The European Journal of Health Economics (2021) 22:1289–1299 https://doi.org/10.1007/s10198-021-01336-7 ORIGINAL PAPER Business cycle and mortality in Spain María Cervini‑Plá1,2 · Judit Vall‑Castelló3,4 Received: 16 September 2020 / Accepted: 8 June 2021 / Published online: 23 June 2021 © The Author(s) 2021, corrected publication 2021 Abstract In the last couple of decades, there has been a lot of interest on the impact of macroeconomic fluctuations on health and mortality rates. Many studies, for different countries, find that mortality is procyclical. However, studies examining the effects of more recent recessions are less conclusive, finding mortality to be less procyclical, or even countercyclical. In this paper, using data of Spanish provinces from 1999 to 2016, we investigate how this relationship works in the context of a country that is subject to extreme business cycle fluctuations. Furthermore, we analyze the impact of unemployment for different mortality causes and we explore differences by sex, age group and level of education. In general terms, we find mortality to be procyclical so that when the economy is in a recession, mortality falls. When exploring mortality causes, we show that deaths from cardiovascular disease, cancer, senility, transport accidents and homicides are procyclical. By sex, we find procyclicality for both men and women. By age, mortality is procyclical for all age groups; however, the causes of death that result in this procyclical behavior are specific to each age group. By educational level, suicide appears as a countercyclical cause for individuals with intermediate levels of education. Keywords Mortality · Health · Recessions · Business cycle JEL Classification E32 · I10 · I12 · I14 · J14 · J16 Introduction his paper, Ruhm combined time-series with cross-sectional data to better control for omitted variable bias. There is a widely interest among social science researches In the last couple of decades, many studies have attempted and policymakers in understanding the relationship between to replicate these findings with the aim of understanding macroeconomic fluctuations and health, with a particular the specific causes that could explain this relationship using focus on mortality. multiple techniques and datasets. However, the results of Aggregated time-series data by Brenner [1, 2] disclosed this large literature show ambiguous conclusions: on the one that deaths by cardiovascular disease, cirrhosis, suicide hand, there are many papers supporting the idea that mor- and homicide, and infant mortality rates decrease when tality rates decrease during temporary economic recessions economic conditions improve. Nevertheless, the effects of [6, 12–14, 21, 23, 24] so that “recessions are good for your business cycle conditions on mortality have become more health”. On the other hand, fewer studies reveal the oppo- controversial since the work by Ruhm [15], where he showed site finding pointing towards health improvements during that mortality increases in periods of economic growth. In booms as suggested by the first papers on the literature in the 1970’s [5, 20]. * María Cervini‑Plá In 2008 Edwards [4] pointed towards the importance maria.cervini@uab.cat of the level of aggregation used for the analysis. He used both micro and macrolevel data and found that, when using 1 Universitat Autònoma de Barcelona, Edifici B. Campus aggregated data, the results show a procyclical relationship UAB, Bellaterra (Cerdanyola del Vallès), 08193 Barcelona, Spain between mortality and business cycle fluctuations while at 2 the individual level the results are more mixed depending EQUALITAS, Madrid, Spain on the subgroup of the population studied. Therefore, his 3 Universitat de Barcelona, Barcelona, Spain findings highlight the fact that the level of analysis plays a 4 CRES (UPF), Barcelona, Spain 13 Vol.:(0123456789) 1290 M. Cervini‑Plá, J. Vall‑Castelló crucial role in the sign of the relationship precisely because contribution of each specific cause of death in order to fully business cycle effects are heterogeneous across individu- understand the relationship between mortality and business als. Therefore, the composition of the population plays an cycle conditions. important role in defining the association at the macrolevel. Nonetheless, the factors explaining the divergence of Another more recent paper that explore the opposite- findings in the literature are not entirely clear yet. Some direction effects of contextual and individual unemploy- authors have provided some examples: Ruhm [16] suggests ment on mortality is Tapia Granados et al. [24]. They show that changes over time in the relationship between business how an increase in the unemployment rate correlates with a cycle conditions and mortality may be driven by the fact decrease in overall mortality or mortality due to most causes that economic instability over time is poorly measured when (procyclicality). However, at the same time, at the individ- using short periods of analysis. Another possible reason is ual level, to suffer unemployment (because of being fired, the fact that the mechanisms behind the association between plant closure or whatever) seems clearly connected with an macroeconomic conditions and mortality may not be stable increase in the risk of death. Using US data they find that over time due to the changes