potential non-stationary of the data by using logarithmic first differences. He confirms the early findings of Ruhm. Aggregate state-level data have the limitation that potentially relevant individual characteristics cannot be incorporated in the analysis. Ettner (1997) addresses this issue by using cross-sectional microdata from the 1988 National Health Interview Survey.1 He finds that involuntary unemployment increases alcohol consumption but reduces alcohol dependence. Dee (2001), using data from 1984 to 1995 of the BRFSS, confirmed Ruhm’s procyclicality results for the number of drinks consumed and chronic drinking participation. However, he showed that binge drinking was countercyclical.2 Ruhm and Black (2002) analyze this issue further by considering that business cycle conditions might affect alternative drinking behaviors differently.3 Thus, they use different definitions of light, binge, and heaving drinking. They find a consistent procyclical relationship for heavy and binge drinking and a countercyclical relationship for light and moderate drinking. Thus, all these studies reach the common finding that, while economic downturns reduce alcohol consumption for heavy drinkers, they increase consumption for light drinkers (although the results regarding binge drinking are less conclusive).

On the other hand, Arkes (2007) studies the effect of business cycle conditions on teenage alcohol consumption using data from the National Longitudinal Survey of Youth from 1996 to 2004. He finds that teenagers react differently to macroeconomic conditions than adults. Thus, for this younger group of individuals, the number of days of alcohol consumption reacts countercyclically to business cycle fluctuations.

Contrary to the results of Ruhm (1995), Dee (2001), and Ruhm and Black (2002), Charles and DeCicca (2008) and Jiménez-Martín et al. (2006) find that labor market fluctuations are not statistically correlated to different measures of alcohol consumption. Charles and DeCicca (2008) use data from the National Health Interview Surveys from 1997 to 2001 and show that the metropolitan statistical area-level unemployment rate is not statistically correlated with binge drinking. Jiménez-Martín et al. (2006) estimate the same specifications as Dee (2001) and Ruhm and Black (2002) but using a pseudo-panel of the BRFSS and find no relationship between business cycle conditions and alcohol consumption. Finally, Dávalos, Fang, and French (2012) use panel data from the National Epidemiological Survey on Alcohol and Related Conditions and show that binge drinking and alcohol abuse or dependence increase with the unemployment rate. Similarly, driving after having drunk too much alcohol also increases during recessions.

Even if the crisis of 2008 was more severe than most previous recessions, very few studies have focused on the relationship between economic conditions and alcohol consumption during the Great Recession. Bor, Basu, Coutts, McKee, and Stuckler (2013) show that during the period of 2008 to 2009, there was an increase in both drinking abstention and binge drinking. Frijters, Johnston, Lordan, and Shields (2013), using searches for alcohol-related terms on Google, show that a rise in unemployment is followed by an increase in searches in the next 12 months. Finally, Carpenter, McClellan, and Rees (2017) do not find a consistent effect of unemployment rates on alcohol consumption.

Other studies have examined this relationship in other countries. Johansson, Böckerman, Prättälä, and Uutela (2006) evaluate the impact of business cycle conditions on alcohol consumption and alcohol-related mortality in Finland. They use both aggregated mortality data from 1975 to 2001 as well as micro-level data from 1982 to 2001 from national surveys at the province level. Their results show that the regional employment rate is not statistically significantly related to the probability of being a drinker or to the amount of alcohol consumed. However, they observe a procyclical relationship between the regional real GDP and the number of drinks consumed.4 Bassols and Castelló (2016) analyze the impact of economic conditions on alcohol consumption for individuals aged 15–64 during the Great Recession in Spain using large-scale survey data from 2005 to 2011. They find that alcohol consumption is procyclical.

Smoking

Fewer studies have addressed the effect of economic conditions on smoking behavior. In the United States, Ruhm (2005), using BRFSS data from 1987 to 2000, finds that an increase in the percentage of the state population that is employed is associated with increases in cigarette smoking. Moreover, this procyclical relationship is disproportionately concentrated among heavy smokers (more than 40 cigarettes a day). Charles and DeCicca (2008) confirm the same result as Ruhm (2005) but using data from the National Health Interview Surveys from 1997 to 2001 and metropolitan statistical area-level unemployment rates. Tekin et al. (2013) use the BRFSS from 1990 to 2014 to confirm the procyclicality of smoking. However, they also report that this relationship weakens during the Great Recession. Similarly, Xu (2013) take a structural approach to investigate the effects of wages and working hours on cigarette smoking for low-educated individuals. They find that smoking is positively associated with the number of hours worked. However, a recent study in Spain by Bassols and Castelló (2016) finds that the probability of smoking tobacco daily reacts countercyclically to changes in economic conditions.

Drugs

A lack of data is probably the main reason why there are only five studies that analyze the impact of economic recessions on illicit drug use. Four of these focus on drug consumption behavior of teenagers during the Great Recession in the United States. Arkes (2007) and Arkes (2011) use the National Longitudinal Surveys of Youth 1997 to examine the relationship between business cycle conditions and drug use among teenagers (15–19 years old), and young adults (20–24 years old). He shows that marijuana and cocaine are countercyclical for teenagers, while the impact of economic conditions on cocaine is less clear for young adults. Pabilonia (2014) uses data from the Youth Risk Behavior survey from 2003 to 2011 and confirms that marijuana is countercyclical only for 15- to 17-year-old black males. Yet she does not find any significant impact of economic downturns on marijuana use for white teenagers. Carpenter et al. (2017), using data from the National Surveys on Drug Use and Health from 2002 to 2013, identifies, for the first time, the impact of economic conditions on the consumption of different types of illicit drugs and prescription drugs for adults (such as cocaine, crack, stimulants, methamphetamines, analgesics, oxycodone, heroin, hallucinogens, LSD, PCP, ecstasy, sedatives and tranquilizers, and inhalants). They highlight the importance of analyzing separately the different types of drugs as they find that the use of some drugs (such as LSD) is procyclical while the use of other drugs (ecstasy) is countercyclical. They also show a strong relationship between economic downturns and higher consumption of hallucinogens and prescription pain relievers. Finally, Bassols and Castelló (2016) use survey data from 2005 to 2011 in Spain to analyze the impact of economic conditions on consumption of drugs for individuals aged 15 to 64. Unlike the previous papers, they report a positive relationship between local economic conditions and the consumption of marijuana and cocaine.

Hollingsworth, Ruhm, and Simon (2017) focused instead on opioids and other drug-poisoning behaviors in the United States. Particular attention is given to opioids, as they are one of the major drivers of the recent fatal drug epidemic. In 2014, 53% of all fatal drug overdose deaths were caused by opioids. In their paper they use data on mortality rates from 1999 to 2014 and emergency department visits from 2002 to 2014. They find that opioid death rates and emergency department visits for opioid overdoses are countercyclical, similar to what the previous literature has found for the consumption of illicit drugs.

Empirical Methods

An important challenge of the literature is to compare the main results across studies. This is because studies differ not only in the countries and time spans employed but also in the empirical methods used, making the generalization of the findings extremely complicated.

Level of Aggregation

Benner’s pioneering work, using aggregate time-series data, suggested that health deteriorates during temporary economic slowdowns (Brenner, 1973, 1975, 1979). As time-series techniques suffer from important omitted variable biases, subsequent studies used instead time fixed-effect models to estimate the within-area variation in individuals’ health relative to the changes occurring in other areas. Yet the level of geographic aggregation chosen to perform the fixed-effect model differs between studies. While many studies consider states as the level of aggregation (Gonzalez & Quast, 2011; McInerney & Mellor, 2012; Miller et al., 2009; Neumayer, 2004; Ruhm, 2000, 2003, 2005, 2015; Stevens et al., 2015; Tekin et al., 2013), others choose countries (Economou et al., 2008; Gerdtham & Ruhm, 2006; Lin, 2009), metropolitan statistical areas (Charles & DeCicca, 2008), regions (Bellés-Obrero et al., 2016; Svensson, 2007), provinces (Ariizumi & Schirle, 2012; Granados, 2005; Latif, 2014), or counties (An & Liu, 2012; Gerdtham & Johannesson, 2005).

The level of aggregation chosen is important, as it can affect the interpretation of the results (Lindo, 2015). First, changes in the economic conditions will have an impact on health through different mechanisms depending on the level of geographic aggregation used in the analysis. The effect of changes in individuals’ employment on their health is estimated more directly when considering as the level of aggregation the local area where individuals work. However, at the local level there might be some other mechanisms that can play a role, such as migration, pollution, government policies, etc. Thus, more aggregated areas will include in the estimated coefficients any spillover effects across more disaggregated areas. If spillovers are important, the estimation of the effects using more disaggregated areas may be biased. For instance, migration is normally influenced by economic conditions. If those individuals that migrate during recessions are healthier, the improvements in health during recessions could be underestimated when using more disaggregated areas. However, estimates using more disaggregated analyses will be more precise as they can capture variation in economic conditions that are disguised using more aggregated measures. For example, using counties as the level of aggregation will capture variations in the severity of contractions or expansions experienced in different parts of the state, while state-level analysis will not. Finally, studies using more disaggregated-level data will be more subject to measurement error in their economic indicators if these are based on household surveys.

In particular, Lindo (2015) replicates earlier work by Ruhm (2000), Stevens et al. (2015), and Dehejia and Lleras-Muney (2004) and conducts various analyses using different levels of geographical aggregation to investigate whether the level of aggregation influences the estimated effects of economic conditions on individuals’ health. He finds that the results are sensitive to the level of geographic aggregation used in the analysis. The effect of business cycle conditions on health is larger if more aggregated indicators of economic conditions are used, which indicates that these include the spillover effects across more disaggregated areas. In fact, Lindo (2015) shows that economic conditions of other counties in the same state have a significant effect over the mortality rate of the county. However, he also shows that estimations using more disaggregated economic indicators are more precise. He finds that the procyclicality of mortality persists even when considering recent years of data, contrary to what Ruhm (2015) finds using state-level unemployment rates.

