Dobewall et al.

BMC Public Health (2019) 19:1447

[Link]

RESEARCH ARTICLE Open Access

Health and educational aspirations in

adolescence: a longitudinal study in
Finland
Henrik Dobewall1,2* , Pirjo Lindfors1,2, Sakari Karvonen3, Leena Koivusilta4, Mari-Pauliina Vainikainen5,6,
Risto Hotulainen6 and Arja Rimpelä1,2,7

Abstract
Background: The health selection hypothesis suggests that poor health leads to low educational attainment during
the life course. Adolescence is an important period as poor health might prevent students from making the best
educational choices. We test if health in adolescence is associated with educational aspirations and whether these
associations persist over and above sociodemographic background and academic achievement.
Methods: Using classroom surveys, a cohort of students (n = 5.614) from the Helsinki Metropolitan Region was
followed from the 7th (12–13 years,) up to the 9th grade (15–16 years) when the choice between the academic and
the vocational track is made in Finland. Health factors (Strengths and Difficulties Questionnaire (SDQ), self-rated health,
daily health complaints, and long-term illness and medicine prescribed) and sociodemographic background were self-
reported by the students. Students’ educational aspirations (applying for academic versus vocational track, or both) and
their academic achievement were obtained from the Joint Application Registry held by the Finnish National Agency for
Education. We conducted multilevel multinomial logistic regression analyses, taking into account that students are
clustered within schools.
Results: All studied health factors were associated with adolescents’ educational aspirations. For the SDQ, daily health
complaints, and self-rated health these associations persisted over and above sociodemographic background and
academic achievement. Students with better health in adolescence were more likely to apply for the academic track,
and those who were less healthy were more likely to apply for the vocational track. The health in the group of those
students who had applied for both educational tracks was in between. Inconsistent results were observed for long-
term illness. We also found robust associations between educational aspirations and worsening health from grade 7 to
grade 9.
Conclusions: Our findings show that selection by health factors to different educational trajectories takes place at early
teenage much before adolescents choose their educational track, thus supporting the health selection hypothesis in
the creation of socioeconomic health inequalities. Our findings also show the importance of adolescence in this
process. More studies are needed to reveal which measures would be effective in helping students with poor health to
achieve their full educational potential.
Keywords: Health selection, Social causation, Adolescents, Educational trajectories, School survey, aspirations

* Correspondence: [Link]@[Link]
1
Faculty of Social Sciences (Health Sciences), Tampere University, Po Box 20,
(Arvo Ylpön katu 34), 33014 Tampere, Finland
2
PERLA—Tampere Centre for Childhood, Youth and Family Research,
Tampere University, 33014 Tampere, Finland
Full list of author information is available at the end of the article

© The Author(s). 2019 Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0
International License ([Link] which permits unrestricted use, distribution, and
reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to
the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver
([Link] applies to the data made available in this article, unless otherwise stated.
Dobewall et al. BMC Public Health (2019) 19:1447 Page 2 of 15

Introduction Educational aspirations are the first step in the process

Years of schooling and the level of education are associ- of the formation of one’s educational path. They are de-
ated with virtually all health outcomes: the higher educa- fined as abstract statements and beliefs about students’
tional attainment, the better health [1–4]. Two main future plans such as the level of education one wishes to
mechanisms to explain these relationships have been achieve [18, 19]. They are a strong predictor of future
presented: the social causation hypothesis and the hy- educational trajectories and through that their adult so-
pothesis of health selection that can differ in importance cioeconomic position [18, 20, 21]. Poor health, however,
at different periods of life course [5–7]. In this paper, might distort the development of educational aspirations
our focus is the health selection in adolescence. Adoles- and consequently prevent students from realizing their
cence is a sensitive period from the point of view of fu- full educational potential. Health disadvantage and lower
ture educational plans as well as for the development of levels of education in combination might thus lead to di-
health and risk factors for health [8]. minished economic returns in the form of labor earnings
Prospective cohort studies that investigate the effect of in adulthood [22]. Only a few studies have investigated
health in adolescence on educational attainment are ac- how health in adolescence is related to educational aspi-
cumulating slowly. Some studies support the selection rations. One of the few is a Canadian study which
hypothesis. Studies from Finland and the USA have showed that fewer adolescents with physical disabilities
shown that diverse health factors, e.g. self-rated health, had plans for education after high school [23]. Another
psychosomatic symptoms, and long-term illness in ado- study from Slovakia showed that self-rated health was
lescence predict later educational outcomes [6, 9–13]. not related to educational aspirations among students in
Studies that controlled for unobserved person or family three different school tracks [24]. It is therefore cur-
characteristics have shown that the education-health rently not known which health factors might influence
gradient is largely shaped by health selection in adoles- adolescents’ plans for further education.
cence [6, 11]. Some studies have not found support for Academic achievement is a strong predictor of a stu-
the health selection hypothesis. Depressive symptoms in dent’s educational trajectory, but even in a Nordic welfare
adolescence were not related to life-course trajectories state like Finland, parents’ education and employment
of education and work in a Swedish study [14], and predict their children’s academic achievement and choice
hardly any association was found between timely gradu- of educational tracks [25–27]. In addition, other sociode-
ation from secondary education and health records in a mographic factors such as gender, immigrant background,
Dutch study [15]. A study from New Zealand showed and family structure are known to be associated with edu-
that social problems but not the psychological ones were cational choices [24, 28, 29]. When studying the inde-
associated with later educational attainment [16]. In pendent effect of health on educational aspirations, the
summary, the findings of these prospective studies test- sociodemographic background and the academic achieve-
ing if health in adolescence influences education at a ment of the student need to be controlled for.
later age are mixed. The differences can be based on dif- Health selection in adolescence can be a pathway to
ferent samples, studied health factors, or which educa- future health inequalities. With this study, we want to
tional outcomes have been used. generate knowledge on whether health in adolescence
Also, the educational context differs between countries. patterns educational aspirations and through that educa-
We study here the process of health selection in Finland, a tional trajectories. Based on the above, we hypothesize
Nordic welfare state with a 9-year comprehensive school that health in adolescence is related to educational aspi-
with a national curriculum. In grades 7 to 9 (lower rations so that students with better health are more
secondary school) most subjects have a subject teacher likely to apply for the academic track and those who are
while the lower grades 1 to 6 are taught by a class teacher. less healthy, are more likely to apply for the vocational
Compared to many other countries [17], tracking to track. It is well-known that adolescents’ sociodemo-
different school paths takes place quite late, in the 9th graphic background and particularly academic achieve-
grade (age 16) when compulsory schooling ends. Virtually ment strongly predict educational trajectories. In
all adolescents apply to upper secondary education, and accordance with the health selection hypothesis, we
do that through a national Joint Application System hypothesize, however, that adolescents’ health has an ef-
([Link] fect over and above these predictors. The research ques-
tion-system), following their educational aspirations for tions are: Are health factors associated with adolescents’
schools of the academic track, the vocational track, or both. educational aspirations and do these associations persist
The selection of students is based on their preferences and over and above sociodemographic background and aca-
grade point average – GPA –. This makes Finland an ideal demic achievement? Does health matter already at the
context for studying the relationship between health and beginning of 7th grade (age 12–13 years) when students
educational aspirations in adolescence. start lower secondary education or does health matter
Dobewall et al. BMC Public Health (2019) 19:1447 Page 3 of 15

