A.  The Structure of the Finnish Education System before and after the Reform

Finland had a selective two-track school system until the 1970s. The reform replaced the old two-track system with a uniform comprehensive school system (Somerkivi 1982). Figure 1 describes the structures of both systems (see Sahlberg 2014).

The reform postponed the tracking age from 11 to 16 years. Both before and after the reform, school starts at age 7 and is compulsory until age 16. In the old system, pupils were taught together in the same class for only four years, from age 7 to age 11. Then they were placed into academic or vocational tracks for the remaining five years. In contrast, in the new system, there is an almost uniform curriculum for all nine years, until age 16.⁴ At its core, the reform significantly affected the composition of the peers to whom pupils were exposed between ages 11 and 16.⁵ There were no systematic changes in classroom size, gender composition of classes, or teacher quality due to the reform.

The comprehensive school reform was rolled out gradually across Finnish municipalities over the period 1972–77 (fig. 2). The timing of the reform in the different municipalities was decided by the National Board of Education (NBE). Municipalities made suggestions regarding the timing of the reform, but it was the NBE that finally approved and ratified them. Municipalities were in charge of the practical implementation of the reform (in collaboration with the surrounding municipalities). There is spatial correlation in the timing of the reform, because the NBE wanted to make sure that within larger areas students would get equal opportunities to move to secondary education (Somerkivi 1982). Nevertheless, there is meaningful variation in exposure to the new comprehensive school system, both across birth cohorts and across municipalities (see table 1 below).⁶ This variation provides a quasi-experimental research setting.

Previous research on the Finnish comprehensive school reform has shown that it decreased the intergenerational correlation between earnings of fathers and sons (Pekkarinen, Uusitalo, and Pekkala Kerr 2009). The reform also slightly improved the verbal and mathematical test scores of boys belonging to low socioeconomic backgrounds (Pekkala Kerr, Pekkarinen, and Uusitalo 2013). In addition, the reform increased the gender difference in the probability of choosing an academic track and obtaining tertiary education (Pekkarinen 2008). Finally, Ravesteijn et al. (2017) study the effect of the reform on all-cause mortality. For males, using an 11% random sample of the Finnish population, they find occasionally negative or positive effects, depending on the subsample. For females, they do not find any effect on mortality. Using total population data, we focus on the effects of the reform on mental health.

B.  Expected Effects on Mental Health

A key feature of the Finnish reform is that it did not affect the length of compulsory schooling. Thus, the reform did not change the minimum school-leaving age of 16. This allows us to focus on examining the effects of change in the age of tracking on mental health outcomes while keeping the number of years of compulsory education fixed.

School tracking reform may positively affect education and income in adulthood. In turn, better human capital and availability of financial resources may improve health (Galama, Lleras-Muney, and van Kippersluis 2018). For this reason, we analyze the potential mediating role of education and income on mental health in section VI.

Besides education and income effects, other potential mechanisms are also at play. Interactions with peers in school during childhood and adolescence are important determinants of mental health outcomes in adulthood (WHO 2014). In theory, the predicted effects of the reform on health outcomes, and especially on mental health, are ambiguous. After the reform, students aged 11–16 now have a common set of peers rather than peers from their specific track only. Low-achieving students, who would have been assigned to the vocational track, are being exposed to higher-achieving peers in the postreform system. This change is expected to improve education and related economic outcomes for low-achieving students, which in turn may improve their health outcomes in the long run. It is also possible that low-achieving students will start adopting and imitating health behaviors of their higher-achieving peers. Conversely, higher-achieving students may be adversely affected by being exposed to lower-achieving peers.⁷

The comprehensive school reform also affected the ability ranking in the classroom. Vocational track students who scored high in the ability ranking in the prereform system are now, on average, ranked lower in comprehensive-school classrooms. Cicala, Fryer, and Spenkuch (2018) have shown that students’ academic achievements and disruptive behaviors depend on their ordinal ranking among their peers. Moreover, Elsner and Isphording (2018) provide evidence that a student’s ordinal ability ranking in a high school cohort is an important determinant of engaging in risky behaviors (such as smoking, drinking, and proneness to physical fights). This evidence suggests that pupils who would have been tracked into vocational schooling without the reform could have lower health, especially poor mental health, after the reform. It contradicts the positive effect of being exposed to higher-achieving peers.

Moreover, the fact that students in the postreform system follow a common curriculum from age 11 to age 16, rather than a track-specific curriculum, also means that, after the reform, students in a given classroom are less homogeneous. This makes it more difficult for teachers to tailor their pedagogical approaches to a more heterogeneous mix of students, thereby resulting in negative education and health outcomes (Betts 2011). Students’ achievement may be better when they are surrounded by peers with similar characteristics. Indeed, the Finnish experience suggests that mixed-ability groups led to learning difficulties; disruptive behaviors increased after the reform, and the number of pupils in special education more than doubled between school years 1974–75 and 1979–80 (Somerkivi 1982, 40).