in institutions or changes in compared to be employed, for those who experienced unem- health behaviors. Cutler et al. [3] also point in this direction, ployment the hazard of death was increased by an amount since they find that the intensity and even the direction of equivalent to 10 extra years of age, and, at the same time, the relationship between mortality rates and macroeconomic each percentage-point increase in the state unemployment conditions can be directly related to government spending rate (contextual unemployment) reduced the mortality haz- levels. Going one step further, Stevens et al. [17] presents ard in all individuals by an amount equivalent to a reduction evidence that staff in nursing homes in the USA has a coun- of 1 year of age. In line with this result at the microlevel, ter-cyclical behavior so that (at least part) of the mortality Sullivan and von Wachter [19] find that the risk of death is increases during booms can be explained by fluctuations in higher in individuals exposed to unemployment. the quality of the health care provided. The recent recession that affected the global economy Our paper contributes to this literature by examining in 2008 marked a new opportunity to study this relation- the relationship between macroeconomic conditions and ship. The new wave of papers looking at the relationship mortality in Spain, a country that has been severely hit by between mortality and business cycle conditions find even the recent economic crisis. Using very rich administrative more diverging conclusions: some authors point towards a data (mortality registers) covering the period 1999–2016, reduction in the previously documented procyclicality when we link local (provincial) level unemployment rates to local using post-2008 data as compared to papers using pre-2008 level mortality rates while controlling for the age structure data [7, 8, 13, 14, 17] another group of authors find the of the population, regional and time fixed effects, as well relationship to be countercyclical [9, 11] while still a third as regional trends. The strong variability in business cycle group of authors report mortality to be unrelated to macro- conditions over time in Spain (as seen in Fig. 1) represents economic conditions [16]. a powerful element that strengthens the identification strat- Ruhm [16] suggests that changes in health behaviors such egy and potentially allows us to increase the precision of as the reduction in drinking, obesity, smoking and physical the estimates. We believe that strong variations in the main inactivity during recessions represent a credible and power- explanatory variable in Spain vis-à-vis the relatively smaller ful mechanism that backs up the procyclicality argument. variation in other developed countries previously studied Cutler et al. [3] find that pollution also has explanatory in the literature is a comparative advantage that will help power in the relationship between mortality and business us better identify the relationship between business cycle cycle conditions as they report that, in places where pol- conditions and mortality rates. We also explore the impacts lution is low (i.e. agricultural economies), mortality rates for specific causes of death as well as differences by age, sex are countercyclical. However, when pollution is important and educational level. and is highly determined by the level of economic activity, As far as we are aware, the Spanish case has been previ- mortality rates become procyclical. This finding is consist- ously studied in Tapia Granados [22] and Regidor et al. [13, ent with the work of Ionides et al. [10] showing that pro- 14]. The paper by Tapia Granados [22] explores the relation- cyclical mortality is driven mostly by respiratory diseases ship between economic conditions and mortality for Spain and traffic injuries. Similarly, Strumpf et al. [18] estimate using data from 1980 to 1997. His results show a procyclical that increases in the local unemployment rates are associ- relationship for the Spanish case although the size of the ated with reductions in all-cause mortality rates and, 60% effects are smaller when compared to other countries: more of this decrease, is explained by cardiovascular diseases. specifically, he estimates a drop of 0.11% in mortality for a They also find reductions in motor vehicle fatalities during one percentage point increase in the local unemployment economic recessions, especially for men under age 65. All rate. Our paper presents several novelties with respect to these new papers strongly suggest the need to explore the Tapia Granados [22]. First, we include observations until 13 Business cycle and mortality in Spain 1291 Fig. 1  Unemployment rates (1999–2013). Eurostat 2016 which allows us to capture the impacts of the recent Great Recession, using Census data and following individu- economic crisis, which represented an unprecedented shock als until 2011. Regidor et al. [14] analyse mortality trends for the Spanish economy that raised unemployment rates in different socioeconomic groups, quantifying the change over 25%, as well as periods with an extremely booming within each group. They classified individuals by socioeco- economy (such as 2005–2007) with unemployment rates nomic status (low, medium, or high) using two indicators reaching minimum historical values below 10%. Therefore, of