Hollingsworth et al. (2017) also check whether the effect of the county-level unemployment rate on opioid/drug mortality rates and emergency department visits is replicable when using state unemployment rates. They find that both results are largely consistent, although the state-level estimates are larger in magnitude, similar to Lindo’s findings. Interestingly, they find that results for blacks are more sensitive to the choice of the aggregation level, which suggests that precision or spillover effects might be more important for this group.

Given the previous findings, we conclude that there are a number of trade-offs associated with performing analysis with more or less aggregated economic indicators. Thus, it will be advantageous for future studies to use economic conditions at different levels of aggregation as it could help understanding the mechanisms behind the estimated relationships. At the same time, it could shed some light on the existence of spillovers that could be underestimating or overestimating the findings.

Time Spans

We have already pointed out that the cyclicality of health, and mortality in particular, seems to be changing with the inclusion of the most recent data available. This suggests that the time span chosen in different studies could be another potential explanation for the divergence in results.

In fact, Ruhm (2015) illustrates that his estimates are sensitive to the change in starting and ending dates. More importantly, he points out that this volatility is more important when the time spans chosen are short (less than 15 years). Also, Tekin et al. (2013), as a robustness check, estimate models for different alternative sample windows of 5, 10, and 15 years. They also find that the coefficients of the model are more instable when using 5- and 10-year intervals, instead of 15 years.

This finding raises concerns about all studies that have used short time spans (Arkes, 2007, 2009; Bassols & Castelló, 2016; Bor et al., 2013; Bozzoli & Quintana-Domeque, 2014; Charles & DeCicca, 2008; Colman & Dave, 2013; Dávalos et al., 2012; Frijters et al., 2013; Pabilonia, 2014; Ruhm, 2000, 2003). In these studies, the relationship between economic conditions and health might be too noisy to be informative. Thus, any reanalysis made using longer periods of time will be beneficial for providing more credible estimates.

Proxies of Business Cycle Conditions

The unemployment rate is the most common measure of economic conditions used in the literature as it measures the direct ability of individuals in the labor market to find a job and it is also associated with other country-specific labor market characteristics (i.e., the rigidity of institutions and labor market policies, labor taxation, human capital, etc.).

Yet, some studies (Clark & Summers, 1982) have claimed that the employment-to-population ratio is a more stable measure of the labor market conditions. This is especially true for individuals who are more likely to enter and exit the labor market over time, such as young workers or women. In addition, changes in the unemployment rate could underestimate or overestimate the real economic conditions. For example, some discouraged individuals may re-enter the labor market as a consequence of an improvement in the economic perspective, which would lead to an increase in the unemployment rate. On the other hand, the unemployment rate may decrease when individuals leave the labor market after losing hope of finding a job during economic downturns. As a consequence, some studies have analyzed the consistency of their results using this alternative proxy of business cycle conditions. For instance, Ruhm (1995) uses both state unemployment rates and the employment-to-population ratio as a proxy for economic conditions to estimate its effect on drinking and traffic fatalities. He finds that the procyclicality of traffic fatalities and drinking behavior is robust to changes in the measure of economic conditions. In particular, he estimates that a one percentage point increase in the population employed raises alcohol consumption by 0.8%, while a one percentage point increase in the unemployment rate decreases it by 0.6%. If we look at other studies (Freeman, 1999; Granados, 2005; Tekin et al., 2013), the main conclusion is that results are strongly consistent between these two measures of economic conditions (with the exception of Svensson, 2007).

It also very common to use the GDP (either deviations from the trend or the change in per capita GDP) as an alternative measure of business cycle conditions. The GDP better reflects the economic activity of the country when there is a significant percentage of the population that is self-employed, has long-term or fixed labor contracts, or are public employees (as these groups of people have a higher probability of experiencing a drop in income and/or consumption rather than an increase in unemployment during recessions). Similarly, the GDP is also a better measure in countries that experience large migration flows of unemployed individuals during recessionary periods. Migration of unemployed individuals during recessions will mitigate the unemployment rate rise during downturns. Finally, the time span used and the rigidity of the labor market will influence the suitability of using GDP or the unemployment rate as a measure of the business cycle condition. Countries with more flexible labor markets (like the United States) tend to experience greater increases in unemployment rate in the short run during recessions compared with countries with more rigid labor markets (like France). Johansson et al. (2006), analyzing the effect of business cycle conditions on alcohol-related mortality and drinking behavior, find substantial differences when using the employment rate or the change in the regional real GDP. While an increase in the employment rate is significantly associated with a decrease in alcohol-related mortality, changes in the regional real GDP are not unrelated to this outcome. On the other hand, procyclicality of drinking is stronger when economic conditions are measured by the regional GDP growth than employment rates.

Carpenter et al. (2017), in their robustness check section, also explore models using state employment-to-population ratios and log GDP per capita instead of the unemployment rate. Their findings are robust, in sign and magnitude, for the different proxies of business cycle conditions. However, the countercyclicality of substance use disorders involving analgesics and hallucinogens lose their significance with the alternative measures. Finally, Neumayer (2004) also indicates that his results are quite similar when replacing the unemployment rate by the growth rate in real GDP. The major differences are in deaths due to pneumonia/influenza and suicide, which are not procyclical when using the growth rate in real GDP.

Gerdtham and Johannesson (2005) use six different business cycle indicators to analyze the effect of economic conditions on mortality risk in Sweden: the unemployment rate, the notification rate, the deviation from the GDP trend, the GDP change, the industry capacity utilization, and the industry confidence indicator.5 They show that their results are highly sensitive to the business cycle indicator used. Mortality risk is only significantly countercyclical for men when using the notification rate, capacity utilization, confidence indicator, and GDP change. These proxies are characterized by the fact that they all measure changes from the current level rather than deviations of the current conditions from the average level (as the unemployment rate and the deviation from the GDP trend).

Finally, Frijters et al. (2013) use insured unemployment rates in addition to unemployment rates. Insured unemployment rates measure the quantity of continued unemployment insurance claims divided by the number of workers that qualify for unemployment insurance. They claim that this measure is less noisy as it is not affected by small sample measurement errors that could be present in the labor surveys. They find that the measure of unemployment used does not affect the main conclusions about the effect of business cycle conditions on alcoholism-related Google searches. However, insured unemployment rates have a larger effect than the unemployment rates. They show that one potential explanation for this divergence in the magnitude of the effects is that the insurance unemployment rate is more representative of individuals that have a higher risk of being affected by unemployment (those who became unemployed involuntarily). Very similarly, Ettner (1997) uses involuntary unemployment rates as an alternative measure of economic conditions when examining its effects on alcohol use and symptoms of dependence. They claim that involuntary unemployment could have a stronger effect on alcohol abuse than non-participation in the labor force as it can be perceived as a more stressful event out of the individual’s control. In fact, he finds that non-employment reduces both alcohol consumption and dependence symptoms, while involuntary unemployment increases consumption of alcohol (although it reduces dependence symptoms for single individuals).

Summing up, there seems to be some divergence in the conclusions about the relationship between business cycle conditions and health when different measures of economic conditions are used. Nonetheless, very little is known about the sources of these differences and their interpretation. Further research is needed in order to understand the indicators that best reflect economic conditions at the local level. For now, the conclusions seem to be very case-specific. Looking at the literature, it seems to largely depend on the particular characteristics of the data available (existence of measurement errors), the labor market conditions of the country studied (flexibility, many people entering and exiting the labor market, availability of public unemployment insurance), or the importance of migration patterns, among others. Thus, the recommendation for future scholars is to further investigate the impact of different proxies of economic conditions on their findings using alternative indicators, when available, as robustness checks.

Other Model Specifications

Other model specification characteristics are very important to take into account when drawing comparisons across studies.

Time Trends

Many of the studies attempt to control for unobserved factors that vary over time within areas and that may be important determinants of health trends (e.g., education, social norms, healthcare services) by including area-specific time trends. The estimation using these additional controls is less likely to suffer from omitted variables bias.

However, the inclusion of these area-specific time trends could absorb a larger part of the within-area variation, making the estimates unreliable and more sensitive to measurement error. This is especially important if there is substantial collinearity between the area time trends and the measure of economic conditions used. To check if the collinearity problem is important, some studies (McInerney & Mellor, 2012) regress the area unemployment rate on the area and year fixed effects and the area-specific time trends and calculate the variation inflation factor. If this variation inflation factor is smaller than 10 (the standard rule of thumb), then the multicollinearity is less of a concern. Nevertheless, most papers choose to report models with and without area-specific time trends and then discuss the results accordingly.

Another example of how the modeling assumptions can influence the results can be found in Ariizumi and Schirle (2012). They estimate the relationship between business cycle conditions and age-specific mortality rates. They report different findings allowing or not for age-specific intercepts. If they assume the same intercept for all groups, they do not find any cyclical patter in mortality rates. However, when they allow for different age-specific intercepts, they show a negative and significant effect of the unemployment rate.

Other Controls

The inclusion or exclusion of some controls in the main regressions could also potentially explain the lack of consistency in the findings of different studies. The inclusion of some explanatory variables that can represent potential mediating factors could limit the mechanism through which business cycle conditions have an impact on health, estimating only a partial effect. Moreover, if those variables are highly correlated, it could cause multicollinearity problems and bias the estimates. This is the case with the inclusion of variables related to individual’s employment or personal income, for instance. For example, Economou et al. (2008) find countercyclical effects of business cycle conditions on several cause-specific mortality rates in 13 European Union countries. Their results contradict previous findings in the literature. However, their estimates include potential confounders such as medical intervention indicators, per capita daily consumption of alcohol, etc. Thus, they limit the mechanisms through which economic conditions could be affecting health. Their findings have to be interpreted as the effect of economic conditions on health, beyond their effect over alcohol consumption and medical usage (that can not longer be a cause).

