Social Psychiatry and Psychiatric Epidemiology (2020) 56:857–866

https://doi.org/10.1007/s00127-020-01934-5

ORIGINAL PAPER

Mental health and academic performance: a study on selection

and causation effects from childhood to early adulthood
Sara Agnafors1 · Mimmi Barmark2 · Gunilla Sydsjö1

Received: 7 February 2020 / Accepted: 7 August 2020 / Published online: 19 August 2020
© The Author(s) 2020, corrected publication 2023

Abstract
Purpose An inverse relationship between mental health problems and academic achievement is a well-known phenomenon
in the scientific literature. However, how and when this association develops is not fully understood and there is a lack of
longitudinal, population-based studies on young children. Early intervention is important if associations are to be found
already during childhood. The aim of the present study was to investigate the development of the association between mental
health and academic performance during different developmental periods of childhood and adolescence.
Methods Data from a longitudinal birth cohort study of 1700 children were used. Child mental health was assessed through
mother’s reports at age 3, and self-reports at age 12 and 20. Academic performance was assessed through teacher reports on
educational results at age 12 and final grades from compulsory school (age 15–16) and upper secondary school (age 18–19).
The association between mental health and academic performance was assessed through regression models.
Results The results indicate that social selection mechanisms are present in all three periods studied. Behavioral and emo-
tional problems at age 3 were associated with performing below grade at age 12. Similarly, mental health problems at age 12
were associated with lack of complete final grades from compulsory school and non-eligibility to higher education. Academic
performance at ages 15 and 19 did not increase the risk for mental health problems at age 20.
Conclusion Mental health problems in early childhood and adolescence increase the risk for poor academic performance,
indicating the need for awareness and treatment to provide fair opportunities to education.

Keywords Children · Education · Mental health · Socio-economic status · Social selection · Social causation

Introduction educational attainment (social selection) [1, 2, 5]. Naturally,

this knowledge comes mostly from research on the adult
An inverse relationship between mental health and edu- population as the level of SES and educational attainment
cational attainment is a well-known phenomenon in the are established over the years. However, there are several
scientific literature of sociology, epidemiology and social reasons to investigate the association between mental health
psychiatry [1–3]. Despite nearly a century of research into and academic performance already during childhood. Social
the matter, no consensus has been reached about how the causation processes during childhood are dependent on the
association develops and persists. Several studies have level of SES and educational tradition of the family of ori-
demonstrated the effect of educational attainment on men- gin. Equally important, schooling is a central part of all of
tal health (social causation) [1, 3, 4] and likewise, there childhood, with the educational path starting at an early age.
is support for the influence of mental health problems on If an association between mental health and academic per-
formance can be found already during childhood and ado-
lescence, early recognition and interventions are warranted.
* Sara Agnafors
sara.agnafors@liu.se However, there are few longitudinal, population-based stud-
ies on children and adolescents. As academic performance
1
Division of Children’s and Women’s Health, Department is associated with future educational attainment [6], and
of Biomedical and Clinical Sciences, Linköping University, mental health problems during childhood increases the risk
581 85 Linköping, Sweden
for subsequent mental health problems [7], early interven-
2
Department of Sociology, Lund University, 221 00 Lund, tion is highly valuable. Given the multifactorial etiology of
Sweden