only at the end of the 9th grade (age 15–16 years) at the no parental consent was required when the study was
time when they apply to upper secondary education? Fi- conducted as part of the students’ normal schoolwork.
nally, we want to find out whether health improvement Two of the 14 municipalities had adopted a policy that a
or worsening from the 7th to 9th grade is associated written parental consent is always required. These were
with adolescents’ plans for education after compulsory collected. In the other municipalities, information letters
schooling. were sent to the parents who had the possibility to with-
draw their child from the study. The students were
Methods instructed about the purpose of the study and that par-
Procedure and setting ticipation was voluntary and that they can decline to an-
The study was conducted as part of the project “Redefin- swer any question or withdraw from the survey at any
ing adolescent learning: A multilevel longitudinal cohort time. This was mentioned at the beginning of the ques-
study of adolescent learning, health, and well-being in tionnaire at the first pageRegistry data on students’ edu-
educational transitions in Finland” – Metropolitan cational aspirations were obtained from the Finnish
Longitudinal Finland (MetLoFin) –. It follows a large co- National Agency for Education, covering the period from
hort of students from the Helsinki Metropolitan Region Spring 2014 to Spring 2017. In Finland, students can
from the lower secondary education to the end of upper apply via the Joint Application System to a maximum of
secondary education. In 2011, all 7th graders (12–13 five study places in upper secondary schools, ranked in
years old) were invited to participate. The recruitment the order in which they wish it to be selected. There are
occurred through the educational authorities of all 14 two general application rounds—Spring and Autumn—
municipalities of the Helsinki Metropolitan Region, each which are followed by an additional application round in
of which gave a permission for the study. A follow-up which students can apply for vacant study places. Com-
survey was fielded in 2014 when the students were in bining the survey answers with the Joint Application
the 9th grade (15–16 years old). Registry was done by a data manager who does not
The study protocol was approved by the Ethical analyze the data himself.
Committee of the Finnish Institute for Health and In total, 13,012 students belong to the baseline sample
Welfare. In line with the instructions of the Finnish of the MetLoFin project (for a flow diagram representing
National Board on Research Integrity (TENK) in 2009, the formation of the study population, see Fig. 1). In

Fig. 1 Flow diagram representing the formation of the study population. The numbers in the final analyses differ due to missing information in
the predictor variables
Dobewall et al. BMC Public Health (2019) 19:1447 Page 4 of 15

total, 9.078 students (50.0% female) answered the health provided on a 4-point fully labeled Likert scale. Students
questionnaire in the 7th grade (response rate of 69.8%). with severe health complaints nearly every day were
Of these students, 5.741 also participated in the 9th classified as “No symptoms,” “One symptom,” and “Two
grade (attrition rate of 36.8%). We excluded from the or more.”
analyzes, those students who never applied via the Joint
Application System (n = 50), who had applied for special
Long-term illness
education at some point (n = 41) [30], or who came from
Long-term illness was assessed with two “Yes/No” ques-
schools where less than five students gave valid answers
tions. The students were asked whether they had a long-
[31]. The analyzed sample consists of the remaining
term illness or disability and whether they regularly used
5.614 students from 116 schools who responded to both
medicine prescribed by a doctor. The answers were cate-
surveys and fulfilled our inclusion criteria. The numbers
gorized into a single variable: “No long-term illness,”
in the final analyses differed due to missing information
“Long-term illness,” and “Medicine prescribed.”
in the predictor variables.

Dependent variable: educational aspirations Self-rated health

The information available in the Joint Application Students’ subjective evaluation of their health was
System was used to construct an objective measure of assessed with a single question [36]. The answers were
students’ educational aspirations. The resulting variable provided on a 5-point Likert scale. The self-rated health
had three categories: students who “Applied for the aca- scale was dichotomized comparing students who an-
demic track” (58.0%, n = 3.258), “Applied for the voca- swered “Good” to those who answered “Average or
tional track” (19.8%, n = 1.111), or were undecided about poor.”
their future plans and “Applied for both educational
tracks” (22.2%, n = 1.244). We treated the recordings of
Missing values and change from grade 7 to grade 9
students’ choices as educational aspirations regardless
To report analyses that are as representative as possible,
some of the students when applying for a study place
we have filled missing values in the health factors using
did not know whether their GPA will be good enough to
the second or previous measurement (21–149 missing
be selected, and some of them did not acquire any place
values were replaced, respectively). To assess the within-
to study. Nevertheless, these were their aspirations.
person change in health from grade 7 to grade 9, we
calculated for each of the health factors a difference
Health factors
score [37]. The resulting variables contrasted students
Strengths and difficulties questionnaire
who remained stable with those whose health improved
The Strengths and Difficulties Questionnaire (SDQ) ver-
or worsened over time (for frequencies, see Additional
sion suitable for adolescents was administered [32, 33].
file 1: Table S1).
It measures emotional symptoms, conduct, hyperactiv-
ity/inattention, and peer relationship problems with five
questions each. The students marked on a 3-point fully Background variables
labeled Likert scale (0 = “Not true” 1 = “Somewhat true,” Sociodemographic background
2 = “Certainly true”) which of the twenty attributes We used students’ gender to account for potential differ-
described them best over the past 6 months. The an- ences between “Girls” and “Boys.” We further used
swers were summed together to generate a total difficul- parental employment (“Both parents working” versus
ties score of psychosocial problems that was categorized “Other”), parental education (“Low” versus “High,” that
to “Normal” (score < 13), “Slightly raised” (14–19), and is at least one parent being highly educated with ma-
“High” difficulty score (20–40). Previous work using the triculation examination or university degree), immigrant
same data as in the current study had found good psy- background (Finnish−/Swedish-speaking “Natives” were
chometric properties for the SDQ [34]. compared to “Immigrants,” who had moved to Finland
and/or had at least one parent who was born abroad),
Daily health complaints and family structure (“Nuclear family” versus “Other”) as
Daily health complaints were assessed with the fre- control variables. Although already 11-year-olds were
quency of ten psychosomatic symptoms (headache, neck found to provide valid and detailed information about
and shoulder pain, lower back pain, stomach aches, ten- their parents’ economic activity and occupation [38], we
sion and nervousness, irritability or outbursts of anger, gave preference to students’ answers to their sociodemo-
trouble falling asleep or waking at night, feeling tired or graphic background provided in the 9th grade. Only in
weak, feeling dizzy, trembling of hands) experienced the case of missing data, answers provided by the stu-
daily over the past 6 months [35]. Answers were dents in the 7th grade were used.
Dobewall et al. BMC Public Health (2019) 19:1447 Page 5 of 15

