Cattan

The Economic Journal, 133 (February), 888–903 https://doi.org/10.1093/ej/ueac078 C The Author(s) 2022.
Published by Oxford University Press on behalf of
Royal Economic Society. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.or
g/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
Advance Access Publication Date: 12 November 2022
THE LONG-TERM EFFECTS O F STUDENT ABSENCE:

EVIDENCE FROM SWEDEN ∗
Sarah Cattan, Daniel A. Kamhöfer, Martin Karlsson and Therese Nilsson
Despite the relatively uncontested importance of promoting school attendance in the policy arena, little
evidence exists on the causal effect of school absence on long-run outcomes. We address this question
Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

by combining historical and administrative records for cohorts of Swedish individuals born in the 1930s.
We find that elementary school absence significantly reduces contemporaneous academic performance, final
educational attainment and labour income throughout the life cycle. The findings are consistent with a dynamic
model of human capital formation, whereby absence causes small immediate learning losses, which cumulate
to larger human capital losses over time and lead to worse labour market performance.
Student absence from school is pervasive around the world. While raising school attendance has
long been the focus of policy in developing countries, the issue has also gained prominence in
developed countries over the past decade. State and national governments have started taking
concrete measures to reduce absenteeism, ranging from better monitoring and public awareness
campaigns to monetary fines.
Despite the relatively uncontested importance of promoting school attendance in the policy
arena, there is little causal evidence of the effect of absence on socio-economic outcomes. The
few papers that credibly establish such evidence find that absences in elementary and secondary
school have a detrimental impact on academic achievement and school graduation (Goodman,
2014; Aucejo and Romano, 2016; Liu et al., 2021). While these papers are important in going
beyond the correlational evidence, they all focus on educational outcomes in the United States.
Much remains to be known as to whether (i) impacts on educational outcomes reflect human
capital losses that translate into long-term outcomes and (ii) whether these impacts generalise to
other contexts.
∗ Corresponding author: Sarah Cattan, Institute for Fiscal Studies, 7 Ridgmount Street, WC1E 7AE London, UK.
Email: sarah [email protected]
This paper was received on 1 March 2022 and accepted on 17 October 2022. The Editor was Barbara Petrongolo.
The authors were granted an exemption to publish their data because access to the data is restricted. However, the
authors provided the Journal with temporary access to the data, which allowed the Journal to run their codes. The
codes are available on the Journal repository. The data and codes were checked for their ability to reproduce the
results presented in the paper. The replication package for this paper is available at the following address: https://
doi.org/10.5281/zenodo.7064130.
For valuable comments we are grateful to Esteban Aucejo, Sonia Bhalotra, Arnaud Chevalier, Paul Devereux, Martin
Dribe, Martin Fischer, Petter Lundborg, Teresa Molina, Erik Plug, Martin Salm, Kjell Salvanes, Hendrik Schmitz, Nina
Schwarz, Guido Schwerdt and Matthias Westphal. We would also like to thank seminar participants at CINCH (Essen),
IFN (Stockholm), Lund University, University of Essex as well as participants of EEA 2015, ESPE 2015, IWAEE 2015,
VfS 2015 and the Essen Health Conference 2015. For collecting and digitising the data used here we are indebted to
our colleagues in Essen and Lund as well as a vast team of research assistants. Earlier versions of this paper circulate as
IZA Discussion Paper 10995 (September 2017), ISF Working Paper 06/21 (February 2021) and DICE Discussion Paper
383 (March 2022). Sarah Cattan gratefully acknowledges financial assistance from the British Academy Postdoctoral
Fellowship pf140104 and the ESRC-funded Centre for the Microeconomic Analysis of Public Policy (ES/T014334/1).
Therese Nilsson gratefully acknowledges financial support from the Swedish Research Council (dnr 2019-03553), the
Gyllenstierna Krapperup Foundation, KEFU and the Crafoord foundation (dnr 20190685). Daniel Kamhöfer is grateful
to the Institute for Fiscal Studies, London, and Lund University for hospitality, as well as to the German Academic
Exchange Service (DAAD) and the University of Duisburg-Essen–ERASMUS+ Mobility Program for financial support.
[ 888 ]
2023] the long-term effects of student absence 889
This paper provides evidence of the impact of student absence on educational achievement,
labour market outcomes and mortality in the context of Sweden. We construct a unique panel
following a representative sample of cohorts born 1930–5, which links digitised records of
absence and educational performance in elementary school with census and tax register data on
socio-economic outcomes throughout the life cycle. To our knowledge, this is the first dataset
that allows for an analysis of the long-term impact of absences. This is an important innovation
in light of evidence showing early career advantages either fading relatively fast (such as the
effect of the business cycle on earnings; cf. Genda et al., 2010; Oreopoulos et al., 2012; Altonji
et al., 2016) or becoming more pronounced at higher ages (such as the effects of schooling; cf.

Bhuller et al., 2017).
Even with rich data, analysing the short- and long-term impacts of student absences requires
a robust strategy to tease out their causal impacts from the vast array of unobserved confounding
factors. Indeed, students who miss school may have poorer health, less engaged parents and/or
less inspiring teachers, which could all lead to spurious correlations between absence, education
and labour market outcomes. To deal with the endogeneity of absence, we exploit two features
of our data. For the short-term analysis of student absences on school performance, we exploit
within-student, between-grade variation in absence and performance at two time points (grade 1
and grade 4). For the long-term analysis, we exploit the presence of siblings in our dataset, and use
within-family variation in absence and long-term outcomes to purge the correlation between them
from all family-level time-invariant factors. We present a series of robustness tests supporting
the identification strategy and estimate bounds following Oster (2019), suggesting that biases
associated with our estimates are unlikely to be either statistically or economically large.
Our analysis yields two main findings. First, we find a negative and significant impact of
student absence on academic performance in elementary school equivalent to 4.5% of a standard
deviation (SD) for ten days of absence. This impact is similar to estimates presented for more
recent cohorts of elementary school children growing up in the United States (Goodman, 2014;
Aucejo and Romano, 2016). Linking school performance to later-life earnings, we also express
the effect of student absence in terms of the long-term earnings potential.
Second, estimates of the long-term effects of absence are most pronounced for our income
measures: income measured at ages 35–40 and pension income, which is a proxy for lifetime
earnings. Ten annual days of absence in elementary school are estimated to decrease these income
measures by 1%–2%. The estimates of the impacts on other long-term outcomes are similar in
size, but not as precisely estimated, with the exception of a significant negative impact on men’s
likelihood to complete secondary education. The results are consistent with a model of skill
accumulation, where early losses in human capital grow into later skill deficiencies, which affect
educational achievement, employment and income, but are only large enough to be picked up
precisely on later outcomes in the life cycle.
Our paper relates to a broad literature examining the impact of instructional time on educational
achievement and adult outcomes. Although school absence is an important determinant of the
total amount of time spent in school, most studies exploit exogenous variation in the length of
the school year as a source of exogenous variation in instructional time (see, e.g., Pischke, 2007;
Leuven et al., 2010; Agüero and Beleche, 2013; Fischer et al., 2019 using laws and law changes;
Fitzpatrick et al., 2011; Carlsson et al., 2015 using variation in test dates; Marcotte and Hemelt,
2008; Marcotte and Hansen, 2010; Hansen, 2011 using unscheduled school closures).
Regarding individual absence from school, three studies, Goodman (2014), Aucejo and
Romano (2016) and Liu et al. (2021), analyse the effects of school absence on educational

