You are on page 1of 25

SCANDINAVIAN JOURNAL OF EDUCATIONAL RESEARCH, 2017

VOL. 61, NO. 4, 479–502


http://dx.doi.org/10.1080/00313831.2016.1172498

Physical Activity and School Performance: Evidence from a Danish


Randomised School-Intervention Study
A. Quinto Romani and T.B. Klausen
Department of Sociology and Social Work, Aalborg University, Aalborg, Denmark

ABSTRACT ARTICLE HISTORY


It has been claimed that physical activity has a positive effect on not only Received 3 July 2014
health but also on school performance. Using data from a randomised Accepted 8 February 2016
school-intervention study, this paper investigates whether different
KEYWORDS
interventions promoting physical activity affect school performance in School performance; physical
primary school children. The results indicate that on average, the activity; randomised school-
interventions have a very limited beneficial impact on the pupils’ based intervention;
performance. The effects are mostly insignificant, and in some cases difference-in-differences
even negative. These results are relevant when considering how general
school resources and resources targeted towards physical activity in
particular should be invested to improve school performance.

In the wake of a prolonged lockout of Danish primary school teachers in the spring of 2013, the Dan-
ish government launched an extensive school reform. A precondition for the reform was to introduce
longer working hours for teachers, which was the main focal point in the conflict between the tea-
chers and the local authorities. The reason for the reform – which has been gradually implemented
since 2014 – is the mediocre results Danish pupils obtain in international comparative tests such as
PISA. The test results located approximately at the mean of comparable OECD member countries in
reading, mathematics and science have been on the political agenda since the first PISA test in 2000
(OECD, 2013).
In order to obtain better results and give more children an adequate basis for further education,
the school reform increases the total number of weekly lessons by 10–15%, the pupils receive an ear-
lier introduction to foreign languages (e.g., English from form 1 instead of form 4), and more atten-
tion is paid to core subjects like Danish and mathematics. The overall aim is that when the reform is
fully implemented, pupils in their 9th school year will have the same level of competence as the
pupils today have in their 10th and final year of primary school. Thus, the idea seems to be that
the approximately 10% extra input should result in a 10% extra outcome.
Longer school days potentially result in more sedentary hours. Thus, the reform has taken
measures to secure 45 minutes of extra physical activity during the prolonged school day. It is up
to the local councils and individual schools to implement the reform, which could result in more
physical education or include more physical activity in the teaching of other subjects or a combi-
nation of the two.
The main argument in favour of more physical activity during the school day is that this could
affect the pupils’ health in a positive way, that is, improving fitness, reducing overweight, and thereby
making a contribution to preventing future lifestyle deceases. But it has been argued that physical
activity might have a positive effect on school performance too. In the implementation phase of

CONTACT A. Quinto Romani aqr@socsci.aau.dk Department of Sociology and Social Work, Aalborg University, Aalborg,
Denmark
© 2016 Scandinavian Journal of Educational Research
480 A. QUINTO ROMANI AND T. B. KLAUSEN

the reform, the potential positive effects that physical activity might have on school performance did
receive increased attention in the public debate. Especially boys – who are performing at a substan-
tially poorer level than girls in primary school – are believed to benefit from a more physically active
learning environment. Likewise, the inactive pupils are believed to benefit from additional manda-
tory physical activity in a school context.
The argument in favour of a positive association between physical activity and cognitive capacity
is backed by several theoretical perspectives, which according to Swedish researchers can be divided
into two major sub-categories: sensory-motoric and neurophysiological perspectives (Ericsson &
Karlsson, 2014). In the sensory-motoric perspective, the main focus is on the role a child’s early
motoric experiences play for its sensory and perceptual development, which is believed to be crucial
for the cognitive processes. Slow motoric development could impede learning, as it could disturb the
ability to concentrate in learning situations (Cratty, 1997; Gjesing, 1997). According to the neuro-
physiological perspective, motoric training changes the structure and functioning of the nervous sys-
tem. To learn complicated movement-patterns, stimulation of the same areas of the brain as those
used to solve problems and for cognitive learning is essential (Jensen, 1998; Shephard, 1997).
Besides the mechanisms highlighted in the Swedish study, three other mechanisms might imply a
beneficial impact on school performance. First, according to the biomedical approach, physical
activities increase the amount of oxygen delivered to the brain, which increases children’s capacity
to learn. Further, physical activity has been argued to raise the levels of norepinephrine, which
reduces stress levels (Ploughman, 2008). Second, according to the psychological perspective also
described by Ericsson and Karlsson (2014), the association between physical activity and cognition
is of a more indirect character. Physical activity is believed to promote self-efficacy (Bandura, 1997;
Steptoe & Butler, 1996), and self-efficacy has a positive impact on cognitive capacity as it stimulates
the desire to learn. Third, according to a more sociological approach, physical activity might have a
positive indirect effect, for example, as children learn to corporate and abide by the rules of the activi-
ties, which could affect school satisfaction (Hanks & Eckland, 1976). Whichever of these mechan-
isms might be present, we would expect school-based interventions aimed at increasing physical
activity to have a positive effect on school performance.
According to economic tradition, the impact of increasing physical activity in school is more
ambiguous. As learning is a cumulative process in which skills acquired in one area could have a
reinforcing impact on skills acquired in another (Holland & Andre, 1987), interventions could
improve school performance even without having any significant effect on fitness. However, time
spent on physical activity may conflict with time spent on academic subjects (Coleman, 1961).
Thus, insofar as the time constraint is an issue, the intervention may have a negative effect on school
performance. For example, if pupils are more engaged in physical education and organised sport,
there is less time and energy left for academic subjects. This might decrease school performance.
Finally, publicly financed physical activity in school may simply replace the private provision of
physical activity without resulting in any health improvement. As the interventions take place in a
school context, this may force a shift in the pupils’ allocation of their physical activity from out-of-
school activities, including structured and unstructured leisure time activities, to in-school physical
activities. Thus, in-school interventions may make out-of-school investments less likely, thereby
reducing the possible positive effect on school performance.
The aim of this paper is to explore if increased physical activity has an effect on school perform-
ance in Denmark. Earlier research has often failed to establish a causal relationship. With a school-
based intervention study we can approach the potential endogeneity of physical activity and provide
new evidence about the relation between physical activity and school performance.
The outline of the paper is as follows: The section immediately below provides a short description
of the conditions of physical activity among children in Denmark, both within and outside of school.
The next section gives an overview of research on physical activity and school performance;
then follows a section describing the research aims of the current paper. In the subsequent
section, the data and methods are presented, followed by a section presenting the empirical results.
SCANDINAVIAN JOURNAL OF EDUCATIONAL RESEARCH 481

In the final section, the results are discussed and suggestions for further research on physical activity
and school performance are presented.