13
Vol.:(0123456789)
858 Social Psychiatry and Psychiatric Epidemiology (2020) 56:857–866

mental illness, and the number of factors predicting educa- health problems and assessments of whether mental health
tional attainment, there is reason to assume that the relation problems as early as pre-school age predict educational out-
between the two is complex [8]. comes have not yet been done.
Studies on the impact of educational achievement and The present study uses a sample from a longitudinal
academic performance on mental health in the younger cohort study to investigate whether mental health predicts
population mainly include adolescents. In a meta-analysis academic performance and, vice versa, whether academic
of 17 original works, early school drop-out was found to performance predicts mental health during different devel-
be associated with substance abuse, depression and exter- opmental periods in childhood and adolescence. The study
nalizing problems [4]. Moreover, academic performance in offers the opportunity to control for a variety of variables
adolescence has been associated with suicide in men but potentially impacting mental health and academic perfor-
not in women [9]. In a recent Swedish study, a low Grade mance, including maternal mental health and parental edu-
Point Average (GPA) at age 16 was found to be associ- cation level. Given the evidence presented above, it is likely
ated with depression in early adulthood, and this associa- that there are mechanisms in both directions but that these
tion was attenuated by externalizing comorbidity [10]. In vary between populations, national/cultural contexts, age
a large population-based cohort, Jonsson et al. [11] found groups, types of mental illness, different aspects of social
that a low GPA was associated with hospitalization due to class/educational achievement, etc. Further studies are,
depression in adolescence. However, academic performance therefore, warranted.
was assessed at age 16 and depression between ages 12 and
17 making a definite conclusion about the direction of the Aim
association precarious. With regard to younger children,
Deighton et al. [12] found support for the effect of poor The aim of the present study was to investigate the devel-
academic performance on subsequent internalizing problems opment of the association between mental health and aca-
in middle childhood. In sum, a couple of studies investigated demic performance during different developmental periods
the effect of academic performance on mental health in ado- in childhood and adolescence. The following five hypotheses
lescence and early adulthood, although most studies focused are used to test the associations empirically:
mainly on depression.
Several population-based studies investigated the effect 1. Internalizing and/or externalizing problems at age 3
of mental health on academic performance in children and increase the risk for poor academic performance at age
adolescents. In a longitudinal study, Fletcher showed that 12.
adolescent depression was linked to years of schooling, con- 2. Internalizing and/or externalizing problems at age 12
trolling for psychiatric comorbidity and sociodemographic increase the risk for incomplete final grades from com-
factors [5]. In a study on 800 children followed from age 6 pulsory school (age 15–16).
to 18, externalizing but not internalizing problems predicted 3. Internalizing and/or externalizing problems at age 12
poor academic performance [13]. Breslau and colleagues increase the risk for non-eligibility to higher education
found that attention problems at age 6 predicted math and (age 18–19).
reading achievement at age 17, while no effect was seen for 4. Incomplete final grades from compulsory school (age
externalizing and internalizing problems [14]. In a cohort 15–16) increase the risk for internalizing and/or exter-
study of 400 children, McLeod and Kaiser found that inter- nalizing problems at age 20.
nalizing and externalizing problems at age 6–8 strongly 5. Non-eligibility to higher education increases the risk for
diminished the chance of accomplishing a high-school internalizing and/or externalizing problems at age 20.
degree [15]. Deighton et al. [12] investigated the associa-
tion between internalizing and externalizing problems and
academic performance during middle childhood and early Methods
adolescence, confirming the effect of externalizing problems
on later academic achievement. In a longitudinal popula- Subjects and procedures
tion-based study, Miech et al. [8] found that externalizing
problems had direct negative effects on adolescent school Data from a longitudinal birth cohort study, the SESBiC
performance, while neither causation nor selection processes study, were used [16]. All women who gave birth to chil-
applied for depression and SES. The authors conclude that dren during 20 consecutive months 1995–1996 in five geo-
disorder-specific models are required. In sum, previous graphically adjacent municipalities in southern Sweden were
research shows that mental health problems predict aca- asked to take part in the overarching study. Of those, 88%
demic performance from middle childhood and up to adult- (n = 1723) accepted participation. For an overview of the
hood. Results seem, however, to vary by the type of mental study population and the waves of data collection, see Fig. 1.

13
Social Psychiatry and Psychiatric Epidemiology (2020) 56:857–866 859

n=1452 n=1183 n=695 a problem”. The CBCL has been used previously in Scan-
n=1723
(84 %) (69 %) (42 %)
dinavian population-based studies and has shown good pre-
AP AP AP cision when screening for child psychiatric disorders [20].
MH MH MH The mothers filled out the Swedish version of the CBCL 2/3
at the 3-year follow-up. In logistic regression, the 90th per-
centile was used as cut-off. The variable was, thus, dichoto-
3 months 3 years 12 years 16 years 19 years 20 years
mized, and children with a score within the 90th percentile
Note: MH: mental health, AP: academic performance, = social selection, were compared to children with a lower score (reference).
= social causation The Strengths and Difficulties Questionnaire (SDQ) is
a screening instrument [21] consisting of 25 items divided
Fig. 1  Model of the study outline and participants between four problem subscales (emotional-, conduct-,
hyperactivity- and peer problems) and one strength subscale
(prosocial behavior). The self-report version was filled out
Among participating children, 52.8% were boys, and there by the children at the 12-year follow-up. In logistic regres-
were 27 pairs of twins. The baseline study and the 3-year fol- sion, the 90th percentile was used as cut-off. The variable
low-up were conducted at Child Welfare Centers (CWC’s), was, thus, dichotomized, and children with a score within
in connection with the routine age-based examination. At the 90th percentile were compared to children with a lower
baseline, the mothers were interviewed by a psychologist, score (reference).
and at the 3-year follow-up, they filled out questionnaires The Symptom Checklist (SCL-25) was completed by the
on mental health and well-being for themselves as well as mothers at the 12-year follow-up [22]. The form consists
for their children. of 25 items scoring on a scale of 1–4, from “not at all” to
The 12-year follow-up, was carried out at school as most “extremely” and is designed to measure anxiety and depres-
of the children still lived in the catchment area. Research sion during the most recent 14 days. When used in logistic
assistants supervised the children as they filled out ques- regression, a cut-off was set to mean value 1.75 which has
tionnaires on mental health and well-being and helped with been used previously [23].
questions. A package of questionnaires including standard- The teacher’s report form (TRF) [24] is a screening
ized instruments regarding mental health and well-being was instrument for child behavior problems including informa-
sent to parents by ground mail. Teachers rated academic tion on academic performance. The TRF 5–18 was answered
performance for reading, mathematics and English language by the teachers at the 12-year follow-up and in this study,
respectively at the 12-year follow-up. At the 20-year follow- only information on child performance in reading, math-
up, the now young adults answered standardized instruments ematics and English language was used. English is the sec-
on mental health and behavior. Parents had to give written ondary language for the study participants, taught from age
consent for the child to be enrolled in the baseline, 3-year 10 at the latest. The teachers rated child performance on a
and 12-year follow-ups. Written consent was obtained from 5-point scale as “far below grade”, “somewhat below grade”,
the young adults themselves at the 20-year follow-up. “at grade level”, “somewhat above grade” and “far above
grade”. In models using logistic regression, the variable was
Instruments dichotomized and “far below grade” and “somewhat below
grade” was compared to “at grade level”, “somewhat above
The Edinburgh Postnatal Depression Scale (EPDS) [17] is grade” and “far above grade” (reference).
a self-report scale designed to screen for postnatal depres- The Adult Self Report [25] is a 126-item form assessing
sion in community samples. It holds 10 items ranged 0–3, mental health divided into the two main domains of inter-
with a total score of 30, a high score indicating postnatal nalizing and externalizing problems. Each item is rated on
depression. The EPDS refers to symptoms of depression and a 3-point scale from “not true” to “somewhat or sometimes
anxiety perceived during the recent week and was filled out true” and “very true or often true”. The ASR has shown
by the mothers at baseline. A cut-off of 10 was set for the good validity and has been used previously in population-
EPDS, which has been done previously for screening pur- based studies [26]. The ASR was answered by the young
poses [18]. In the analysis, the variable was, thus, dichoto- adults at the 20-year follow-up.
mized and women with a score of ≥ 10 were compared to
women with a score of 9 or lower (reference). Register data
The Child Behavior Checklist/2–3 (CBCL) [19] is a well-
known form assessing child behavior into two main domains The Swedish school system is based on 10 years of compul-
of internalizing and externalizing problems. The form holds sory education, followed by 3-year optional upper secondary
100 items, each scored 0–2 from “not a problem” to “often education. Individuals without complete grades from upper