C The Author(s) 2023.
890 the economic journal [february
achievement in the United States. Using Massachusetts data (2003–10) for students attending
grade 3 onwards and North Carolina data (2006–10) for grade 3 to 5 students, respectively,
Goodman (2014) and Aucejo and Romano (2016) showed that school results are negatively
affected by absence. Both studies control for institutional heterogeneity using school, teacher
and individual fixed effects, as we do. Liu et al. (2021) estimated the impact of absences during
secondary school on educational outcomes using data from a Californian school district (2002–
13). They used between-subject, within-individual variation in absence to identify the impact of
absence on contemporaneous achievement, and estimate bounds around the impact of absence
on high school graduation and college enrolment using the method of Oster (2019).

We contribute to the above literature by providing new evidence on the effect of student
absence as one determinant of instructional time. Our paper is the first to present estimates of the
impact of days of absence outside the United States, and on adult and later life outcomes. The
literature examining the effect of region- or school-level changes in instructional time suggests
that institutions and the educational system are relevant factors for observed effects (see, e.g.,
Pischke, 2007; Gathmann et al., 2015; Galama et al., 2018). Sweden makes a particularly
interesting case in comparison to the United States, given that its labour market was characterised
by active labour market policies and compressed wages (Erixon, 2008), and embedded in a Social
Democratic welfare state providing comprehensive social insurance against health and social risks
that workers face (Bergh, 2014).
Our results show that these innovations to the literature matter for our understanding of the
impact of school absences. Considering effects throughout the life cycle sheds new light on
previous findings regarding the role of school absence, and indicate that the short-term human
capital losses manifested as impacts on test scores translate into long-term penalties in the labour
market.
1. Background and Data

1.1. The Swedish Education System
In the 1930s and 1940s, all children in Sweden were required to attend public school, Folkskola,
starting at the age of seven. Education was free and co-educational. The first four grades of
Folkskola, which we refer to as elementary school, were mandatory. Admissions to secondary
school were very selective and depended on academic performance, and the minority of students
who progressed to secondary school generally matriculated after grade 4. Students not pursuing
post-compulsory schooling had to remain in Folkskola for another two years.1
The responsibility of providing elementary education was decentralised to the 2,400 school
districts, but the Ministry of Ecclesiastical Affairs provided nationwide standards, for example for
curriculum design and grading. Three academic subjects were taught in elementary school: math,
reading and speaking, and writing. The government established grading principles, dictating that
teachers should reward the quality of knowledge and regularly take notes to ensure that grades
reflected performance through the year (SOU, 1942). Online Appendix A provides details on the
school system.
1 During elementary school, our cohorts were exposed to two educational reforms, rolled out across municipalities
between 1936 and 1948. One reform expanded instructional time per school year from 34.5 to 36.4 and 39 weeks. We
control for the number of days of the school year in all specifications. The second reform introduced a mandatory seventh
grade of Folkskola. While about half of the individuals in our sample were affected by the reform, only 6% of the sibling
pairs were affected differently. Controlling for the reform does not change our point estimates; see column (6) of Table 4.

As their main organisational tool, teachers kept daily records of students’ performance, absence
and reasons for absence in an exam catalogue (Online Appendix Figure A1). For every school
year, teachers noted the total days of absence by type and final grades by subject. To conduct our
analysis, we digitised end-of-year information from these catalogues and linked this information
to administrative data on long-term outcomes.
1.2. Data Sources

The base data are a census of children born 1930–5 in a representative sample of 133 out of
about 2,400 Swedish parishes; see Bhalotra et al. (2017). We construct our analysis datasets

by combining this information with a number of historical and administrative sources described
below. We provide information about the representativeness of the sample in Online Appendix B.
Online Appendix Figure B1 shows the spatial distribution of our parishes, and Online Appendix
Table B1 compares them to all parishes in Sweden in terms of 1930 population census information.
Online Appendix B.2 defines all the variables included in the analysis.
1.2.1. Church records

The church records contain the child’s name, gender, the date, parish and location (hospital or
home) of birth, whether the birth was a singleton and the mother’s marital status at the child’s
birth. These records also include information on the parents’ names and occupations at the time
of the child’s birth. We create an indicator for maternal employment status and indicators for
whether the father was an agricultural, a production or a service worker. We use all these variables
as controls in our analysis (see summary statistics in Online Appendix Table B2).
1.2.2. School records
Schooling information comes from exam catalogues kept in historical archives, with yearly
student-level absence by type (sickness and non-sickness) and grade points by subject (math,
reading and speaking, writing) in grades 1 and 4 (years 1937–47). Grades 1 and 4 are pivotal, as
grade 1 is the first occasion academic performance that can be observed, and grade 4 represents
the last before some students proceed to secondary schooling. Grade points range from 1 (lowest)
to 15 (highest); see Online Appendix Table A1. To facilitate interpretation, we standardise the
grade points to have mean 0 and SD 1.
School records were matched to church records using information on parish, birth date and full
name. We mitigate data lost on account of migration out of the birth parish by tracking migrants
and collecting school records from destination parishes.2
1.2.3. Final education
Information on highest educational level completed comes from the 1970 population census
(SCB, 1972), when individuals are aged 35–40. We create a binary indicator for whether an
individual attains more than Folkskola.
1.2.4. Labour market outcomes
Information about employment at ages 25–30 and 35–40 comes from the 1960 and 1970 popu-
lation censuses (SCB, 1962; 1972). For each age group, we construct employment variables that
2 Online Appendix B.3 describes the matching and how we trace individuals who move to another school parish.
Online Appendix Figure B2 gives the degree of selection bias possibly arising from attrition. Online Appendix Table B3
shows balancing results for the full sample and the exam catalogue sample in terms of socio-demographic characteristics.
Online Appendix Table B4 indicates that linear regression estimates of absence on grade points do not differ substantially
across samples.

