2.1. Background on the Educational System and the Reform
Compulsory school attendance was introduced in Sweden in 1842 and applied to all resident children. The compulsory school attendance implied both the right to cost-free schooling in compulsory schools and the obligation to take part in the schooling offered. The central management of the education system was practiced by the
Ecklesiastikdepartementet, the Ministry for Ecclesiastical affairs. The country was divided into more than 2,000 school districts, and the local administration of compulsory education in these school districts was the responsibility of a school board (education policies were consequently not designed at the county (regional) level, which is the level we use in our empirical analysis). With the exception of girls’ schools, private schools were always insignificant in Sweden [
20].
At the beginning of the 20th century, the Swedish educational system was highly selective. Schooling started at the age of seven and was compulsory for six years. The vast majority ended school after these six years in
Folkskolan. In 1918, it was further decided that sixth grade pupils that did not continue to the non-compulsory secondary school also had to take two years of vocational courses [
20]. These were all practically oriented one-or two-year courses in domestic science, craft or manufacturing given at local schools. Importantly, the vocational courses were taught with a very low intensity. As described by Fredriksson [
21], the time for vocational courses was only 180 h per year.
Students who chose to follow an academic path continued to lower secondary school to pass the so-called
realexamen (although rare, students could choose to continue onto lower secondary school already at the age of 11,
i.e., leaving compulsory school after the fourth grade, implying that the Swedish educational system had parallel structures). However, secondary schools became widely spread geographically only between the mid-1940s and the early 1960s [
22], and very few students continued to secondary schooling before the 1950s. In 1940, only 10% of the cohort graduating from
Folkskolan continued in secondary school [
23]. After realexamen, students either left school or entered the upper secondary level and had to sit in the prestigious
studentexamen after three more years of studies. This was comparable to the French BAC, German
Abitur and the English A-levels and a prerequisite for matriculating into university [
24]. With a highly selective school system, only 5% of a cohort continued to upper secondary schooling in 1940 [
23].
Students in
Folkskolan were attending full-time schooling approximately eight months per year. In the rural areas, it was, however, also possible for school districts to offer half-time reading, so that children attended school only certain days of the week. This school form only existed in the rural areas of Sweden, where children often helped out in the agrarian sector. The existence of half-time reading was heavily debated in the 1920s, and the extent of half-time reading was reduced during this decade, although still permitted. In 1933, 93% of all pupils took part in full-time schooling [
21].
For a long time, local conditions decided the format and content of primary education, and at the beginning of the 20th century, there was large variation across school districts. The national government issued its directives for the curriculum in so-called
normalplaner (normal plans), but this document was only advisory. In 1919, however, the so-called
Utbildningsplanen (the education plan) was introduced, which came to restructure the school’s work according to the central guidelines of Ecklesiastikdepartementet. Utbildningsplanen was a governing document and included time-tables and syllabuses for compulsory school [
25]. It is, of course, difficult to know about the quality of the education across school districts. Completion rates were, however, high: 90% of all pupils finished Folkskolan with full curriculum [
21]. The implementation of the Utbildningsplan was the first instance in which local autonomy had to give way for national standards, and the Government’s edict was subject to financial compensation [
20].
On 1 July 1936, the national Government decided that seven-year school should be compulsory. Already, in 1925, a clause had been introduced in the primary school code that a seventh school year could be made compulsory [
21]. School districts were also allowed to introduce eight year compulsory schooling, but this was a very rare event, both in urban and rural areas. In 1940, only 0.1% of all schools in the country offered eight years of education [
21]. To extend compulsory schooling with an extra year was then the decision of the school board in a school district, and already, in 1936, several school districts in the urban areas had introduced an extra year of schooling. Furthermore, in the southernmost and mainly rural region of Scania, several school districts had implemented seven years of schooling, but, still, only 16% of all children in rural areas of Sweden attended seven years of schooling in 1936.