However, the inclusion of region-level covariates can be useful to investigate the consistency of results. That is, adding these control variables in the baseline regression can be informative when trying to uncover the potential mechanisms behind the reported effects.

Therefore, additional controls that could be mediating factors through which economic conditions affect health should only be included to investigate which mechanisms are important in that relationship. In a similar vein, it is very important to be aware of the inclusion/exclusion of control variables before comparing the findings across different studies.

Discussion

As shown here, following the pioneering work of Ruhm (2000), most research up until 2008 found that health improves during temporary economic downturns, and this finding has been reported for several countries around the globe. However, studies examining the effects of more recent recessions are less conclusive, finding mortality to be less procyclical, countercyclical, or even unrelated to macroeconomic conditions.

While the association between business cycle conditions and mortality has been extensively documented, the concrete mechanisms behind these associations remain poorly understood. We review some of the most important articles in the literature and provide evidence of this divergence in the findings reported.

There are a number of reasons for the difference in findings across studies. Studies not only differ in the country and time span employed but also in the empirical methods used, making the generalization of the findings extremely complicated.

The following items should be considered when discussing and comparing papers analyzing the impact of business cycle conditions on health outcomes:

  1. 1. Different measures of economic conditions at the local level can be used as the main explanatory variable. There are some divergences in the conclusions about the relationship between business cycle conditions and health when different measures of economic conditions are used. Nonetheless, very little is known about the sources of these differences and their interpretation. Further research is needed in order to understand the indicators that best reflect economic conditions at the local level. For now, the conclusions seem to be very case-specific.

  2. 2. There are a number of trade-offs associated with performing analysis with more or less aggregated economic indicators. Thus it will be advantageous for future studies to use economic conditions at different levels of aggregation as it could help understanding the mechanisms behind the estimated relationships. At the same time, it could shed some light on the existence of spillovers that could be underestimating/overestimating the findings.

  3. 3. Additional controls at the regional level are included in some studies. These controls could be mediating factors through which economic conditions affect health. Thus, they can be used to investigate which mechanisms are important in that relationship.

  4. 4. The inclusion of area-specific time trends can potentially absorb a larger part of the within-area variation, which make it more difficult to identify the effects of local area economic conditions. Most papers in the literature choose to report models with and without area-specific time trends and then discuss the results accordingly.

  5. 5. Estimates are sensitive to the time span of the data included in the analysis. This volatility is more important when the time spans chosen are short (less than 15 years) (Ruhm, 2015).

See Table 1 for a compilation of study findings.

Table 1. Study Findings

Study

Country

Data

Measures

Economic conditions measure

Methods

Additional controls

Findings

An and Liu (2012)

US

BRFSS

Micro-level data 1990–2006

Last month’s:

  1. 1) Hours of physical activity

  2. 2) Moderate or vigorous exercise participation

  3. 3) Hours of exercise

  4. 4) Hours among regular exercisers

  1. 1) County-level unemployment rate

FE model with month, year, and county FE

  1. 1) Demographics

  2. 2) Education, marital status

Hours are countercyclical

No relation with exercise participation

Aparicio and González (2014)

Spain

Birth and death-certificate data

Micro-level data 1981–2010

  1. 1) Birthweight in grams

  2. 2) Low and very low birthweight (<2,500, <1,500 g)

  3. 3) Late fetal death (<24 h)

  4. 4) Neonatal mortality (1–28 days)

  5. 5) Post-neonatal mortality (28 days–1 year)

  6. 6) Fertility rates

  7. 7) Composition of the families giving births (mother’s age, marital status, fraction of babies with no registered father, parental occupation, birth order and multiplicity)

  8. 8) Mother’s employment status during pregnancy

  9. 9) Women’s health or health-related behavior

  1. 1) Province unemployment rate

  2. 2) Province non-employment rate

  1. 1) FE model with province and year fixed effects

  2. 2) To control for selection, add parents fixed effects (subsample of siblings)

In some specifications:

  1. 1) Quadratic trend instead of the year fixed effects

  2. 2) Province-specific linear trends

Procyclical neonatal, post-neonatal mortality, and late fetal death, low and very low birthweight, fertility, first births, multiple births, and babies with no registered father

Countercyclical birthweight, mothers who are married

Results are stronger for the subsample of low-skill parents, and the birthweight effects are driven by the low-income provinces, while the effects on mortality are stronger in high-income regions

Recessions leads to lower fertility among low-skilled parents

Effect on maternal age and parental occupation are mixed

Selection plays a role but do not account for the procyclicality of infant health

Maternal employment during pregnancy is not affected by the cycle, but one possible mechanism is engaging in more health-enhancing behaviors

Ariizumi and Schirle (2012)

Canada

Canada’s Population Statistics

Aggregate-level longitudinal data 1977–2009

  1. 1) Age-specific mortality rates (all, females and males)

  1. 1) Province unemployment rate

FE model with age, province, year FE, and province-specific time trends

None

Procyclical mortality for middle-aged individuals (30–39)

No relationship on mortality for infants and seniors

Arkes (2007)

US

National Longitudinal Survey of Youth

Micro-level data 1996–2004

Teenage’s consumption of marijuana, cocaine/hard drugs, and alcohol:

  1. 1) Last year

  2. 2) Past 30 days (except for cocaine/hard drugs)

  3. 3) Heavy use

  4. 4) Frequency of use of the substance (last 30 days for marijuana and alcohol, and last year for cocaine)

  5. 5) Sell marijuana or hard drugs

  1. 1) State-level unemployment rate

FE model with state and time FE

Data weighted

  1. 1) Demographics

  2. 2) Parent’s education

  3. 3) Parental figures the teenager lived with

  4. 4) Variables for siblings

  5. 5) Urbanity of the teenager’s residence

  6. 6) Interview characteristics

Countercyclical for all variables

Arkes (2009)

US

US National Longitudinal Survey of Youth from 1997

Micro-level data 1997–2004

Teenage (15–18 years)

  1. 1) Overweight

  2. 2) Obesity

  3. 3) Underweight

  4. 4) BMI

  1. 1) State-level unemployment rate

FE model with state and year FE

  1. 1) Demographics

  2. 2) Parent’s education

  3. 3) Family composition

  4. 4) Educational enrollment

Countercyclical overweight and BMI for girls, obesity for boys

Procyclical overweight and BMI for boys, underweight for girls

Arkes (2011)

US

National Longitudinal Survey of Youth

Micro-level data 1996–2007

Young adult’s consumption of marijuana, other drugs:

  1. 1) Last year

  2. 2) Past 30 days (except for cocaine/hard drugs)

  3. 3) Heavy use

  4. 4) Frequency of use of the substance (last 30 days for marijuana, last year for cocaine)

  1. 1) State-level unemployment rate

FE model with state and time FE

Data weighted

  1. 1) Demographics

  2. 2) Parent’s education

  3. 3) Parental figures the teenager lived with

  4. 4) Variables for siblings

  5. 5) Urbanity of the teenager’s residence

  6. 6) Interview characteristics

Countercyclical for marijuana consumption

Weak evidence of countercyclical other drug (only significant for frequency)

Baird et al. (2011)

59 developing countries

123 demographic and health surveys covering 59 countries

Micro-level data 1975–2003

  1. 1) Mortality during the first year of life

  1. 1) Natural log of real per capita GDP in 2000 US dollars (adjusted in PPP)

FE model with country fixed effects and flexible, country-specific time trend

  1. 1) None

  2. 2) Children’s and mother’s characteristics: mother’s education, maternal age at the time of birth, birth order, place of residence at birth, gender of the child, and whether the child was a multiple birth

  3. 3) Mothers’ fixed effects and gender and multiple births (sample limited to mothers with at least two live births)

  4. 4) Lagged and lead per capita GDP

  5. 5) Per capita GDP first month of life and the next 11 months

Countercyclical infant mortality

Changes in the composition of women do not account for the countercyclical infant mortality

Economic conditions around birth appear to matter most for infant survival

Mortality of girls is much more sensitive to changes in economic circumstances than that of boys (especially during economic contractions)

Bellés-Obrero et al. (2016)

Spain

Spanish National Health Survey

Micro-level data 1987–2012

Children’s (2–15 years):

  1. 1) BMI

  2. 2) Overweight

  3. 3) Obesity

  4. 4) Underweight

  5. 5) Exercise

  6. 6) Fruit daily

  7. 7) Mediterranean Diet

  8. 8) Breakfast with protein

  9. 9) Sweets every day

  1. 1) Regional-level unemployment rate

FE model with region, trimester FE, and regional- specific linear time trends

  1. 1) Demographics

  2. 2) Parent’s education

Only effects for children <6 or >12 years

Obesity and exercise is procyclical

Underweight is countercyclical

No relation with BMI, overweight

Mixed results for diet

Bhalotra (2010)

India

National Family Health Survey of India

Micro-level data 1970–1997

  1. 1) Mortality during the first year of life

  2. 2) Parent’s characteristics

  3. 3) Health-seeking behaviors

  4. 4) Parent’s labor supply

  1. 1) State-year logarithm of per capita net domestic product deflated by the consumer price index for agricultural workers

FE model with mother and year fixed effects, state-specific time trends (or cubic state trends), quarter-year fixed effects

  1. 1) Children variables: gender, birth order, birth month, and age of mother at the birth

  2. 2) Quarter-year fixed effect

  3. 3) Cubit state trends (instead of state-specific time trend)

  4. 4) Deviations of state-level rainfall from its time-series mean

  5. 5) Log and the change in the log of population, an indicator for urbanization, income inequality, and poverty

Countercyclical infant mortality in rural areas

In rural areas, births to uneducated women, women with uneducated husbands, and women of scheduled tribes are significantly underrepresented in a recession