Table 1. Descriptive Statistics on Long-Term Outcomes.
(1) (2) (3) (4) (5) (6)
Mean
Age range All Female Male # obs % female
Education
More than Folkskola (in %) 11.8 12.1 11.4 5,976 49.6
Employed (in %)
In 1960 25–30 65.7 36.6 94.1 7,434 49.4
In 1970 35–40 74.3 55.9 92.6 5,976 49.6

Earnings (in Swedish krona, current values)
Labour market income 1970 35–40 19,275 10,039 28,475 5,886 49.9
Pensions 2003–8, if > 0 68–78 160,237 138,568 183,468 4,772 51.7
Notes: The numbers refer to the individuals in the siblings sample (second row of Online Appendix Table B5) we are
able to match to the long-term outcomes. The age range gives the individual’s age at which the variable is measured.
Education is taken from the 1970 census, but likely reflects completed schooling for most individuals. The education
indicator takes the value 1 if the individual has more than compulsory Folkskola education, and 0 otherwise. Employment
in 1960 and 1970 is taken from the census information in these years and takes the value 1 if the individual is employed,
and 0 otherwise. Labour market income 1970 is based on the 1970 census; unemployment enters the labour market
income as zero. Pensions information is taken from tax registers and averaged over 2003–8. Zero pensions are dropped
(very few cases). Labour market income and pensions are measured in Swedish krona (current values).
equal 1 if the individual works at least part time, and 0 otherwise. From the 1970 census we also
have labour market income when individuals were in prime working age. The income measure
has imputed zeros for individuals who were not employed.
1.2.5. Pension income

From tax registers we measure average annual pensions 2003–8, when our youngest cohort is
aged 68–79. Because. for our cohorts, full pensions require 30 years of contributions and is based
on the 15 highest income years (Sundén, 2006), this measure is a proxy for lifetime earnings.
That pension income is less sensitive to year-to-year fluctuations in labour supply than annual
earnings is a desirable feature, especially for women.3
1.2.6. Death records

Church records and the Swedish Death Index (Federation of Swedish Genealogical Societies,
2014) provides the exact date of death for all individuals that passed away. We generate indicators
for whether the individual died before 1960, 1970 and 2003.
For adult outcomes, we match individuals in the matched schooling data to the different regis-
ters using either full name, sex, birth date and parish of birth, or a unique social security number
(see Online Appendix B.4 for details about the matching procedure). Online Appendix Table B5
reports the numbers of individuals, siblings and families included in our analyses.
1.3. Descriptive Statistics

Table 1 presents descriptive statistics for our long-term outcomes, and Online Appendix C.1
presents descriptive statistics for absence by type and school achievements. In grade 1, the
average number of missed days is 11.1 days (median seven days) versus 11.6 (median eight days)
3 Online Appendix D provides details about the pension system and related rules.

in grade 4. Despite a very different context, the distribution of total days of absence is comparable
with that reported in recent US studies of absence in elementary school (Goodman, 2014; Aucejo
and Romano, 2016). We observe a slightly higher density of very high numbers of absence days
(see Online Appendix Figure C1 for the overall variation, and Online Appendix Figure C2 for
the within-family and within-individual variations), but unlike Goodman (2014), who excluded
observations with more than 60 days of absence, we do not cap absence days.4 Most absences
are sickness absences. The average number of missed days for other reasons than sickness is 1.7
in grade 1 and 3.3 in grade 4.
Average grade point in math, reading and writing, and speaking is higher in grade 4 than in

grade 1. In line with national guidelines (cf. SOU, 1942), only few students receive a very low or
a very high grade point, and the variance of the grade points is higher in grade 4 than in grade 1
(Online Appendix Figures C3–C4). The selective nature of the education system is reflected by
the fact that only 12% of our sample has more than Folkskola education.
The cohorts studied were born between 1930 and 1935, and entered the labour market around
1950.5 Online Appendix C.2 provides insights into the most prevalent occupations of our cohorts,
and shows that there was a high degree of gender segmentation in the labour market (see Online
Appendix Table C2 and Online Appendix Figure C6). Employment and earnings measured at
ages 25–30 and 35–40 reflect such segmentation. As expected, pensions are more equal between
genders than earnings are.
Our main outcomes are all negatively correlated with total days of absence (see Online
Appendix Figures C5 and C7). While these correlations point to a potentially negative effect of
school absences on short- and long-term outcomes, they are obviously not evidence of a causal
link. To start exploring the extent of selection, we look at how total days of absence vary across
groups defined by different observable characteristics (see Online Appendix Figure C9). Overall,
these statistics show limited signs of selection based on these observables. Whether students
positively or negatively select into absence is also unclear. As we describe in the next section,
our identification strategies exploit within-individual or within-family variation in absence and
outcomes. We show in Online Appendix Table C1 that there is sufficient variation of this kind.
2. Empirical Strategy
2.1. Identification of Short-Term Effects
While the main focus of this paper is on the long-term effects of school absence, a natural
starting point is to estimate the contemporaneous effect of absence on school performance. This
also allows us to compare effects with those found for the current US context.
Our aim is to identify the effect of the number of days of absence in one year on academic
achievement measured at the end of that year. We exploit the fact that we observe performance
4 Instead we winsorise our data at the top 98th percentile of the distribution of days of sickness and non-sickness
absence.
5 The cohorts under review were born during the Great Depression and were in elementary school during World War
II. Although the depression was not as severe in Sweden as it was in many other countries, we re-estimate our model for
long-term outcomes, controlling for local economic conditions (poverty rate, taxable income per capita and the annual
change of the latter) in the year of birth; see column (7) of Table 4. These estimates are extremely similar to our main
estimates and, if anything, more precise. WWII is unlikely to have disrupted either the education or the archiving of
school records. Sweden was neutral in WWII and there was an oversupply of teachers (Paulsson, 1946). We have not
found historical sources suggesting that WWII caused disruptions in education and the probability that we found exam
catalogues in archives was the same for years during and not during the War.

