2.1 Spanish EPL and fixed-term contracts liberalisation

The Workers’ Statute introduced in Spain in 1980 designed temporary contracts to be used only for temporary reasons such as seasonal jobs. These restrictions on temporary contracts meant that they were rarely used in the early-1980s. However, aiming job creation due to high unemployment rates, the Spanish government implemented a labour liberalisation reform in 1984. The reform regulating the temporary contracts was passed in August 1984 but was implemented on October 17th by the Real Decreto 1989/1984 published in the BOE (Spanish Official State Bulletin). It consisted of the liberalisation of temporary contracts and a reduction of dismissal costs. This meant that temporary contracts were no longer required to be only used for temporary reasons but could be signed for all activities and their duration was limited between a minimum of six months and a maximum of three years. Moreover, fixed-term contracts had low dismissal severance payments compared to permanent contracts. On one hand, firing costs at termination were 8 days per salary per year of job tenure in the case of temporary contracts. On the other hand, permanent contracts severance payments depended on whether the cause of dismissal was considered fair or unfair. Dismissal was only considered as fair if the firm was able to argue that the employee was not capable of performing the job tasks or if it was required due to economic or technological reasons. While unfair cost firing was equal to 45 days of salary per year of job tenure with a maximum of 42 months of wages, for fair dismissals cost firings amount to 20 days with a maximum of one year’s wages. As the 1984 labour reform did not modify permanent contracts, incentives for firms to employ fixed-term contracts increased and there was a strong substitution of permanent by temporary workers [27].

Temporary contracts represented a low percentage of the total number of non-agricultural private sector employees until the labour reform was passed on the last trimester of 1984 [28] After the reform, the percentage of temporary employees over total employment increased over time, especially for young workers, and Spain reached soon the first position in the proportion of temporary contracts among OECD countries. For instance, the percentage of temporary employment in Spain was 32.2% in 2000, while the EU (28 countries, UK included) average was 12.7%. Temporary employment rates are particularly high among young workers and were already around 70% in the early-1990s. Moreover, temporary contracts were widely adopted very soon after the reform, and their use heavily affected young employees.

In a recent publication, García-Pérez et al. [12] argue that the Spanish 1984 labour reform and the liberalisation in the regulation of fixed-term contracts had a negative long-term impact on the number of days worked and earnings of less educated workers. They find evidence that workers entering the labour market just after the implementation of the reform had a higher chance of having a fixed-term contract than workers entering the labour market a trimester before, and this enhanced a vicious chaining cycle of temporary jobs without upgrading to a permanent one. While the impact was larger in the first 10 years of these workers’ careers, the yearly earning losses after 27 years on less educated workers still amount to 7.3% due to the liberalisation of fixed-term contracts. At the same time that fixed-term contracts enhanced less educated young workers to land a first job, it worsened their career paths. The negative long-term effects of the liberalisation of fixed-term contracts are not limited to labour outcomes but other negative side effects, such as an impact on suicide rates of less educated workers, are present.

Therefore, to estimate these long-term effects we use a RDD strategy comparing individuals who reach the labour market entry age of 16, or in their graduation date if required, before and after the reform. This empirical strategy allows us to evaluate the long-term impact of the Spanish 1984 labour reform on completed suicides for different educational attainment level groups. Our paper aims to understand whether entering the labour market under different situations (wide availability or not of temporary contracts) increases long-term vulnerability to suicidal behaviour.

2.2 Suicide in Spain

Premature death from suicide in Spain is the current leading external cause of death and responsible for more than twice the traffic accidents deaths. According to OECD latest statistics from 2017, Spain has a yearly suicide rate of 7.0 suicides per 100.000 persons. However, considerable gender differences are common in suicides, and suicides rates in Spain reach 10.9 suicides per 100.000 men while falling to 3.6 for women. Suicide rates differ considerably across developed countries and Spain has a relatively low number of suicides compared to other OECD countries. For instance, the United States has a yearly suicide rate of 14.0 suicides per 100.000 persons, France has a suicide rate of 12.3, and Belgium of 15.9 suicides per 100.000 persons.