Income effect is higher for uneducated parents, mothers younger than 18, and girls

Procyclical home births, antenatal care and immunization, and treatment for illnesses

Countercyclical mother’s labor supply

Bockerm et al. (2007)

Finland

National Public Health Institute

Micro-level data 1978–2002

  1. 1) BMI

  2. 2) Overweight

  3. 3) Obese

  4. 4) Underweight

  5. 5) Severely obese

  1. 1) Regional-level employment-to-population ratio

  2. 2) Change in real GDP

FE model with region and year FE (robustness with region-specific time trends)

  1. 1) Demographics

  2. 2) Education, marital status

Obesity and severe obesity are countercyclical

No relation with overweight and BMI

Bor et al. (2013)

US

BRFSS

Micro-level data 2006–2010

Past month’s:

  1. 1) Drinking participation

  2. 2) Number of drinks

  3. 3) Binge drinking

  4. 4) Heavy drinking (>60)

  5. 5) Light drinking

  6. 6) Moderate drinking

  7. 7) Frequent binge drinking

None

OLS comparing the mean variables two years before vs. two years after

  1. 1) Demographics

  2. 2) Education, marital status

  3. 3) State of residence

  4. 4) Month of interview

Procyclical drinking participation and light drinking

Countercyclical frequent binge drinking, moderate and heavy drinking

Bozzoli and Quintana-Domeque (2014)

Argentina

National Registry of Live Births

Micro-level data 2000–2005

  1. 1) Birthweight

  2. 2) Low birthweight

  1. 1) Month-level deviations of the log index of economic activity with respect to its long-term trend (expressed in log units)—average during the period of gestation (1–9 months), or average in each of the three quarters of the pregnancy

  1. 1) FE mode month of birth, mother’s province of residence, year FE, province-specific year, and province-specific month-of-birth fixed effects

  2. 2) Event study with pregnancies before August 2001 (do not expected the crisis) to control for selection

  1. 1) Postnatal economic fluctuations (the average of the monthly cyclical component during the first nine months after birth)

  2. 2) Gender of the child FE

  3. 3) Mother’s age, birth order, mothers’ education, mothers’ marital status, and the interaction of these last two variables

Procyclical birthweight and the prevalence of low birthweight is countercyclical

The statistically relevant periods of pregnancy are the first and third trimesters for low-educated mothers and only the third for more educated mothers—nutritional deficiencies affecting only low-educated mothers, while stress associated with economic downturns affects both low- and high-educated mothers

Bradford and Lastrapes (2014)

US

National Ambulatory Medical Care Survey

Aggregate-level longitudinal data 1989–2009

For all the population and for patients age 19 to 64

  1. 1) Number of prescriptions for anti-depressants or anti-anxiety drugs

  2. 2) Number of prescriptions specifically for anti-depressants

  3. 3) Number of prescriptions for anti-anxiety drugs

  4. 4) Total number of drug prescriptions of any kind

  5. 5) Total number of doctor visits

  6. 6) Number of doctor visits resulting in a drug prescription

  7. 7) Number of visits for mental health issues

  1. 1) Regional unemployment rate

  2. 2) Regional level of employment

Time-series regressions and vector autoregression models

  1. 1) Regions’ population each month

Countercyclical mental health drug prescriptions for the Northeast region, also countercyclical but weaker for doctor visits with mental health diagnoses

Countercyclical total drug prescriptions and doctor visits for all regions

Buchmueller, Jusot and Grignon (2007)

France

Mortality Statistics

Aggregate-level longitudinal data 1982–2002

  1. 1) Total mortality rate

  2. 2) 20–44, 45–64, ≥65 mortality rates

  3. 3) Deaths from malignant neoplasms, cardiovascular diseases, liver disease, motor vehicle accidents, other accidents, or suicide

  1. 1) “Département” unemployment rate

FE model with “département” and year FE

  1. 1) Average household income

  2. 2) National unemployment rate

Procyclical total mortality rate, deaths from cardiovascular disease, and accidental deaths

No relationship with cancer, liver diseases, deaths, or suicides

Byron and Fertig (2012)

US

Child Development Supplement of the Panel Study of Income Dynamics

Micro-level data 1997–2007

Children’s (2–18 years):

  1. 1) BMI for children living in households

    • –with no debt

    • –with any debt

    • –with a high debt-to-income ratio

  1. 1) State-level unemployment rate

FE model with state and month-year FE and state-specific linear time trends

In other models FE with month-year and child FE, and state-specific linear time trends

unemployment interacted with debt

  1. 1) Demographics of the parents and the children

  2. 2) Parents’ education

  3. 3) Number of children

  4. 4) Family income

  5. 5) Own home

  6. 6) Parents’ hours of work

Countercyclical BMI for all type of households

When controlling for children FE, no relation for households with no debt, countercyclical in households that take debt, and procyclical for households that pay the debt

Carpenter et al. (2017)

US

National Surveys on Drug Use and Health

Micro-level data 2002–2013

Alcohol, marijuana, any illicit drug, cocaine, crack, stimulants, methamphetamines, analgesics, oxycodone, heroin, hallucinogens, LSD, PCP, ecstasy, sedatives, and tranquilizers Inhalants:

  1. 1) Past month participation

  2. 2) Past year participation

  3. 3) Past year disorder

  4. 4) Past year disorder conditional on past year use

  1. 1) State-level unemployment rate

FE models with state and time FE (robustness with state-specific linear time trends)

  1. 1) Demographics

  2. 2) Education, marital status

Analgesic and hallucinogen participation countercyclical

No relation with marijuana, any illicit drug, sedatives, tranquilizers, inhalants, methamphetamines, oxycodone

LSD participation procyclical, but ecstasy participation countercyclical

Ambiguous relationship of alcohol participation and disorders

Charles and DeCicca (2008)

US

National Health Interview Survey

Micro-level data 1997–2001

Men’s:

  1. 1) Current smoker, all days

  2. 2) Current smoker, some days

  3. 3) Any days with 5+ drinks

  4. 4) Days with 5+ drinks

  5. 5) Mental health in the past month (sad, hopeless, worthless, restless, nervous, everything was effort)

  6. 6) BMI

  7. 7) Overweight

  8. 8) Obesity

  9. 9) Severe obesity

  10. 10) Very severe obesity

  11. 11) Underweight

  12. 12) Normal weight

  13. 13) Moderate exercise, any time or 3+ times

  14. 14) Vigorous exercise, any time or 3+ times

  15. 15) Strength training, any time or 3+ times

  1. 1) Metropolitan statistical area-level unemployment rate

FE model with MSA, quarter and year FE

  1. 1) Demographics

  2. 2) Education, marital status

  3. 3) Income relative to poverty line

No relation with alcohol or being a current smoker some days

Countercyclical smoking every day for those least likely to be employed, but procyclical for those in the highest employment decile

Procyclical mental health

(stronger for individuals with low employment probabilities, for African American and white men and less educated men)

No relation with any type of exercise, BMI, severe obesity, and very severy obesity

Countercyclical overweight, obesity, and underweight

Procyclical normal weight

Colman and Dave (2013)

US

American Time Use Survey

Micro-level data 2003–2010

  1. 1) Work minutes excluding job search

  2. 2) Work minutes × Metabolic Equivalent of Task (MET)

  3. 3) Exercise > 10 min

  4. 4) Exercise min

  5. 5) Exercise × MET

  6. 6) Min exercise < 4 METS

  7. 7) Min exercise > 4 METS

  8. 8) Total minutes × MET

    MET-adjusted time use:

  9. 9) Sleep

  10. 10) Personal care

  11. 11) Housework

  12. 12) Childcare

  13. 13) Care of HH adults

  14. 14) Purchasing goods and services

  15. 15) Eating and drinking

  16. 16) Socializing and relaxing

  1. 1) State-level employment-to-population ratio

  2. 2) Gender-specific state-level employment-to-population ratio

FE model with region, day and month FE (excluding year FE)

  1. 1) Demographics

  2. 2) Education, marital status

Exercise, sleep, childcare, and television is countercyclical

Time spend at work and purchasing goods and services is procyclical

Physical exertion is procyclical

Davalos et al. (2012)

US

Waves 1 and 2 of the National Epidemiological Survey on Alcohol and Related Conditions

Micro-level data 2001–2005

Year prior to interview:

  1. 1) Binge drinking

  2. 2) Number of binge-drinking days

  3. 3) Drinking and driving

  4. 4) Alcohol abuse and/or dependence

  1. 1) State-level unemployment rate

First difference (for continuous variables) or conditional fixed-effects logit technique (for binary measures) with month FE

As robustness:

  1. 1) Demographics

  2. 2) Education, marital status

  3. 3) Health status

  4. 4) Individual economic and employment characteristics

  5. 5) Beer tax

  6. 6) Union representation (% employed)

  7. 7) Alcohol consumption (gallons per capita)

All alcohol measures are countercyclical

Dave and Kelly (2012)

US

BRFSS

Micro-level data 1990–2009

Month’s consumption of:

  1. 1) Healthy products (fruit, fruit juice, carrots, green salad, vegetables)

  2. 2) Unhealthy products (snacks, hamburgers, hot dogs, french fries, fried chicken, doughnuts)

  1. 1) State-level unemployment rate

FE model with month, year, and region FE

  1. 1) Demographics

  2. 2) Education, marital status

Procyclical consumption of healthy foods

Countercyclical relationship with unhealthy food but only significant for females and individuals with good health

Dee (2001)

US

BRFSS

Micro-level data 1984–1995

Past month’s:

  1. 1) Drinking participation

  2. 2) Number of drinks

  3. 3) Binge drinking

  4. 4) Heavy drinking (>60)

  1. 1) State-level unemployment rate

FE model with state FE, linear month time trends, and state-year FE

  1. 1) Demographics

  2. 2) Education, marital status

  3. 3) Per capita income (state level)