household wealth: household floor space and number of the effects that we find are much larger than those reported cars owned by the residents of the household. They find in Tapia Granados [22] as our results show that mortal- that in all socioeconomic groups, the all-cause mortality rate ity reduces by 0.42% for a once percentage point increase declined more during the first 4 years of the crisis period in the local unemployment rate. Furthermore, having this (2008–11) than in the 4 years receding the crisis (2004–07), updated information is also an important contribution to the except for women in the high socioeconomic group, whose recent literature as several papers have reported potential mortality declined less during the crisis. Exploring cause- changes in the relationship between business cycle condi- specific mortality rates, they show a similar pattern, except tions and mortality in recent years. Our results show that, for cancer in all women and for traffic and other uninten- for the Spanish case, mortality is still strongly procyclical tional injuries in women in the high socioeconomic group. and this relationship has not changed in the recent years, as Furthermore, the acceleration of the downward linear trend has been documented for other countries. On the contrary, in mortality was higher in the lowest socioeconomic group the relationship has become larger as the coefficient in our than in the highest group for all-cause deaths and for most of estimates is larger than that reported in previous papers for the specific causes of death analysed. Regidor et al. [13] ana- Spain. Furthermore, we present a detailed analysis of the lyse mortality rates for employed, unemployed and inactive impact for several sources of death which allows us to better individuals. They find that for employed and unemployed understand the mechanisms driving this relationship. Finally, men, mortality rates increased until 2007 and then declined, we also focus on sex, age and educational differences that whereas in employed and unemployed women, mortality were previously undocumented. rates increased and then stabilized during 2008–2011. The More recently, we find two papers Regidor et al. [13, 14] mortality rate among inactive men and women decreased that analyse the mortality rate in Spain before and during the throughout the follow-up. Thus, they find that during the 13 1292 M. Cervini‑Plá, J. Vall‑Castelló economic crisis, the upward trend in the rate of mortality is an important determinant of total mortality and mortality from all diseases was inverted in the groups of employed and due to specific causes. Thus, to control for the demographic unemployed subjects at baseline. Compared with these two structure of the population and any potential changes over interesting most recent papers, our article provides a longer time (that might not be fully captured by the province-spe- period, a record of all the people who have died in Spain and cific linear trends), we also include as demographic controls a detailed analysis by age and educational level. the proportion of the population above 65 years old A65jt Our results suggest that, when the economy is in a reces- as well as the proportion of the population below 15 years sion, mortality rate falls. This result is consistent with the old A15jt . In that way, the results of the model reflect how findings in Tapia Granados [22] and suggests that the rela- age-adjusted mortality responds to the changes in economic tionship between business cycle conditions and mortality conditions. has not changed in the last 30 years. Additionally, our study To estimate the relative effects we can estimate, following detects that, while cardiovascular deaths, senility diseases, the same idea than Ruhm [16]: cancer and deaths by transport accidents are procyclical, countercyclical patterns emerge for suicide and diabetes. (e𝛾̂ k − 1) ∗ 100% With respect to sex differences, our results show that men This expression gives us the predicted percentage change are significantly more affected by transport accidents than in mortality from source k resulting from a one percent- women in Spain. Although mortality is procyclical for all age point increase in the unemployment rate. These relative age groups , the causes that lead this pro-cyclicality are dif- effects are translated into absolute numbers through esti- ferent in each age segment. Finally, suicide appears as a mates of: countercyclical cause for intermediate education levels. This paper is organized as follows. "Methodology" sec- (eΔ𝛾̂ k − 1) ∗ 𝜋k D, tion presents our methodology. "Data and descriptive sta- tistics" section explains the dataset used and our sample where Δ𝛾̂ k = 𝛾̂2k − 𝛾̂1k and 𝜋k is the share of deaths due to selection. "Results" section shows the main results and "Het- source k and D is the average annual number of deaths erogeneous effects" section explores heterogeneities by sex, (384,710). age and education. Finally, "Conclusion" section concludes. In particular, we estimate Eq (1) for the total mortality rate and by specific mortality causes using annual data by province. Furthermore, we explore differences in the impact Methodology of business cycle conditions on annual mortality rates by sex, age groups and educational level using group-specific To analyse the relationship between macroeconomic