During the last decades, the Spanish deaths due to other external causes such as drug consumption or traffic accidents have been successfully prevented through several campaigns. For instance, the Spanish government implemented national strategic plans to deal with traffic accidents in 1980 (Plan Nacional de Seguridad Vial) and drug consumption in 1985 (Plan Nacional sobre Drogas). However, Spain has experienced a gradual, but steady, increase in suicides over the years, given the lack of an effective suicide strategy plan. Spain has historically integrated suicide prevention into their national mental health plan (Estrategia en Salud Mental del Sistema Nacional de Salud). In recent years, suicide prevention has been considered by healthcare professionals and policy-makers as a main priority to deal with in following mental health strategic plans. As we can see in Fig 1, suicides are the current leading external cause of death and of particular concern for men, whereas traffic accidents have undergone a substantial and constant decrease over the years.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Evolution of deaths due to suicide and traffic accident by sex (1980–2018).

Source: Death statistics according to cause of death from the Spanish Statistical Office (INE).

https://doi.org/10.1371/journal.pone.0252077.g001

On the other hand, global effort to address suicide has surged in recent years. In 2014, the WHO published a suicide prevention report: Preventing suicide: a global imperative [29], making a call to action for countries to adopt national suicide prevention strategies and suggesting the need for a multisectoral approach in health promotion. Steps for developing suicide prevention strategies followed and around 40 countries have developed or revised their national strategy. Moreover, attention on the matter was restated by WHO in 2018 by publishing another report aiming to support countries in continuing the progress in preventing suicide and to encourage establishing or revising national suicide prevention strategies: National suicide prevention strategies: Progress, examples and indicators [30]. It seems clear that the implementation of suicide prevention through action plans and defined national strategies is necessary to cope with a long-term neglected important health issue. Moreover, these prevention strategies have been recommended to focus on the population as a whole and vulnerable persons in particular. Therefore, understanding the mechanisms impacting on the vulnerability to suicidal behaviour as the aim of this study will help to boost the effectiveness of prevention policies and action plans taken in place.

Turning to Spain, although Spanish suicide rates are lower than most other developed countries, it is still an important cause of death in all age groups, and particularly concerning working-age people. Suicide rate increases with age, especially for men, and the number of completed suicides peaks for individuals 40–59 years old, age range comprising 41% of completed suicides in 2018, as it can be seen in Fig 2. However, youth suicide has also received plenty of worldwide attention as is one of the leading causes of death in adolescence and youth adulthood. In Spain, completed suicide is the second leading cause of death among young people but an important serious public health issue across all working-age groups.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Number of suicides by age and gender between 2014–2018 (Spain).

Source: Deaths Statistics according to Cause of Death (INE).

https://doi.org/10.1371/journal.pone.0252077.g002

Therefore, we consider that understanding and estimating the EPL impact on completed suicides is relevant for suicide prevention policies and of interest to healthcare institutions. In Spain, the National Health System is highly decentralized to the 17 Spanish Autonomous Communities, which are in charge of the management and provision of health care. While, the management and provision of health care of the two Autonomous Cities, Ceuta and Melilla, are in charge of the Spanish government. Among their functions, the regional governments are in charge of the design and implementation of health-related legislation. However, the Spanish Ministry of Health is responsible for the coordination of national plans, and therefore the development of effective prevention strategy plans will require a multi-actor collaboration at different administrative levels.

3.1 Data

We use the Death Registers conducted by the Spanish National Institute of Statistics (INE). Under an agreement with the INE, we were granted access to the restricted licence version of the Death Statistics which contains individual death data by the cause of death. This dataset registers all deaths occurred in Spain and, from 2012 onwards, includes individuals educational level attainment. We use data from all available years, 2012 to 2018, and differentiate between two periods depending on macroeconomic conditions. According to the Encuesta Población Activa (EPA), the Spanish Labour Market started registering an increase of employed individuals from the first quarter of 2014 onwards, after six years of decreasing absolute numbers. For each death, the dataset contains personal information on the date of death, the country of birth, gender, year and month of birth, the Spanish Autonomous Community of birth and death, educational level attainment and the cause of death. In our sample, we select the cohorts born between 1964 and 1972 so individuals are aged 41–54 during 2012–2018. We exclude foreign-born from the data and, therefore, restrict the analysis to the Spanish native sample.