and absence in two grades (1 and 4) in order to control for individual-specific time-invariant
unobservable factors. We estimate the equation
Yistg = β0 + τ Dig + Q stg ’β2 + Ssg + Ttg + δi + u istg , (1)
where Yistg is grade-g performance of a student i attending school s taught by teacher t, Dig is the
number of days student i was absent from school in a grade, Q stg is a set of school/grade-specific
controls, Ssg is a grade-g school fixed effect (FE), Ttg is a grade-g teacher FE, and δi is an
individual FE.6
As argued by Bond and Lang (2013), which measure of educational achievement is used can

significantly alter the conclusion of the analysis. To address this issue and provide economi-
cally meaningful interpretations of our estimated effects of absence on achievement, we anchor
the grade point scale to pensions, a policy-relevant outcome measured in a meaningful metric
(Swedish krona, SEK). Online Appendix E provides details on the anchoring procedure and
Appendix Table E1 reports estimates of the anchoring equations.
The main identifying assumption in model (1) is that unobserved grade- and child-specific
factors affecting student i’s achievement in grade g are uncorrelated with the child’s absence,
conditional on the observables and fixed effects included in the model. The assumption is the
same as in Aucejo and Romano (2016), but stronger than the assumption required in the design
of Liu et al. (2021), who exploited within-grade, between-subject variation in absences to control
for time-varying individual-level unobserved shocks.7 Given the main focus of the paper on
long-term outcomes and the fact that this strategy has been used in several related papers, we
refer the reader to Online Appendix F.3 for a discussion of threats to its validity and robustness
checks.
2.2. Identification of Long-Term Effects

Our main analysis focuses on the effect of school absences in elementary school on adult
outcomes. We define our treatment variable as the average number of days of absence in grades 1
and 4. Since the average days of absence in elementary school are fairly stable across all grades,
we can think of the treatment variable as closely measuring the average yearly number of days
of absence in each year throughout elementary school.8
Our research design exploits the fact that we can identify siblings in our data. We propose to
identify the impact of school absence on long-term outcomes from within-family variation and
estimate the equation
Wi f = β0 + τ Di + X i ’β1 + ω f + v i f ,
where Wi f is an outcome of individual i of family f ; X i is a set of individual-specific time-
invariant controls, including demographic characteristics, grade-1 school identifier, grade-1
6 The specification assumes a homogeneous effect of absence in grades 1 and 4 in order to interpret τ as the marginal
effect of a day of absence. This assumption cannot be tested in the context of the individual FE model, but it can be tested
in the sibling FE model. We report results of such a test in the lower panel of Online Appendix Table F6, which shows
that the effects of absence in grade 1 and in grade 4 are statistically indistinguishable.
7 On the other hand, Liu et al. (2021) required an absence of spillover effects across subjects for identification.
8 We have not collected information on grades 2 and 3 systematically, but some exam catalogues cover students in
several grades. Average days of absence in grades 2 (10.7 days) and 3 (11.9) are close to the averages for grades 1 (11.1)
and 4 (11.6). We also estimate the model allowing for a different effect of grade-1 and grade-4 absences on long-term
outcomes. In the sibling FE model, estimates of these two effects are not statistically different from each other (Online
Appendix Table G2).

teacher identifier, as well as information about class size, lowest and highest grade taught to
students in the same classroom and length of the school year in weeks, averaged across grades 1
and 4. We also control for an indicator for whether grade-1 teacher is the same as grade-4 teacher
and an indicator for whether grade 1 school is the same as grade 4 school. Here ω f is a family
FE.9
The advantage of this strategy is that, in contrast with the bounding method used in Liu et al.
(2021), it allows us to point identify the effect of absence on long-term outcomes. However, for
the parameter τ to have a causal interpretation, it requires that any child-specific unobservable
that does not have the same additive effect on outcomes for both siblings is uncorrelated with the

child’s absence. We discuss possible threats to identification in Subsection 3.3, where we present
the results of a number of exercises to gauge the research design’s validity.
3. Results
3.1. Short-Term Effects of School Absence
Column (1) of Table 2 reports ordinary least square (OLS) estimates of parameter τ in (1), and
shows that one additional day of absence is significantly associated with a 0.51% of an SD
decrease in performance. The individual FE estimate in column (2) is slightly smaller (0.0045),
but also statistically significant at the 1% level.10
Assuming linearity, the effect of ten days of absence—the average in our sample—corresponds
to 4.5% of an SD decrease in performance. Despite analysing absence in a very different context
and literally in another century, our results measured in SD units are comparable to those in the
related literature. Using recent US data, Goodman (2014) found an effect of 0.8% of an SD in
math and English, and Aucejo and Romano (2016) found effects of 0.55% of an SD in math
and 0.29% in reading. Online Appendix Figures F1 and F2 present evidence strongly suggesting
that the relationship between absence and academic performance is linear. Moreover, we show
that there are no significant differences in the short-term impact of absences between gender and
across individuals from different socio-economic groups (Online Appendix F.2).
By anchoring student performance to pension incomes, we translate the short-term effect of
absence on school performance into its effect on earnings potential. Table 2 (second and third
rows) reports the results. In the individual FE specification, the impact of ten additional absence
days on school performance translates into a non-significant decrease in earnings potential of
SEK 16, equivalent to a 0.1% decrease in average pension income. We obtain similarly sized
effects when anchoring by grade and gender (third row), and when using earnings at age 35–40,
as the anchor.
Online Appendix F.3 presents results of tests we conduct to probe the validity of the individual
FE strategy to recover the causal effect of absence on performance. These results strongly suggest
that our short-term estimates of absence are likely robust to the presence of individual-level time-
varying unobserved heterogeneity.
9 Online Appendix Table C1 shows that a large fraction of variation in absence and performance in the data comes
from within-family variation, which provides confidence that there is sufficient variation to consider using a sibling FE
strategy.
10 Online Appendix Table F1 reports estimates for control variables. Online Appendix Table F2 states negative and
significant estimates of the effect of absences on performance for each of the three subjects. Online Appendix Table F3
shows that the estimates are unchanged when grade points are standardised before the aggregation. Online Appendix
Table F4 gives the estimates when averaging over the math grade points and reading and speaking grade points (without
writing). Online Appendix Table F5 gives the estimates when grades are coded on a seven-point scale.

Table 2. Estimates of the Short-Term Effect of School Absence on Academic Performance.
(1) (2)
Individual
OLS fixed effects
Panel A: average grade points in units of SD (mean 0, SD 1)
Days of absence −0.0051∗∗∗ −0.0045∗∗∗
(0.0007) (0.0013)
Panel B: average grade points in units of pension (anchoring by grade, mean
pension in sample: 160,237 Swedish krona)
Days of absence −54.7344∗∗∗ −16.1192