In-School and Out-of-School Physical Activity in Denmark


In Denmark, mandatory physical education (PE) is the main in-school physical activity. The central
education authorities decide the minimum number of PE lessons in compulsory education. The rec-
ommended minimum number of lessons during a year is 90 in primary school and 60 in lower-sec-
ondary school. Nevertheless, a tendency has emerged for schools not to comply with the PE
minimum recommendations. In the 4th and 5th forms, 41 and 52% of the schools, respectively,
did not fulfil the recommendations (Danish Evaluation Institute, 2004).
Outside of school, Danish children are fairly active. On average, they spend 4 hours and 46 min-
utes per week on organised sports, and 30% spend more than 6 hours per week on these (Laub &
Pilgaard, 2013). Moreover, they participate on average in 3.2 different types of activities and 71%
of all children aged 7–15 participate regularly in sports competitions (Laub & Pilgaard, 2013). Sports
are highly subsidised, and the fees are usually fairly low. Financial support is not means tested, thus
the families’ financial resources in regard to the most popular sports like football, handball, and gym-
nastics should not be a major barrier for participation. However, fees for some types of exclusive
sports such as horse-riding and tennis are high.
Unstructured leisure time physical activity is also quite popular among Danish youth – 86% of the
children aged 7–15 are regularly physically active, 46% are physically active on their own, and 21%
participate in after-school physical activities (Laub & Pilgaard, 2013). Finally, Danish children are
fairly mobile, as they do not have to travel long distances to access parks or sports clubs. Good walk-
ing and bicycle paths mean that parental chauffeuring is only necessary to a limited extent. Thus,
children in Denmark often bike or walk to school and sports, just as they have the freedom to
move around and play outside without adult supervision.
The 2014 school reform has increased physical education and physical activity in school. How-
ever, the lenghthened school days might have some unintended consequences regarding out-of-
school physical activity, both because there is more physical activity in the timetable and because
the schooldays are longer – inevitably resulting in less leisure time.

Previous Research on Psychical Activity and School Performance


Recent research has taken three broad approaches to identifying the effect of physical activity on
school performance: cross-sectional, experimental, and longitudinal/econometric methods.

Cross-Sectional Studies
A large share of the research in this field uses cross-sectional data (Tomporowski, 2003; Trudeau &
Shephard, 2008). Research based on this approach generally concludes that physical activity has a
positive effect on academic achievement. Nelson and Gordon-Larsen (2006) studied 11,957 adoles-
cents, finding that physically active pupils are more likely to receive higher grades, especially in math-
ematics and English. In their study of 89 high school pupils, Field, Diego, and Sanders (2001) also
found that physical activity improves grades. Dwyer, Sallis, Blizzard, Lazarus, and Dean (2001),
who studied 7,961 pupils aged 7–15 years, found a beneficial effect on academic achievement too.
In contrast, Tremblay, Inman, and Willms (2000), studying 6,923 pupils attending 6th form,
found that physical activity had a negative effect on school performance. In their studies of 232 Eng-
lish pupils aged 13–16 years, Daley and Ryan (2000) also found a negative association between phys-
ical activity and academic achievement. Finally, some studies find no impact of physical activity on
school performance, among them studies conducted by Dollman, Boshoff, and Dodd (2006) inves-
tigating 117 Australian pupils in forms 3, 5, and 7.
482 A. QUINTO ROMANI AND T. B. KLAUSEN

However, the overall impression from the research based on cross-sectional data is that physical
activity has a substantial beneficial impact on school performance. The ambiguity that still exists in
these studies may emerge because cross-sectional studies do not take into consideration the potential
endogeneity problem arising as the error term is correlated with the explanatory variable, which
undermines causal inferences. One problem is the omitted variable bias that may arise if physically
active pupils have a more privileged family background, causing them to perform better in school.
Another problem is the selection bias that may arise if more ambitious and self-confident pupils self-
select into physical activity.

Experimental Studies
To deal with the endogeneity problem, a number of studies have been based on experimental inter-
ventions in school settings. Results from these types of studies are more mixed. Some research seems
to find a positive effect (Donnelley et al., 2009, 2013; Ericsson & Karlsson, 2014; Shephard et al.,
1984), whereas others find no impact of physical activity on school performance (Coe, Pivarnik,
Womack, Reeves, & Malina, 2006; Sallis et al., 1999). A review of these studies is presented in a
Table A1 in Appendix A, indicating whether the study was a randomised controlled trial (RCT),
a cluster RCT, or a quasi RCT.
Using 1970–1977 data from the Trois-Riviéres region in Quebec, Shephard et al. (1984) explored
the impact of 5 hours (260 minutes) of PE per week on 546 pupils, compared to the standard PE
lesson of 40 minutes per week. The extra PE lessons were taken away from academic teaching.
Results indicate that the increase in PE had a positive effect on school performance. Donnelley
et al. (2009) explored the impact of 90 minutes of additional PE per week for three years and
found that increased physical activity improved test scores.
Inspired by Shephard et al. (1984), Ericsson and Karlsson (2014), carried out a single-school nine-
year intervention study following 251 Swedish pupils during all their years at primary and lower-sec-
ondary school, giving the intervention group a 45 minute extra daily lesson of PE in addition to the
weekly 2 hours in the control group. This prolonged the school day by one lesson, meaning that both
groups received the same number of lessons in the other subjects during a week. The extra PE lessons
had a significant impact on school performance among boys, but not girls.
In contrast, Sallis et al. (1999) studied 655 Californian pupils in 5th and 6th forms who received
27–42 minutes of additional PE per week for two years. The results indicate that increased PE time
had no impact on academic achievement. Coe et al. (2006) explored the impact of 55 minutes of PE
per day on 214 pupils attending 6th form (11-year-olds). They found that the increase in PE had no
effect on academic achievement except for the pupils who exercised intensively.
Careful reading of this literature suggests that the impact of physical activity varies by differ-
ences in intensity, frequency, and duration, implying there might be a threshold at which PE starts
to be beneficial. If correctly done, random assignment solves issues regarding selection bias. This
will make alternative explanations less likely as causes of the observed outcome of the
interventions.
Nevertheless, random assignment in RCT does not always have the intended effect, as baseline
imbalance might be an issue (Angrist & Pischke, 2009; Heckman, Jeffrey, & Smith, 1995). Imperfect
randomisation implies that the correlation between physical activity and pupils’ school performance
could partly be measuring the effect of observed and unobserved confounding variables related to,
for example, family socioeconomic background.