13
860 Social Psychiatry and Psychiatric Epidemiology (2020) 56:857–866

secondary education may attend municipal adult education, Table 1  Frequency characteristics of the study population
with the possibility to achieve the eligibility requirements Variable N (%) Median/range
for university studies. In 2019, 84.3% of students received
complete final grades from compulsory school [27], and Baseline
approximately 80% completed upper secondary education. Gender 1723
Final grades from compulsory and upper secondary school Girl 814 (47.2%)
were obtained from the national statistics office. Lack of Parental immigration status 1704
complete grades (and thereby inability to continue upper One parent born abroad 243 (14.3%)
secondary school) was compared to complete final grades Maternal school drop-out at baseline 1626
from compulsory school (reference). Non-eligibility to uni- Yes 157 (9.7%)
versity studies/higher education was compared to eligibility Maternal symptoms of PPD 1679 4/0–23
to higher education (reference). EPDS ≥ 10 204 (12.0%)
3-year follow-up
Socio‑demographic factors Behavioral problems 1428 7/0–34
CBCL externalizing ≥ 90th percen- 134 (9.4%)
tile
Parental immigration status was noted at the baseline study,
Emotional problems 1428 3/0–32
and children of one or both parents born abroad were com-
CBCL internalizing ≥ 90th percen- 138 (9.5%)
pared to children of parents born in Sweden (reference). tile
Information on school drop-out of the mothers was obtained 12-year follow-up
at the baseline survey. Mothers who did not complete com- Mathematics performance 964
pulsory school or upper secondary school were compared Below grade 166 (17.2%)
to mothers who did (reference). Information on parental Reading performance 983
education was obtained at the 12-year follow-up. Mothers Below grade 181 (18.4%)
and fathers reported their highest level of education, and English language performance 960
compulsory/upper secondary school (≤ 12 years of school- Below grade 200 (20.8%)
ing) was compared to post-secondary education (> 12 years Behavioral problems 1183 1/0–9
of schooling, reference level), based on the parent with the SDQ conduct ≥ 90th percentile 116 (9.8%)
highest level of education. Emotional problems 1183 2/0–9
SDQ emotion ≥ 90th percentile 89 (7.5%)
Data analysis Maternal symptoms of depression 885 33/25–94
SCL25 ≥ M 1.75 161 (18.2%)
For frequencies of included variables, that is, mental health Parental education level 919
parameters (CBCL, SDQ, ASR), academic performance > 12 years of schooling 402 (43.7%)
(performance in reading, mathematics and English language Age 16
at age 12, final grades from compulsory school and upper Compulsory school grades 1668
secondary school) and control variables (maternal school Incomplete 183 (11%)
drop-out, maternal mental health, parental education level, Age 19
gender and parental immigrations status), see Table 1. Miss- Eligibility to higher education 1314
ing data ranged from 0% (gender) to 58.4% (ASR exter- Non-eligibility 297 (22.6%)
nalizing). To test the hypotheses of an association between 20-year follow-up
mental health and education, bivariate linear regression or Behavioral problems 693 9/0–55
logistic regression was performed in five separate models. ASR externalizing ≥ 90th percentile 65 (9.4%)
Linear regression was used for continuous dependent vari- Emotional problems 695 11/0–57
ables, and logistic regression was used for binary outcome ASR internalizing ≥ 90th percentile 68 (9.8%)
variables. Then, for each model, multivariate linear or logis-
tic regression was carried out controlling for maternal school PPD Postpartum depression, EPDS Edinburgh Postnatal Depression
Scale, CBCL Child Behaviour Checklist, SDQ Strengths and Diffi-
drop-out, maternal mental health, parental education level,
culties Questionnaire, SCL25 Symptom Checklist 25 ASR Adult Self
gender and parental immigration status, and when possible Report
also for mental health at a previous data collection point.
Stepwise regression was performed, excluding the control
variable with the highest p value until all remaining control tables the beta coefficients (B) and the 95% confidence inter-
variables showed statistical significance. Gender was, how- vals (CI) are presented, and in the logistic regression tables
ever, included in all final models. In the linear regression the odds ratios (OR) and their respective CI are presented.