The information available on the cause of death follows the classification criteria by the 10th revision of the International Statistical Classification of Diseases and Related Health Problems, ICD-10 [31]. Moreover, INE provides a reduced list of groups of cause (CIE-10) aggregating by ICD-10 subdivisions. In our analysis, we use codes related to the external causes of death (90 to 102) of the INE reduced list CIE-10. Specifically, code 98 is used for Suicides (classified as X60-X84 in ICD-10), 95 for Drugs Abuse (classified as X41, X42, X44 and X45 in ICD-10) and 92, 93 and 97 are jointly used as a placebo test incorporating external deaths causes attributed to accidental reasons.

In order to obtain the mortality rates by educational level attainment groups, we combine the deaths dataset with the 2011 Census. The reference date of the Population Census is the 1st November 2011, therefore we assume the 1st January 2012 population to be similar to the reported in the 2011 Census. For each subsequent quarters, we extract the number of deaths that occurred during that particular quarter to the previous population by education, gender and quarterly cohort group. Then, we match the obtained populations to the corresponding deaths and calculate the quarterly cohorts mortality rates per 100.000 individuals.

The reform was fully implemented by October 1984 and the quarterly cohorts selected for our analysis depend on the individual educational level attainment and the corresponding age of labour market entry. In the case of high-school dropouts, an entry-age of 16 is considered, as the minimum working age in Spain is set at 16-years-old. However, first-level high school graduates and second-level high school graduates entering the labour market the last trimester of 1984, must have graduated in the summer of 1984. On one hand, first-level high school graduates are those who studied until the course corresponding to the 16 years of age (currently ESO in Spain). On the other hand, second-level high school graduates correspond to college-preparatory high school or basic Vocational and Technical Schools (Bachillerato and CFGM, respectively). Hence, the first trimester cohorts graduating in 1984 are considered as the entry-age to the labour market for these specific education groups. For instance, the cohort born in the first trimester of 1968 is the first cohort affected by the 1984 reform in the case of first-level high school graduates. In contrast, for second-level high school graduates, the first cohort affected is the cohort born in the first trimester of 1966.

Therefore, we have a total of 53,730 deaths recorded in the sample, 36,438 native males and 17,292 native females, who entered the labour market within four years before or after the 1984 reform could affect them. Of these recorded deaths, 33,017 individuals were born before they could be affected by the reform whereas 20,713 individuals after. A further description of the number of deaths by sex, educational attainment level and the causes of deaths can be found in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Sample summary.

https://doi.org/10.1371/journal.pone.0252077.t001

3.2 Empirical strategy

We analyze the effect of the 1984 labour reform on the long-term suicide mortality rates outcomes. The key explanatory variable is the trimester of birth cohort, which determines if those individuals entered the labour market before or after being affected by the liberalisation of fixed-term contracts. The cutoff on the trimester of birth depends on the education group. For high school dropouts, the cutoff is set as the fourth trimester of 1968. Therefore, treated cohorts are those who turn 16 from October to December in 1984 or after, while older cohorts are left untreated as they enter the labour market before the full implementation of the reform. As for first-level high school graduates, the trimester of birth cutoff is defined as the first trimester of 1968 due to their graduation date, whereas for second-level high school graduates is set at the first trimester of 1966. We measure the reform impact by estimating the jump in the suicide mortality rates for the cutoff cohort. It must be stated that the untreated cohorts are also affected by the labour reform some quarters after entering the labour market. However, García-Pérez et al. (2019) [12] argument that the impact on these cohorts is very small because the conditions at labour market entry are of particular relevance on long term career outcomes [32–33]. Moreover, they argue that there is a discontinuous jump in the availability and use of fixed-term contracts in Spain before and after the implementation of the reform.

As existing literature suggests that macroeconomic conditions have an impact on suicide risk, we restrict the analysis to an economic upswing period in Spain (2014–2018). Then, we collapse the individual-level data by the trimester of birth control (c) and the year of death (t). We run a cohort regression discontinuity design [34], including different trends on each side of the cutoff, and add the years of death as a fixed effect to capture macro-level factors: (1) where C is the treatment cut-off set at the first cohorts affected by the reform. Y_ct is the cause-specific mortality rate per 100,000 individuals for each cohort and year of death. A linear cohort trend is included by subtracting the cutoff from the cohort’s birth trimester (BirthTrim_c − C). Then, we interact the trend with a dummy variable (after_c) that captures the cohorts affected by the reform, allowing us to include a different trend only for these cohorts. ϕ_t is a year of death fixed effect and ϵ_ct is the error term. Standard errors are clustered by cohort.