Studies Based on Econometric Methods


Nevertheless, it is not always possible to use RCT in social sciences; for this reason, non-experimental
or econometric methods have been applied to solve the endogeneity problem. Compared to the RCT,
SCANDINAVIAN JOURNAL OF EDUCATIONAL RESEARCH 483

which provides a simple solution to the selection problem, non-experimental methods are often
more advanced and based on crucial identification assumptions.
Rees and Sabia (2010) claim that research results are method-dependent. Simple ordinary least
square regression models tend to find that physical activity has a beneficial academic effect. However,
if more advanced methods like fixed effects and instrumental variable methods are used, the relation-
ship is weaker. This indicates that the positive association between physical activity and school per-
formance can be subject to endogeneity problems.
Dills, Morgan, and Rotthoff (2011) and Cawley, Frisvold, and Meyerhoefer (2013) studied
pupils attending 1st–5th form. In order to address the selection problem, they explored the natural
experiment of variation in mandatory PE – using it as an instrument variable. The omitted variable
bias is taken into account when adjusting the model for a rich set of control variables, assuming
that the unobserved and observed characteristics are correlated. Here, an increase in PE does
not come at the cost of academic time, but extends the school days. The results indicate that
increased PE reduced obesity in the 5th form but had no impact on academic score. Hence, studies
that apply econometric methods tend to find that an increase in PE has at most a neutral effect on
school performance.
Contradictive results in earlier research are found both within and between different methodo-
logical approaches. The variation within a certain method seems to arise due to the differences in
intensity, frequency, and duration of physical activity, whereas the variation between different
methods seems to depend on whether the endogeneity problems are approached. Researchers
using econometric methods tend to find more modest results implying that the positive relationship
between physical activity and school performance may be subject to endogeneity.

Research Aims
One important purpose of this paper is to investigate the impact of interventions directed towards
physical activity on school performance in Denmark, where such studies have so far been virtually
non-existent. Further, improvements compared to previous research will be made on several
dimensions.
First, the state-of-the-art methodology of RCT is followed and extended by combining it with
econometric methods. Simple differences in mean are often used when evaluating RCT; however,
if randomisation does not successfully balance pupils’ characteristics across intervention assign-
ments, such results could be biased. One way to deal with this issue is following Donnelley et al.
(2009), who present adjusted simple differences in mean. We adjust for different parental character-
istics. Another way to avoid the criticism of imperfect randomisation to which earlier RCT has been
subject is to apply the more advanced adjusted difference-in-differences approach, which, by taking
advantage of the panel structure of the data, captures unobserved time-invariant determinants of
school performance.
Second, we consider whether the lack of consensus in earlier research may have been caused by a
focus on mean impact, concealing major heterogeneity in participation rate. Thus, the mean may be
biased since the parameter estimated is the average intervention impact on all pupils selected for the
intervention, neglecting the fact that not all pupils may participate or participate at the intended
level. To our knowledge, none of the earlier studies in this field have considered the “take-up
rate.” Hence, we address the impact of the interventions on school performance for a subsample
of pupils who improved and did not improve their physical fitness.
Finally, the impact of the school interventions will focus not only on the biological mechanisms.
The idea is that the actual content of the intervention could affect school performance differently
when not only the intended but also the unintended consequences are included. Hence, an interven-
tion may simply be “crowding out” the private provision of after-school physical activity. We
approach this issue by considering the impact of the interventions on the after-school activity level.
484 A. QUINTO ROMANI AND T. B. KLAUSEN

Data and Methods


Data
The data set used to investigate the association between physical activity and school performance is
based on a randomised school-based intervention study, Project 3A (randomised school-based inter-
vention study), in the Municipality of Aalborg, Denmark. Data were collected when the pupils
attended 6th form (aged 11–13 years of age) in 2008 and again when they attended 8th form
(aged 13–15 years of age) in 2010.
Figure 1 illustrates the design of the study, which is a cluster RCT. All public schools in the Muni-
cipality of Aalborg with lower-secondary school programmes were included (37 out of a total of
50 schools). The schools were stratified into upper, middle and lower schools based on average

Figure 1. Design of randomization.


Note: N capture number of students, School capture number of schools. IHC “individual health coaching”, HIT “high-intensity train-
ing”, PHT “physical test.”, “including organised sports”.
SCANDINAVIAN JOURNAL OF EDUCATIONAL RESEARCH 485