(7.4068) (20.0252)
Panel C: average grade points in units of pension, by gender (anchoring by
grade and gender, mean pension in sample: 138,568 Swedish krona for
women and 183,468 Swedish krona for men)
Days of absence −69.2581∗∗∗ −37.5660∗∗
(9.2558) (18.4195)
# observations 14,066 8,934
# individuals/families 4,467
Notes: Each cell reports the coefficient associated with days of absence in separate regressions where the dependent
variable is the variable indicated in the first row of each panel. The dependent variable in panel A is average performance
over math, reading and speaking, and writing, standardised to have mean 0 and SD 1. In panel B, it is a measure of average
grade points in units of pensions, where the relationship between grade points and pensions is estimated separately for
grade 1 and grade 4; see Online Appendix E. The third panel repeats the anchoring of grade points in units of pensions,
but estimates the relationship between grade points and pensions separately by school grade and gender. Both OLS and
individual FE models control for grade, range of grades instructed in the same classroom and length of the school year
in weeks. The OLS specification also controls for gender, born out of wedlock, twin birth, mother employed at the time
of birth, born in hospital, full sets of fixed effects for the year and month of birth, year and month interactions, age,
parent’s year of birth and family’s socio-economic status based on the first digit of the Historical International Standard
Classification of Occupations (HISCO) code (see van Leeuwen et al., 2002) of the father. Standard errors clustered at the
parish level are given in parentheses. Significance: ∗∗ p ≤ 0.05, ∗∗∗ p ≤ 0.01.
3.2. Long-Term Effects of School Absence

Table 3 reports the effects on long-term outcomes of the average days of absence across grades
1 and 4 in columns (1) (OLS) and (3) (sibling FEs). To ease interpretation, we also report the
effect of ten days of absence relative to the mean of the outcome considered in columns (2) and
(4). We calculate this by multiplying the coefficient by five (the treatment variable measures the
average number of days of absence across two grades) and dividing it by the outcome mean.
A first observation is that the OLS and sibling FE estimates are similar both in terms of
magnitude and statistical significance. This aligns with one of the conclusions from the short-term
analysis that selection into absence based on unobservables may not be particularly prominent
in our context.
Looking at the sibling FE estimates of Table 3, we find that ten days of absence in elemen-
tary school leads to a statistically insignificant 2.2% reduction in secondary school completion
relative to baseline. Given the gender differences in educational attainment that existed during
this period, impacts could vary by gender. Indeed, the effect of absence on secondary school
completion is negative and statistically significant for men, while it is close to zero for women
(Online Appendix Table G1). The negative effect of elementary school absence on secondary
schooling for men is consistent with secondary school admissions depending on elementary
school performance.

Table 3. Long-Term Effects of School Absence.
(1) (2) (3) (4)
OLS Sibling fixed effects
Coefficient Rel. size Coefficient Rel. size
More than Folkskola (1 = yes)
Absence (average, grades 1 and 4) −0.0007 −0.0297 −0.0005 −0.0215
(0.0005) (0.0009)
Employment 1960 (1 = yes)
Absence (average, grades 1 and 4) −0.0012∗ −0.0088 −0.0007 −0.0053
(0.0007) (0.0013)

(0.0007) (0.0014)
Labour market income 1970
Absence (average, grades 1 and 4) −54.2973∗∗ −0.0141 −80.7042∗∗ −0.0209
(22.7088) (37.2151)
Pensions 2003–8
Absence (average, grades 1 and 4) −330.0940∗∗∗ −0.0103 −396.0887∗∗ −0.0124
(71.2631) (169.3640)
Notes: Each panel reports the coefficient associated with the total days of absence (average over grades 1 and 4) in
separate regressions where the dependent variable is indicated in the left column. Controls include gender, born out of
wedlock, twin birth, born in hospital, mother employed at the time of birth and sets of fixed effects for the year and month
of birth, year and month interactions. Specifications also controls for class size, the lowest and highest grade taught to
students in the same classroom and length of the school year in weeks, averaged across grades 1 and 4. Models also
control for school and teacher fixed effects, and a dummy if the individual changes school or teacher between grades 1 and
4. The OLS models further control for parent’s year of birth and family’s socio-economic status based on the first-digit
HISCO code of the father. Number of observations: more than Folkskola 5,976, employment 1960 7,434, employment
1970 5,976, income 1970 5,886, pensions 4,772. Standard errors clustered at the parish level are given in parentheses.
Significance: ∗ p ≤ 0.1, ∗∗ p ≤ 0.05, ∗∗∗ p ≤ 0.01.
Next, we turn to the effects of absence on labour market outcomes. Impacts on employment
are not statistically significant, but the point estimates suggest that impacts on employment may
grow slightly with age, from a 0.5% reduction in employment at ages 25–30 to a 1.3% reduction
ten years later. As shown in row 3 of Table 3, a one-day increase in the average number of days
of absence in elementary school leads to an SEK 80 reduction in earnings in 1970 and an SEK
396 reduction in pensions;11 both estimates are significant at the 5% level. Relative to the means
of these variables, the impacts are fairly similar in magnitude to each other, between 1%–2%
reduction in income. To put these magnitudes in perspective, consider that Fredriksson et al.
(2013) estimated an earnings increase of 1.2% at age 27–42 when class size in Swedish schools
is reduced by one student, Chetty et al. (2014) estimated an earnings rise of 1.3% at age 28 from
a 1-SD improvement in teacher quality in the United States, while Carrell et al. (2018) drew on
data for a Florida county and showed that exposure to a disruptive peer during elementary school
reduces earnings at age 26 by 3%–4%.
As with the short-term outcomes, we explore whether the effects of school absence on long-
term outcomes are non-linear, but find no strong evidence that this is the case (Online Appendix
11 Online Appendix Table D1 shows that pension results are robust when controlling for the receipt of a widow
pension.

Figure G1). We also find no strong evidence that the effect of absence depends on whether
absence happened in grade 1 or grade 4 (Online Appendix Table G2).
Overall, our estimates point to a consistently negative effect of absences in elementary school
on economic outcomes through the life cycle, with more pronounced and statistically significant
impacts on educational attainment for men and on income for both genders. In terms of mag-
nitude, the effects are larger than what is implied by the short-run effects of absence on school
performance.12 Moreover, the effects become slightly larger and more significant when measured
further into the life cycle. Taken together, these results are consistent with a dynamic model of
skill accumulation, where skill deficits resulting from absences in elementary school translate

into lower skill levels progressively accumulating over the life cycle. These results also underline
the importance of measuring impacts of absence at various points in the life cycle.
3.3. Robustness Checks