Academic achievement variables but gender. Participants were more likely to

In the Finnish education system, students both apply to have highly educated and working parents (p < .001) and
the upper secondary education and are accordingly to live in a nuclear family (p < .05), and were less likely
sorted into educational tracks mostly by their grade immigrants (p < .001) than non-participants.
point average – GPA – which results from performance
in different study subjects graded by the subject Results
teachers. Grades from the school leaving certificate The proportions of students in relation to the study vari-
(from 9th grade) are also included in the Joint Applica- ables are presented in Table 1 grouped by the students’
tion Registry. The GPA of each student was calculated educational aspirations.
based on his/her grade in mother tongue, foreign lan- The results of the multilevel multinomial logistic re-
guage, mathematics, and science (averaging grades in gression analyses are presented in Tables 2, 3 and 4. The
biology, geography, physics, and chemistry). Academic differences between schools accounted for 0.64 variance
achievement (GPA) was categorized as “High” (9–10 which translates into an intra-class correlation [40] of
(excellent) points), “Medium” (7.5–8.5 points), and 16.2%.
“Low” (4 (fail) - 7 points).
Health in the 7th grade
Analytical strategy Already in 12–13 year-olds (the 7th grade), all health
Multilevel multinomial logistic regression analyses with factors were associated with students’ educational aspira-
random effects were estimated with generalized struc- tions recorded more than 2 years later (Table 2). Con-
tural equation modeling using Stata Version 15. First, we cerning the SDQ, having a slightly raised or a high
calculated the variance in educational aspirations attrib- difficulty score, compared to having normal levels of
utable to differences between schools which the students psychosocial problems, was associated with a propor-
attended in the 7th grade. Second, we regressed stu- tionally higher likelihood to apply for the vocational ra-
dents’ choices between the educational tracks on their ther than the academic track. Students with psychosocial
health in the 7th grade (12–13 year-olds) and repeated this problems were also more likely to be undecided about
analysis with students’ health in the 9th grade (15–16 their plans for upper secondary education (i.e., applied
year-olds). Third, we controlled for the sociodemographic for both tracks). The associations were robust to ac-
background of the students. Fourth, students’ academic counting for both sociodemographic background and
achievement was entered into the models. Finally, we academic achievement. There was, however, one excep-
looked at within-person changes in health factors over tion: The associations became non-significant for the
time. Students’ health factors in the 7th grade were in- high difficulty score category. Having one psychosomatic
cluded in this analysis to account for starting levels and symptom nearly every day, compared no daily health
potential ceiling effects. The results of this analysis of complaints, was associated with applying for the voca-
within-person change, however, should not be interpreted tional track. Also, students who were undecided in their
as fixed effects estimates because our outcome variable future educational plans were more likely to report one
educational aspirations did not change over time [39]. In psychosomatic symptom. Both associations remained
all models, we controlled for gender differences. The significant after including sociodemographic background
model parameters were presented as odds ratios (OR) variables and academic achievement in the model. Hav-
with 95% confidence intervals (CI). Akaike (AIC) and ing medicine prescribed by the doctor, compared to no
Bayesian (BIC) information criteria were reported for long-term illness, was positively associated with applying
comparing the fit of the models to the data. Interaction for the vocational track. In the models that controlled
effects between gender and health factors were not signifi- for all other predictors, having a long-term illness was
cant (results not shown). no longer significantly associated with educational aspi-
rations. Students who reported average or poor health,
Attrition analyses compared to good health, were more likely to belong to
An independent samples t-test revealed that students the group that had not decided yet and had thus applied
that answered the survey in both the 7th and the 9th for both educational tracks and these associations were
grade were more likely to have better grades than those robust to controlling for sociodemographic background
who dropped out (p < .001). Chi-squared tests revealed and academic achievement.
that, in the 7th grade, non-participants were also more
likely to have psychosocial problems, long-term illness Health in the 9th grade
and medicine prescribed, and more daily health com- We also found significant associations when health was
plaints (p < .001). There were also statistically significant assessed in the same year (in the 9th grade, at the age of
differences in frequencies for all sociodemographic 15–16 years) in which Finnish students have to decide
Dobewall et al. BMC Public Health (2019) 19:1447 Page 6 of 15

Table 1 Health Factors, Sociodemographic Background, and Academic Achievement: Descriptive Statistics by Educational
Aspirations, % (n)
Applied for Applied for Applied for Total
vocational track both tracks academic track
Strengths and Difficulties Questionnaire = SDQ, 7th grade
Normal 78.7 (873) 80.6 (1.002) 89.1 (2.900) 85.2 (4.775)
Slightly raised 14.1 (156) 13.7 (170) 8.3 (270) 10.6 (596)
High difficulty score 7.2 (79) 5.7 (71) 2.6 (85) 4.2 (235)
Strengths and Difficulties Questionnaire = SDQ, 9th grade
Normal 63.0 (698) 66.8 (830) 80.0 (2.603) 73.7 (4.131)
Slightly raised 19.6 (217) 19.3 (240) 13.2 (431) 15.8 (888)
High difficulty score 17.4 (193) 13.9 (173) 6.8 (221) 10.5 (587)
Daily health complaints, 7th grade
No symptoms 73.7 (818) 73.6 (915) 79.3 (2.583) 76.9 (4.316)
One symptom 15.6 (173) 15.4 (191) 12.6 (409) 13.8 (773)
Two or more 10.7 (119) 11.1 (138) 8.2 (266) 9.3 (523)
Daily health complaints, 9th grade
No symptoms 69.2 (768) 68.9 (857) 73.2 (2.386) 71.5 (4.011)
One symptom 13.6 (151) 14.0 (174) 14.5 (471) 14.2 (796)
Two or more 17.2 (191) 17.1 (213) 12.3 (401) 14.3 (805)
Long-term illness,7th grade
No long-term illness 69.1 (767) 71.4 (888) 72.8 (2.370) 71.7 (4.025)
Long-term illness 12.9 (143) 12.9 (160) 12.5 (409) 12.7 (712)
Medicine prescribed 18.0 (200) 15.7 (196) 14.7 (478) 15.6 (874)
Long-term illness, 9th grade
No long-term illness 58 (639) 63 (778) 60 (1.958) 60 (3.375)
Long-term illness 14 (159) 14 (174) 16 (507) 15 (840)
Medicine prescribed 28 (312) 23 (292) 24 (792) 25 (1.396)
Self-rated health, 7th grade
Good 86.7 (963) 84.6 (1.051) 91.2 (2.968) 88.8 (4.982)
Average or poor 13.3 (148) 15.4 (191) 8.8 (288) 11.2 (627)
Self-rated health, 9th grade
Good 78.7 (874) 81.6 (1.014) 87.6 (2.851) 84.5 (4.739)
Average or poor 21.3 (237) 18.4 (228) 12.4 (405) 15.5 (870)
Gende
Girl 34.8 (387) 45.7 (569) 57.0 (1.857) 50.1 (2.813)
Boy 65.2 (724) 54.3 (675) 43 (1.401) 49.88 (2.800)
Immigrant background
Native 94.5 (1.050) 92.9 (1.155) 93.4 (3.044) 93.5 (5.249)
Immigrant 5.5 (61) 7.2 (89) 6.6 (214) 6.5 (364)
Parental employment
Both parents working 84.7 (914) 88.4 (1.078) 91.1 (2.946) 89.3 (4.938)
Other 15.3 (165) 11.6 (141) 8.9 (288) 10.7 (594)
Parental education
High 43.0 (478) 65.8 (818) 82.4 (2.684) 70.9 (3.980)
Low 57.0 (633) 34.2 (426) 17.6 (574) 29.1 (1.633)
Dobewall et al. BMC Public Health (2019) 19:1447 Page 7 of 15