Finally, one of the main reasons to favor the delay in tracking is that the likelihood of a student being placed in the “wrong” track is reduced and the anxiety associated with tracking lessens, since the amount of pretracking information about students’ abilities is higher at the time of the tracking decision (Brunello, Giannini, and Ariga 2007). This suggests that the reform improves mental health. However, Pekkarinen (2008) argues that for boys, this benefit is offset by the fact that tracking now occurs during puberty, in contrast to girls, for whom it occurs after puberty.

In short, previous research does not clearly predict the reform’s effect on mental health, since different mechanisms are pushing in different directions. However, it does suggest that the effect likely differs by gender and academic ability.

A.  Census Data

We use the population register data of permanent residents of Finland. The data originate from the Longitudinal Population Census Files from Statistics Finland. Demographic and labor market information are available for the years 1975 and 1985 and after that annually over the period 1987–2014. The municipality of residence is recorded annually since 1971. The data contain almost complete household and parental links. The date and municipality of birth are also recorded. Furthermore, the data also contain detailed information about degrees completed from 1970 onward.

The core data include the individuals born in Finland between 1962 and 1966, following Pekkala Kerr, Pekkarinen, and Uusitalo (2013, 586). We start with birth cohort 1962 and end with birth cohort 1966 to increase the homogeneity of the cohorts under study.⁸ This sample restriction also allows us to follow all birth cohorts up to the maximum age of 45.⁹ As shown in table 1, there are between 72,248 and 74,248 persons in each birth cohort. Hence, we have approximately 366,000 individuals in total. Annual information about the municipality of residence, together with the birth date, determines whether a pupil attended the tracked or comprehensive school system. The reform was effective for students who were at most 11 years old, that is, entering fifth grade, at the end of the year in which the reform was implemented in their region of residence. For instance, people who turned 11 in the region in which the reform was implemented in 1975 received postreform schooling if they were born in 1964, 1965, or 1966 and prereform schooling if they were born in 1962 or 1963.

We exclude foreign-born individuals (most of whom immigrated to Finland after the reform) and those living in the Åland Islands (in total, 8% of the original sample) from the estimation sample, retaining individuals living in 465 different municipalities. We also exclude a small number of individuals who, at ages 11–15, migrated between municipalities with a different year of adoption of the reform (less than 2.6% of the original sample),¹⁰ since the reform indicator cannot be assigned unambiguously for these people. Finally, we exclude a very small number of emigrants from the original data.

Since measures of academic ability before tracking are not available in the data, we investigate heterogeneity by parents’ education.¹¹ The variable is categorized into three values: (1) low educated (53%), if neither of the parents completed postcompulsory education, that is, if they both had a maximum of nine years of schooling; (2) mid-educated (27%), if either or both of the parents completed a vocational degree but neither studied further; and (3) highly educated (20%), if at least one of them completed high school or a higher level of (tertiary) education (see fig. 1 for a reminder of the prereform system).

B.  Mortality and Mental Health Disorders

To relate our findings to previous research on the education-health nexus (Lleras-Muney 2005; Clark and Royer 2013; Meghir, Palme, and Simeonova 2018), we examine mortality outcomes. We use data regarding the year and cause of death from the comprehensive death certificates (until 2013). All diagnoses of the causes of death pass a routine validation conducted by Statistics Finland, and unclear cases are judged by a panel (Lahti and Penttilä 2001). Our mortality outcomes (suicides, mental health–related deaths, and all-cause mortality) are measured until age of 45. We picked age 45 so that all birth cohorts could be followed over the same window.

We use suicides as the primary mortality variable (a dummy for occurrence before age 45). Suicides are defined by the codes X60–X84 and Y87.0 in the International Classification of Diseases, 10th revision (ICD-10), which is the standard diagnostic tool for clinical purposes. Suicide is a relevant outcome for three reasons. First, suicides are closely related to mental health problems. For example, approximately 90% of suicides are associated with psychiatric disorders (Henriksson et al. 1993; Pirkola et al. 2009). Second, approximately 25% of all deaths by age 45 are suicides. Third, the suicide mortality of young Finns is among the highest in the world (Lahti et al. 2011).

Following Alexander and Schnell (2019), we also consider a broader measure of mental health–related deaths, which includes not only suicides but also injuries of undetermined intent (i.e., fatal injuries about which it is not known whether they occurred accidentally or were purposely inflicted) and accidental deaths involving poisonings, drownings, and deaths involving firearms and trains. In our data, 69.3% of mental health–related deaths by age 45 are suicides, 3.6% are injuries of undermined intent, 19.8% are accidental poisonings,¹² 5.7% are accidental drownings, and 1.5% are accidents involving firearms or trains. ICD codes that are used to identify these causes of death are given in appendix A. We utilize this broader measure of mental health–related mortality because not necessarily all deaths caused by mental health disorders are classified as suicides. Finally, all-cause mortality by age 45 is measured by a dummy variable (table 2). In our data, 4.6% of men and 1.8% of women die by age 45. Moreover, the mental health–related death rate for men is higher (1.9%) than that for women (1.2%).