grades five years before the intervention; random allocation was determined for each stratum, and
randomisation took place at school level. The total sample in 2008 included 1,638 pupils. The
schools were randomly selected for one of the four interventions: (1) “high-intensity training”
(HIT), (2) “physical test” (PHT), (3) “including organised sports” (IOS) and (4) “individual health
coaching” (IHC). The core of all interventions was physical activity. As the IHC was a targeted inter-
vention directed only towards overweight and inactive pupils at seven schools, this intervention will
be analysed separately. Thus, the participants in this group are excluded from the analyses in this
paper, as the results are not comparable with the other interventions. Of the remaining 1,337 pupils,
93 did not participate and we lack information about parents and grades from registers for 87. This
leaves us with 1,157 pupils from 30 schools (86.5% of the total sample in 2008), of which 18 were
randomly selected for intervention schools, and 12 were randomly selected for control schools. After
a period of 24 months, data were re-collected. In 2008 the 1,157 pupils who participated included
283 pupils in HIT, 201 pupils in PHT, 210 pupils in IOS and 463 in the control group (C). Of the
925 pupils participating in the 2010 follow up, 223 were in HIT, 169 in PHT, 162 in IOS and 371 in
C (cf. Figure 1).
The HIT intervention took place exclusively in school. The core element of the intervention was
high intensity and frequency of the physical activity; it implies exercise at about 75–85% or more of
maximum heart rate for 20 minutes twice a week. The training could be either a part of the PE or
included in other subjects. The biological approach, which claims that physical activity increases the
amount of oxygen delivered to the brain, implying increased learning capability, is tested by this
intervention. However, according to Heckman and Carneiro (2003), this intervention might cause
at least a modest “crowding out” effect on the private provision of physical activity. If this is the
case, the expected positive effect might be modest.
The PHT intervention was supposed to examine the potential behavioural consequences of phys-
ical tests. The tests took place in school, but it was left to the pupils to improve their fitness by exer-
cising after school. A series of fitness tests was completed with the aim of making the pupils more
aware of their own fitness, but it did not train them in any way. The intervention took place once
a month during a PE class and lasted 30–45 minutes. According to the psychological perspective,
self-efficacy could be the mechanism behind a positive effect of this intervention. According to Hol-
land and Andre (1987), another explanation might be that learning is a cumulative process, implying
that learning in one area has a reinforcing impact on learning in another area. Regardless of the
underlying pathways, if the intervention does have an effect on school performance, it is expected
to be positive.
The IOS intervention was supposed to be targeted towards inactive pupils in “real life.” In the
actual intervention, the participants were, of course, randomly selected. The IOS offered sports
opportunities and moderate physical activity to pupils by embedding sports in school, expecting
it to have a positive effect on out-of-school physical activity. The core element of the intervention
was moderate intensity and long duration. The intervention took place once a month, replacing
PE classes and lasting 90–135 minutes. According to the sociological approach, this intervention
is expected to affect school performance through social spill-over, which will increase school satis-
faction and school connectedness. However, according to Coleman (1961), it may be expected that
increased participation in PE and organised sports could be a competing activity, taking away time
that could be spent on school activities. Thus, the impact of this intervention could be modest or
even negative.
The “control group” was divided into a genuine control group and a control group that received
sealed personalised letters containing information about their fitness. The control group was struc-
tured in this way to test whether information in itself has an impact on the results. As there were no
significant differences between the two control groups, we treated them as one throughout the rest of
this study.
The data in this study came from three different sources labelled the register data, the test data,
and the survey; these were linked through a unique personal identification number.
486 A. QUINTO ROMANI AND T. B. KLAUSEN

The main focus of this paper is school performance. However, as school performance is only a
secondary outcome, it was not registered in Project 3A in 2008 or in 2010. To measure school per-
formance, grade point average (GPA) was obtained from register data in 2011/2012, when the pupils
were in the 9th form. We have no information about whether the schools did anything to continue
with the intervention after the RCT ended. Thus, one should be cautious about this gap when inter-
preting the results. If possible, the ideal solution would have been to time the RCT so that it ended at
the same time as grades were recorded.
In order to explore whether average GPA conceals important heterogeneity, another outcome is
GPA within specific subjects such as reading, mathematics, the sciences, and languages. These are
categories resembling those used in international test score comparisons like PISA.
For 2008, GPA was obtained at school level. The school-level GPA was computed from the mean
grades of primary school leavers (9th form) at each school in 2008, whereas the 2011/2012 GPA was
that of individuals participating in the RCT. This solution was the only one possible in order to
obtain a GPA measure in 2008, which was necessary to construct a difference measure in the ana-
lyses. It is far from a perfect solution but, according to Dills et al. (2011), it is better than conducting
the analyses without a baseline measure. Even if we had thought of the possibility of studying school
performance at the beginning of the experimental period, we could not have had an individual base-
line measure of school performance simply because most Danish 6th formers have not yet received
any grades at all (most of them get their first grades in their 7th form). The average GPA is a
weighted average of the grades -3, 0, 2, 4, 7, 10, and 12, which constitute the Danish grading
scale, which have the same number of categories as in the European Credit Transfer System grading
scale; 2 and above are pass grades.
Moreover, the register data include parental characteristics such as ethnicity, highest completed
education, and income. This makes it possible to adjust the models for these observable
characteristics.
The test data include information on physical fitness and weight, which were primary outcomes
in Project 3A, implying that information was obtained in 2008 and in 2010 (Quinto Romani, 2014).
We constructed the binary indicator “active,” which is equal to 1 if the physical fitness of the pupil in
2008 was above mean, and 0 if not. We also constructed the binary indicator “increased fitness,”
which is a dummy variable equalling 1 if the pupil did improve their physical fitness from 2008
till 2010, and 0 if not. Changes in fitness is measured in log (ranging from −0.4698 to 0.5308),
and any increase between 2008 and 2010 will give the pupil a value of 1 on the dummy-variable.
In Table 1 we apply the categories good fitness, bad fitness, and very bad fitness, which are age-
and gender-specific cutting points developed by Åstrand (1952) and further supplemented by
data from Andersen, Henckel, and Saltin (1987) and Wedderkopp, Froberg, Hansen, and Andersen
(2004).
The survey data include information on eating and exercise habits. This will be used to try to
uncover any countervailing behaviour. We captured the pupils’ participation in physical activity out-
side of school through dummy variables. The intensity of after-school physical activity was measured
by a question addressing leisure time activities; “intensively active” is a dummy equal to 1 if a pupil
chooses the answer “I practice sports often during the week and I practice very hard,” and 0 if not.
“Sedentary activity” is a dummy equal to 1 if a pupil chooses the answer “ I spend most of my time
watching television, playing computer games, listening to music or other seated activities,” and equal
to 0 if not.
Test data and survey-data were collected from 20 August to 19 November in both 2008 and 2010
under almost identical circumstances, that is, the same date, time, and place, both before and after
the intervention. The data were collected with permission from the Danish Data Protection Agency.
Any publications based on these data protect the anonymity of individuals and schools.
Figure 2 gives an overview of the data in this paper. The total population included 1,638 pupils in
2008 and 1,369 in 2010. This original sample has been reduced by four different processes. First, the
targeted intervention IHC was, as mentioned above, left out as this was not a randomly selected
SCANDINAVIAN JOURNAL OF EDUCATIONAL RESEARCH 487

Table 1. Descriptive statistics.