Table 1 Health Factors, Sociodemographic Background, and Academic Achievement: Descriptive Statistics by Educational
Aspirations, % (n) (Continued)
Applied for Applied for Applied for Total
vocational track both tracks academic track
Family structure
Nuclear family 53.2 (591) 62.1 (772) 73.9 (2.407) 67.2 (3.770)
Other 46.8 (520) 37.9 (472) 26.1 (851) 32.8 (1.843)
School performance
High 1.1 (12) 10.0 (124) 49.7 (1.618) 31.3 (1.754)
Medium 28.9 (321) 56.2 (699) 47.3 (1.542) 45.7 (2.562)
Low 70.0 (776) 33.8 (420) 3.0 (98) 23.1 (1.294)
Note. SDQ Strengths and Difficulties Questionnaire

about their upper secondary education (Table 3). The Changes in health from the 7th to the 9th grade
more psychosocial problems were reported, the more The models that used the data of both surveys simultan-
likely the adolescents were to apply for the vocational eously to assess within-person change from the 7th to
track instead of the academic track or the more often the 9th grade and its association with educational aspira-
they were undecided in their choice between tracks. tions are presented in Table 4. We found a relationship
Similar to the results for the SDQ in the 7th grade, of worsening of health in regards to the SDQ with apply-
all associations persisted over and above sociodemo- ing for the vocational track and applying for both tracks.
graphic background and academic achievement. There Improvement in this health factor, however, was only
was also a weak association between daily health weakly associated with applying for the vocational track.
complaints and educational aspirations in this age Moreover, getting worse health in regards to long-term
group. Having two or more psychosomatic symptoms illness was associated with a decreased likelihood of ap-
nearly every day, compared to no symptoms, was re- plying for both educational tracks. Finally, worsening
lated to applying for the vocational track rather than self-reported health over time increased the likelihood to
the academic track. Having a long-term illness with apply for the vocational track. Remarkably, all associa-
and without medicine prescribed was associated with tions between educational aspirations and increasingly
being less undecided. The association for the use of worse health persisted when the sociodemographic back-
medicine persisted over and above sociodemographic ground and academic achievement were controlled for.
control variables. Educational aspirations were also
robustly associated with self-rated health with one ex- Discussion
ception: reporting average or poor health, compared Educational aspirations, measured by applying for aca-
to good health, was no longer associated with apply- demic versus vocational track or both, were associated
ing for both educational tracks when including socio- with all studied health factors at the age of 12–13 as well
demographic background into the model. the age of 15–16 years. Most associations remained sig-
nificant after controlling for students’ sociodemographic
background and academic achievement. Our results sup-
Socio-demographic background and academic port the health selection hypothesis, i.e. poor health
achievement leads to lower educational attainment; students with bet-
Of the sociodemographic variables, all but immigrant ter health in adolescence were more likely to apply for
background showed associations with students’ future the academic track, and those who were less healthy
plans for upper secondary education (Tables 2 and 3). were more likely to apply for the vocational track. In line
Across the models, boys were less likely than girls to with our expectations, health in the group of undecided
apply for academic track only. Applying for the academic students who had applied for both educational tracks lay
track, instead of the vocational track or to both, was as- in between.
sociated with students’ parental background in terms of In our data, lower educational aspirations were related
higher education and nuclear family structure, while the to having psychosocial problems assessed with the SDQ,
association between applying for vocational track and daily health complaints assessed with the frequency of
parental employment disappeared after accounting for psychosomatic symptoms, and average or poor self-rated
academic achievement. Unsurprisingly, especially aca- health. Previous studies did not provide a clear picture
demic achievement was a very strong and significant of whether poor health distorts educational aspirations
predictor of educational aspirations. [23, 24] and also associations between health and
Table 2 Associations of Educational Aspirations with Health, Sociodemographic Background, and Academic Achievement in the 7th Grade: Multilevel Multinomial Logistic
Regression. Odds Ratio (OR) and 95% Confidence Intervals (CI) are Presented
Health factors (n = 5.600) +Sociodemographics (n = 5.525) +School performance (n = 5.522)
Applied to academic track Vocational track Both tracks Vocational track Both tracks Vocational track Both tracks
is the reference category
OR (CI) OR (CI) OR (CI) OR (CI) OR (CI) OR (CI)
SDQ
Normal 1 1 1 1 1 1
Slightly raised 1.76 (1.39–2.23) 1.57 (1.29–1.98) 1.60 (1.24–2.07) 1.50 (1.18–1.89) 1.37 (> 1.00–1.88) 1.34 (1.04–1.74)
Dobewall et al. BMC Public Health

High difficulty score 2.92 (2.05–4.17) 2.07 (1.45–2.97). 2.50 (1.70–3.67) 1.94 (1.34–2.79) 1.47 (0.91–2.36) 1.43 (0.94–2.15)
Daily health complaints
No symptoms 1 1 1 1 1 1
One symptom 1.35 (1.09–1.68) 1.29 (1.05–1.59) 1.32 (1.04–1.66) 1.28 (1.04–1.58) 1.36 (1.02–1.81) 1.30 (1.03–1.64)
(2019) 19:1447