13
Social Psychiatry and Psychiatric Epidemiology (2020) 56:857–866 861

Statistical significance was defined as (two-sided) p ≤ 0.05. Model 3, social selection

All statistical analyses were performed using IBM SPSS
version 24. For an overview of the study outline, see Fig. 1. In bivariate analysis, both conduct and emotional problems
at age 12 were associated with non-eligibility to higher
Drop‑out rate analysis education (Table 2).
In multivariate analysis, conduct problems at age 12
At the 3-year follow-up, the retention rate was 84.2% increased the risk for non-eligibility to higher education
(n = 1452). At the 12-year follow-up, the corresponding after controlling for gender and parental education level
number was 68.7% (n = 1183). At the 20-year follow-up, at the child’s age 12 (Table 2). The same applied for emo-
two individuals had died, 10 had moved out of the country, tional problems.
25 had incorrect or missing postal addresses, 19 individuals
declined participation due to learning disabilities or autism
resulting in 1667 eligible participants out of whom 41.7% Model 4, social causation
(n = 695) accepted participation. For detailed information
on the drop-out rate analysis, see Supplementary Material. In bivariate analysis, incomplete grades from compulsory
school were associated with externalizing problems at age
20, while no association was found for internalizing prob-
Results lems (Table 3).
In multivariate analysis, no association was found
Model 1, social selection between incomplete grades from compulsory school and
externalizing problems at age 20 after controlling for gen-
In bivariate analysis, externalizing problems at age 3 der, maternal symptoms of postpartum depression, con-
increased the risk for performing below grade in English duct problems at age 12 and emotional problems at age
and mathematics, but not reading, at age 12 (Table 2). Inter- 12 (Table 3). However, when controlling only for gen-
nalizing problems increased the risk for low performance in der and maternal symptoms of postpartum depression,
reading, English and mathematics (Table 2). incomplete grades from compulsory school increased the
In multivariate analysis, externalizing problems at age risk for externalizing problems at age 20 (OR 2.77, CI
3 were still associated with English language performance 0.32–5.22). No association was found between incomplete
at age 12, after controlling for parental education level and grades from compulsory school and internalizing problems
gender (Table 2). When including concurrent conduct prob- at age 20 after controlling for gender, maternal symptoms
lems in the model, the association remained. No associations of postpartum depression and emotional problems at age
were seen for externalizing problems at age 3 on mathemat- 12 (Table 3).
ics and reading performance at age 12 (Table 2). Internal-
izing problems at age 3 were shown to predict performing
below grade in English language and mathematics after con- Model 5, social causation
trolling for parental education level and gender (Table 2).
The associations remained even when including concurrent Non-eligibility to higher education was associated with
emotional problems in the models. externalizing problems at age 20 in bivariate analysis
(Table 3). No association was found for internalizing
Model 2, social selection problems.
In multivariate analysis, no association was found
In bivariate analysis, conduct problems at age 12 was asso- between non-eligibility to higher education and external-
ciated with lack of final grades from compulsory school izing problems at age 20, controlling for gender, maternal
(Table 2). No association between emotional problems at symptoms of postpartum depression, conduct problems at
age 12 and incomplete grades from compulsory school was age 12 and emotional problems at age 12 (Table 3). Like-
found (Table 2). wise, no association was found between non-eligibility to
In multivariate analysis, conduct problems at age 12 higher education and internalizing problems at age 20 con-
predicted incomplete grades from compulsory school after trolling for above mentioned factors (Table 3).
controlling for gender and parental education level at the To further investigate the support for the social causa-
child’s age 12 (Table 2). Likewise, an association was found tion theory using data from the SESBiC study, whether aca-
between internalizing problems at age 12 and incomplete demic performance at age 12 predicted internalizing and
grades from compulsory school after controlling for the externalizing problems at age 20 was investigated; however,
above-mentioned variables (Table 2). no associations were found. Likewise, parental education