I C HIT PHT IOS
Outcomes
GPA 6.3308*** 6.4624 6.2266*** 6.6009* 6.1925***
(0.7441) (0.8124) (0.5274) (0.8729) (0.7826)
Examination GPA 6.3026 6.3689 6.1090*** 6.5984*** 6.2606
(0.7598) (0.7147) (0.5540) (0.9103) (0.7411)
Assessment GPA 6.3472*** 6.5171 6.2944*** 6.6023 6.1536***
(0.7621) (0.8849) (0.5545) (0.8631) (0.8269)
Reading grades 6.0863*** 6.2603 5.9446*** 6.5149*** 5.8343***
(0.7499) (0.8280) (0.5687) (0.7697) (0.7634)
Mathematics grades 6.7818*** 7.1879 6.6362*** 6.8992*** 6.8597***
(1.0340) (0.9639) (0.7020) (1.1788) (1.2217)
Science grades 6.2977** 6.1920 6.0139*** 6.6560*** 6.3147
(0.7939) (0.8742) (0.4850) (0.9425) (0.8167)
Language grades 6.4848*** 6.7017 6.4671*** 6.6873 6.2979***
(0.9415) (0.8963) (0.6896) (1.0984) (1.0253)
Fitness 44.5516*** 46.0578 44.3061*** 43.8677*** 45.6331
(7.5158) (7.7388) (7.8042) (7.3619) (7.1770)
Good fitness 0.0678** 0.1156 0.0657** 0.0641* 0.0748
(0.2516) (0.3202) (0.2483) (0.2457) (0.2640)
Bad fitness 0.3100 0.2687 0.3380* 0.3333 0.2448
(0.4629) (0.4439) (0.4741) (0.4729) (0.4314)
Very bad fitness 0.1259 0.1213 0.1408 0.1410 0.0884
(0.3321) (0.3270) (0.3486) (0.3491) (0.2848)
Parental characteristics
Income very low 0.1877* 0.1428 0.1390 0.2130** 0.2283**
(0.3908) (0.3503) (0.3467) (0.4106) (0.4211)
Income low 0.2075* 0.1590 0.2242** 0.2248** 0.1666
(0.4059) (0.3661) (0.4180) (0.4187) (0.3738)
Income median 0.1805 0.1913 0.1883 0.1834 0.1666
(0.3849) (0.3939) (0.3918) (0.3881) (0.3738)
Income high 0.1931 0.1994 0.2286 0.1656 0.1728
(0.3951) (0.4001) (0.4209) (0.3728) (0.3792)
Income very high 0.2310*** 0.3072 0.2197** 0.2130*** 0.2654
(0.4218) (0.4619) (0.4149) (0.4106) (0.4429)
Immigrant 0.0776 0.0512 0.0582 0.1124*** 0.0679
(0.2678) (0.2207) (0.2348) (0.3168) (0.2523)
Low education 0.1823 0.1455 0.1883 0.1893 0.1666
(0.3864) (0.3531) (0.3918) (0.3929) (0.3738)
Median education 0.4386 0.4097 0.4708 0.3964 0.4382
(0.4966) (0.4924) (0.5002) (0.4906) (0.4977)
High education 0.3790** 0.4447 0.3408*** 0.4142 0.3950
(0.4855) (0.4976) (0.4750) (0.4940) (0.4903)
N 554 371 223 169 162
Note: Present means for the sample in 2008. Standard deviations in parentheses.
*,**, ***indicate that the difference in means for the control and intervention schools is statistically significant at 10, 5, and 1%
levels, respectively. GPA = grade point average. I = a dummy equal to one if the school was selected for any intervention.
C = a dummy equal to one if the school was selected for the control. HIT = a dummy equal to one if the school was selected
for the intervention “high-intensity training.” PHT = a dummy equal to one if the school was selected for the intervention
“physical test.” IOS = a dummy equal to one if the school was selected for the intervention “including organised sports.”

group (Quinto Romani, 2014). This reduced the 2008 sample to 1,337 pupils and the 2010 sample to
1,122 pupils. Second, in 2008, the survey response rate was 91.5%, reducing the 2008 sample to 1,244
pupils. The response rate in 2010 was 88.1%, reducing the 2010 sample to 988 pupils. Third, to be
able to compare 2008 with 2010, only pupils for whom the information was available in both years is
included, reducing the sample to 981 pupils. Finally, as time differences occurred in the collection of
the survey data and the register data, it was not possible to obtain a perfect match. Therefore the
sample size was reduced to 925 pupils.
Those of the pupils in this sample who transferred to another school within the Municipality
during the RCT are allocated to the school they attended in the first wave, whereas pupils who
488
A. QUINTO ROMANI AND T. B. KLAUSEN
Figure 2. Data flows.
Notes: N capture number of students, School capture number of schools. IHC “individual health coaching”.
SCANDINAVIAN JOURNAL OF EDUCATIONAL RESEARCH 489

transfer to another school outside of the Municipality are left out of the analyses. The data have been
tested for attrition bias by the use of a simple t-test. Results indicate that attrition might be influen-
cing the results, as the pupils who left the sample are in general more disadvantaged than the pupils
who stayed in the sample. Hence, it may be a problem that 215 pupils included in the first wave were
not present in the second wave. This is due to the fact that pupils moving to another school outside of
the Municipality were more disadvantaged than their peers, who either did not change schools or
moved to another school within the Municipality. As regards missing values in the control variables,
they are very few and are dealt with by imputing 0 for the missing values and creating dummies to
control for the possibility of non-random imputation.