Two or more 1.20 (0.91–1.58) 1.18 (0.92–1.53) 1.14 (0.85–1.53) 1.15 (0.89–1.50) 1.17 (0.82–1.67) 1.11 (0.83–1.48)
Long-term illness
No long-term illness 1 1 1 1 1 1
Long-term illness 1.01 (0.81–1.26) 0.97 (0.78–1.19) 1.04 (0.82–1.31) 0.97 (0.78–1.20) 1.22 (0.91–1.63) 1.04 (0.82–1.31)
Medicine prescribed 1.22 (> 1.00–1.49) 1.04 (0.86–1.27) 1.19 (0.96–1.47) 1.02 (0.84–1.25) 0.97 (0.74–1.27) 0.94 (0.75–1.18)
Self-rated health
Good 1 1 1 1 1 1
Average or poor 1.20 (0.95–1.52) 1.51 (1.22–1.88) 1.03 (0–80-1.33) 1.41 (1.14–1.77) 1.17 (0.86–1.59) 1.57 (1.23–2.01)
Gender
Girl 1 1 1 1 1 1
Boy 2.82 (2.42–3.29) 1.76 (1.53–2.03) 3.08 (2.61–3.62) 1.81 (1.57–2.09) 1.98 (1.61–2.42) 1.35 (1.15–1.59)
Immigrant background
Native 1 1 1 1
Immigrant 0.88 (0.63–1.24) 1.21 (0.91–1-60) 0.85 (0.56–1.29) 1.22 (0.89–1.68)
Parental employment
Both parents working 1 1 1 1
Other 1.24 (0.98–1.58) 1.04 (0.83–1.32) 1.30 (0.96–1.76) 1.06 (0.82–1.38)
Parental education
High 1 1 1 1
Low 5.27 (4.47–6.21) 2.07 (1.77–2.43) 3.26 (2.66–4.00) 1.48 (1.24–1.77)
Family structure
Nuclear family 1 1 1 1
Page 8 of 15
Table 2 Associations of Educational Aspirations with Health, Sociodemographic Background, and Academic Achievement in the 7th Grade: Multilevel Multinomial Logistic
Regression. Odds Ratio (OR) and 95% Confidence Intervals (CI) are Presented (Continued)
Health factors (n = 5.600) +Sociodemographics (n = 5.525) +School performance (n = 5.522)
Applied to academic track Vocational track Both tracks Vocational track Both tracks Vocational track Both tracks
is the reference category
OR (CI) OR (CI) OR (CI) OR (CI) OR (CI) OR (CI)
Other 1.99 (1.69–2.35) 1.53 (1.32–1.79) 1.60 (1.30–1.96) 1.29 (1.09–1.52)
School performance
High 0.05 (0.03–0.08) 0.18 (0.15–0.23)
Dobewall et al. BMC Public Health

Medium 1 1
Low 35.15 (27.07–45.65) 9.53 (7.40–12.27)
AIC / BIC 10.194.5 / 10.320.5 9483.0 / 9661.7 7454.9 / 7660.0
Notes. Statistically significant associations are marked bold. Strengths and Difficulties Questionnaire = SDQ. Akaike (AIC) and Bayesian (BIC) information criteria are shown
(2019) 19:1447
Page 9 of 15
Table 3 Associations of Educational Aspirations with Health, Sociodemographic Background, and Academic Achievement in the 9th Grade: Multilevel Multinomial Logistic
Regression
Health factors (n = 5.600) +Sociodemographics (n = 5.525) +School performance (n = 5.522)
Applied for academic track is Vocational track Both tracks Vocational track Both tracks Vocational track Both tracks
the reference category
OR (CI) OR (CI) OR (CI) OR (CI) OR (CI) OR (CI)
SDQ
Normal 1 1 1 1 1 1
Slightly raised 1.94 (1.58–238) 1.76 (1.45–2.14) 1.86 (1.50–2.31) 1.69 (1.38–2.06) 1.56 (1.20–2.04) 1.44 (1.16–1.79)
Dobewall et al. BMC Public Health

High difficulty score 3.08 (2.42–3.91) 2.41 (1.89–3.06) 2.49 (1.93–3.23 2.15 (1.68–2.75) 1.59 (1.16–2.19) 1.54 (1.17–2.02)
Daily health complaints
No symptoms 1 1 1 1 1 1
One symptom 1.08 (0.87–1.22) 1.07 (0.87–1.31) 1.04 (0.82–1.32) 1.08 (0.88–1.33) 1.12 (0.84–1.50) 1.13 (0.89–1.42)
(2019) 19:1447

Two or more 1.45 (1.19–1.76) 1.22 (0.98–1.52) 1.07 (0.84–1.38) 1.22 (0.98–1.53) 1.03 (0.76–1.40) 1.19 (0.93–1.52)
Long-term illness
No long-term illness 1 1 1 1 1 1
Long-term illness 0.90 (0.73–1.12) 0.81 (0.66–0.99 0.90 (0.72–1.13) 0.83 (0.67–1.02) 0.99 (0.74–1.31) 0.89 (0.71–1.12)
Medicine prescribed 1.03 (0.86–1.23) 0.82 (0.69–0.98) 1.03 (0.85–1.24) 0.83 (0.69–0.98) 1.07 (0.85–1.36) 0.84 (0.69–1.02)
Self-rated health
Good 1 1 1 1 1 1
Average or poor 1.45 (1.19–1.76) 1.27 (1.04–1-54) 1.35 (1.10–1.67) 1.20 (0.98–1.46) 1.54 (1.18–2.00) 1.29 (1.03–1.61)
Gender
Girl 1 1 1 1 1 1
Boy 2.82 (2.42–3.30) 1.78 (1.54–2.05) 3.04 (2.58–3.59) 1.83 (1.58–2.12) 1.97 (1.60–2.42) 1.37 (1.17–1.61)
Immigrant background
Native 1 1 1 1
Immigrant 0.92 (0.65–1.30) 1.24 (0.94–1.65) 0.87 (0.57–1.32) 1.24 (0.90–1.70)
Parental employment
Both parents working 1 1 1 1
Other 1.30 (1.02–1.65) 1.07 (0.85–1.35) 1.32 (0.98–1.79) 1.07 (0.83–1.39)
Parental education
High 1 1 1 1
Low 5.11 (4.22–6.02) 2.03 (1.73–2.38) 3.21 (2.62–3.95) 1.47 (1.23–1.76)
Family structure
Nuclear family 1 1 1 1
Page 10 of 15
Table 3 Associations of Educational Aspirations with Health, Sociodemographic Background, and Academic Achievement in the 9th Grade: Multilevel Multinomial Logistic
Regression (Continued)
Health factors (n = 5.600) +Sociodemographics (n = 5.525) +School performance (n = 5.522)
Applied for academic track is Vocational track Both tracks Vocational track Both tracks Vocational track Both tracks
the reference category
OR (CI) OR (CI) OR (CI) OR (CI) OR (CI) OR (CI)
Other 1.92 (1.63–2.26) 1.53 (1.31–1.78) 1.55 (1.27–1.91) 1.29 (1.09–1.53)
School performance
High 0.05 (0.03–0.08) 0.19 (0.15–0.24)
Dobewall et al. BMC Public Health