In Sweden, child labor laws and compulsory attendance laws have generally been coordinated. One basic principle has been that the right to education takes precedence over the demands of the labor market—so that educational requirements with respect to knowledge and time should determine if and when the young were allowed to work. Compulsory school attendance regulations have consequently reinforced the protective labor legislation, and as discussed by Sjöberg [
26], Swedish authorities have generally relied on double protection—age limits and compulsory school attendance. According to the 1931 Labor Act, the minimum age for manufacturing and construction work was 14 years, whereas the limit for “light work” was 13 years. A child was allowed to work from the beginning of the calendar year in which they would reach the age limit. After the implementation of the compulsory school reform, most pupils left school in the middle of the year they turned 14, whereas before they would have left the year, they turned 13. This means that the reform reduced the time a child could spend in “light work” by one year, whereas the corresponding reduction for “hard” (industrial) work would have been 5–6 months only. The 1949 Labor Act increased the age limit by one year, in turn harmonizing the age when a majority of students finished schooling and started to work. Notably, the legal documents generally regulated full-time work, but not the part-time work of young people [
26].
The main motive for the reform was that six years was considered too short for achieving the learning objectives that were stated for the
Folkskolan. This motive was also mentioned in the decision allowing school districts to implement seven years of compulsory schooling in 1925. Additional arguments on the importance of a change in the compulsory schooling legislation mentioned in various investigations were the increasing youth unemployment among those who had just finished elementary school, but also that another year of education was of importance to maintain a democratic society [
20]. In the debate preceding the introduction of the reform, politicians were also often benchmarking with other Western countries, and it was noted that the number of school years was the most striking difference of compulsory education in Sweden compared with Denmark, Norway, Germany and Great Britain.
In line with the underlying motive for the parliamentary decision, the reform did not require any fundamental changes with respect to learning outcomes to be achieved or curricula, but instead emphasized the goal of achieving more long-lasting results of schooling. The recommendation from the central administration was that the school districts should distribute the pre-reform compulsory school curricula over seven years instead of six [
20].
The reform was not implemented at the same time in all school districts, but, instead, it was stipulated that it had to be implemented in all school districts before 1949. The compulsory seventh year was consequently introduced during a twelve-year transition period. In 1936–1941, an implementation of an extra year was completely at the discretion of the school district. From 1942 and onwards, a school district could be assigned to implement the reform, but according to official sources, this only happened once. The national school authority
Skolöverstyrelsen, monitored the implementation. Initially, the reform led to a relatively rapid transition. In 1940, 33% of the rural and 80% of the urban schools had implemented a seventh year [
21]. After 1940, the implementation rate, however, seems to have decreased somewhat.
Figure 1(a) shows the trends in implementation and reports the proportion of school districts that had at least seven years of compulsory schooling at the end of each school year, and
Figure 1(b) shows the number of students affected by the reform by birth cohort.
Figure 1.
(a) Proportion of school districts with seven years of compulsory schooling; (b) number of students affected.
Figure 1.
(a) Proportion of school districts with seven years of compulsory schooling; (b) number of students affected.
Due to the soft transition rules, the reform does not seem to have caused any major difficulties in the school districts, and the implementation was also facilitated by the fact that the responsibility for funding of school buildings, teaching materials and teachers’ salaries was the responsibility of the central government and not the school districts [
27].
From 1936 until the 1950s, not much else other than the implementation of the seven year reform changed in the Swedish educational system. In 1948, a parliamentary school committee proposed a new school reform that implied nine-year compulsory and comprehensive schooling. Additionally, this reform, starting in 1949 and evaluated, e.g., in Meghir and Palme [
28] and Meghir
et al. [
14], was gradually implemented across school districts over time. The major consequences of this reform were additional years of schooling, but also a reshaping of the entire school system.
As regards the wider institutional context, we are unaware of any reform relevant to public health that might have coincided with the school reforms at the local level. Such a coincidence is unlikely, since decisions related to healthcare were taken at either the county level or at the health district level, both of which typically consisted of several school districts. Moreover, most of the expansion of the Swedish healthcare system came after the introduction of uniform social health insurance in 1955 [
29]. Thus, access to services was limited and did not change much during the period we consider. For example, a survey made by the National Board of Health (
Medcinalstyrelsen) in 1927 concluded it was very rare that pregnant women had been examined by any kind of medical staff, including midwifes, before delivery [
30]. This lack of coverage was perceived as a problem and led to a national roll-out of preventative services for pregnant women and newborns from 1937 onwards [
31]. Moreover, as discussed by Heidenheimer
et al. [
32], the physician density in Sweden was significantly lower compared to, e.g., Norway and Denmark, in the 1930–1950 period, and official statistics suggest that the number of hospitals and hospital beds
per capita remained more or less constant [
33]. One trend that did, however, affect the cohorts included in this study was the transition to institutional delivery, which became the standard between the years 1920–1940 [
34]. Even though this general trend could theoretically confound our analysis, it is very unlikely that it does so in practice. The expansion in hospital births was very smooth and mainly driven by individual-level demand-side decisions.