The interpretation of the above results as causal effects of absences on long-term outcomes relies
on the assumption that school absences are uncorrelated with unobservable determinants of long-
term outcomes that vary across siblings. There are several threats to the validity of this assumption.
First, long-term outcomes could be affected by child-specific unobserved endowments, which
may also affect children’s likelihood to miss school. For example, one sibling may be born
frailer, have lower grit and work ethics than the other, which would make her more likely to both
be absent and have worse adult outcomes than her sibling. Second, long-term outcomes could
be affected by idiosyncratic, child-specific shocks that directly affect the incidence of absence
and have long-lasting effects on adult outcomes (over and beyond the absence). An obvious
example are health shocks during elementary school, which make one sibling (but not the other)
particularly sick one year and have long-lasting effects on his/her adult outcomes. Other types
of relevant shocks may include changes to the institutional and/or economic environments over
time, which affect siblings differently because they are born in different years.
Below we discuss a series of empirical exercises to help us gauge the likely salience of these
different threats to the validity of our research design. First, we re-estimate the short-term effect
of absence on educational achievement using the sibling FE strategy and compare it to the
individual FE estimates of Table 2. As reported in Online Appendix Table G4, the two estimators
yield very similar estimates of the impact of absence on achievement in elementary school.
This suggests that child-specific unobservable determinants of educational achievement that vary
between siblings are unlikely to be correlated with absences. While this does not guarantee
that the same is true of unobservable determinants of long-term outcomes, it does reinforce the
message that selection into absence based on unobservables is unlikely to be strong.
We then estimate the effects of grade-4 absence on grade-1 performance. If sibling FEs
fully account for unobserved child-specific time-invariant factors, future absence should have no
power in predicting grade-1 performance. Online Appendix Table G5 shows that, while grade-
4 absence and grade-1 performance are negatively correlated in the OLS model, the grade-4
absence coefficient drops and becomes insignificant in the sibling FE model.
12 Specifically, the (sibling FE) effect of school performance, measured as average grade points in grades 1 and 4, on
secondary schooling completion is about 0.07 (Online Appendix Table G3). Multiplying this estimate by the estimated
effect of absence on school performance of −0.0045 (Table 2), we would expect an effect of absence on enrolment of
−0.0003 in the sibling FE specification. In contrast, our direct sibling FE estimate in Table 3 (−0.005) is considerably
larger than the indirect estimate.

The next exercise uses the idea that, if the sibling FE specification does not control for an
important unobservable (fixed or time-varying) determinant of long-term outcomes, the sibling
FE estimate of grade-4 absence on long-term outcomes would change through the inclusion of
grade-1 performance, to the extent that grade-1 performance captures these unobservables (e.g.,
idiosyncratic work ethic or health shock occurring before grade 1). The results are reported in the
second and third columns of Table 4, and show that the estimated effect of grade-4 absence on
long-term outcomes remains virtually unchanged through the inclusion of grade-1 performance.
We then conduct two exercises to assess the possibility that individual-specific health shocks
during childhood confound our long-term estimates. First, we test whether absence due to

sickness affects long-term outcomes differently from absence due to other reasons. If health
shocks simultaneously determined absence and long-term outcomes, we would likely find that
absence due to sickness is more detrimental than absence due to other reasons. We test this
hypothesis by re-estimating our model, also controlling for the average number of non-sickness
days across grades 1 and 4. We report the coefficients for any absence in the fourth column
of Table 4 (see Online Appendix Table G7 for non-sickness absence coefficients and Online
Appendix Table F7 for the short-term estimations by reason of absence). Across all long-run
outcomes, we cannot reject that the coefficient on the average days of non-sickness absence is
equal to the coefficient on the average days of (any) absence, which suggests that the long-term
effects of absence in Table 3 are more likely to be driven by the loss of instructional time rather
than by a health shock.
Our second test for confounding health shocks looks directly at whether school absences have
an effect on long-term health, as measured by mortality. To do so, we re-estimate our sibling
FE model, this time using indicators for mortality at three time points (1960, 1970 and 2003) as
outcomes. We cannot reject that the coefficient on absence in any of these regressions is zero
(see panel B of Table 4). Again, this suggests that the effect of absence we identify through the
sibling FE strategy is likely to capture the effect of losing instructional time rather than the effect
of a health shock associated with the absence.
Given that the labour market was gender segmented, we also test whether sibling sex compo-
sition matters. Column (5) of Table 4 reports estimates for the sample of families with same-sex
siblings, which are very similar to our main results and suggest that low female labour force
participation is unlikely to bias the analysis using the entire sample. The last two columns of
Table 4 show that the baseline results are also robust to controlling for siblings’ differential
exposure to the instructional time school reform and to economic environments at birth.
3.4. Bounds
The above robustness checks all suggest that the identifying assumption underlying the sibling FE
strategy is unlikely to be violated. However, given that it does remain an untestable assumption, we
conclude our analysis by bounding our estimates, using the approach of Oster (2019), which is also
implemented in Liu et al. (2021). As we describe in Online Appendix H, this approach exploits
observables in order to gauge how unobservables may affect the estimates, and is therefore
only helpful to the extent that the selection on observables is informative of the selection on
unobservables one might suspect. As our empirical strategy controls for a breadth of factors that
are likely to affect absence and long-term outcomes, the bounding approach seems useful here.
Column (1) (‘short regression’) of Table 5 reports the coefficients associated with absence in a
regression without any controls. Column (2) (‘intermediate regression’) replicates the sibling FE

900
Table 4. Robustness of the Effect of Absence on Long-Term Outcomes.

(1) (2) (3) (4) (5) (6) (7)
Baseline (i) Grade-4 effect with (ii) Differentiat- (iii) Same- (iv) Contr. for (v) Contr. for
results and without controlling ing for reason sex siblings compulsory local conditions
(Table 3) for grade-1 performance of absence schooling at birth
Avg., Grade-4 Grade-4 Avg., Avg., Avg., Avg.,
Reported coefficient on: grades 1 and 4 absence absence grades1 and 4 grades 1 and 4 grades 1 and 4 grades 1 and 4
Panel A: estimates of the effect of absence in different sibling fixed effect specifications
More than Folkskola (1 = yes) −0.0005 −0.0010 −0.0011 −0.0006 −0.0011 −0.0005 −0.0004
(0.0009) (0.0008) (0.0008) (0.0007) (0.0016) (0.0009) (0.0012)
Employment 1960 (1 = yes) −0.0007 0.0016 0.0016 −0.0005 −0.0023 −0.0007 −0.0016
(0.0013) (0.0011) (0.0011) (0.0010) (0.0019) (0.0013) (0.0014)
Employment 1970 (1 = yes) −0.0020 −0.0016 −0.0016 −0.0015 −0.0048 −0.0019 −0.0039∗∗∗
(0.0014) (0.0010) (0.0010) (0.0012) (0.0030) (0.0014) (0.0013)
Labour market income 1970 −80.7∗∗ −61.1 −62.1 −50.0 −117.4 −81.1∗∗ −85.9∗
(37.2) (45.4) (45.2) (32.7) (85.14) (37.3) (45.9)
Pensions 2003–8 −396.1∗∗ −284.9∗∗ −248.2∗∗ −319.1∗∗∗ −689.2∗ −395.8∗∗ −572.1∗∗∗
(196.4) (119.4) (105.0) (121.5) (405.4) (169.2) (199.8)
Panel B: additional results—effects of absence (average, grades 1 and 4) on mortality
the economic journal
Passed away before 1960 (1 = yes) 0.00001