13
862 Social Psychiatry and Psychiatric Epidemiology (2020) 56:857–866

Table 2  Social selection OR (95.0% CI) p-value aOR (95.0% CI) p-value N

Model 1
Reading 12 years below grade
Internalizing 3 years ≥ 90th percentile 1.85 (1.09–3.13) 0.022 1.81 (0.96–3.43) 0.069 706
Gender 0.47 (0.31–0.71) < 0.001
Parental education level 12 years 0.44 (0.28–0.68) < 0.001
Externalizing 3 years ≥ 90th percentile 1.57 (0.91–2.68) 0.104 1.20 (0.62–2.34) 0.586 706
Gender 0.47 (0.31–0.72) < 0.001
Parental education level 12 years 0.44 (0.28–0.69) < 0.001
English language 12 years below grade
Internalizing 3 years ≥ 90th percentile 1.88 (1.11–3.17) 0.018 2.15 (1.17–3.95) 0.013 688
Gender 0.93 (0.63–1.38) 0.721
Parental education level 12 years 0.38 (0.25–0.58) < 0.001
Externalizing 3 years ≥ 90th percentile 2.37 (1.42–3.94) 0.001 2.65 (1.47–4.79) < 0.001 688
Gender 0.96 (0.65–1.42) 0.837
Parental education level 12 years 0.40 (0.26–0.61) < 0.001
Mathematics 12 years below grade
Internalizing 3 years ≥ 90th percentile 1.89 (1.11–3.23) 0.020 1.92 (1.04–3.56) 0.038 695
Gender 0.98 (0.65–1.47) 0.926
Parental education level 12 years 0.49 (0.32–0.76) 0.001
Externalizing 3 years ≥ 90th percentile 1.86 (1.09–3.17) 0.023 1.43 (0.75–2.72) 0.273 695
Gender 1.00 (0.66–1.49) 0.979
Parental education level 12 years 0.50 (0.32–0.77) 0.001
Model 2
Incomplete grades compulsory school
SDQ emotion 12 years≥ 90th percentile 1.51 (0.77–2.93) 0.229 2.35 (1.07–5.12) 0.032 901
Gender 0.82 (0.48–1.42) 0.477
Parental education level 12 years 0.34 (0.18–0.63) 0.001
SDQ conduct 12 years≥ 90th percentile 2.42 (1.42–4.11) 0.001 2.60 (1.30–5.22) 0.007 901
Gender 1.00 (0.58–1.73) 0.990
Parental education level 12 years 0.34 (0.18–0.64) 0.001
Model 3
Non-eligibility higher education
SDQ emotion 12 years≥ 90th percentile 1.81 (1.06–3.09) 0.031 1.95 (1.06–3.59) 0.031 760
Gender 0.51 (0.36–0.74) < 0.001
Parental education level 12 years 0.36 (0.25–0.52) < 0.001
SDQ conduct 12 years≥ 90th percentile 2.53 (1.57–4.07) < 0.001 2.24 (1.29–3.89) 0.004 760
Gender 0.57 (0.40–0.82) 0.002
Parental education level 12 years 0.35 (0.24–0.52) < 0.001

Odds ratios in predicting academic performance

Bivariate and multivariate logistic regression. 0 is used as a reference level. N presented for adjusted mod-
els. Model 1: Dependent variables: academic performance in reading, English language and mathematics
(0 = “at grade level”, “somewhat above grade “and “far above grade”, 1 = far below grade”, “somewhat
below grade”). Independent variables: CBCL internalizing and externalizing (0 < 90th percentile, 1 ≥ 90th
percentile), gender (0 = boys, 1 = girls), parental education level (0 ≤ 12 years of schooling, 1 > 12 years
of schooling). Model 2: Dependent variables: final grades compulsory school (0 = complete grades,
1 = incomplete grades). Independent variables: SDQ emotion and conduct (0 < 90th percentile, 1 ≥ 90th
percentile) gender (0 = boys, 1 = girls), parental education level (0 ≤ 12 years of schooling, 1 > 12 years of
schooling). Model 3: Dependent variables: non-eligibility to higher education (0 = eligibility, 1 = non-eligi-
bility). Independent variables: SDQ emotion and conduct (0 < 90th percentile, 1 ≥ 90th percentile) gender
(0 = boys, 1 = girls), parental education level (0 ≤ 12 years of schooling, 1 > 12 years of schooling)
CI confidence interval, OR odds ratio, CBCL Child Behaviour Checklist, SDQ Strengths and Difficulties
Questionnaire

13
Social Psychiatry and Psychiatric Epidemiology (2020) 56:857–866 863

Table 3  Social causation B (95.0% CI) p-value Adjusted B (95.0% CI) p-value N

Model 4
ASR internalizing 20 years
Compulsory school grades 3.24 (− 0.16–6.65) 0.062 2.56 (− 0.54–6.67) 0.220 560
Gender 5.38 (3.62–7.14) < 0.001
EPDS (maternal, baseline) 0.38 (0.14–0.61) 0.002
SDQ emotion 12 years 1.27 (0.84–1.72) < 0.001
ASR externalizing 20 years
Compulsory school grades 3.24 (0.83–5.65) 0.009 0.81 (− 2.21–3.83) 0.600 551
Gender 1.05 (− 0.16–2.25) 0.088
EPDS (maternal, baseline) 0.21 (0.05–0.37) 0.009
SDQ conduct 12 years 0.84 (0.39–1.30) < 0.001
SDQ emotion 12 years 0.39 (0.07–0.71) 0.015
Model 5
ASR internalizing 20 years
Eligibility higher education 0.04 (− 2.12–2.20) 0.970 − 0.44 (− 2.68–1.79) 0.698 503
Gender 5.61 (3.83–7.39) < 0.001
EPDS (maternal, baseline) 0.32 (0.08–0.56) 0.009
SDQ emotion 12 years 1.24 (0.80–1.69) < 0.001
ASR externalizing 20 years
Eligibility higher education 1.51 (0.08–2.94) 0.038 0.55 (− 0.94–2.05) 0.467 498
Gender 0.80 (− 0.43–2.02) 0.201
EPDS (maternal, baseline) 0.22 (0.06–0.38) 0.008
SDQ conduct 12 years 0.71 (0.24–1.19) 0.003
SDQ emotion 12 years 0.45 (0.12–0.78) 0.007