Another potential concern relates to World War II (1939–1945) and the fact that compulsory schooling generally protects individuals from getting involved in armed conflicts. If Swedes joined armed forces and died in battle, this could generate a misinterpretation of the results as an effect of education, while rather capturing an education gradient in enlistment. Notably, however, Sweden was neutral in World War II, and less than 9,000 individuals (almost all of them men and all older than 18 years) fought in the war as volunteers [
35], suggesting this should be a minor problem in our analysis.
In summary, the reform in 1936 introduced an exogenous change in the extent of compulsory schooling in Sweden. The timing of implementation in individual school districts was based on a mixture of local and national decisions; and there is no reason to suspect that the implementation of the reform was driven by health differences. Thus, it will be our working assumption that the reform was exogenous from the individual point of view so that it can be used to identify the effects of schooling on health. As regards the exact definition of treatment, the above account has made clear that affected pupils faced no significant increase in the curriculum to be covered. Thus, effects, if present, will be driven by changes in the amount of time spent in education.
2.2. Data and Sample Selection
We combine three different datasets in our analysis. Our school reform indicator is based on unpublished material taken from the archive of the national school authority [
36]. After the national government had decided to make a seventh year compulsory in 1936, the national school authority,
Skolöverstyrelsen, monitored the implementation. Data are available from 1931 onwards for rural areas and from 1938 onwards for urban areas. These tables were based on reports from individual school districts, and they provide the number of districts that had introduced seven (and in some rare cases, eight) years of schooling. In a small number of cases, the classification of districts was challenging, since the seventh year was introduced at different times for different schools within the district. In the original data, such districts have been coded as having seven years of schooling. Given the low number of school districts where this is an issue, this is unlikely to lead to measurement error.
A more serious concern is that we do not have the identity of the school districts that have extended compulsory schooling—only in exceptional cases (i.e., when there is only one city in the county) is it possible to unambiguously identify the reform year. Thus, we use the proportion of reformed districts as our treatment variable.
Our second source of data is the census of 1935 [
37]. Traditionally, Sweden had a complete census every ten years; however, in 1935 and 1945, there were additional censuses, which provided basic demographic information on the whole population and, then, studied some parts of the country in more detail. The 1935 census gives us information on cohort sizes, which are used for calculating mortality rates at different ages. For each of the 25 counties, we collect cohort sizes of males and females from urban and rural areas separately—the exception being the City of Stockholm, which was the only county without rural areas. Thus, the total number of cells in each year is 98.
Our third data source consists of the Swedish Death Index, which is a digital data source provided by Federation of Swedish Genealogical Societies [
38]. The dataset is based on official records, such as church books, and it is complete from 1947 onwards. For earlier years, 25% of deaths are missing at random throughout the country—with the sole exception of the county of Värmland, which is complete. Since our cohort size data give cohort sizes after infancy, the missing deaths, thus, apply to ages at which mortality rates approach their lifetime minimum. For the cohorts included in our dataset, the cumulative mortality rates between the baseline year and the first year with complete records (
i.e., for the period 1936–1946) vary between 1.2% and 2.2%, according to the life tables of the period (our calculations from the census tables). Therefore, the measurement error due to the missing information is likely to be small, but in order to correct for it we multiplied annual cohort-specific mortality rates in relevant years by a factor of 1.33 (
i.e., the factor necessary to compensate for the missing deaths).
The Swedish Death Index contains information on gender, date of birth, date of death, parish of birth and parish of death. Since all parishes can be attributed to either the urban or the rural category (and some of them were recategorized during the time covered), we can add them up for each region to get the numerator of the mortality rate. The Swedish Death Index has the advantage that we are able to calculate death rates for the short run, as we can practically observe all deaths in the whole of Sweden for each year.
However, a slight complication arises, due to the coding of birth parishes. Prior to 1946, it was common to attribute hospital births to the place of birth (
i.e., the hospital parish) and not to the place of residence (
cf., [
39]). This is a serious concern for cohorts born in the 1940s, when institutional delivery became the norm. However, during the period we consider, hospital births were still exceptional and, also, predominantly affected children from the cities. Consequently, the measurement error arising is limited, but we circumvent this problem by aggregating urban and rural areas in our baseline specification—which brings the number of cells down to 50 per year.