(0.00030)
(0.00048)
(0.00078)
Notes: Each panel reports the coefficient associated with the total days of absence (average over grades 1 and 4) in separate regressions where the dependent variable is indicated
in the left column. Controls as reported in Table 3. Panel A shows estimates in different sibling fixed effect specifications: (i) grade-4 effect with and without controlling for
C

grade-1 performance (see Online Appendix Table G6 for grade-1 performance coefficients), (ii) differentiating for different reasons of absence (see Online Appendix Table G7 for
non-sickness absence coefficients), (iii) estimates using same-sex siblings, (iv) controlling for school reforms and (v) controlling for local conditions at birth (poverty rate, taxable
income per capita, economic development (the year-to-year change in the taxable income per capita), all assessed at the time of birth). Panel B shows estimates on absence on
mortality by 1960, 1970 and 2003. Standard errors clustered at the parish level are given in parentheses. Significance: ∗ p ≤ 0.1, ∗∗ p ≤ 0.05, ∗∗∗ p ≤ 0.01.
[february
The Author(s) 2023.

Table 5. Oster Bounds for the Long-Term Effects of Absence in School.
(1) (2) (3) (4)
Coefficient of absence Selection bias
Short Intermediate Same Opposite
Dependent variable regression regression direction direction
More than Folkskola (1 = yes)
Absence (average, grades 1 and 4) −0.0009∗ −0.0005 −0.73850 −0.00014
(0.0005) (0.0009)
[0.00] [0.41]

Absence (average, grades 1 and 4) −0.0013 −0.0007 0.00306 −0.00177
(0.0008) (0.0013)
[0.00] [0.60]
(0.0008) (0.0014)
[0.00] [0.50]
Labour market income 1970
Absence (average, grades 1 and 4) −65.4768∗∗∗ −80.7042∗∗ −107.88081 −73.15848
(22.3882) (37.2151)
[0.00] [0.64]
Pensions 2003–8
Absence (average, grades 1 and 4) −384.5674∗∗∗ −396.0887∗∗ −90.55889 −423.49521
(81.8270) (169.3640)
[0.00] [0.59]
Notes: Column (1) gives the coefficient associated with absence in the short regression where the outcome variable (stated
on the left) is regressed on absence and an intercept (without any controls), denoted by β̇ in Online Appendix H. The
intermediate regression in column (2) reiterates the sibling FE specification estimates of the coefficient associated with
absence (reported in Table 3 and corresponding to β̃ in the model). Columns (3) and (4) report the bounds (β ∗ ) when
selection on unobservables is of the same magnitude as selection on observables and goes in the same direction (δ = 1)
and the opposite direction (δ = −1), respectively. The R 2 of the short and intermediate models are reported in brackets
in columns (1) and (2), respectively. To calculate the bounds, we use the Stata ado-file psacalc provided online by
Emily Oster. All errors are our own responsibility. Number of observations as in Table 3. Standard errors clustered at the
parish level are given in parentheses.
estimates from Table 3. Column (3) (column (4)) reports the bound under the assumption that the
unobservables move the point estimate in the same (opposite) direction and by the same magnitude
as the introduction of the observables does. Despite the R 2 of the intermediate regression being
higher than 0.5 for all outcomes, there is remarkably little change between columns (1) and (2).
This modest influence of selection on observables is reflected in the estimated bounds: all bounds
are negative, with the exception of the bound for the effect of absence on 1960 employment in
column (3). Thus, even under the extreme scenario that our sibling FE model fails to account for
as much selection on unobservables as it does account for selection on observables, we cannot
reject there would be a negative effect of absence on long-term outcomes.13
4. Conclusion
School absences are an important, but often overlooked, determinant of instructional time. To
date, little is known about the long-run impact of students missing school, and the only studies
13 Bounds for short-term effects in Online Appendix Table H1 underline this conclusion.

providing causal evidence of the impact of student absence on academic performance focus on
the United States. The contribution of this paper is to estimate the impact of student absence in
elementary school on short- and long-term outcomes for a non-US context by using a unique
combination of historical records and administrative datasets from Sweden.
Our analysis shows that absence in elementary school has a significant and negative impact
on student performance: increasing total absence in one grade by ten days leads to a reduction
in grade point average of 4.4% of an SD, an effect of comparable magnitude to that found in the
United States. For men, this immediate impact on school performance spills over into secondary
school admissions, which were based on elementary school performance. This effect is at least

as large as one would expect based on the effect of absence on performance—even though we
are unable to attribute it to a certain school grade.
For other long-term outcomes, we find consistent evidence that there is a penalty to absence
in elementary school: estimates have the expected negative sign for all long-term outcomes, and
they are statistically significant for earnings along the entire life cycle. Together, the short- and
long-term effects of absence suggest that a key mechanism underlying these results is the effect
of instructional time losses on early levels of skills, which accumulate over the life cycle and
eventually create non-negligible income penalties.
Our research starts filling an important gap in the evidence base on the long-term impacts of
school absence and thereby informs policy discussions about high rates of absences around the
world. Our findings hone in on the impact of individual absences, as opposed to school closures.
In this light, they may only partly be relevant to predict the long-term effects of the school
closures during the early phases of the COVID-19 pandemic. But, as school absenteeism becomes
increasingly driven by individual students self-isolating, our results can provide useful evidence
that associated learning losses may have a long-term impact if not appropriately compensated.
Institute for Fiscal Studies, London, UK & IZA—Institute of Labor Economics, Bonn, Germany
Heinrich Heine University Düsseldorf, Germany & IZA—Institute of Labor Economics, Bonn,
Germany
University of Duisburg-Essen, Germany & IZA—Institute of Labor Economics, Bonn, Germany
Lund University, Sweden & Research Institute of Industrial Economics (IFN), Stockholm, Sweden
Additional Supporting Information may be found in the online version of this article:
Online Appendix
Replication Package
References
Agüero, J. and Beleche, T. (2013). ‘Test-Mex: Estimating the effects of school year length on student performance in
Mexico’, Journal of Development Economics, vol. 103(C), pp. 353–61.
Altonji, J.G., Kahn, L.B. and Speer, J.D. (2016). ‘Cashier or consultant? Entry labor market conditions, field of study,
and career success’, Journal of Labor Economics, vol. 34(S1), pp. S361–401.
Aucejo, E.M. and Romano, T.F. (2016). ‘Assessing the effect of school days and absences on test score performance’,
Economics of Education Review, vol. 55, pp. 70–87.
Bergh, A. (2014). Sweden and the Revival of the Capitalist Welfare State, Cheltenham, UK: Edward Elgar Publishing.
Bhalotra, S., Karlsson, M. and Nilsson, T. (2017). ‘Infant health and longevity: Evidence from a historical intervention
in Sweden’, Journal of the European Economic Association, vol. 15(5), pp. 1101–57.
Bhuller, M., Mogstad, M. and Salvanes, K.G. (2017). ‘Life-cycle earnings, education premiums, and internal rates of
return’, Journal of Labor Economics, vol. 35(4), pp. 993–1030.