We then study the effects of the reform on mental health–related hospitalizations, using inpatient data.¹³ We focus on serious mental health–related hospitalizations for two reasons. First, their treatment costs are particularly high in the universal health care system. Second, severe mental illnesses cause substantial indirect costs in terms of absenteeism, weak long-run labor market attachment, and early disability pensions (Hakulinen et al. 2019). Therefore, undoubtedly, society cares about these outcomes.

We use register-based measures that are free from the potential measurement error inherent to self-reported mental health symptoms (Ritter et al. 2001). Information about mental health disorders is extracted from the Hospital Discharge Register (HDR) compiled by the National Institute for Health and Welfare for the period 1969–2013. The data include dates of admission to the hospital, dates of discharge, and the primary reason for hospitalization. Diagnosis codes are from the eighth, ninth, and 10th revisions of the ICD. Spells due to mental health disorders correspond to a diagnosis code starting with the letter F in the ICD-10 classification and to codes 290–319 in ICD-8 or ICD-9. Validation studies have shown that the HDR data are of high quality from 1972 onward (Sund 2012).

In the baseline model, we use a dummy variable to indicate whether the individual had any mental health–related hospitalizations between ages 16 and 45.¹⁴ According to the data, there are 138,800 mental health–related hospitalization spells between ages 16 and 45 among 28,700 individuals (representing 7.8% of the total population). Approximately 10% of men and 6% of women had mental health problems that resulted in hospitalization (table 2). Conditional on having a mental health–related hospitalization, the average time spent in the hospital between ages 16 and 45 is 151 days for men and 193 days for women.¹⁵ We then also consider separately whether the individual had any spell starting at ages 11–15, 16–25, 26–35, or 36–45. Again, we record new hospitalization spells until age 45 so that all birth cohorts can be followed during the same age window.

To get a comprehensive picture, we also examine the effect of the reform on different types of mental health disorders (Suvisaari et al. 2009; Santavirta et al. 2015): (1) schizophrenia, a mental disorder characterized by hallucinations, delusions, and cognitive deficits; (2) other psychoses that are not related to emotions or moods (nonaffective psychosis); (3) bipolar disorder, an affective psychosis involving emotional and mood abnormalities (and manic episodes); (4) depressive disorder, which can include repeated episodes of severe depression or chronic mild-grade depression (dysthymia); (5) severe anxiety, stress, and neurotic disorders, which can interfere with daily activities, such as job performance, school work, and social relationships; and (6) substance-use disorder, which includes all psychiatric hospitalizations related to alcohol or substance abuse or addiction. Appendix A contains details about the codes used to define these categories.

Table 2 provides the summary statistics for the main outcomes of interest, broken down by gender, and table B1 provides these same statistics, broken down by socioeconomic background, in addition to some summary statistics on additional outcomes.

A.  Graphical Presentation of the Main Results

To check the validity of our research design, as well as to graphically preview the key findings, we estimated the baseline specification using lead and lag year dummies around the reform and omitting the year before the reform ($t=\mathrm{-1}$), as in Pekkala Kerr, Pekkarinen, and Uusitalo (2013):

The results reported in figures 3, 4, B3, and B4 suggest that our main findings are not affected by prereform trends, which provides evidence in favor of our identification assumption.¹⁹ They also preview that our main findings are going to be mostly null effects. We now turn to regressions where we pool together all the postreform (and prereform) years to precisely estimate the magnitude of these (null) effects.

B.  Mortality

Table 3 reports the estimates for the effects of the reform on the incidence of suicide by age 45. Regardless of the sample, we find no evidence that the reform affected suicides by age 45. For instance, for males, we can rule out increases of 0.14 percentage points or decreases of 0.18 percentage points, relative to the mean outcome of 1.3%. Table B7 shows that the results for the full sample are robust to the variations of the model (e.g., adding controls and excluding some observations).²⁰

Next, we consider the effects of the reform on mental health–related deaths, which include suicides, injuries of undetermined intent, and accidental deaths. The results, reported in table B2, are similar to those found for suicides. The only notable exception is that for this broader measure of mental health–related deaths, improved survival in males from low-educated families can be seen. The estimate indicates reduction in mortality by 0.3 percentage points (significant at the 5% level), relative to the mean outcome of 1.4%. However, after adjustment for multiple-hypothesis testing, its significance level is marginally outside the 10% level (table B11). The results for males from low-educated families remain qualitatively intact to the alternative specifications of the model, but the significance levels vary by model (table B8). The estimated effects remain insignificant in other subsamples.