2.4. Empirical Strategy
In our basic specification, the probability of dying within the next
x-years from 1935 for an individual
is assumed to be given by a linear probability model:
where
c indicates the cohort,
g the region,
Zigs,c+14 is an indicator of whether an individual has been affected by the extension of compulsory schooling,
Xisgc is a vector of covariates, including a dummy of whether an individual resides in an urban area in 1935,
µc and
νg are fixed cohort and regional effects,
δs is a gender-specific constant and
fg(
c) is a region-specific cohort trend. Given the small number of cohorts, we approximate the regional trend by polynomials of the first and second order.
gives the number of observations within a specific gender-cohort-region cell. In the basic regressions without the inclusion of cohort trends,
β1 is identified by deviations from a statewide cohort trend and regional specific intercepts and, therefore, constitutes a simple difference-in-difference estimate. This specification has been shown to be sensitive concerning the inclusion of regional specific trends (see Mazumder [
16]).
As mentioned before, we do not observe the treatment indicator for the individual, but only the share of schools with extended compulsory schooling within a region,
g, for a specific cohort,
c. Averaging Equation (2) over the observations in each gender-cohort-region cell produces:
The treatment is now given by the ratio of students affected by extended compulsory schooling,
. In general, the constructed treatment share,
, as
, does not distinguish between gender and weighs all school districts equally by construction. This creates a (possibly non-classical) measurement error in the instrument. In the following, we will carefully state the assumptions that are necessary for inference on the effects of the treatment,
β1.
Assuming (conditional) random assignment of extended compulsory schooling to school districts (or schools) within counties, the measurement error in our instrument has the expectation of zero. The estimates then represent a lower bound for β1. This is unproblematic with respect to the gender aggregation. However, as we do not know the size of the school districts, non-random assignment with respect to the number of students could occur. Larger districts could, for example, introduce the extended minimum schooling first. If the state trends and the covariates insufficiently control for this possibility, the measurement error would not necessarily lead to an attenuation bias. Instead, the sign of the bias is unclear.
Most importantly, inference on individual behavior, as described by Equation (2), is achieved from aggregates. Therefore, we are subject of committing an
ecological fallacy. The problems associated with inferring individual behavior from aggregates is well-known Robinson [
40] (in his seminal work, Robinson investigated the relationship between the ratio of foreigners and the literacy rate in U.S. states. The positive relationship on the aggregate level is a spurious correlation from foreigners living mostly in areas with higher literacy rates and disappears if confronted with individual data (Freedman [
41])). The problem occurs, as the (conditional) joint distribution of the mortality rates and the treatment is the aim of inference, while only the marginal distributions are observed. We do not know the exact combinations of mortality and schooling for each individual. This makes regressions on aggregate levels especially prone to spurious results.
Without imposing strong assumptions, it is not possible to draw conclusions about the joint distribution using only marginal distributions from aggregated data. Though this issue cannot be solved without individual level data, the reliability of the estimates can be nevertheless strengthened. Given that we use variation between cohorts within a county, we can reasonably correct, to some extent, for possible confounding. Prior studies indeed confirmed that the aggregation bias can be sufficiently relaxed in this way. For example, Lleras-Muney [
13] uses the same specification to model the effects of education on mortality rates. Comparing analysis from individual survey data and aggregated data from official statistics, her results only differ slightly. This indicates the absence of a serious aggregation bias.
Note that we face only partial compliance, as not all students change their educational decision, due to an extension of the years of compulsory schooling. Therefore, Equation (2) can be interpreted as a reduced form for the effects of education on mortality. Under random assignment of the instrument and the absence of an aggregation bias, the estimation yields a lower bound for an intention to treat a parameter of compulsory schooling extensions. It is a lower bound, due to the measurement error in the share of districts treated relative to the number of students.
In Meghir and Palme [
28], for cohorts born between 1940 and 1957, the share of fathers having more than compulsory schooling can be found. These fathers could potentially be treated in our application. The proportion of men having more than compulsory schooling is less than 25% in their sample. This would indicate a large compliance rate for the reform. Without the reform, it is very likely that most of the individuals would have experienced less schooling. The estimates should, therefore, be close to the effects on those kept longer in school, only due to the extension of compulsory schooling.