condi- outcomes. tions and mortality rates, we follow very close Ruhm [16]. We consider 52 geographic units that, for simplicity, we Our dependent variable is the mortality rate calculated as the call provinces. However, strictly speaking what we include number of deaths in a particular cell (we use different defini- corresponds to 50 provinces plus Ceuta and Melilla, that tions of a cell) divided by the population in that particular are autonomous cities. All regressions are weighted by the cell. More specifically, we estimate the following equation: squared root of the population in each geographic unit and time, and standard errors are clustered by geographic unit. ln(Mkjt ) =𝛼kj + Ukjt 𝛾 + 𝜆kt + Hj ∗ t + A15jt 𝜂 (1) + A65jt 𝛿 + 𝜖kjt Data and descriptive statistics where Mkjt is the mortality rate from diseases k in province j at certain moment of time t, U is the unemployment rate As an indicator of macroeconomic conditions, we use in that province at that time (a proxy of the economic con- the unemployment rate. We obtain annual and quarterly ditions), 𝛼 are province fixed-effects that control for time- unemployment rate information for each province from the invariant province characteristics (such as persistent lifestyle National Institute of Statistics in Spain (INE). disparities between residents of two different provinces), 𝜆 We use mortality registers for the period 1999–2016 accounts for calendar time fixed effects (which capture any also from the INE. The restricted access version of the files determinant of mortality that vary over time but in the same allow us to include information on the type of diseases that way across provinces: e.g. advances in widely used medical caused the death. The registers are collected at the individual technologies or behavioural norms), Hj ∗ t is a province- level and include all deaths in Spain from 1990 to 2016, the specific linear trend and 𝜖 is the error term. Therefore, 𝛾 is diseases that caused the death and some individual char- our parameter of interest and captures the effects of the cycle acteristics such as sex, age, education and province of the on mortality. Finally, we acknowledge that the age structure person that died. We collapse the individual information by 13 Business cycle and mortality in Spain 1293 province for each year to get the number of individuals dying not clear; while in the first period it seems that both vari- in each province and time period. We aggregate the informa- ables move in parallel, after the crisis it looks like they go tion for total mortality but also for different causes of death. in opposite directions, recovering a the parallel path in the To do the analysis of cause-specific mortality we classify the last years of the analyzed period. different mortality causes in 12 categories: cardiovascular, Table 6 in Appendix shows some descriptive statistics cancer, respiratory, senility, Alzheimer, diabetes, digestive, of our sample. With respect to the causes of death, we can other diseases, transport accidents, other accidents, suicides observe that the most common causes of death are: cardio- and homicides. To calculate mortality rates, we divide the vascular problems, cancer and respiratory problems. As for number of deaths by the population (and multiply it by age, we can see that the proportion of individuals dying is 1000), using information on population numbers in the same highly concentrated in the oldest age group. More than 70% province and year. of the deaths occur for people who have more than 75 years In addition to total annual mortality rates for each prov- old. Regarding the level of education, most of individuals in ince, we construct sex-specific death rates, mortality rates our sample have a relatively low level of education. This is for five age groups (< 25, 25–44, 45–64, 65–74, > 74), as due to the fact that the education level of the elderly is much well as by level of education for those aged over 25 years lower than the average for the entire population. old (we only have information about education from 2012 Figure 3 in the Appendix shows the evolution of the dif- onwards). ferent mortality causes in the period analyzed. On the left Table 5 in Appendix presents information, for each of panel we can see a slight reduction in deaths from cardio- the years included in our database, on the total number of vascular problems, an increase in cancer mortality and dis- people dying (total mortality), the total number of individu- eases associated with an aging population such as senility als (population) and the mortality rate. As can be seen in the and Alzheimer. On the right hand side of the panel, we can table, both the number of people who die and the total popu- see a continuous reduction in traffic accidents throughout lation is increasing in our period. However, the mortality the period, which coincides with an increased governmental rate has more volatility and it is increasing in some periods effort to raise public awareness on the importance of enforc- and decreasing in others. Figure 2 shows the evolution of ing traffic rules. Other types of accidents are increasing dur- the mortality rate and the unemployment rate in Spain from ing the period and suicides appear to have increased after 1999 to 2016. The relationship between the two variables is the crisis. Fig. 2  Mortality rate and unemployment