To evaluate the overall health effects of the reform, table B3 reports the findings on all-cause mortality. Echoing the results for suicides, we find no evidence that the reform affected all-cause mortality by age 45. The estimated effects are close to zero.²¹ For males, we can rule out increases of 0.17 percentage points or decreases of 0.5 percentage points at the 5% significance level, relative to the mean outcome of 4.6%. For females, we can rule out increases of 0.4 percentage points or decreases of 0.15 percentage points, relative to the mean outcome of 1.8%. Despite the universal health care system, the baseline differences in mortality by parental education are substantial—males from low-educated families have a three times higher mortality than those from highly educated families—but the treatment effect is insignificant across groups. Robustness of these results is documented in table B9.

The conclusions are also not sensitive to the use of a Cox proportional hazards model (table B16). These results provide opportunity to compare the precision of our estimates to prior evidence. Meghir, Palme, and Simeonova (2018) report the effects of the reform on hazard rates of death by age 45 in their appendix table A5 (sample of males and females). Their estimate from a Cox regression is 1.0006 (SE of 0.0217). We report a comparable estimate in the note to table B16, which indicates a combined reform effect of 0.985 (SE of 0.030). Thus, Meghir, Palme, and Simeonova (2018) can rule out (with 95% confidence) changes in the risk of death by age 45 that are outside −4.2% and +4.3%, whereas we can rule out changes that are outside −7.3% and +4.6%. In both studies, standard errors are clustered at the municipal level.

C.  Hospitalization

Table 4 reports the results that use the incidence of mental health–related hospitalizations as an outcome variable over the age range 16–45. The average of the outcome variable is 10% and 6% for males and females, respectively. These figures indicate that these incidents are not rare in our data.

The reform led to an increase in mental health–related hospitalizations, but for only some subgroups. The first finding is that males from low-educated families who were exposed to late tracking have a 0.74 percentage points higher probability of having severe mental health disorders that result in hospitalization than those who were educated in the prereform system, that is, engaged in early tracking. The size of the effect represents a 7% increase relative to a mean outcome of 10.8%. The effect is robust to using an outcome measure that captures also the intensive margin and not just the extensive one, that is, the number of days spent in the hospital for mental health reasons between ages 16 and 45 (see table B17). Importantly, the effect for males from low-educated families is no longer statistically significant after the standard errors are adjusted for a Romano-Wolf correction for multiple-hypothesis testing (Romano and Wolf 2005).

The second finding is that, because of the postponement of the tracking age from 11 to age 16, for females from highly educated families the probability of severe mental health disorders increased by 0.95 percentage points. We highlight this finding for several reasons. First, the quantitative magnitude of the effect is rather large, given the low baseline probability of mental health disorders for women from highly educated families (5.2%). Second, the effect remains significant (at 10% level) even after the standard errors are adjusted for Romano-Wolf correction for multiple-hypothesis testing (Romano and Wolf 2005; table B11). Third, the effect remains intact regardless of whether we control for the incidence of prior mental health disorders and/or parents’ mental health status during the prereform period (table B10). Fourth, the finding is also robust to several other sensitivity checks, for example, accounting for the full set of region-specific linear time trends (table B10). Fifth, the estimated dynamic response models show that the finding is not driven by prereform trends (fig. 4). Therefore, it is unlikely that the result is affected by unobserved regional characteristics that are potentially correlated with the rollout of the reform. Sixth, the finding remains intact even if individuals are classified into treatment on the basis of their municipality of birth (table B18) or municipality at age 10 (table B19) instead of municipality at age 11 (and not eliminating individuals who migrate between ages of 11 and 16).

The historical timing of the reform, together with the availability of longitudinal data, enables us to examine the effects of the reform over the life cycle. We report the results for relevant age categories in table 5. We find that at ages 36–45, for females from high-educated families, mental health disorders are significantly more likely to occur in those who had been exposed to the postreform school system than in those who were educated in the prereform system. However, for this affected group, we do not find significant effects during the school years. Additionally, for males from low-educated families, the positive effect of the comprehensive school reform on the probability of hospitalization for mental health reasons peaks at ages 26–35, but there is also a significant effect at later ages, that is, between the ages of 36 and 45.

Table 6 examines whether any specific disorder drives the estimated treatment effects. We observe a significant increase in the probability of depressive disorders for females from highly educated families. Conversely, for males from low-educated families, the overall increase in hospitalizations is driven by the increase in alcohol-related mental disorders. Alcohol abuse may be caused by self-medication related to perceived stress (Enoch 2011). The gender pattern that we observe is plausible, since in our data, alcohol-related mental disorders and depressive disorders are the most prevalent mental health problems for males and females, respectively (table 2).