No effect on drinking participation

Number of drinks and heavy drinking procyclical

Binge drinking countercyclical

Dehejia and Lleras-Muney (2004)

US

Vital Statistics Natality and Centers for Disease Control (for mortality)

Micro-level data 1975–1999

  1. 1) Birthrates

  2. 2) Characteristics of the mothers (age, education, marital status)

  3. 3) Prenatal visits

  4. 4) Smoking and drinking during pregnancy

  5. 5) Birthweight <1,500, <2,500 g

  6. 6) Apgar score below 5

  7. 7) Congenital defects

  8. 8) Infant, neonatal, postneonatal mortality rates

  1. 1) State-level unemployment rate in the year of conception and the year prior to conception (only for mortality)

FE model with state, year FE (some include

state-specific time trend)

None

No significant effect over birthrates but procyclicality in % of black babies births

Procyclicality in the probability of having birthweight <1,500, <2,500 g, infant and postneonatal mortality rates

No effect on neonatal mortality rate, congenital effects, Apgar score, probability of smoking or drinking

During recessions mothers and fathers are more educated, older, do more prenatal visits

Economou, Nikolaou and Theodossiou (2008)

European Union

Health for All Database, WHO

Aggregate-level longitudinal data 1977–1996

  1. 1) Total mortality rate (male and female)

  2. 2) 20–34, 35–44, 45–54, 55–64, 65–74, and 75–84 years mortality rates

  3. 3) Deaths from malignant neoplasms, cardiovascular diseases, cancer of trachea, bronchus and lung cancer, motor vehicle accidents, homicide, suicides, and infant mortality

  1. 1) National unemployment rate

FE model with country and year FE

  1. 1) Age and education distribution in the population

  2. 2) Inpatient care admissions (%)

  3. 3) Population density

  4. 4) Number of cigarettes consumed per person, per day

  5. 5) Pure alcohol consumption, liters per capita

  6. 6) Average number of calories available per person, per day

  7. 7) Carbon dioxide emissions, metric tons per capita

Procyclical (not significant) total mortality rate and for the groups 25–35, 35–44, and 65–74 years when only controlling for age and education

Countercyclical total mortality and mortality for 45–54 years group when controlling for everything else

Countercyclical mortality for all death causes in both models

Ettner (1997)

US

National Health Interview Survey

Micro-level data 1988

  1. 1) Average daily consumption of ethyl alcohol during the past two weeks

  2. 2) Alcohol dependent/abuse

  1. 1) State-level unemployment rate

  2. 2) State-level involuntary unemployment rate

Two-stage instrumental variables methods

  1. 1) Demographics

  2. 2) Education, marital status

  3. 3) Family records of alcoholism

  4. 4) Area-level alcohol prices

  5. 5) Cost of living (state level)

With involuntary unemployment, countercyclical alcohol consumption, but cyclical alcohol dependence

With unemployment, both procyclical

Freeman (1999)

US

U.S. Brewers’ Association

Aggregate state-level

1970–1995

  1. 1) Total alcohol consumed

  1. 1) State-level unemployment rate

  2. 2) Percent of the state population employed

FE model with state FE, linear time trends, and logarithmic first differences

  1. 1) Per capita income

  2. 2) State tax on beer

  3. 3) Minimum legal drinking age

Alcohol consumption procyclical

Frijters, Johnston, Lordan and Shields (2013)

US

Google Insights for Search

Micro-level data 2004–2011

  1. 1) Likelihood a random user will search for “alcohol or alcoholic or alcoholics or alcoholism or aa” during a month and they restrict searches to those related to health

  1. 1) State-level unemployment rate

  2. 2) State-level insured unemployment rate

FE model with state, month, and year FE and state-specific linear and quadratic time trend

  1. 1) Yahoo searches

  2. 2) Hotmail searches

  3. 3) Cancer searches

Contemporaneous unemployment is unrelated to searches

Countercyclical relationship of searcher with previous 12 months unemployment rates

Gerdtham and Johannesson (2005)

Sweden

Statistic Sweden’s Survey of Living Conditions and National Causes of Death Statistics

micro-level longitudinal data

1980/1986–1996

  1. 1) Mortality risk (probability of dying during a given year)

  1. 1) Unemployment rate

  2. 2) Notification rate

  3. 3) Deviation from the GDP trend

  4. 4) GDP change

  5. 5) Industry capacity utilization

  6. 6) Industry confidence indicator

FE model with year FE

  1. 1) Age control of individual each year

  2. 2) County FE and county-specific time trends

Countercyclical mortality risk for men when using notification rate, capacity utilization, confidence indicator, and GDP change

No effect when using unemployment rate or deviation from the GDP trend or for women using any measure

Gerdtham and Ruhm (2006)

OECD

OECD Health Data 2000

Aggregate-level longitudinal data 1960–1997

  1. 1) Total mortality rate

  2. 2) Deaths from malignant neoplasms, cardiovascular diseases, pneumonia or influenza, liver disease, motor vehicle accidents, other accidents, homicide, infant mortality

  1. 1) Standardized national unemployment rate

FE model with country, year FE, and country-specific linear trends

  1. 1) Age and sex distribution of population

  2. 2) OECD-wide unemployment rate

  3. 3) National per capital income

Procyclical total mortality, deaths from cardiovascular disease, influenza/pneumonia, liver disease, and accidental deaths

No relationship with cancer mortality

Countercyclical suicides (not significant) and homicides

Procyclical fluctuation stronger in countries with weak social insurance programs

Gonzalez and Quast (2011)

Mexico

Population statistics (Secretaria de Salud)

Aggregate-level longitudinal data 1993–2004

  1. 1) Total mortality rate

  2. 2) 20–49, 50–64, ≥65 years mortality rates

  3. 3) Deaths from malignant neoplasms, cardiovascular diseases, diabetes, pneumonia or influenza, liver disease, motor vehicle accidents, other accidents, homicide, suicides, infant mortality

  1. 1) GDP per capita

FE model with state, year FE, and state-specific time trends

  1. 1) State population characteristics (age, education)

  2. 2) Number of doctors

  3. 3) Government health spending per capita

  4. 4) Net international migration rate

  5. 5) Net interstate migration rate

  6. 6) GDP per capita

Procyclical total mortality, stronger for individuals 20–49 years old

Procyclical suicides, homicides, and vehicle accidents

Countercyclical mortality due to cancer

No relation for other causes of death

Hollingsworth et al. (2017)

US

National Vital Statistics System of the Centers for Disease Control and Prevention Multiple Cause of Death (Micro-level data, 1999–2014)

State Emergency Department Databases (Micro-level data for 5 states, 2002–2014)

AHRQ’s Healthcare Cost and Utilization Project (Aggregate-level data for 15 states, 2000–2013)

  1. 1) Opioid-involved drug death rates (per 100k)

  2. 2) All drug mortality rate (per 100k)

  3. 3) Opioid overdose ED visit rates (per 100k)

  4. 4) Drug overdose ED visit rates (per 100k)

  1. 1) County or state-level unemployment rate

  2. 2) Employment-to-population ratios or percent changes in manufacturing employment or import exposure between 1990 and 2007

FE models with county (or state), year, and state-by-year or county-by-year FE

None

Opioid and other drug mortality rate and opioid ED visits are countercyclical

Not enough power to detect the relationship for other drug overdose ED visits

Jiménez-Martín, Labeaga and Vilaplana (2006)

US

BRFSS pseudo- panel data from the different cross sections available

Micro-level data 1987–2003

Past month’s:

  1. 1) Drinking participation

  2. 2) Number of drinks

  3. 3) Binge drinking

  4. 4) Heavy drinking (>60 and >100)

  5. 5) Drinking and driving

  6. 6) Number of drinks-to-drinkers ratio (in logs)

  1. 1) State-level unemployment rate

Cohort FE model with state, month, and year FE

Data weighted

  1. 1) Demographics

  2. 2) Education, marital status

  3. 3) Per capita income (state level)

  4. 4) State alcohol tax

No relationship to any of the definitions

Johansson et al. (2006)

Finland

Death Statistics Finland (Aggregate-level data, 1975–2001)

National Public Health Institute (Micro-level data, 1982–2001)

  1. 1) Number of alcohol-related deaths

  2. 2) Drinking participation during the survey week

  3. 3) Number of drinks during the last week

  1. 1) Regional-level employment rate

  2. 2) Change in regional real GDP

FE model with state and year FE

  1. 1) Demographics

  2. 2) Education, marital status

Alcohol-related mortality countercyclical for the whole period but procyclical during 1990 to 1996

Regional employment and drinking not related

Procyclical number of drinks with real GDP

Lam and Piérard (2017)

US

National Vital Statis-tics System of the National Center for Health Statistic

Aggregate-level data

1961–2004

  1. 1) Total mortality rate

  2. 2) 15–24, 25–34, 35–44, 45–54, 55–64, 65–74, 75–84, ≥85 years mortality rates

  3. 3) Deaths from

    cardiovascular and circulatory problems and motor vehicle accidents

  1. 1) Deviation of the unemployment rate from its trend level

TVP model

None

Total mortality remains strongly procyclical for the 15–24 and 85 and over age groups. For the other age groups, total mortality has become less procyclical over time

Deaths due to motor vehicle accidents are procyclical and remain procyclical, while in deaths due to cardiovascular diseases the procyclicality weakens over time

Latif (2014)

Canada

National Population Health Survey

Micro-level data 1994–2007

  1. 1) Obesity

  2. 2) Overweight

  3. 3) Severe obese

  4. 4) BMI

  1. 1) Province-level unemployment rate

FE model with province, year, and individual FE, with province-specific time trends

  1. 1) Demographics

  2. 2) Education, marital status

  3. 3) Wealth

  4. 4) Income

No relation with obesity, overweight

BMI and severe obesity are countercyclical

Lin (2009)