Methods
When analysing the relationship between intervention and school performance, randomisation into
interventions makes it possible to approach the selection bias, which enables us to address a causal
effect of physical activity on school performance. The simple differences model can be described as:
Ais = b0 + rIs + 1is (1)
Ais denotes the school performance of pupil i, attending school s, in 2011/2012. The impact is
measured by using the difference in means between pupils attending schools receiving intervention
and controls r = (h1 − h0 ); hI defines the mean performance in group I, where I indicates whether
schools receiving the intervention (I= 1) or not (I= 0); 1is is the idiosyncratic error term attempting
to capture any difference between the intervention and the control schools that might occur because
of unobserved confounding variables. The assumption that makes the model valid is that no differ-
ences exist between the intervention and control schools besides the intervention. This assumption is
very strong and rarely met. Thus, an estimation strategy with weaker assumptions is preferable. One
possible alternative strategy is to use a difference-in-differences model, which takes advantage of the
panel structure of data:
(Aist − Ast−1 ) = b0 + tIs + (1ist − 1st−1 ) (2)
Aist denotes the school performance of pupil i, attending school s, in 2011/2012. Ast−1 is school
performance of school s in 2008. The effect is measured by using the difference-in-differences esti-
mator t = (h11 − h10 ) − (h01 − h00 ), which measures the effect of the interventions when con-
trolled for any prior differences. (1ist − 1st−1 ) are the differences in the idiosyncratic error term
(Quinto Romani, 2014).
This model is based on the assumption that any changes between the intervention and control
group are the results of the intervention, implying that the trend in GPA is identical for the inter-
vention and control schools. Besides taking into account the selection bias, the model addresses
unobserved characteristics that are fixed over time, but the model does not take into account unob-
served characteristics that are not fixed over time, for example self-efficacy. When exposed to inter-
vention, self-efficacy may increase. However, if pupils are not perfectly randomised into intervention
and control groups, and if self-efficacy is related to socioeconomic background, a “self-efficacy bias”
might occur.
Table 1 provides descriptive statistics obtained in 2008 across different groups, indicating that
randomisation was not perfect. Columns 1 and 2 present the descriptive statistics of pupils selected
for intervention or control schools, whereas columns 3 to 5 display the descriptive statistics of pupils
selected for a specific intervention. The results indicate that the sample was imbalanced at baseline,
implying that, despite applying a cluster RCT, the randomisation did not work as intended. On aver-
age, the pupils attending the intervention schools have a lower socioeconomic background than the
pupils attending the control schools.
490 A. QUINTO ROMANI AND T. B. KLAUSEN

A plausible explanation for the imperfect randomisation is that randomisation took place at the
school level and not at an individual level. Hence, randomising at a few numbers of schools makes it
unlikely that any differences between the groups will level out. We will approach this imperfect ran-
domisation by controlling for parental socioeconomic status in 2008, implying that the models we
use will be adjusted simple difference and adjusted difference-in-differences models.
In 2011/2012, grades were collected at individual level after the experimental period, while – as
stated in the data section – school average GPA was used to capture baseline GPA in 2008. Thus,
it is important to be certain that 2008 is a good baseline year. Appendix B presents Table B1, showing
the results from analyses based on two other years (2009 and 2010), which are fairly similar to the
analysis with 2008 as baseline year in Table 3, panel B.
Cluster standard errors were applied in all models to correct for the correlation between obser-
vations, since pupils in the same school are subject to the same environmental and family back-
ground influences and therefore tend to have correlated GPAs (Angrist & Pischke, 2009).

Empirical Results
Table 2 compares mean pre and post school performance for the pupils exposed to intervention and
control, respectively. Results from the simple comparison of group means, which are normally
shown in RCT, indicate that a significant improvement in GPA occurred between 2008 and 2011/
2012. However, the improvement was largest for the control group for all the GPA estimates, except
for language grades, on which HIT and PHT had a more positive effect. But also within the other
subject areas the HIT group was, with the exception of science and assessment GPA, the best per-
former among the intervention groups and was nearly level with the control group in mathematics
and reading. The fact that the control group at baseline performed better than the intervention
groups combined indicates that the randomisation was not perfect. However, it is worth mentioning
that on most baseline GPA measures the HIT group has a somewhat higher score than the control
group. It is also worth mentioning that the results in Table 2 reveal quite substantial grade inflation
for 9th formers from 2008 to 2011/2012.
Table 3, panel (A), shows the regression results, using the adjusted simple differences when com-
paring mean school performance in 2011/2012 for the pupils exposed to intervention and control.
Panel (B) presents the adjusted difference-in-differences model, which measures the impact of the
intervention when controlled for any prior differences. Both models were adjusted for parental socio-
economic background to improve precision. The results from the two models in Table 3 are fairly
similar in magnitude. Thus, despite the fact that neither of the models is problem-free, the results
are fairly reliable. Throughout the remainder of this paper, only results from the adjusted differ-
ence-in-differences model will be presented, as this model includes a baseline measure.
Column 1 presents the mean effect of the interventions, whereas columns 2 to 4 present the effect
of each of the interventions. In general, the coefficients are small in size and statistically insignificant,
suggesting that school-based interventions approaching physical activity are not associated with a
major improvement in school-performance. Only one result indicates a positive effect: HIT improves
language grades by 0.75 points compared to the control group. This is quite a substantial improve-
ment. Although nearly all other coefficients are positive as regards the interventions, their statistical
insignificance means that the school-based interventions do not have the expected beneficial impact
on school performance.

Robustness Check
It might be argued, however, that the mean values in the regressions above possibly conceal hetero-
geneity, which calls for robustness checks to verify the results. We will approach this by making sep-
arate analyses for both boys and girls and for active and inactive pupils.
SCANDINAVIAN JOURNAL OF EDUCATIONAL RESEARCH 491