rate 13 1294 M. Cervini‑Plá, J. Vall‑Castelló Results across several countries and this is particularly the case for transport accidents and cardiovascular diseases. At the mac- In the first row of Table 1, we examine the relationship rolevel, our data (and the majority of the previous literature) between economic conditions and total mortality rates with seems to support the idea that in times of economic booms, annual information. Our parameter of interest, the local level and higher employment probabilities, there are more car- unemployment rate, has a significant and negative coeffi- diovascular deaths. As we mentioned in the introduction, cient. Thus, we find a procyclical relationship between Tapia Granados et al. [24] explore the paradox between what business cycle conditions and mortality in Spain during our is found at the micro and macroeconomic levels. Although sample period. That is, once long-term trends are adjusted it is true that higher unemployment at the individual level for, when the local unemployment rate increases because can generate more mortality associated with greater stress, the economy is in a recession, mortality falls. We find that at the macrolevel, a higher unemployment rate is associated a one percentage point increase in the local unemployment (in most countries) with lower mortality from most causes rate reduces the mortality rate by 0.42%. The estimation (including cardiovascular). For example, the reduction in include year and province fixed effect, linear trends, and mobility leads to a lower number of transport accidents. age controls. In the following rows of Table 1, we present separate results for each of the 12 categories of cause-specific mortal- Heterogeneous effects ity. We find that some causes are strongly and significantly procyclical, such as cardiovascular diseases, cancer, senility In the last two rows of Table 1, we present results separately diseases, transport accidents and homicide. Another group for men and women using annual data. We find a statistically of diseases have a negative sign but the coefficient is only significant procyclical coefficient for both, men and women. significant at the 10% level, such as respiratory diseases or For the case of men, we estimate that a one percentage point other accidents. We also find that a number of diseases show increase in unemployment leads to a reduction in mortality a counter-cyclical relationship with business cycle condi- by 0,41%. For women the coefficient is also negative but tions, but not significant, such as Alzheimer and diabetes. somewhat smaller; a one percentage point increase in the Some of these relationships have been found in papers look- unemployment rate is associated with a reduction of 0.25% ing at the same question in other countries [15, 16]. There- in mortality. fore, some of the sources of deaths seem to be procyclical Table 7 in the Appendix presents the results by sex for the different causes. For men, we find that the coefficients that appear most significant are cardivascular, cancer, senility Table 1  Effect of a one percentage point increase in the unemploy- and transport accidents. These coefficients have a negative ment rate on the mortality rate due to all causes sign, which shows a procyclical relationship. When unem- ployment increases in a recession, deaths due to cardio- Mortality rate Effect Standard error vascular problems decrease partly because individuals are All causes − 0.0042*** (0.0009) exposed to less job-related stress and eating habits improved Cardiovascular − 0.0072*** (0.0017) due to more home cooked meals. At the same time, as mobil- Cancer − 0.0022*** (0.0008) ity is reduced during unemployment spells, there are lower Respiratory − 0.0025* (0.0015) deaths due to transport accidents. For the case of women, we Senility − 0.0146*** (0.0036) observe the same pattern but the estimated coefficients are Alzheimer 0.0001 (0.0023) somewhat smaller. Some causes are only significant either Diabetes 0.0006 (0.0061) for men or for women. For example, we find that the coef- Digestive − 0.0039 (0.0027) ficient of homicides and other accidents is significant and Other diseases − 0.0021 (0.0018) negative (showing procyclicality) for men. While in the case Transport accidents − 0.0147*** (0.0039) of women we find procyclicality in deaths from digestive Other accidents − 0.0088* (0.0044) diseases. Suicide − 0.0035 (0.0025) A second exploration exercise that we are interested in Homicide − 0.0239** (0.0110) this article is to analyse the differences in the cyclicality Men − 0.0041*** (0.0008) of mortality by age group. As we explained in the descrip- Women − 0.0025*** (0.0007) tive section, we have five age groups (< 25, 25–44, 45–64, 65–74, > 74). In Table 2 we can see that, for all age groups, Number of observations: 936. All estimations include Year FE, Prov- ince FE, Linear trends and demographic controls we find strong evidence that mortality is procyclical and Robust standard errors in parentheses, *** p < 0.01, ** p < 0.05, * significant. That means that, when unemployment increases, p < 0.1 mortality decreases and vice versa. However, there are 13 Business cycle and mortality in Spain 1295 Table 2  Effects on age-specific Mortality rate 0–25 25–44 45–64 65–74 > 74 mortality rates associated with a one-percentage point increase in Unemployment − 0.0079** − 