Medium 1 1
Low 35.10 (27.02–45.60) 9.49 (7.37–12.23)
AIC / BIC 10.109.6 / 10.235.6 9431.2 / 9609.9 7442.5 / 7647.6
Notes. Odds Ratio (OR) and 95% Confidence intervals (CI) are presented. Statistically significant associations are marked bold. Strengths and Difficulties Questionnaire = SDQ. Akaike (AIC) and Bayesian (BIC) information
criteria are shown
(2019) 19:1447
Page 11 of 15
Table 4 Associations between Educational Aspirations and Change in Health from the 7th–9th Grade, with Health Factors in the 7th Grade Included in the Analysis to Account
for Starting Levels and Potential Ceiling Effects (not Shown): Multilevel Multinomial Logistic Regression. Odds Ratio (OR) and 95% Confidence Intervals (CI) are Presented
Health factors (n = 5.600)1 +Sociodemographics (n = 5.525)1,2 +School performance (n = 5.522)1,2,3
Dobewall et al. BMC Public Health

Applied for academic track is the Vocational track Both tracks Vocational track Both tracks Vocational track Both tracks
reference category
OR (CI) OR (CI) OR (CI) OR (CI) OR (CI) OR (CI)
SDQ
Improved 0.66 (0.45–0.95) 0.77 (0.54–1.10) 0.86 (0.57–1.27) 0.86 (0.60–1.25) 1.03 (0.63–1.67) 0.94 (0.62–1.42)
(2019) 19:1447

Stable 1 1 1 1 1 1
Worse 2.11 (1.75–2.54) 1.82 (1.52–2.19) 2.02 (1.66–2.27) 1.76 (1.46–2.12) 1.62 (1.27–2.07) 1.45 (1.18–1.78)
Daily health complaints
Improved 1.04 (0.76–1.42) 1.04 (0.78–1.40) 1.04 (0.74–1.44) 1.00 (0.74–1.34) 1.12 (0.75–1.68) 1.10 (0.79–1.53)
Stable 1 1 1 1 1 1
Worse 1.11 (0.91–1.36) 1.12 (0.93–1.35) 1.04 (0.84–1.29) 1.10 (0.80–1.52) 1.10 (0.84–1.44) 1.16 (0.94–1.44)
Long-term illness
Improved 1.09 (0.82–1.45) 1.01 (0.77–1.33) 1.08 (0.80–1.46) 0.98 (0.74–1.30) 0.91 (0.62–1.33) 0.87 (0.64–1.19)
Stable 1 1 1 1 1 1
Worse 0.98 (0.82–1.18) 0.78 (0.65–0.93) 0.98 (0.81–1.19) 0.77 (0.65–0.93) 0.98 (0.77–1.25) 0.77 (0.63–0.94)
Self-rated health
Improved 0.90 (0.59–1.39) 1.04 (0.70–1.54) 0.82 (0.52–1.30) 1.03 (0.69–1.53) 0.79 (0.45–1.38) 0.97 (0.63–1.52)
Stable 1 1 1 1 1 1
Worse 1.48 (1.17–1.87) 1.21 (0.95–1.53) 1.38 (1.08–1.77) 1.14 (0.89–1.45) 1.57 (1.15–2.14) 1.19 (0.91–1.56)
AIC / BIC 10.100.9 / 10.333.0 9708.4/ 9708.4 7753.4/ 7753.4
Notes. Analyses were adjusted for 1gender, 2immigrant background, parental employment, and education, and family structure, and 3school performance. Statistically significant associations are marked bold. Strengths
and Difficulties Questionnaire = SDQ. Akaike (AIC) and Bayesian (BIC) information criteria are shown
Page 12 of 15
Dobewall et al. BMC Public Health (2019) 19:1447 Page 13 of 15

educational attainment in adolescence have been found The significant proportion of the variance attributable
to be mixed [6, 9–16]. to differences between schools suggests that the role of
The associations of the SDQ, daily health complaints, student composition and contextual factors cannot be
and self-rated health with students’ educational aspira- ignored in the complex relationship between health and
tions stayed but attenuated after controlling for sociode- educational aspirations [34, 41].
mographic background and academic achievement As expected, students’ educational aspirations were re-
which supports the independent effect of health factors lated to their parents’ education and employment as well
in the creation of socioeconomic health inequalities. The as their academic achievement. Both this result and the
plausibility of the health selection hypothesis was further fact that educational aspirations and health in adoles-
strengthened by the finding that the group of those stu- cence showed an association over and above students’
dents whose health worsened over time in terms of the academic achievement might point to the bidirectional
SDQ and self-rated health, had on average lower aca- nature of the relationships [4, 12, 42]. Health and aca-
demic aspirations. This makes it less likely that an demic achievement are most likely interconnected since
unobserved third factor that influences both health and performance at school already reflects students’ earlier
educational aspirations had confounded the observed health, and perceptions related to academic success and
associations. Previous studies that were able to examine failure are probably intertwining with health perceptions
fixed effects estimates, similarly, found that the over the school years [11, 34, 43]. It is also well known
education-health gradient is largely shaped by health se- that even in the Finnish welfare state social factors of
lection [6, 11]. the family influence students’ educational choices and
With this study we wanted to find out whether stu- trajectories [25–27, 29]. Thus, the interplay between the
dents’ plans after compulsory schooling are already mechanisms of health selection and social causation in
patterned by their health in the 7th grade (age 12– the production of health inequality was visible in our
13) when students start lower secondary education or data which highlights that they can have different influ-
whether health matters only at the end of the 9th ences at different periods of life course [5, 8].
grade (age 15–16 years) at the time when they apply
to upper secondary education. On average, the effect
of health was weaker at age 12–13 than at age 15–16. Limitations and strengths
As the differences fall within the respective CIs, how- We cannot exclude bias in our results due to selective
ever, these associations do not seem to be signifi- attrition. Without attrition, however, the observed effects
cantly modified by being assessed in the 7th or 9th of the studied health and social factors on educational
grade. Thus, both times seem to be crucial for deter- aspirations might have been even stronger because those
mining students’ successful educational paths into who were less healthy and from a more disadvantaged
adulthood. At the same time, the results indicate that family background were less likely to participate in the
health in adolescence influences students’ future plans second survey.
even if assessed years before the choice between aca- Among the considerable strengths of the research is
demic and vocational track has to be made in the fact that we used a longitudinal multilevel design to
Finland. This finding aligns well with research on the understand how health in adolescence links to the
influence of health disadvantage in early childhood on choice between educational tracks that took into ac-
later educational attainment [4] and shows the im- count the significant effect of the school attended on
portance of adolescence as a formative period of life. educational aspirations. Very few, if any other large ado-
Inconsistent results were observed for long-term ill- lescent cohorts have covered health and education as
ness, which related to lower educational aspirations comprehensively both in terms of health indicators and
when being assessed in the 7th grade but instead to the opportunity to follow the same individuals over the
higher educational aspirations when being assessed in transition to further education after the compulsory
the 9th grade. Adolescents that reported worsening of schooling ends. Assessing health longitudinally enabled
health between the measurement points in regards to us to identify those periods in adolescence which are
long-term illness also applied proportionally less often sensitive for their successful paths into adulthood and to
for both educational tracks instead of the academic track examine the effects of within-person change in health
only. That the associations had the opposite sign at dif- over time. Educational aspirations were assessed object-
ferent ages matches the mixed results obtained in previ- ively by obtaining from the national registry covering all
ous work on adolescents with long-term illness and students in the country, the choices they have made
educational attainment [9, 10, 15]. Our results further when applying to upper secondary education. Using na-
show that health-related selection may work differently tional registry data further reduced measurement error
for different health factors [13]. and the amount of missing data due to nonresponse.
Dobewall et al. BMC Public Health (2019) 19:1447 Page 14 of 15