Table 2. Pre and post school performance differences in mean. Control and intervention schools.
I C HIT PHT IOS
GPA
Pre 6.3308 6.4624 6.2266 6.6009 6.1925
(0.74415) (0.8124) (0.5274) (0.8729) (0.7826)
Post 6.7764 7.1250 6.7190 7.0770 6.5420
(0.8793) (1.0486) (0.6168) (1.0130) (0.9515)
Dif 0.4456*** 0.6626*** 0.4923*** 0.4760*** 0.3495***
Examination GPA
Pre 6.3026 6.3689 6.1090 6.5984 6.2606
(0.7598) (0.7147) (0.5540) (0.9103) (0.7411)
Post 6.8484 7.1704 6.7757 7.0863 6.7002
(0.9078) (1.0961) (0.7665) (0.9510) (0.9939)
Dif 0.5457*** 0.8014*** 0.6667*** 0.4878*** 0.4396***
Assessment GPA
Pre 6.3472 6.5171 6.2944 6.6023 6.1536
(0.7621) (0.8849) (0.5545) (0.8631) (0.8269)
Post 6.7283 7.0944 6.6825 7.0704 6.4343
(0.9028) (1.0524) (0.5855) (1.0762) (0.9515)
Dif 0.3811*** 0.5772*** 0.3880*** 0.4681*** 0.2807***
Readings grades
Pre 6.0863 6.2603 5.9446 6.5149 5.8343
(0.7499) (0.8280) (0.5687) (0.7697) (0.7634)
Post 6.7355 7.0895 6.6699 7.1009 6.4445
(0.9375) (1.1166) (0.6533) (1.0074) (1.0679)
Dif 0.6491*** 0.8291*** 0.7252*** 0.5860*** 0.6102***
Mathematics grades
Pre 6.7818 7.1879 6.6362 6.8992 6.8597
(1.0340) (0.9639) (0.7020) (1.1788) (1.2217)
Post 6.8918 7.4303 6.8308 7.0015 6.8614
(1.1188) (1.1412) (1.0334) (1.0923) (1.2503)
Dif 0.1100* 0.2423*** 0.1945** 0.1023 0.0016
Sciences grades
Pre 6.2977 6.1920 6.0139 6.6560 6.3147
(0.7939) (0.8742) (0.4850) (0.9425) (0.8167)
Post 6.7843 6.8685 6.4898 7.2908 6.6613
(1.0691) (1.0039) (0.9546) (1.0189) (1.0871)
Dif 0.4865*** 0.6765*** 0.4758*** 0.6348*** 0.3465***
Language grades
Pre 6.4848 6.7017 6.4671 6.6873 6.2979
(0.9415) (0.8963) (0.6896) (1.0984) (1.0253)
Post 7.1774 7.1586 7.4125 7.3409 6.6833
(1.0307) (1.2711) (0.6670) (1.1938) (1.0947)
Dif 0.6926*** 0.4569*** 0.9454*** 0.6536*** 0.3853***
N 554 371 223 169 162
Note: Present mean GPA for the sample in 2008 and 2011/2012 as well as mean differences. Standard deviation in parentheses.
Simple differences for intervention and control schools according to the different grades outcome.
*, **, *** indicate that the coefficients are statistically significant at 10, 5, and 1% levels, respectively. I = a dummy equal to one if
the school was selected for any intervention. C = a dummy equal to one if the school was selected for the control. GPA = Grade
Point Average. Dif = mean differences. HIT = a dummy equal to one if the school was selected for the intervention “high-intensity
training.” PHT = a dummy equal to one if the school was selected for the intervention “physical test.” IOS = a dummy equal to one
if the school was selected for the intervention “including organised sports.”

The main reason to check for a gender-specific outcome is the widely held view that boys could
benefit more from physical activity than girls. As shown in Table 4, substantial gender heterogeneity
is in fact present in the current data. On average, there is no significant mean effect of the interven-
tions between girls or boys. However, the estimates are more positive among the girls. The effect of
HIT on language grades is also quite similar within the two genders. Language grades for girls and
boys improved by 0.77 and 0.72 points, respectively. However, the improvement is only statistically
significant for girls. That which constitutes the biggest difference among the two genders is the effect
of the exposure to the IOS intervention. For girls, all coefficients are positive. However, they are not
statistically significant. For boys, nearly all coefficients are negative, and about half of them are
492 A. QUINTO ROMANI AND T. B. KLAUSEN

Table 3. School performance.


I HIT PHT IOS N
Panel A. Adjusted simple differences
GPA −0.0565 −0.1112 0.3270 −0.3745 925
(0.2229) (0.2378) (0.2680) (0.2329)
Examination GPA −0.0702 −0.1259 0.2513 −0.3237 924
(0.2178) (0.2733) (0.2306) (0.2285)
Assessment GPA −0.0687 −0.1204 0.3620 −0.4368* 923
(0.2298) (0.2395) (0.2923) (0.2372)
Reading grades −0.1030 −0.1621 0.3289 −0.4647 925
(0.2509) (0.2714) (0.2831) (0.3014)
Mathematics grades −0.2375 −0.3065 0.0071 −0.3939* 925
(0.2406) (0.3605) (0.2812) (0.2090)
Science grades 0.2404 −0.0199 0.7832*** 0.0471 923
(0.2568) (0.3690) (0.2415) (0.2265)
Language grades 0.3672 0.5653 0.6628* −0.2070 922
(0.3363) (0.3696) (0.3572) (0.3708)
Panel B. Adjusted difference-in-differences
GPA 0.0428 0.0903 0.1457 −0.1275 925
(0.2198) (0.2555) (0.3122) (0.2810)
Examination GPA −0.0390 0.0977 −0.0255 −0.2402 924
(0.2119) (0.2697) (0.3119) (0.2670)
Assessment GPA 0.0706 0.0693 0.2352 −0.0951 923
(0.2339) (0.2637) (0.3307) (0.2909)
Reading grades 0.0405 0.1197 0.0333 −0.0603 925
(0.2104) (0.2564) (0.2742) (0.3460)
Mathematics grades 0.1293 0.2051 0.2422 −0.0898 925
(0.3284) (0.3964) (0.4356) (0.3988)
Science grades 0.1070 0.1402 0.2828 −0.1177 921
(0.2581) (0.3619) (0.3443) (0.2346)
Language grades 0.5345* 0.7496* 0.6060 0.1649 921
(0.3041) (0.3834) (0.3818) (0.3514)
Note: Panel A presents estimated effect using the adjusted simple differences model. Panel B presents estimated effects using the
adjusted difference-in-differences regression controlled for parental income, education and ethnicity. Each row represents sep-
arate regressions. Robust standard errors corrected for school cluster are in parentheses.
*, *** indicate that the coefficients are statistically significant at 10 and 1% levels, respectively. GPA = Grade Point Average. I = a
dummy equal to one if the school was selected for any intervention. C = a dummy equal to one if the school was selected for the
control. HIT = a dummy equal to one if the school was selected for the intervention “high-intensity training.” PHT = a dummy
equal to one if the school was selected for the intervention “physical test.” IOS = a dummy equal to one if the school was selected
for the intervention “including organised sports.”