0.0024** − 0.0032** − 0.0042*** − 0.0029*** the unemployment rate (0.0026) (0.0011) (0.0015) (0.0011) (0.0010) Observations 936 936 936 936 936 R-squared 0.8773 0.8618 0.9889 0.9960 0.9994 All estimations include Year FE, Province FE, Linear trends Robust standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1 Table 3  Effects on the death Mortality rate Less primary Primary Secondary High school University rate of groups of different educational level associated Unemployment 0.0007 (0.0021) 0.0023 (0.0014) 0.0001 (0.0016) 0.0021 (0.0029) 0.0034 (0.0026) with a one-percentage point Observations 260 260 260 260 260 increase in the unemployment rate R-squared 0.9939 0.9960 0.9966 0.9633 0.9716 All estimations include Year FE, Province FE, Linear trends and demographic controls Robust standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1 important differences in the causes that lead this procycli- Table 4  Predicted number of deaths cality, which vary for each age group. Population sub-group Number of deaths 95%Conf. interval In Table 8 in the Appendix, we explore the results by age group and for different causes of deaths. There is consider- All death − 1612*** − 936, − 2288 able heterogeneity across different sources of deaths. For Males − 801*** − 495, − 1107 those under the age of 25 (which we can observe in more Females − 472*** − 213, − 730 detail in the first column of the table), there are three coef- Cardiovascular for all − 805*** − 434, − 1176 ficients that have a negative and significant sign (evidence of Cancer − 232*** − 67, − 398 prociclicality): deaths due to cardiovascular diseases, other Transport accidents − 28*** − 14, − 43 diseases and transport accidents. However, it is important Homicide − 9** − 1, − 17 to mention that cardiovascular diseases are not important in To calculate the predicted number of death we use (e𝛾̂ k − 1) ⋅ 𝜋k D this age group. In the age group 25–44, as we can observe where 𝜋k is the share of deaths in the second column, we do not find any significant coeffi- due to source k and D is the average annual number of deaths cient. In the third column we show the results for individuals (384,710) aged 45–64 and we find a negative coefficient in the follow- 95%confidence intervals are estimated as (e𝛾̂ k±1.96⋅sk − 1)x𝜋k D for sk ing mortality causes: cardiovascular, respiratory diseases, standard error diabetes, other diseases and transport accidents. For the age *** p<0.01, ** p<0.05 group 65–74 (column 4), we can see a negative coefficient in deaths due to cardiovascular problems and respiratory dis- eases. Finally, we present detailed results for those over 74 old. Unfortunately we only have this information from 2012 and we obtain a significant and negative coefficient for: car- onwards. In our dataset, we can distinguish between five diovascular problems, digestive and other diseases. There- education levels: less than primary school, primary school, fore, although mortality is procyclical for all age groups, the secondary, high school and university. As we can see in causes that lead this procyclicality are somewhat different Table 3, we do not obtain any significant coefficient for the for the different age segments. regressions of all cause mortality by different educational A third heterogeneity exercise that we implement dis- groups. However, when analyzed by cause, in Table 9 in the tinguishes across different levels of education. We classify Appendix, we observe that suicides show a countercyclical individuals according to the highest level of education that pattern, as they increase during crises periods and decrease they have obtained. As we only want to include those indi- in moments of expansion, for individuals with an intermedi- viduals that have already completed their desired level of ate level of education. education, we restrict the sample to those aged over 25 years 13 1296 M. Cervini‑Plá, J. Vall‑Castelló Finally, an important exercise is to calculate how much more recent period, which allows us to capture the impacts these coefficients represent in terms of the number of deaths. of the recent economic crisis. As our results are consistent Table 4 shows that a one percentage point increase in the with his findings, we are confident when concluding that the unemployment rate decreases the total number of deaths by, sign and significance levels of the relationship between busi- on average, 1612 per year (with a 95 % confidence interval ness cycle conditions and mortality in Spain has not changed ranging from 936 deaths to 2288). By sex, the prediction in the last 30 years. However, we report that the effect of the of the effect of a 1 point increase in the unemployment rate business cycle on mortality has become larger in more recent for men is higher (801 fewer deaths) than for women (472 years, as the coefficient in our estimates is larger than that fewer deaths). By causes, the main cause that decreases in a reported in the previous papers for Spain. recession is cardiovascular deaths; a 1 point increase in the Compared with Regidor et al. [13, 14] our article uses unemployment rate decreases cardiovascular deaths by 805 administrative data derived from the mortality registers, per year. Finally, we can also predict that a 1 point increase which is available for every year included in our study and, in the unemployment rate decreases cancer