Asia-Pacific countries

Statistical Yearbook for Asia and the Pacific (UN), 2005 World Development Indicators (WB) and vital statistical data

Aggregate-level longitudinal data 1976–2003

  1. 1) Total mortality rate

  2. 2) Deaths due to cardiovascular diseases, motor vehicle accidents, suicides, and the infant mortality rates

  1. 1) Country unemployment rate

FE model with country, year fixed effects, and country-specific time trends

  1. 1) Age and sex distribution of population

  2. 2) Percentage of population living in urban areas

  3. 3) Number of physicians and hospitals per 10,000 population in each country

  4. 4) GDP per capita

Procyclical total mortality rate, deaths from cardiovascular diseases, motor vehicle accidents

No significant effect over suicides and infant mortality

Martín and Castelló (2016)

Spain

National Plan on Drugs

Micro-level data (2005, 2007, 2009, 2011)

Consumption in the last month and year:

  1. 1) Alcohol

  2. 2) Tobacco, smoke every day, number of cigarettes a day

  3. 3) Marijuana

  4. 4) Hard drugs: cocaine, crack, heroin, inhalants, hallucinogens, and ecstasy

  5. 5) Cocaine

  6. 6) Ecstasy

  1. 1) Province-level unemployment rate

FE models with province and year FE

  1. 1) Demographics

  2. 2) Education, marital status

  3. 3) Having good health

Tobacco, marijuana, and cocaine are countercyclical

Alcohol participation procyclical only for the year consumption

No relation with hard drugs and ecstasy

McInerney and Mellor (2012)

US

Medicare Current Beneficiary Survey, Community Tracking Study Physician Survey and National Cancer Institute’s Surveillance Epidemiology and End Results program

Aggregate and micro-level data 1976–2008

  1. 1) Mortality rate for the elderly (≥65 years)

  2. 2) Measures of general health for the elderly (poor or fair health, health limits activity, mental disorders)

  3. 3) Elderly’s health behaviors (smoking, weight disorders)

  4. 4) Healthcare utilization by the elderly

  1. 1) State unemployment rate

FE model with state, year FE, and state-specific time trends

  1. 1) State population characteristics (age, gender, ethnic, employment in construction and manufacturing)

Procyclical elderly mortality in 1976–1991 but countercyclical in 1994–2007

Procyclical physical and mental health

No effect on smoking and procyclical BMI

No significant effect over healthcare utilization

Miller and Urdinola (2010)

Colombia

Annual coffee prices and intensity of county coffee production (1970–2006) National Federation of Coffee Growers and the International Coffee Organization

1993 Colombian population census

Time-intensive child health investments: Colombia’s Demographic and Health Surveys (1986, 1990, 1995, and 2000)

Parental time use: Colombia’s Familias en Acción evaluation survey (2002, 2003, 2005)

  1. 1) Cohort size (ages 0–1 and 1–2)

  2. 2) Mothers’ labor outcomes: whether or not worked the week prior to the survey and hours of work in the past month

  3. 3) Infant/Child Morbidity

  4. 4) Adult Time Use

  5. 5) Health behavior

  1. 1) Interaction of internal birth year coffee price with the economic importance of coffee in one’s birth county

FE model with birth county and birth cohort fixed effects

Include year fixed effect for the labor market outcomes

  1. 1) County-specific linear trends

  2. 2) maternal characteristics: mother’s age, education, number of household members, number of preceding births, age at first birth, and age at first marriage

Countercyclical cohort size, procyclical mortality

Procyclical mother’s labor outcomes and children’s morbidity

Countercyclical child care (when the child is ill), health investment, vaccination use

Miller et al. (2009)

US

Centers for Disease Control and Prevention, and Health Interview Survey

Aggregate-level longitudinal data 1972/78–2004

  1. 1) Total mortality rate

  2. 2) Mortality rate for every age

  3. 3) Deaths due to cardiovascular disease, cancer, respiratory infections, degenerative brain disease, kidney disease,

    motor vehicle accidents, other accidents, suicide, homicide, other

    nutrition, birth defects, gastrointestinal, all non–motor vehicle accidents

  1. 1) State unemployment rate

  2. 2) “Own-group” and state unemployment rate

  3. 3) Employment- to-population ratios

Poisson count data model with state, year FE and state-specific linear trends

  1. 1) Demographic controls

Procyclical total mortality, stronger for those age 20–34 and young children

Procyclical deaths due to all causes except cancer and suicides

“Own-group” unemployment rate positive but insignificant while state remains negative

Neumayer (2004)

Germany

Health Statistical Database

Aggregate-level longitudinal data 1980–2000

  1. 1) Total mortality rate

  2. 2) 20–44, 45–64, ≥65 years mortality rates

  3. 3) Deaths due to malignant neoplasms, cardiovascular diseases, pneumonia or influenza, liver disease, motor vehicle accidents, other accidents, homicide, suicide, infant mortality, and neonatal mortality

  1. 1) State-level unemployment rate

  2. 2) Growth rate in real GDP

FE model with state and year FE

  1. 1) Demographic characteristics

  2. 2) Per capita personal income

  3. 3) Gini coefficient

Procyclical overall mortality rate, cardiovascular disease, suicide mortality rates

Procyclical mortality by motor vehicle accidents only for females and from pneumonia and influenza only for males

No effect on malignant neoplasms liver diseases, homicide, other accidents, infant and neonatal mortality

Pabilonia (2014)

US

Youth Risk Behavior Survey

Micro-level data 2003–2011

Teenagers’ in the past 30 days:

  1. 1) Smoke

  2. 2) Smoke daily

  3. 3) Alcohol use

  4. 4) Marijuana use

  1. 1) State-level unemployment rate

FE model with state and year FE

  1. 1) Demographics

  2. 2) Grade at school

  3. 3) State maximum compulsory schooling age

  4. 4) Logarithm of minimum wage

  5. 5) State minimum age for getting a driver’s license

  6. 6) Real state cigarette price

Countercyclical smoking every day for girls

Countercyclical alcohol for Hispanic boys

Countercyclical marijuana use for black boys

Paxson and Schady (2005)

Peru

Demographic and Health Surveys and Peru Living Standards Measurement Study

Micro-level data 1986, 1991/1992, 1996 and 2000—DHS; 1985/86 and 1991—LSMS

  1. 1) Mortality during the first year of life

  2. 2) Neonatal mortality (1 month or younger)

  3. 3) Mortality during the first 6 months of life

  4. 4) Antenatal care

  5. 5) Home births

  6. 6) Consumption patterns of food, semi-durables, and services

  1. 1) Log of per capita GDP

For mortality, correlation with time trends

For healthcare use variables, differences across years, controlling for maternal and child characteristics with mother-specific random effects or mother-specific fixed effects

For maternal selection, oaxaca-type decomposition of the changes in infant mortality over the years (time effect and selection effects)

None

Countercyclical infant mortality

Procyclical use of health services (prenatal care and home births), healthcare speeding, and medicines

No effect on patterns of food consumption

Selection does not account for the year-to-year changes in infant mortality

Ruhm (1995)

US

U.S. Brewers’ Association and the National Highway Traffic Safety Administration’s Fatal Accident Reporting System

Aggregated state-level

1975–1988

  1. 1) Total alcohol consumed (natural log of gallons)

  2. 2) Highway vehicle fatalities (total for 15–20, 21–24 years; nighttime

    for 15–20 and 21–24) years

  1. 1) State-level unemployment rate

  2. 2) Percent of the state population employed

FE model with state FE and linear time trends

  1. 1) Per capita income

  2. 2) State tax on beer

  3. 3) Minimum legal drinking age

Alcohol consumption and traffic deaths are procyclical

Ruhm (2000)

US

Vital Statistics of the US BRFSS

Aggregate-level longitudinal and Micro-level data 1972–1991, 1987–1995

  1. 1) Total mortality rate

  2. 2) 20–44, 45–64, ≥65 years mortality rates

  3. 3) Deaths due to malignant neoplasms, cardiovascular diseases, pneumonia or influenza, liver disease, motor vehicle accidents, other accidents, homicide, infant mortality, and neonatal mortality

    Last month’s:

  4. 4) Current smoking

  5. 5) Number of cigarettes

  6. 6) Drinking participation

  7. 7) Number of drinks

  8. 8) BMI

  9. 9) Overweight

  10. 10) Obese

  11. 11) Underweight

  12. 12) Any type of exercise

  13. 13) Regular exercise

  14. 14) Daily servings of fruits and vegetables

  15. 15) Grams of fat consumed daily

  1. 1) State-level unemployment rate

  2. 2) National unemployment rate

  3. 3) Personal income

FE model with state, year FE, and state-specific time trends

  1. 1) Demographics

  2. 2) Education, marital status

  3. 3) In some specifications: state-level median family income

Total mortality procyclical (larger for age group of 20–44 years)

Procyclical mortality for all causes except cancer (no effect) and suicides (countercyclical)

Smoking is procyclical

No relation to alcohol consumption

BMI, overweight, obesity, underweight, and consumption of fat are procyclical

No relation to consumption of fruits and vegetables

Exercise is countercyclical

Ruhm (2003)

US

National Health Interview Surveys

Micro-level data 1972–1981

  1. 1) At least one medical condition, chronic condition, or acute condition at the date of the survey

  2. 2) One or more restricted-activity day or bed day in the preceding two weeks

  3. 3) At least one hospitalization or doctor visit during the last year

  1. 1) State-level unemployment rate

FE model with state, year FE, and state-specific linear trends

  1. 1) Demographic characteristics

  2. 2) Per capita personal income

Procyclical medical, chronic, and acute conditions, restricted activity and bed days (although not significant the last two) and hospitalizations and doctor visits

Effect stronger for male and prime–working age adults

Countercyclical effect will be stronger without the protective effect of income

Ruhm (2005)