Bivariate and multivariate linear regression predicting mental health

Bivariate and multivariate linear regression. N presented for adjusted models. Model 4: Dependent vari-
ables: ASR internalizing and externalizing scores. Independent variables: final grades compulsory school
(0 = complete grades, 1 = incomplete grades), gender (0 = boys, 1 = girls), EPDS total score, SDQ emo-
tion and conduct subscales. Model 5: Dependent variables: ASR internalizing and externalizing scores.
Independent variables: Eligibility to higher education (0 = eligibility, 1 = non-eligibility), gender (0 = boys,
1 = girls), EPDS total score, SDQ emotion and conduct subscales
CI confidence interval, OR odds ratio, EPDS Edinburgh Postnatal Depression Scale, SDQ Strengths and
Difficulties Questionnaire, ASR Adult Self Report

level at the child’s age 12 did not predict internalizing or risk for incomplete final grades from compulsory school
externalizing problems at age 20. (data not shown). These findings are in line with previous
studies investigating the association between externalizing
problems during middle childhood and adolescence and
Discussion subsequent academic achievement [8, 14, 15]. Externalizing
problems include Conduct Disorder and Oppositional Defi-
The aim of the present study was to investigate whether ant Disorder which implicates behavioral disturbances that
mental health predicts academic performance and, vice could affect the adjustability in the classroom, and thereby
versa, whether academic performance predicts mental health the performance. Early behavioral problems might also trig-
during different developmental periods in childhood and ger child–teacher conflicts and social exclusion leading to
adolescence. The main findings are discussed below. negative experiences of the school environment.
Externalizing problems at age 3 predicted academic per- In the present study, selection effects were also found
formance (English language) at age 12, after controlling for a for internalizing problems on academic achievement. Men-
number of relevant factors. Conduct problems at age 12 were tal health at pre-school age was associated with academic
also found to increase the risk for incomplete grades from performance (English and mathematics) at age 12. Simi-
compulsory school and non-eligibility for higher education. larly, internalizing problems at age 12 increased the risk
Additionally, externalizing problems at age 3 increased the for incomplete grades at ages 15 and 19. Previous research

13
864 Social Psychiatry and Psychiatric Epidemiology (2020) 56:857–866

shows, however, mixed results. A recent study found strong which in turn has a strong effect on mental health in adult-
selection effects for internalizing problems such as anxiety hood. It is, therefore, still, from a mental health perspective,
and depression in an adult twin-population [2]. Internalizing of utterly importance that effort is being made to support
problems in early school age has been shown to diminish the children who perform “below grade level” to prevent health
chances of completing a high-school degree [15]. In con- risks associated with low SES later in life.
trast, other studies found no effects of internalizing problems Girls had a fivefold increased risk for internalizing prob-
on academic performance from childhood to adolescence lems at age 20 compared to boys; however, no gender differ-
[13], or from adolescence to early adulthood [8]. With the ences were seen for externalizing problems. Previous results
background of previous conflicting results, the present study support gender differences in mental health with a pattern
adds to the literature confirming social selection processes of higher frequency of internalizing problems in girls and
for the association between internalizing problems and aca- higher prevalence of externalizing problems in boys [30]. In
demic performance. the present study, girls participated in the 20-year follow-up
Interestingly, boys were more likely to perform below to a greater extent than boys (51.4% compared to 30.5%,
grade in reading at age 12 and to be non-eligible for higher p < 0.001), possibly impacting the results.
education compared to girls. No gender differences were
noted for lack of compulsory school grades. In fact, the
impact of gender on eligibility to higher education in the Limitations
present study was considerable; 61% of non-eligible individ-
uals were boys. Gender differences in academic performance While the present study is strengthened by the large num-
with an advantage for girls are a well-known phenomenon ber of participants and the longitudinal design, a few
[28]. limitations need to be considered when interpreting the
Incomplete grades from compulsory school (age 15–16) results. First, using the social selection and social causa-
were not associated with mental health at age 20. When tion approach in a child population means that possible
controlling only for gender and maternal symptoms of post- associations need to be interpreted carefully. While the
partum depression, incomplete grades from compulsory social causation hypothesis generally implies that lower
school predicted externalizing problems at age 20. When education level brings lower income, unhealthier life style
controlling for mental health problems at age 12, however, and more life stressors, this does not readily apply dur-
the association diminishes, indicating that the association ing childhood and adolescence. Rather, children are under
is a result of social selection rather than social causation influence of the environmental circumstances impacting
mechanisms. Regarding the effect of academic performance their parents and thus, a number of possibly confound-
on internalizing problems, no associations were found nei- ing factors need to be considered. We have included sev-
ther in bivariate nor in multivariate analysis. eral such factors in the models but cannot be certain that
These findings stand in contrast to the previous findings some important variable has not been left out. Second,
on the adult population including a meta-analysis, where a long period of time elapsed between the measures of
the risk of reporting depression was almost doubled in low mental health problems and academic performance, pos-
SES groups [29]. Similarly, Miech et al. found effects of sibly diluting the effects. Narrower timespans between the
adolescent school performance on anxiety conditions [8], follow-ups, or additional follow-ups would have been of
while Sörberg Wallin et al. demonstrated an increased risk great value. Moreover, while the influence of mental health
for depressive symptoms in early adulthood following poor on academic performance was modelled with three sepa-
academic performance in adolescence [10]. The social cau- rate outcomes of academic performance during adoles-
sation models in the present study were limited by a large cence, the influence of academic performance on mental
drop-out rate, and a higher retention rate might have ren- health was based on one single measurement of mental
dered other results. Also, at the age of 20, many individu- health at age 20. The effect of academic performance on
als have not established a stable level of SES; university mental health might have been stronger if measured for
students not yet even graduated. For a child, poor academic example within the year after receiving incomplete grades
performance does not have the same direct consequences from compulsory school. Third, there was a considerable
as low educational attainment might in adulthood (possibly drop-out (58.3%) at the 20-year follow-up. In general, dif-
lower income, lesser ability to compete on the labor market). ferences between participants and non-participants were
Moreover, the Swedish society offers additional possibilities noted on educational variables rather than mental health
to complete upper secondary education, which could influ- parameters. Moreover, fewer 20-year olds whose parents
ence the effect of poor academic performance on mental were born abroad and fewer men than women chose to
health during adolescence. Academic performance in child- take part in the 20-year follow-up. Since immigration sta-
hood is, however, related to later educational attainment [6], tus often is associated with lower SES [31], the skewed