death by 232, therefore, allows us to include a longer period of time than transport accidents death by 28 and homicides by 9 per year. Regidoret al. [13, 14], which use Census data. Furthermore, we are able to present a detailed analysis of the impact for several sources of death, which allows us to better under- Conclusion stand the mechanisms driving this relationship, as well as a detailed analysis by sex, age and education, that was previ- This paper shows that, during the period 1999–2016, mortal- ously undocumented. ity exhibits a procyclical pattern in Spain; that is, when the local unemployment rate increases mortality falls. Exploring the impact of unemployment for different mortality causes Appendix we show that deaths from cardiovascular disease, cancer, senility and transport accidents are strongly procyclical. See Appendix Table 5, Fig. 3, and Tables 6, 7, 8, 9. However, countercyclical patterns have emerged for diabe- tes and suicide. When we study the effects by sex, we see that, for both men and women the coefficient is negative (procyclicality) Table 5  Total number of deaths, population and mortality rate (per although the effects for men are somewhat higher. Again, thousand population) cardiavascular, senility and transport accident causes seem Year Deaths Population Mortality rate to lead that procyclicality. Mortality is procyclical for all the age groups analysed; however, the causes that foster this 1999 371,102 40,202,160 9.23 procyclicality are slightly different in each age group. For 2000 360,391 40,499,791 8.90 example, for those between 45 and 65 respiratory problems 2001 360,131 41,116,842 8.76 and transport accidents explain the procyclicality pattern. 2002 368,618 41,837,894 8.81 On the other hand, procyclicity of deaths due to cardiovas- 2003 384,828 42,717,064 9.01 cular problems appear in a significant manner in all age 2004 371,934 43,197,684 8.61 groups (except those of 25–45). By educational level, suicide 2005 387,355 44,108,530 8.78 appears as a countercyclical cause (increasing in recessions 2006 371,478 44,708,964 8.31 and decreasing in the growing seasons) for individuals with 2007 385,361 45,200,737 8.53 intermediate education levels. 2008 386,324 46,157,822 8.37 We believe that our paper provides new evidence on the 2009 384,933 46,745,807 8.23 relationship between business cycle conditions and mor- 2010 382,047 47,021,031 8.13 tality with respect to the previous literature. More specifi- 2011 387,911 47,190,493 8.22 cally, our data allows us to study this relationship for a more 2012 402,950 47,265,321 8.53 recent period and for a country that has been exposed to an 2013 390,419 47,129,783 8.28 extremely strong recession in the mid-2008. With respect to 2014 395,830 46,771,341 8.46 the article of Tapia Granados [22], we are able to explore a 2015 422,568 46,624,382 9.06 2016 410,611 46,557,008 8.82 13 Business cycle and mortality in Spain 1297 Fig. 3  Evolution mortality causes Table 6  Proportions of deaths in total mortality by cause of death Table 7  Effect on different causes by sex or characteristic of the deceased Source: Spanish mortality registers Diseases Men Women 1999–2016, Spanish National Institute of Statistics Characteristic Percentage Cardiovascular − 0.0066*** − 0.0050*** Cancer − 0.0019** − 0.0014** Sex Respiratory − 0.0024 − 0.0004 Men 50.9 Senility − 0.0148*** − 0.0101*** Women 49.1 Alzheimer − 0.0013 0.0003 Causes Diabetes − 0.0002 0.0026 Cardiovascular 29.2 Digestive − 0.0023 − 0.0042** Cancer 27.5 Other diseases − 0.0038** − 0.0009 Respiratory 11.4 Transport accidents − 0.0188*** − 0.0145*** Senility 4.9 Other accidents − 0.0080** 0.0020 Alzheimer 3.6 Suicide − 0.0042 0.0053 Diabetes 2.3 Homicide − 0.0352** − 0.0217 Digestive 2.9 Other diseases 14.4 Number of observations: 936 for each estimation. All estimations include Year FE, Province FE, Linear trends and demographic con- Transport accidents 0.5 trols Other accidents 2.4 Robust standard errors in parentheses, *** p < 0.01, ** p < 0.05, * Suicides 0.9 p < 0.1 Homicides 0.1 Age < 25 0.6 25–44 1.8 45–64 11.5 65–74 13.3 > 74 72.8 Education Less than primary school 27.8 Completed primary school 34.9 Compulsory secondary education (ESO) 19.9 High school and vocational education 8.1 University degree 9.3 13 1298 M. Cervini‑Plá, J. Vall‑Castelló Table 8  Different cause by age Diseases 0–25 25–44 45–64 65–74 > 74 group Cardiovascular − 0.0288* − 0.0035 − 0.0079*** − 0.0090*** − 0.0067*** Cancer 0.0040 0.0014 0.0025 0.0010 0.0003 Respiratory − 0.0027 − 0.0049 − 0.0067** − 0.0043* − 0.0008 Senility − 0.0045 0.0043 0.0113 − 0.0024 − 0.0016 Alzheimer 0.0000 − 0.0136 0.0184 − 0.0051 0.0020 Diabetes − 0.0039 0.0259 − 0.0109* − 0.0047 − 0.0011 Digestive − 0.0257 0.0140 − 0.0009 − 0.0002 − 0.0039* Other diseases − 0.0100** − 0.0033 − 0.0082*** − 0.0049** − 0.0031** Transp Acc. − 0.0142** 0.0059 − 0.0143*** − 0.0093 − 0.0098* Other Acc. − 0.0113 − 0.0010 − 0.0063 0.0027 0.0064 Suicide − 0.0019 − 0.0005 − 0.0100** − 0.0054 − 0.0073* Homicide − 0.0102 − 0.0088 − 0.0141 − 0.0077 0.0074 Number of observations: 936 for each estimation. All estimations include Year FE, Province FE, Linear trends Robust standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1 Table 9  Different cause by Diseases Less primary Primary Secondary High school University education Cardiovascular 0.0010 0.0030 − 0.0004 − 