Conclusions Ethics approval and consent to participate

Our findings show that selection by health factors into The study protocol was approved by the Ethical Committee of the Finnish
Institute for Health and Welfare. The educational authorities of all 14
different educational trajectories takes place already at municipalities of the Helsinki Metropolitan Region gave a permission for the
the early teenage much before the adolescents need to study. In line with the instructions of the Finnish National Board on Research
choose which educational track – if any – they wish to Integrity (TENK) in 2009, no parental consent was required when the study
was conducted as part of the students’ normal schoolwork. Two of the 14
apply after the compulsory schooling. Our findings sup- municipalities had adopted a policy that a written parental consent is always
port the health selection hypothesis in the creation of required. These were collected. In the other municipalities, information
health inequalities: those whose health is worse, more letters were sent to the parents who had the possibility to withdraw their
child from the study. The students were instructed about the purpose of the
often had lower educational aspirations than those study and that participation was voluntary and that they can decline to
whose health is better. That health factors had an effect answer any question or withdraw from the survey at any time.
over and above sociodemographic background, and
Consent for publication
school performance shows that health in adolescence is
Not applicable.
independently associated with the plans of students for
their further education. Our findings also show the im- Competing interests
portance of adolescence in the creation of inequalities. The authors declare that they have no competing interests.
More studies are needed to reveal which measures Author details
would be effective in helping students with poor health 1
Faculty of Social Sciences (Health Sciences), Tampere University, Po Box 20,
to achieve their full educational potential. (Arvo Ylpön katu 34), 33014 Tampere, Finland. 2PERLA—Tampere Centre for
Childhood, Youth and Family Research, Tampere University, 33014 Tampere,
Finland. 3Social Policy Research Unit, Finnish Institute for Health and Welfare,
Supplementary information 00271 Helsinki, Finland. 4Department of Social Research, Faculty of Social
Supplementary information accompanies this paper at [Link] Sciences, University of Turku, Turku, Finland. 5Faculty of Education, Tampere
1186/s12889-019-7824-8. University, 33014 Tampere, Finland. 6Centre for Educational Assessment,
University of Helsinki, 00014 Helsinki, Finland. 7Department of Adolescent
Psychiatry, Pitkäniemi Hospital, Tampere University Hospital, 33380 Nokia,
Additional file 1: Table S1. Change in Health from the 7th–9th Grade,
Finland.
% (n).
Received: 21 January 2019 Accepted: 21 October 2019
Abbreviations
AIC: Akaike information criteria; BIC: Bayesian information criteria;
CI: Confidence intervals; GPA: Grade point average; MetLoFin: Metropolitan References
Longitudinal Finland; OR: Odds ratios; SDQ: Strengths and Difficulties 1. Crimmins EM, Saito Y. Trends in healthy life expectancy in the United States,
Questionnaire; TENK: Finnish National Board on Research Integrity 1970-1990: gender, racial, and educational differences. Soc Sci Med. 2001;
52(11):1629–41. [Link]
Acknowledgments 2. Kröger H, Pakpahan E, Hoffmann R. What causes health inequality? A
We are grateful to all the students and schools that took part in the systematic review on the relative importance of social causation and health
MetLoFin study. We thank Laura Kortesoja, Mikko Asikainen (both University selection. Eur J Pub Health. 2015;25(6):951–60. [Link]
of Helsinki), and Lasse Pere (Tampere University) for their support of the eurpub/ckv111.
study. Any views expressed in this paper are solely those of the authors. 3. Moor I, Spallek J, Richter M. Explaining socioeconomic inequalities in self-
rated health: a systematic review of the relative contribution of material,
psychosocial and behavioural factors. J Epidemiol Community Health. 2017;
Authors’ contributions
71(6):565–75. [Link]
HD, PL, and AR conceived and designed the study. HD and AR performed
4. Conti G, Heckman J, Urzua S. The education-health gradient. Am Econ Rev.
data analysis and wrote the first draft of the manuscript. HD, PL, SK, LK, M-PV,
2010;100(2):234–8. [Link]
RH, and AR interpreted the data, commented on the manuscript, and con-
5. Hoffmann R, Kröger H, Pakpahan E. Pathways between socioeconomic
tributed to all revision stages by substantively revising the manuscript. HD,
status and health: does health selection or social causation dominate in
PL, SK, LK, M-PV, RH, and AR read and approved the final version of the paper
Europe? Adv Life Course Res. 2018;36:23–36. [Link]
and have agreed both to be personally accountable for the author’s own
2018.02.002.
contributions and to ensure that questions related to the accuracy or integ-
6. Lynch JL, von Hippel PT. An education gradient in health, a health gradient
rity of any part of the work, even ones in which the author was not person-
in education, or a confounded gradient in both? Soc Sci Med. 2016;154:18–
ally involved, are appropriately investigated, resolved, and the resolution
27. [Link]
documented in the literature.
7. West P. Rethinking the health selection explanation for health
inequalities. Soc Sci Med. 1991;32(4):373–84. [Link]
Funding 0277-9536(91)90338-D.
This study was supported by the Academy of Finland (Skidi-Kids Research 8. Sweeting H, Green M, Benzeval M, West P. The emergence of health
Programme, the project 264276/2012, and the project 288774/2015), inequalities in early adulthood: evidence on timing and mechanisms from a
Competitive State Research Funding of the Expert Responsibility Area of West of Scotland cohort. BMC Public Health. 2015;16(1):41. [Link]
Tampere University Hospital. The funding bodies had no role in the design 10.1186/s12889-015-2674-5.
of the study, collection, analysis, interpretation of data, and writing the 9. Acacio-Claro P, Doku D, Koivusilta L, Rimpelä A. How socioeconomic
manuscript. circumstances, school achievement and reserve capacity in adolescence
predict adult education level: a three-generation study in Finland. Int J
Availability of data and materials Adolesc Youth. 2018;23(3):382–97. [Link]
The access for data for external researchers may be granted based on a 1389759.
written request to AR and RH including a research plan and a data 10. Champaloux SW, Young DR. Childhood chronic health conditions and
management plan. AR is responsible for the health part of the school data. educational attainment: a social ecological approach. J Adolesc Health.
RH is responsible for the education part of the school data. 2015;56(1):98–105. [Link]
Dobewall et al. BMC Public Health (2019) 19:1447 Page 15 of 15