The impact of the reform is captured by the after_c variable, a binary variable equal to 1 for cohorts born at the cutoff trimester or after, and equal to 0 for cohorts born before. Therefore, the coefficient of interest is β₃, which estimates the differential effect of the reform in the cause-specific mortality rates for the treated and untreated groups. Under the assumption of continuity around both sides of the reform cutoff point, β₃ measures the long-term causal effect of the liberalisation of fixed-term contracts on cause-specific mortality rates (i.e., suicide rates). This key identification assumption was already addressed by García-Pérez et al. (2019) [12] through a differences-in-differences approach for labour market outcomes, and they also provide a descriptive analysis of the running variable by education. We also address it through two placebo test in Section 4.

The bandwidth used for the regression discontinuity analysis is selected to include 16-trimester cohorts before and 16-trimester cohorts after the cutoff. While a larger bandwidth yields more precise estimates, since more cohorts are used, it can lead to a biased estimation of the treatment effect. Therefore, this concern is addressed by reducing the bandwidth in Table 3. Specifically, we report a bandwidths of 6 and 10 cohorts respectively, the former being selected through the data-driven bandwidth point estimator proposed by Calonico et al. (2017) [35]. The larger results we find due to reducing the bandwidth are indicative that the possible bias in the 16 cohort-bandwidth is towards zero. Therefore we report both results in the discussion and the conclusion. The number of bins is limited by the size of the bandwidth as trimester cohorts bins are chosen for a more straightforward interpretation. As a result of the low number of cohorts (32 cohorts), we calculate the standard errors using the Wild Bootstrapping method proposed in Cameron et al. (2008) [36] and p-values are reported in brackets.

5.1 Robustness tests

Our identification strategy, based on a cohort regression discontinuity design, identifies the causal effects conditional on some assumptions that we test in this section. More specifically, the issues that we address in this section are: (i) the right size of the bandwidth, (ii) the functional form, and (iii) the continuity assumption at the cutoff point.

In our baseline results, we include 16-trimester cohorts before and 16-trimester cohorts after the implementation of the reform. As a robustness test, we check for the potential risk of including observations far from the cutoff point by reducing the bandwidth. In this nonparametric method, we limit the analysis to a closer surrounding of the cutoff, where local linear regressions are more likely, and estimate the treatment effect as local randomization. We reduce the bandwidth to 10-trimester cohorts and 6-trimester cohorts as shown in Table 3. Specifically, the later being selected through the data-driven bandwidth point estimator proposed by Calonico et al. (2017) [35].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Robustness checks—Suicide mortality rates (2014–2018).

https://doi.org/10.1371/journal.pone.0252077.t003

In both cases, the impact of the reform on suicide rates is positive and of a similar size for males. The after coefficient for the 10 trimester-bandwidth estimates a 26.7% increase in the suicides rates (column (1)). Alternatively, the after coefficient for the 6 trimester-bandwidth estimates a 41.8% increase in the suicides rates (column (2)). Although all estimates indicate a large impact of the liberalisation in the regulation of fixed-term contracts on the suicide rates in the long run, the stronger impact shown when the 6 trimester-bandwidth is used reinforces the causal interpretation. While the rectangular kernel used in our RDD has a more straightforward interpretation, in the light of the results of the robustness test carried out, a triangular kernel weighting could be considered a good model fit too. The results of this first robustness test allows us to conclude that the large impact of the liberalisation in the regulation of fixed-term contracts on suicide rates is robust to bandwidth selection and that the impact ranges between an increase of 25.3% and a higher bound of 41.8%.

We next focus on the functional form of the trend variable used in the main results specifications. In our analysis, we use linear trends to estimate the local average treatment effect on suicides around the cutoff point. As a robustness test, we test different functional forms for the trend variable to minimize specification bias from the possibility of an incorrect linear functional form, which are reported in Table 3. Specifically, we use a parametric strategy changing the order of the polynomials to a second and third degree for trimester cohort trends. The different orders of the trend variable are interacted with the after dummy variable. Thus, the terms of the polynomial (slopes) estimated below the threshold are allowed to be different from the slopes estimated above the cutoff point.

Table 3 shows our baseline model specification using second and third degree polynomials for the trimester cohorts trend variable. As can be seen in the table, the large impact of the liberalisation in the regulation of fixed-term contracts on male suicide rates is robust to the inclusion of higher order polynomial trends (columns (3) and (4)). However, we choose the linear regression as our baseline model because it reports the minimum Akaike Information Criteria (AIC) among all the estimated models. Moreover, the second polynomial specification reports similar results than the linear fit.