0.0007 0.0085 Cancer − 0.0040 0.0015 − 0.0042 0.0081* − 0.0015 Respiratory 0.0052 0.0021 − 0.0030 − 0.0049 − 0.0003 Senility 0.0137 0.0046 0.0030 0.0135 − 0.0186 Alzheimer 0.0033 − 0.0052 0.0145 0.0193 0.0134 Diabetes − 0.0004 0.0164 0.0217 − 0.0161 0.0403 Digestive 0.0055 0.0088 0.0111 − 0.0406 0.0109 Other diseases − 0.0016 − 0.0027 0.0031 − 0.0009 0.0021 Transp Acc. − 0.0237 0.0209 0.0043 − 0.0099 − 0.1208* Other Acc. − 0.0078 0.0066 0.0261 0.0013 − 0.0012 Suicide 0.0059 0.0345 0.0175 0.0779** 0.0146 Homicide − 0.1023 0.0564 0.1237 0.1857 0.0254 Number of observations: 936 for each estimation. All estimations include Year FE, Province FE, Linear trends and demographic controls Robust standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1 Funding Open Access funding provided thanks to the CRUE-CSIC References agreement with Springer Nature. 1. Brenner, M.: Trends in alcohol consumption and associated ill- Open Access This article is licensed under a Creative Commons Attri- nesses. Some effects of economic changes. Am. J. Public Health bution 4.0 International License, which permits use, sharing, adapta- 65(12), 1279–1292 (1975) tion, distribution and reproduction in any medium or format, as long 2. Brenner, M.: Mortality and the national economy. A review, and as you give appropriate credit to the original author(s) and the source, the experience of England and Wales, 1936-76. Lancet 2(8142), provide a link to the Creative Commons licence, and indicate if changes 568–573 (1979) were made. The images or other third party material in this article are 3. Cutler, D. M., Huang, W., Lleras-Muney, A.: Economic Condi- included in the article's Creative Commons licence, unless indicated tions and Mortality: Evidence from 200 Years of Data. NBER otherwise in a credit line to the material. If material is not included in Working Papers 22690, National Bureau of Economic Research, the article's Creative Commons licence and your intended use is not Inc (2016) permitted by statutory regulation or exceeds the permitted use, you will 4. Edwards, R.: Who is hurt by procyclical mortality? Soc. Sci. Med. need to obtain permission directly from the copyright holder. To view a 67(12), 2051–2058 (2008) copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/. 5. Gerdtham, U., Johannesson, M.: Business cycles and mortality: results from Swedish microdata. Soc. Sci. Med. 60(1), 205–218 (2005) 13 Business cycle and mortality in Spain 1299 6. Gerdtham, U., Ruhm, C.J.: Deaths rise in good economic times: 15. Ruhm, C.J.: Are Recessions Good for Your Health? Q. J. Econ. Evidence from the OECD. Econ. Hum. Biol. 4(3), 298–316 (2006) 115(2), 617–650 (2000) 7. Gonzalez, F., Quast, T.: Macroeconomic changes and mortality in 16. Ruhm, C.J.: Recessions, healthy no more? J. Health Econ. Mexico. Empir. Econ. 40(2), 305–319 (2011) 42(C), 17–28 (2015) 8. Haaland, V.F., Telle, K.: Pro-cyclical mortality across socioeco- 17. Stevens, A.H., Miller, D.L., Page, M.E., Filipski, M.: The best nomic groups and health status. J. Health Econ. 39(C), 248–258 of times, the worst of times: understanding pro-cyclical mortal- (2015) ity. Am. Econ. J. Econ. Policy 7(4), 279–311 (2015) 9. Halliday, T.J.: Unemployment and mortality: Evidence from the 18. Strumpf, E.C., Charters, T.J., Harper, S., Nandi, A.: Did the PSID. Soc. Sci. Med. 113(C), 15–22 (2014) Great Recession affect mortality rates in the metropolitan 10. Ionides, E., Wang, Z., Tapia Granados, J.: Macroeconomic United States? Effects on mortality by age, gender and cause of effects on mortality revealed by panel analysis with nonlinear death. Soc. Sci. Med. 189(C), 11–16 (2017) trends. Ann. Appl. Stat. 7(3), 1362–1385 (2013) 19. Sullivan, D., von Wachter, T.: Job displacement and mortal- 11. McInerney, M., Mellor, J.M.: Recessions and seniors’ health, ity: an analysis using administrative data. Q. J. Econ. 124(3), health behaviors, and healthcare use: Analysis of the Medicare 1265–1306 (2009) Current Beneficiary Survey. J. Health Econ. 31(5), 744–751 20. Svensson, M.: Do not go breaking your heart: Do economic (2012) upturns really increase heart attack mortality? Soc. Sci. Med. 12. Neumayer, E.: Recessions lower (some) mortality rates: evi- 65(4), 833–841 (2007) dence from Germany. Soc. Sci. Med. 58(6), 1037–1047 (2004) 21. Tapia Granados, J.: Increasing mortality during the expansions of 13. Regidor, E., Ronda, E., Tapia Granados, J.A., Pulido, J., de la the US economy, 1900–1996. Int. J. Epidemiol. 34, 1194–1202 Fuente, L., Barrio, G.: Reversal of upward trends in mortality (2005) during the great recession by employment status at baseline 22. Tapia Granados, J.: Recessions and mortality in spain, 1980–1997. in a national longitudinal study. Am. J. Epidemiol. 188(11), Eur. J. Popul. 21(4), 393–422 (2005) 2004–2012 (2019) 23. Tapia Granados, J.: Macroeconomic fluctuations and mortality in 14. Regidor, E., Vallejo, F., Granados, J.A.T., Viciana-Fernández, postwar Japan. Demography 45(2), 323–343 (2008) F.J., de la Fuente, L., Barrio, G.: Mortality decrease according 24. Tapia Granados, J.A., House, J.S., Ionides, E.L., Burgard, S., Sch- to socioeconomic groups during the economic crisis in spain: oeni, R.S.: Individual joblessness, contextual unemployment, and a cohort study of 36 million people. Lancet 388(10060), 2642– mortality risk. Am. J Epidemiol. 180(3), 280–287 (2014) 2652 (2016) 13