11. Haas SA, Fosse NE. Health and the educational attainment of adolescents: 32. Goodman R. Psychometric properties of the strengths and difficulties
evidence from the NLSY97 *. J Health Soc Behav. 2013;49(2):178–92. https:// questionnaire. J Am Acad Child Adolesc Psychiatry. 2001;40(11):1337–45.
[Link]/10.1177/002214650804900205. [Link]
12. Huurre T, Rahkonen O, Komulainen E, Aro H. Socioeconomic status as a 33. Goodman R, Meltzer H, Bailey V. The strengths and difficulties questionnaire:
cause and consequence of psychosomatic symptoms from adolescence to a pilot study on the validity of the self-report version. Eur Child Adolesc
adulthood. Soc Psychiatry Psychiatr Epidemiol. 2005;40(7):580–7. [Link] Psychiatry. 1998;7(3):125–30. [Link]
org/10.1007/s00127-005-0930-1. 34. Minkkinen J, Lindfors P, Kinnunen J, et al. Health as a predictor of students’
13. Koivusilta L, Rimpela A, Rimpela M. Health status: does it predict choice in academic achievement: a 3-level longitudinal study of Finnish adolescents. J
further education? J Epidemiol Community Health. 1995;49(2):131–8. https:// Sch Health. 2017;87(12):902–10. [Link]
[Link]/10.1136/jech.49.2.131. 35. Haugland S, Wold B. Subjective health complaints in
14. Landstedt E, Brydsten A, Hammarström A, Virtanen P, Almquist YB. The role adolescence—reliability and validity of survey methods. J Adolesc. 2001;
of social position and depressive symptoms in adolescence for life-course 24(5):611–24. [Link]
trajectories of education and work: a cohort study. BMC Public Health. 2016; 36. Idler EL, Benyamini Y. Self-rated health and mortality: a review of twenty-
16(1):1–16. [Link] seven community studies. J Health Soc Behav. 1997;38(1):21–37. [Link]
15. Uiters E, Maurits E, Droomers M, Zwaanswijk M, Verheij RA, Van Der Lucht F. org/10.2307/2955359.
The association between adolescents’ health and disparities in school 37. Salkind NJ. Difference score. In: Salkind NJ, editor. Encyclopedia of
career: a longitudinal cohort study. BMC Public Health. 2014;14(1):1104. Measurement and Statistics. Thousand Oaks California 91320 United States
[Link] of America: Sage Publications, Inc; 2007. p. 258–60. [Link]
16. Miech RA, Caspi A, Moffitt TE, Wright BRE, Silva PA. Low socioeconomic 9781412952644.
status and mental disorders: a longitudinal study of selection and causation 38. West P, Sweeting H, Speed E. We really do know what you do: a
during young adulthood. Am J Sociol. 1999;104(4):1096–131. [Link] comparison of reports from 11 year olds and their parents in respect of
10.1086/210137. parental economic activity and occupation. Sociology. 2001;35(2):539–59.
17. Betts JR. The economics of tracking in education. In: Hanushek E, Machin S, [Link]
Woessmann L, editors. Handbook of the Economics of Education, vol. 3. 39. Allison P. Fixed effects regression models. Thousand Oaks: Sage
Amsterdam: Elsevier; 2011. p. 341–81. [Link] Publications, Inc.; 2009. [Link]
53429-3.00007-7. 40. Snijders TB, Bosker RR. Multilevel analysis: an introduction to basic and
18. Khattab N. Students’ aspirations, expectations and school achievement: advanced multilevel modeling. London: SAGE Publications; 1999.
what really matters? Br Educ Res J. 2015;41(5):731–48. [Link] 41. Lindfors P, Minkkinen J, Rimpelä A, Hotulainen R. Family and school social
1002/berj.3171. capital, school burnout and academic achievement: a multilevel
19. Reynolds JR, Pemberton J. Rising college expectations among youth in the longitudinal analysis among Finnish pupils. Int J Adolesc Youth. 2018;23(3):
United States: a comparison of the 1979 and 1997 NLSY. J Hum Resour. 368–81. [Link]
2001;36(4):703. [Link] 42. Esch P, Bocquet V, Pull C, et al. The downward spiral of mental disorders
20. Croll P. Educational participation post-16: a longitudinal analysis of and educational attainment: A systematic review on early school leaving.
intentions and outcomes. Br J Educ Stud. 2009;57(4):400–16. [Link] BMC Psychiatry. 2014;14(1):237. [Link]
10.1111/j.1467-8527.2009.00445.x. 43. Kinnunen JM, Lindfors P, Rimpelä A, et al. Academic well-being and
21. Sewell WH, Hauser RM. The Wisconsin longitudinal study of social and smoking among 14- to 17-year-old schoolchildren in six European cities. J
psychological factors in aspirations and achievements. In: Kerckhoff AC, Adolesc. 2016;50:56–64. [Link]
editor. Research in sociology and education, vol. 1. Greenwich: JAI Press;
1980. p. 59–99. Publisher’s Note
22. Haas SA. Health selection and the process of social stratification: the effect Springer Nature remains neutral with regard to jurisdictional claims in
of childhood health on socioeconomic attainment. J Health Soc Behav. published maps and institutional affiliations.
2006;47(4):339–54. [Link]
23. Stevens SE, Steele CA, Jutai JW, Kalnins IV, Bortolussi JA, Biggar WD.
Adolescents with physical disabilities: some psychosocial aspects of health. J
Adolesc Health. 1996;19(2):157–64. [Link]
139X(96)00027-4.
24. Madarasova Geckova A, Tavel P, van Dijk JP, Abel T, Reijneveld SA. Factors
associated with educational aspirations among adolescents: cues to
counteract socioeconomic differences? BMC Public Health. 2010;10(1):154.
[Link]
25. Kallio JM, Kauppinen TM, Erola J. Cumulative socio-economic disadvantage
and secondary education in Finland. Eur Sociol Rev. 2016;32(5):649–61.
[Link]
26. Koivusilta L, West P, Saaristo V, Nummi T, Rimpelä A. From childhood socio-
economic position to adult educational level - do health behaviours in
adolescence matter? A longitudinal study. BMC Public Health. 2013;13:711.
[Link]
27. OECD. PISA 2015 Results (Volume I): Excellence and Equity in Education;
2016. [Link]
28. Garg R, Kanuppi C, Lewko J, Urajnik D. A structural model of educational
aspirations. J Career Dev. 2002;29(2):87–109. [Link]
1019964119690.
29. Kalalahti M, Varjo J, Jahnukainen M. Immigrant-origin youth and the
indecisiveness of choice for upper secondary education in Finland. J Youth
Stud. 2017;20(9):1242–62. [Link]
30. Lappalainen K, Hotulainen R. An age group’s nine year long change in
educational and employment position. In: Herz B, Kuorelahti M, editors.
Cross-categorical special education needs in Finland and Germany. Münster:
Waxmann; 2007. p. 111–22.
31. Maas CJM, Hox JJ. Sufficient sample sizes for multilevel modeling.
Methodology. 2005;1(3):86–92. [Link]