13
Social Psychiatry and Psychiatric Epidemiology (2020) 56:857–866 865

drop-out could have an impact on the results. However, Compliance with ethical standards
parental immigration status was not found to be strongly
associated with the outcomes measured in the present Conflicts of interest The authors declare that they have no conflicts
of interest.
study. The skewness regarding men and women might have
contributed to a lower degree of externalizing problems, Ethical considerations The study was approved by the Ethics com-
as this is more common in men than in women. While mittee at the University of Lund in 1994 and 1998, and by the Ethical
no difference between participants and non-participants review board in Linköping in 2007 and 2015.
regarding mental health at age 3 or 12 was noted, we can- Open Access This article is licensed under a Creative Commons Attri-
not completely rule out the possibility that individuals bution 4.0 International License, which permits use, sharing, adapta-
who developed mental health problems after 12 years of tion, distribution and reproduction in any medium or format, as long
age were less likely to participate in the follow-up. If so, as you give appropriate credit to the original author(s) and the source,
provide a link to the Creative Commons licence, and indicate if changes
that would diminish the chance of detecting an association were made. The images or other third party material in this article are
between academic performance in adolescence and mental included in the article's Creative Commons licence, unless indicated
health problems in early adulthood in line with the social otherwise in a credit line to the material. If material is not included in
causation hypothesis. the article's Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will
need to obtain permission directly from the copyright holder. To view a
copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/.

Conclusions
References
The study adds to the existing literature by the use of a
large, two-generational cohort, and longitudinal prospec- 1. Dohrenwend BP, Levav I, Shrout PE, Schwartz S, Naveh G,
tive design with multiple data collection points, investigat- Link BG, Skodol AE, Stueve A (1992) Socioeconomic status
and psychiatric disorders: the causation-selection issue. Science
ing the association between mental health and academic 255(5047):946–952
performance during different developmental periods from 2. Halpern-Manners A, Schnabel L, Hernandez EM, Silberg JL,
age 12 to 20. The results emphasize the necessity to detect Eaves LJ (2016) The relationship between education and mental
externalizing and internalizing problems at a young age health: new evidence from a discordant twin study. Soc Forces
95(1):107–131
and continuously throughout the school years. In practice, 3. Ritsher JE, Warner V, Johnson JG, Dohrenwend BP (2001) Inter-
it means that these kinds of problems need to be noticed generational longitudinal study of social class and depression: a
at preschool age and educational practices adjusted and test of social causation and social selection models. Br J Psychia-
adequate treatment given to promote transition to the try Suppl 40:84–90
4. Esch P, Bocquet V, Pull C, Couffignal S, Lehnert T, Graas M,
school environment and completion of compulsory school. Fond-Harmant L, Ansseau M (2014) The downward spiral of
Knowledge and recognition of the potential effects of mental disorders and educational attainment: a systematic review
internalizing problems on academic performance might on early school leaving. BMC Psychiatry 27(14):237
5. Fletcher JM (2010) Adolescent depression and educational attain-
be especially important as these problems tend to be less
ment: results using sibling fixed effects. Health Econ 19:855–871
explicit to others compared to externalizing problems. 6. Magnuson K, Duncan G, Lee KTH, Metzger M (2016) Early
No support was found for the association between aca- school adjustment and educational attainment. Am Educ Res J
demic performance during adolescence and mental health 53(4):1198–1228
7. Caspi A, Moffitt TE, Newman DL, Silva PA (1996) Behavioral
status in early adulthood in this Swedish context. However,
observations at age 3 years predict adult psychiatric disorders.
this result needs to be interpreted in the light of a consider- Longitudinal evidence from a birth cohort. Arch Gen Psychiatry
able drop-out rate and a long time span between measures 53:1033–1039
of impact and outcome. 8. Miech RA, Caspi A, Moffitt TE, Wright BE, Silva PA (1999) Low
socioeconomic status and mental disorders: a longitudinal study
Electronic supplementary material The online version of this article of selection and causation during young adulthood. Center for
(https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 00127-0​ 20-0​ 1934-5) contains supplementary Demography and Ecology University of Wisconsin, Wisconsin
material, which is available to authorized users. 9. Gunnell D, Löfving S, Gustafsson JE, Allebeck P (2011) School
performance and risk of suicide in early adulthood: follow-up of
Acknowledgements We would like to thank Professor emerita Mari- two national cohorts of Swedish schoolchildren. J Affect Disord
anne Cederblad, and Professor emeritus Carl Göran Svedin for their 131(1–3):104–112
work in the SESBiC study. 10. Sörberg Wallin A, Koupil I, Gustafsson JE, Zammit S, Allebeck
P, Falkstedt D (2019) Academic performance, externalizing disor-
Funding Funding was provided by Skandia research, Swedish ders and depression: 26,000 adolescents followed into adulthood.
Research Council for Health Working Life and Welfare and Professor Soc Psychiatry Psychiatr Epidemiol 54(8):977–986
Bror Gadelius memorial foundation. Open access funding provided by 11. Jonsson U, Goodman A, von Knorring AL, von Knorring L,
Linköping University. Koupil I (2012) School performance and hospital admission
due to unipolar depression: a three-generational study of social