US

BRFSS

Micro-level data 1987–2000

  1. 1) Overweight

  2. 2) Obese

  3. 3) Severely obese

  4. 4) Irregular exercise

  5. 5) Physically inactive

  6. 6) Multiple: two or more (smoking, severe obesity, or physical inactivity)

  1. 1) State-level employment rate

FE model with state, month, and year FE (robustness state-year FE, state-specific linear time trends, and weighted data)

  1. 1) Demographics

  2. 2) Education, marital status

Obesity and severe obesity procyclical

Exercise is countercyclical

No relation with overweight

Ruhm (2015)

US

Centers for Disease Control and Prevention Compressed Mortality Files

Aggregate-level longitudinal data 1976–2010

  1. 1) Total mortality rate

  2. 2) <25, 25–44, 45–64, 65–74, ≥75 years mortality rates

  3. 3) Deaths due to diseases (cardiovascular disease, cancer, other diseases), external causes (transport accidents, other accident, suicides, homicides), and other accidents (falls, drowning/submersion, smoke/fire/flames, poisoning/noxious)

  1. 1) State unemployment rate

  2. 2) State nonemployment rate

FE model with state, year and state-specific time trends

None

Procyclical total mortality (stronger for men and young and middle-aged individuals), deaths from diseases, external causes, and other accidents 1976–1993

Procyclicality of mortality disappears for all ages in the recent years (1991–2010) and becomes countercyclical deaths from external causes and other accidents

Ruhm and Black (2002)

US

BRFSS

Micro-level data 1987–1999

Past month’s:

  1. 1) Drinking participation

  2. 2) Number of drinks

  3. 3) Binge drinking

  4. 4) Heavy drinking (>60 and >100)

  5. 5) Light drinking (<10 and <20)

  6. 6) Drinking and driving

  1. 1) State-level unemployment rate

FE model with state FE and linear month time trends (robustness state-year FE)

Data weighted

  1. 1) Demographics

  2. 2) Education, marital status

  3. 3) Per capita income (state level)

  4. 4) State beer tax

No effect on binge drinking

Drinking participation, number of drinks, alcohol-involved driving, and heavy drinking procyclical

Light drinking countercyclical

Stevens et al. (2015)

US

Vital Statistics’ micro-record, Centers for Disease Control and Prevention, and Health Interview Survey

Aggregate-level longitudinal data 1978–2006

  1. 1) Total mortality rate

  2. 2) Mortality rate for every age

  3. 3) Mortality rate by place of death (nursing home vs. all other)

  1. 1) State unemployment rate

  2. 2) “Own-group” and “Other-group” unemployment rate

  3. 3) Employment- to-population ratios

FE model with state, year FE, and state-specific linear trends

  1. 1) State population characteristics (age, education, ethnic)

Procyclical total mortality, stronger for females, and individuals age 20–24 and young children

“Own-group” unemployment rate insignificant and opposite sign

Deaths occurring in nursing homes are particularly responsive

Svensson (2007)

Sweden

National Board of Health and Welfare

Aggregate-level longitudinal data 1987–2003

Acute myocardial infarction:

  1. 1) Incidence

  2. 2) Mortality

  3. 3) Lethality on day 0 of the incidence

  4. 4) Lethality day first month

  5. 5) Lethality day first year

  1. 1) Regional unemployment rate

  2. 2) Change in regional GDP

  3. 3) Regional employment rate

FE model with region, year FE, and region-specific time trends

  1. 1) Share of individuals with high education, foreign, and men

No relationship with general incidence, mortality, and lethality

Countercyclical for individuals with prime working age

Granados (2005)

Spain

Mortality statistics

Aggregate-level longitudinal data 1980–1997

  1. 1) Total mortality rate

  2. 2) Mortality rates due to cardiovascular disease, cancer, respiratory disease, traffic injuries, infectious disease, suicide, or homicide

  1. 1) Province unemployment rate

  2. 2) National unemployment rate

  3. 3) Province employment rate

FE model with province, year FE, and province-specific time trends

  1. 1) Province age structure (proportion of population under 5 and over 64)

  2. 2) Province real GDP per capita

Procyclical total mortality rate (effect stronger for men) and mortality rate due to traffic injuries

Procyclical, but not significant, mortality rates due to cardiovascular disease, cancer, respiratory disease, and infectious disease

Countercyclical, but not significant, mortality rates due to suicide

No effect on homicides

Tekin et al. (2013)

US

BRFSS

Micro-level data 1990–2014

  1. 1) General health (excellent; poor; fair to poor)

  2. 2) Poor mental health (>10 and >20 days/month)

  3. 3) Current smoker

  4. 4) Daily smoker

  5. 5) Current drinker

  6. 6) Binge drinker

  7. 7) Chronic drinker

  8. 8) Physical exercise

  9. 9) Overweight

  10. 10) Obese

  11. 11) Severely obese

  1. 1) State-level unemployment rate

  2. 2) State-level employment rate

FE model with state, year, month FE, and state-

specific linear time trends

Weighted data

  1. 1) Demographics

  2. 2) Education, marital status

No relationship (estimates small and imprecisely estimated) with self-reported health and mental health

Smoking is procyclical, although the relationship gets weaker during the Great Recession

No robust significant relation with physical exercise and overweight, obesity or severe obesity

Xu (2013)

US

BRFSS combined with Current Population Survey

Micro-level data 1984–2005

  1. 1) Current smoker

  2. 2) Having >10 cigarettes

  3. 3) Having >20 cigarettes

  4. 4) Any alcohol use

  5. 5) Binge drinking

  6. 6) Heavy drinking

  7. 7) Any physical activity

  1. 1) As instrument: state-level unemployment rate and industry mix

Two-sample instrumental variables approach with a FE model with state and year FE

  1. 1) Demographics

  2. 2) Education, marital status

  3. 3) Local cigarette and beer taxes

Procyclicality of smoking

No relation with alcohol consumption

Countercyclical physical exercise with hours of work

No relation with wages

References

Almond, D., Hoynes, H. W., & Schanzenbach, D. W. (2011). Inside the war on poverty: The impact of food stamps on birth outcomes. The Review of Economics and Statistics, 93(2), 387–403.Find this resource:

An, R., & Liu, J. (2012). Local labor market fluctuations and physical activity among adults in the United States, 1990–2009. ISRN Public Health, 2012, 7. Article ID 318610.Find this resource:

Aparicio-Fenoll, A., & Gonzalez, L. (2014). Newborn health and the business cycle: Is it good to be born in bad times? IZA Discussion Papers, 8031.Find this resource:

Ariizumi, H., & Schirle, T. (2012). Are recessions really good for your health? Evidence from Canada. Social Science & Medicine, 74(8), 1224–1231.Find this resource:

Arkes, J. (2007). Does the economy affect teenage substance use? Health Economics, 16(1), 19–36.Find this resource:

Arkes, J. (2009). How the economy affects teenage weight. Social Science & Medicine, 68(11), 1943–1947.Find this resource:

Arkes, J. (2011). Recessions and the participation of youth in the selling and use of illicit drugs. International Journal of Drug Policy, 22(5), 335–340.Find this resource:

Baird, S., Friedman, J., & Schady, N. (2011). Aggregate income shocks and infant mortality in the developing world. Review of Economics and Statistics, 93(3), 847–856.Find this resource:

Bassols, N. M., & Castelló, J. V. (2016). Effects of the great recession on drugs consumption in Spain. Economics & Human Biology, 22, 103–116.Find this resource:

Bellés-Obrero, C., Jiménez-Martín, S., & Vall-Castello, J. (2016). Bad times, slimmer children? Health Economics, 25(S2), 93–112.Find this resource:

Bhalotra, S. (2010). Fatal fluctuations? Cyclicality in infant mortality in India. Journal of Development Economics, 93(1), 7–19.Find this resource:

Böckerman, P., Johansson, E., Helakorpi, S., Prättälä, R., Vartiainen, E., & Uutela, A. (2007). Does a slump really make you thinner? Finnish micro-level evidence 1978–2002. Health Economics, 16(1), 103–107.Find this resource:

Bor, J., Basu, S., Coutts, A., McKee, M., & Stuckler, D. (2013). Alcohol use during the great recession of 2008–2009. Alcohol and Alcoholism, 48(3), 343–348.Find this resource:

Bozzoli, C., & Quintana-Domeque, C. (2014). The weight of the crisis: Evidence from newborns in Argentina. Review of Economics and Statistics, 96(3), 550–562.Find this resource:

Bradford, W. D., & Lastrapes, W. D. (2014). A prescription for unemployment? Recessions and the demand for mental health drugs. Health Economics, 23(11), 1301–1325.Find this resource:

Brenner, M. H. (1973). Mental illness and the economy. Cambridge, MA: Harvard University Press.Find this resource:

Brenner, M. H. (1975). Trends in alcohol consumption and associated illnesses: Some effects of economic changes. American Journal of Public Health, 65(12), 1279–1292.Find this resource:

Brenner, M. H. (1979). Mortality and the national economy: A review, and the experience of England and Wales, 1936–76. The Lancet, 314(8142), 568–573.Find this resource:

Brenner, M. H. (1987). Economic instability, unemployment rates, behavioral risks, and mortality rates in Scotland, 1952–1983. International Journal of Health Services, 17(3), 475–487.Find this resource:

Brenner, M H. 1995. Political economy and health. Society and Health, 211–246.Find this resource:

Buchmueller, T., Grignon, M., & Jusot, F. (2007). Unemployment and mortality in France. Centre for Health Economics and Policy Analysis (CHEPA), McMaster University.Find this resource:

Byron, S., & Fertig, A. (2012). The effect of business cycles and parental indebtedness on childhood obesity. Retrieved from http://www.ucdenver.edu/academics/colleges/CLAS/Departments/economics/Seminars/Documents/Byron%20Fertig%20-%20Business%20Cycles%20on%20Childhood%20Obesity.pdf.