Bond, T.N. and Lang, K. (2013). ‘The evolution of the black-white test score gap in grades K-3: The fragility of results’,
The Review of Economics and Statistics, vol. 95(5), pp. 1468–79.
Carlsson, M., Dahl, G., Öckert, B. and Rooth, D.O. (2015). ‘The effect of schooling on cognitive skills’, The Review of
Economics and Statistics, vol. 97(3), pp. 533–47.
Carrell, S.E., Hoekstra, M. and Kuka, E. (2018). ‘The long-run effects of disruptive peers’, American Economic Review,
vol. 108(11), pp. 3377–415.
Chetty, R., Friedman, J.N. and Rockoff, J.E. (2014). ‘Measuring the impacts of teachers II: Teacher value-added and
student outcomes in adulthood’, American Economic Review, vol. 104(9), pp. 2633–79.
Erixon, L. (2008). ‘The Swedish third way: An assessment of the performance and validity of the Rehn–Meidner model’,
Cambridge Journal of Economics, vol. 32(3), pp. 367–93.
Federation of Swedish Genealogical Societies. (2014). ‘Swedish death index 1901–2013’, Data files, Farsta, Sweden:
Federation of Swedish Genealogical Societies.
Fischer, M., Karlsson, M., Nilsson, T. and Schwarz, N. (2019). ‘The long-term effects of long terms—Compulsory

schooling reforms in Sweden’, Journal of the European Economic Association, vol. 18(6), pp. 2776–823.
Fitzpatrick, M., Grissmer, D. and Hastedt, S. (2011). ‘What a difference a day makes: Estimating daily learning gains
during kindergarten and first grade using a natural experiment’, Economics of Education Review, vol. 30(2), pp.
269–79.
Fredriksson, P., Öckert, B. and Oosterbeek, H. (2013). ‘Long-term effects of class size’, The Quarterly Journal of
Economics, vol. 128(1), pp. 249–85.
Galama, T.J., Lleras-Muney, A. and van Kippersluis, H. (2018). ‘The effect of education on health and mortality:
A review of experimental and quasi-experimental evidence’, Oxford Research Encyclopedia of Economics and
Finance, Oxford University Press, doi: 10.1093/acrefore/9780190625979.013.7.
Gathmann, C., Jürges, H. and Reinhold, S. (2015). ‘Compulsory schooling reforms, education and mortality in twentieth
century Europe’, Social Science & Medicine, vol. 127, pp. 74–82.
Genda, Y., Kondo, A. and Ohta, S. (2010). ‘Long-term effects of a recession at labor market entry in Japan and the United
States’, Journal of Human Resources, vol. 45(1), pp. 157–96.
Goodman, J. (2014). ‘Flaking out: Student absences and snow days as disruptions of instructional time’, Working Paper
20221, National Bureau of Economic Research.
Hansen, B. (2011). ‘School year length and student performance: Quasi-experimental evidence’, Unpublished
manuscript, University of Oregon.
Leuven, E., Lindahl, M., Oosterbeek, H. and Webbink, D. (2010). ‘Expanding schooling opportunities for 4-year-olds’,
Economics of Education Review, vol. 29(3), pp. 319–28.
Liu, J., Lee, M. and Gershenson, S. (2021). ‘The short- and long-run impacts of secondary school absences’, Journal of
Public Economics, vol. 199, 104441.
Marcotte, D. and Hansen, B. (2010). ‘Time for school?’, Education Next, vol. 10(1), pp. 53–9.
Marcotte, D. and Hemelt, S. (2008). ‘Unscheduled school closings and student performance’, Education Finance and
Policy, vol. 3(3), pp. 316–38.
Oreopoulos, P., Von Wachter, T. and Heisz, A. (2012). ‘The short- and long-term career effects of graduating in a
recession’, American Economic Journal: Applied Economics, vol. 4(1), pp. 1–29.
Oster, E. (2019). ‘Unobservable selection and coefficient stability: Theory and evidence’, Journal of Business & Economic
Statistics, vol. 37(2), pp. 187–204.
Paulsson, E. (1946). Om folkskoleväsendets tillstånd och utveckling i Sverige under 1920-och 1930-talen (till omkring
år 1938), Jönköping, Sweden: Länstryckeriaktiebologet.
Pischke, J.S. (2007). ‘The impact of length of the school year on student performance and earnings: Evidence from the
German short school years’, ECONOMIC JOURNAL, vol. 117(523), pp. 1216–42.
SCB. (1962). ‘Population and housing census 1960’, Data files, Stockholm, Sweden: National Bureau of Statistics (SCB).
SCB. (1972). ‘Population and housing census 1970’, Data files, Stockholm, Sweden: National Bureau of Statistics (SCB).
SOU. (1942). ‘Betänkande med utredning och förslag angående betygsättningen i folkskolan, angivet av inom ecklesi-
astikdepartementet tillkalade sakkunniga’, Statens offentliga utredningar 1942:11.
Sundén, A. (2006). ‘The Swedish experience with pension reform’, Oxford Review of Economic Policy, vol. 22(1), pp.
133–48.
van Leeuwen, M., Maas, I. and Miles, A. (2002). HISCO: Historical International Standard Classification of Occupations,
Leuven, Belgium: Leuven University Press.


Cattan

Uploaded by

Copyright:

Available Formats

Cattan

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cattan

Uploaded by

Copyright:

Available Formats

The Economic Journal, 133 (February), 888–903 https://doi.org/10.1093/ej/ueac078 C The Author(s) 2022.

Published by Oxford University Press on behalf of

THE LONG-TERM EFFECTS O F STUDENT ABSENCE:

Sarah Cattan, Daniel A. Kamhöfer, Martin Karlsson and Therese Nilsson

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

1. Background and Data

1.2. Data Sources

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

1.2.1. Church records

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

1.2.5. Pension income

1.2.6. Death records

1.3. Descriptive Statistics

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

2.2. Identification of Long-Term Effects

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

3.2. Long-Term Effects of School Absence

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

3.3. Robustness Checks

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

Table 4. Robustness of the Effect of Absence on Long-Term Outcomes.

Passed away before 1960 (1 = yes) 0.00001

The Author(s) 2023.

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

Downloaded from https://academic.oup.com/ej/article/133/650/888/6825447 by guest on 03 July 2024

You might also like