13
866 Social Psychiatry and Psychiatric Epidemiology (2020) 56:857–866

causation and social selection. Soc Psychiatry Psychiatr Epide- 22. Derogatis LR, Lipman RS, Rickels K, Uhlenhuth EH, Covi L
miol 47(10):1695–1706 (1974) The Hopkins Symptom Checklist (HSCL): a self-report
12. Deighton J, Humphrey N, Belsky J, Boehnke J, Vostanis P, Patalay symptom inventory. Behav Sci 19:1–15
P (2018) Longitudinal pathways between mental health difficulties 23. Nettelbladt P, Hansson L, Stefansson CG, Borgquist L, Nordstrom
and academic performance during middle childhood and early G (1993) Test characteristics of the Hopkins Symptom Check
adolescence. Br J Dev Psychol 36(1):110–126 List-25 (HSCL-25) in Sweden, using the Present State Exami-
13. Van der Ende J, Verhulst FC, Tiemeier H (2016) The bidirectional nation (PSE-9) as a caseness criterion. Soc Psychiatry Psychiatr
pathways between internalizing and externalizing problems and Epidemiol 28:130–133
academic performance from 6 to 18 years. Dev Psychopathol 24. Achenbach TM (1991) Manual for the Teacher’s Report Form and
28(3):855–867 1991 profile. University of Vermont Department of Psychiatry,
14. Breslau J, Miller E, Breslau N, Bohnert K, Lucia V, Schweitzer Burlington
J (2009) The impact of early behavior disturbances on academic 25. Achenbach TM, Rescorla LA (2003) Manual for the ASEBA adult
achievement in high school. Pediatrics 123(6):1472–1476 forms & profiles. Research Center for Children, Youth, & Fami-
15. McLeod JD, Kaiser K (2004) Childhood emotional and behav- lies, University of Vermont, Burlington
ioral problems and educational attainment. Am Sociol Rev 26. van der Ende J, Verhulst FC, Tiemeier H (2012). Agreement of
69(5):636–658 informants on emotional and behavioral problems from childhood
16. Agnafors S, Svedin CG, Oreland L, Bladh M, Comasco E, Sydsjö to adulthood. Psychol Assess 24(2): 293–300.
G (2017) A biopsychosocial approach to risk and resilience on 27. Final grades in compulsory school, spring 2019. [Slutbetyg i
behavior in children followed from birth to age 12. Child Psychia- grundskolan, våren 2019.] Dnr: 5.1.1 -2019:1342. Skolverket 2019
try Hum Dev 48(4):584–596 (In Swedish)
17. Cox JL, Holden JM, Sagovsky R (1987) Detection of postnatal 28. Voyer D, Voyer SD (2014) Gender differences in scholastic
depression. Development of the 10-item Edinburgh Postnatal achievement: a meta-analysis. Psychol Bull 140(4):1174–1204
Depression Scale. Br J Psychiatry 150:782–786 29. Lorant V, Deliege D, Eaton W et al (2003) Socio-economic inequal-
18. Norhayati MN, Hazlina NHN, Asrenee AR, Emilin WMAW ities in depression: a meta-analysis. Am J Epidemiol 157:98–112.
(2015) Magnitude and risk factors for postpartum symptoms: a
literature review. J Affect Disord 175:34–52 30. Merikangas KR, Nakamura EF, Kessler RC (2009) Epidemiology
19. Achenbach TM (1992) Manual for the Child Behavior Check- of mental disorders in children and adolescents. Dialogues Clin
list/2-3 and 1992 profile. University of Vermont, Department of Neurosci 11:7–20
Psychiatry, Burlington 31. Honkaniemi H, Juárez SP, Katikireddi SV, Rostila M (2020) Psy-
20. Bilenberg N, Petersen DJ, Hoerder K, Gillberg C (2005) The prev- chological distress by age at migration and duration of residence
alence of child-psychiatric disorders among 8–9-year-old children in Sweden. Soc Sci Med 20(250):112869. https://​doi.​org/​10.​
in Danish mainstream schools. Acta Psychiatr Scand 111:59–67 1016/j.​socsc​imed.​2020.​112869 (Online ahead of print)
21. Goodman R (1997) The strengths and difficulties questionnaire:
a research note. J Child Psychol Psychiatry 38(5):581–586

13