Camacho, A. (2008). Stress and birth weight: Evidence from terrorist attacks. The American Economic Review, 98(2), 511–515.Find this resource:

Carpenter, C. S., McClellan, C. B., & Rees, D. I. (2017). Economic conditions, illicit drug use, and substance use disorders in the United States. Journal of Health Economics, 52, 63–73.Find this resource:

Charles, K. K., & DeCicca, P. (2008). Local labor market fluctuations and health: Is there a connection and for whom? Journal of Health Economics, 27(6), 1532–1550.Find this resource:

Clark, K. B., & Summers, L. H. (1982). Labour force participation: Timing and persistence. The Review of Economic Studies, 49(5), 825–844.Find this resource:

Colman, G., & Dave, D. (2013). Exercise, physical activity, and exertion over the business cycle. Social Science & Medicine, 93, 11–20.Find this resource:

Dávalos, María E., Fang, H., & French, M. T. (2012). Easing the pain of an economic downturn: Macroeconomic conditions and excessive alcohol consumption. Health Economics, 21(11), 1318–1335.Find this resource:

Dave, D. M., & Kelly, I. R. (2012). How does the business cycle affect eating habits? Social Science & Medicine, 74(2), 254–262.Find this resource:

Dee, T. S. (2001). Alcohol abuse and economic conditions: Evidence from repeated cross-sections of individual-level data. Health Economics, 10(3), 257–270.Find this resource:

Dehejia, R., & Lleras-Muney, A. (2004). Booms, busts, and babies’ health. The Quarterly Journal of Economics, 119(3), 1091–1130.Find this resource:

Economou, A., Nikolaou, A., & Theodossiou, I. (2008). Are recessions harmful to health after all? Evidence from the European Union. Journal of Economic Studies, 35(5), 368–384.Find this resource:

Ettner, S. L. (1997). Measuring the human cost of a weak economy: Does unemployment lead to alcohol abuse? Social Science & Medicine, 44(2), 251–260.Find this resource:

Freeman, D. G. (1999). A note on economic conditions and alcohol problems. Journal of Health Economics, 18(5), 661–670.Find this resource:

Frijters, P., Johnston, D. W., Lordan, G., & Shields, M. A. (2013). Exploring the relationship between macroeconomic conditions and problem drinking as captured by Google searches in the US. Social Science & Medicine, 84, 61–68.Find this resource:

Gerdtham, U. G., & Johannesson, M. (2005). Business cycles and mortality: results from Swedish microdata. Social Science & Medicine, 60(1), 205–218.Find this resource:

Gerdtham, U.-G., & Ruhm, C. J. (2006). Deaths rise in good economic times: Evidence from the OECD. Economics & Human Biology, 4(3), 298–316.Find this resource:

Gonzalez, F., & Quast, T. (2011). Macroeconomic changes and mortality in Mexico. Empirical Economics, 40(2), 305–319.Find this resource:

Granados, J. A. T. (2005). Recessions and mortality in Spain, 1980–1997. European Journal of Population/Revue européenne de Démographie, 21(4), 393–422.Find this resource:

Hollingsworth, A., Ruhm, C. J., & Simon, K. (2017). Macroeconomic conditions and opioid abuse. Journal of Health Economics, 56, 222–233.Find this resource:

Jiménez-Martín, S., Labeaga, José M., & Prieto, C. V. (2006). Further evidence about alcohol consumption and the business cycle. DOCUMENTO DE TRABAJO, 2006(06).Find this resource:

Johannesson, M. (2005). Business cycles and mortality: Results from Swedish microdata. Social Science & Medicine, 60(1), 205–218.Find this resource:

Johansson, E., Böckerman, P., Prättälä, R., & Uutela, A. (2006). Alcohol-related mortality, drinking behavior, and business cycles. The European Journal of Health Economics, 7(3), 212–217.Find this resource:

Lam, J.-P., & Piérard, E. (2017). The time-varying relationship between mortality and business cycles in the USA. Health Economics, 26(2), 164–183.Find this resource:

Latif, E. (2014). The impact of recession on drinking and smoking behaviours in Canada. Economic Modelling, 42, 43–56.Find this resource:

Lin, S.-J. (2009). Economic fluctuations and health outcome: A panel analysis of Asia-Pacific countries. Applied Economics, 41(4), 519–530.Find this resource:

Lindo, J. M. (2015). Aggregation and the estimated effects of economic conditions on health. Journal of Health Economics, 40, 83–96.Find this resource:

Martin, B. N., & Vall, C. J. (2016). Effects of the great recession on drugs consumption in Spain. Economics and Human Biology, 22, 103.Find this resource:

McInerney, M., & Mellor, J. M. (2012). Recessions and seniors’ health, health behaviors, and healthcare use: Analysis of the Medicare Current Beneficiary Survey. Journal of Health Economics, 31(5), 744–751.Find this resource:

Miller, D. L., Page, M. E., Stevens, A. H., & Filipski, M. (2009). Why are recessions good for your health? The American Economic Review, 99(2), 122–127.Find this resource:

Miller, G., & Urdinola, B. P. (2010). Cyclicality, mortality, and the value of time: The case of coffee price fluctuations and child survival in Colombia. Journal of Political Economy, 118(1), 113–155.Find this resource:

Neumayer, E. (2004). Recessions lower (some) mortality rates: Evidence from Germany. Social Science & Medicine, 58(6), 1037–1047.Find this resource:

Pabilonia, S. W. (2014). The effects of the great recession on teenagers’ risky health behaviors and time use. IZA Discussion Paper No. 8204.Find this resource:

Pacula, R. L. (2011). Substance use and recessions: What can be learned from economic analyses of alcohol? International Journal of Drug Policy, 22(5), 326–334.Find this resource:

Paxson, C., & Schady, N. (2005). Child health and economic crisis in Peru. The World Bank Economic Review, 19(2), 203–223.Find this resource:

Rehm, J., Mathers, C., Popova, S., Thavorncharoensap, M., Teerawattananon, Y., & Patra, J. (2009). Global burden of disease and injury and economic cost attributable to alcohol use and alcohol-use disorders. The Lancet, 373(9682), 2223–2233.Find this resource:

Ruhm, C. J. (1995). Economic conditions and alcohol problems. Journal of Health Economics, 14(5), 583–603.Find this resource:

Ruhm, C. J. (2000). Are recessions good for your health? The Quarterly Journal of Economics, 115(2), 617–650.Find this resource:

Ruhm, C. J. (2003). Good times make you sick. Journal of Health Economics, 22(4), 637–658.Find this resource:

Ruhm, C. J. (2005). Healthy living in hard times. Journal of Health Economics, 24(2), 341–363.Find this resource:

Ruhm, C. J. (2015). Recessions, healthy no more? Journal of Health Economics, 42, 17–28.Find this resource:

Ruhm, C. J. (2016). Health effects of economic crises. Health Economics, 25(S2), 6–24.Find this resource:

Ruhm, C. J., & Black, W. E. (2002). Does drinking really decrease in bad times? Journal of Health Economics, 21(4), 659–678.Find this resource:

Stein, A. D., & Lumey, L. H. (2000). The relationship between maternal and offspring birth weights after maternal prenatal famine exposure: The Dutch famine birth cohort study. Human Biology, 72(4), 641–654.Find this resource:

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.Find this resource:

Substance Abuse and Mental Health Services Administration. (2013). Results from the 2012 National Survey on Drug Use and Health: Summary of national findings, NSDUH Series H-46, HHS Publication No. (SMA) 13–4795. Rockville, MD: Substance Abuse and Mental Health Services Administration.Find this resource:

Svensson, M. (2007). Do not go breaking your heart: Do economic upturns really increase heart attack mortality? Social Science & Medicine, 65(4), 833–841.Find this resource:

Tekin, E., McClellan, C., & Minyard, K. J. (2013). Health and health behaviors during the worst of times: Evidence from the great recession. Retrieved from http://www.nber.org/papers/w19234.

Torche, F. (2011). The effect of maternal stress on birth outcomes: Exploiting a natural experiment. Demography, 48(4), 1473–1491.Find this resource:

U.S. Department of Health and Human Services. (2014). The health consequences of smoking—50 years of progress. A report of the Surgeon General. Atlanta: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health.Find this resource:

Volkow, N. D. (2014). Prescription opioid and heroin abuse. National Institute on Drug Abuse. The Science of Drug Abuse and Addiction. Bethesda, MD: National Institutes of Health.Find this resource:

World Health Organization, & World Health Organization. Management of Substance Abuse Unit. (2014). Global status report on alcohol and health, 2014. World Health Organization.Find this resource:

Xu, X. (2013). The business cycle and health behaviors. Social Science & Medicine, 77, 126–136.Find this resource:

Notes:

(1.) Ettner (1997) uses a two-stage instrumental variables method using unemployment rates to instrument individual labor market status. Yet there are some doubts about the validity of the exclusionary restriction.

(2.) Binge drinking is defined by Dee (2001) as consuming five or more drinks on a single occasion.

(3.) Ruhm and Black (2002) use similar data to Dee (2001). Yet they argue that early waves of the BRFSS data contain only 15 states, so they only use data from 1987 on. Moreover, Ruhm and Black (2002), unlike Dee (2001), weighted the data to account for differences in sampling probabilities. Finally, Ruhm and Black (2002) further control for alcohol taxes or prices.

(4.) Pacula (2011) argues that these contradictory findings could be a consequence of the important differences in the level of economic development across different provinces in Finland, which might have biased the results.

(5.) The notification rate is defined by Gerdtham and Johannesson (2005) as the ratio between workers who receive advance notice of impending dismissal from their jobs and the total number of persons in the labor force. In Sweden there is a mandatory notice period for employers that ranges from 2 to 6 months—or longer if the workers are covered by collective agreements made by the unions.