It should also be noticed that our empirical strategy addresses possible anticipation effects. The last graduate untreated cohorts were born in the last quarter of the year and had between nine and twelve months to look for a permanent job before the reform was implemented on October 17th 1984. Whereas the first graduate treated cohorts were born in the first quarter of the year but were not able to work as they were attending school until the summer before the reform. In the summer of 1984 the reform had already been passed while traditionally being a season reporting high rates of seasonal temporary employment, which was already allowed before the reform. Therefore, concerns of spillover effects to untreated are severely mitigated.

Finally, we also perform placebo tests in order to address concerns on the identification assumption and, hence, to check the random assignment around the threshold.

We simulate the implementation of the reform to take place in a pre-reform period (4 years before) in Table 4. This allows us to examine if discontinuities are present at a different unexpected point of the assignment variable. As expected, the local estimates are insignificant and the placebo reform does not show an impact for the period analysed. While other fake thresholds can be checked, we consider this falsification test and the descriptive corroboration from Fig 3 to be evidence of the lack of discontinuities at other trimester cohorts.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Falsification test: Pre-reform four years—Suicide mortality rates (2014–2018).

https://doi.org/10.1371/journal.pone.0252077.t004

Furthermore, as a second placebo test, we check whether other outcomes that are unlikely related to the reform are affected at the cutoff point. We consider deaths due to external accidents as an ideal candidate for the placebo test. Specifically, the placebo outcome comprises numbers 92, 93 and 97 of the INE reduced list CIE-10. They consist of falls, accidental drowning and submersion, other accidental threats to breathing and other types of unclassified accidents. These accidents are unlikely to be affected by the reform as INE list also includes specific codes for accidents related to drugs, alcohol, mental health or traffic, which we do not included in the placebo test. Table 5 reports the estimates for the placebo test on the alternative outcome. The impact of the reform on the placebo outcome is insignificant for all specifications including the educational group clearly affected by the reform (column (5)).

These placebo tests reinforce our conclusion that the liberalisation in the regulation of fixed-term contracts had a large causal impact on the suicide rates of high school first-level male graduates in the long run.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Placebo test: External causes due to accidental mortality rates (2014–2018).

https://doi.org/10.1371/journal.pone.0252077.t005

5.2 Drugs Abuse mortality

While the main focus of interest in our paper are suicides (as it is currently the leading external cause of death), the reform can have long-term effects on other cause-specific mortality rates as well. Therefore, in Table 6, we present a preliminary analysis of the mortality rates due to drug overdose, which can be linked to suicides.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Drugs Abuse mortality rates (2014–2018).

https://doi.org/10.1371/journal.pone.0252077.t006

It is important to note that mortality rates due to drug overdose are much lower compared to suicides for the cohorts analysed. Moreover, there are large differences in drug consumption patterns between women and men, and the male mortality rate due to drug overdose is much higher than for women. In Spain, a national strategic plan to deal with drug consumption, Plan Nacional sobre Drogas, was adopted by the Spanish government in 1985. However, during the decade of the 1990s Spain register record-high drug overdose deaths because of a heroin epidemic, in which young male adults were a particularly susceptible target.

We estimate the same baseline specification than in Table 2 but changing the dependent variable to drug overdose quarterly mortality rates. Specifically, the analysed outcome coresponds to the code 95 of the INE reduced CIE-10 list (psychoactive substances and drug abuse), which is composed of following ICD-10 codes: X41, X42, X44, X45. Different specifications according to the educational attainment level and gender are reported as for suicides mortality rates. A further preliminary descriptive evidence can be found in S4 Fig.

The positive and significant after coefficient in column (5) indicates that there is a 71.4% jump in the mortality rate due to drug overdose of the high school first-level male graduates cohorts after the implementation of the liberalization reform. This is a really strong impact but it has to be noted that the number of deaths due to this cause is relatively small so that, potentially, a larger number of observations would be needed for a more precise estimation. In any case, it represent another indicator that the liberalisation in the regulation of fixed-term contracts had a negative impact on health outcomes in the long run. Thus, even if the specification estimates are less precise, it reinforces our main baseline results pointing towards the role of Employment Protection Legislation (EPL) on long-run health outcomes.