statistically significant. The average grade score is reduced by 0.61 compared to the control group,
examination GPA is reduced by 0.77 and assessment GPA by 0.56. Further, reading grades were
reduced by 0.61 more than in the control group. A possible explanation, in line with Coleman’s
(1961) “zero-sum thesis,” could be that boys increase their participation in sports or PE at the
expense of time and energy available for academic purposes. This should be further investigated
in a follow-up analysis.
The other robustness check investigates whether the impact of the interventions depends on the
physical fitness level of the pupils at baseline. On the one hand, inactive pupils could benefit the most
as they were at the lowest level from the outset. On the other hand, the active pupils could benefit the
most as they are most familiar with and positive towards physical activity. However, it is also possible
that the active pupils did not benefit because they already were very active in their leisure time.
The results presented in Table 5 indicate that HIT had a beneficial impact on school performance,
especially on the pupils who were inactive at baseline. Language grades increased by 1.34 points. In
contrast, PHT had a mainly positive effect on school performance among the pupils who were active
at baseline, but only the result for language grades is statistically significant. Language grades
increased by 0.74 points. Finally, IOS had a negative effect on the pupils who were active at baseline
where GPA was reduced by about 0.5 grade points and a positive effect on the pupils who were inac-
tive at baseline, for whom language grades increased by 0.79 grade points.
SCANDINAVIAN JOURNAL OF EDUCATIONAL RESEARCH 493

Table 4. Difference-in-differences. Gender.


I HIT PHT IOS N
Panel A. Female
GPA 0.1305 0.1904 −0.0451 0.2581 441
(0.3156) (0.3540) (0.3531) (0.4673)
Examination GPA 0.0782 0.2181 −0.1737 0.1906 441
(0.3107) (0.3774) (0.3553) (0.4470)
Assessment GPA 0.1721 0.1916 0.0531 0.2843 440
(0.3298) (0.3627) (0.3665) (0.4911)
Reading grades 0.1248 0.2677 −0.2239 0.3463 441
(0.2827) (0.2938) (0.2979) (0.4896)
Mathematics grades 0.3163 0.1783 0.2535 0.5696 441
(0.4276) (0.5012) (0.5118) (0.4996)
Science grades 0.2410 0.2308 0.2287 0.2684 440
(0.3743) (0.4656) (0.4307) (0.4229)
Language grades 0.5012 0.7709* 0.3917 0.2619 440
(0.3595) (0.4084) (0.3531) (0.5269)
Panel B. Male
GPA −0.1414 −0.0209 0.1698 −0.6124** 483
(0.2708) (0.3135) (0.3676) (0.2463)
Examination GPA −0.2388 −0.0187 −0.0277 −0.7671*** 482
(0.2586) (0.3103) (0.3610) (0.2331)
Assessment GPA −0.1183 −0.0607 0.2628 −0.5588** 482
(0.2897) (0.3297) (0.3947) (0.2683)
Reading grades −0.1861 −0.0441 0.0360 −0.6070** 483
(0.2597) (0.3456) (0.3196) (0.2353)
Mathematics grades −0.0627 0.2302 0.2119 −0.7607 483
(0.4047) (0.4404) (0.5567) (0.4920)
Science grades −0.0671 0.0361 0.2783 −0.5444 480
(0.3362) (0.4328) (0.4210) (0.3454)
Language grades 0.4590 0.7226 0.6205 −0.0839 480
(0.3636) (0.4581) (0.4982) (0.3283)
Notes: The unstandardised regression coefficients present adjusted difference-in-differences regression controlled for parental
income, education and ethnicity. Each row represents separate regressions. Robust standard errors corrected for school cluster
are in parentheses. Panel A presents estimated effect of the intervention for females. Panel B presents the estimated effect of the
interventions for males.
*, **, ***indicate that the coefficients are statistically significant at 10, 5 and 1% levels, respectively. GPA = grade point average. I a
dummy equal to one if the school was selected for any intervention. C = a dummy equal to one if the school was selected for the
control. HIT = a dummy equal to one if the school was selected for the intervention “high-intensity training.” PHT = a dummy
equal to one if the school was selected for the intervention “physical test.” IOS = a dummy equal to one if the school was selected
for the intervention “including organised sports.”

The overall results indicate a somewhat heterogeneous response to the school-based intervention.
The few significant results within the HIT and IOS groups were found among the pupils who were
inactive at baseline; especially their language grades increased substantially. However, the positive
results within the PHT group are found among the pupils who were active at baseline, who improved
their language grades quite substantially.

Specification Check
Having found little causal effect of the interventions on the pupils’ school performance, this section
reports the results from a number of specification checks that have been made to address two poten-
tial concerns regarding intervention studies. First, pupils may vary in their “take-up” of the interven-
tion, meaning that not all pupils selected for the intervention may actually be participating. Second,
“crowding out” of the private provision of physical activity could be a possibility, with no improve-
ment in school performance as a result despite an increase in in-school physical activity.
A heterogeneous take-up rate may occur if, despite being mandatory, participation in physical
exercise is partly voluntary. If some pupils opt out from the RCT, the group receiving the assigned
494 A. QUINTO ROMANI AND T. B. KLAUSEN

Table 5. Difference-in-difference. Pupils’ activity level.


I HIT PHT IOS N
Panel A. Active pupils
GPA −0.1613 −0.0590 0.1758 −0.5380* 477
(0.2564) (0.3128) (0.3709) (0.3007)
Examination GPA −0.1479 0.0277 0.0914 −0.5308* 477
(0.2415) (0.3314) (0.3700) (0.2865)
Assessment GPA −0.1772 −0.1169 0.2202 −0.5537* 477
(0.2805) (0.3163) (0.3980) (0.3206)
Reading grades −0.1422 −0.0821 0.1255 −0.4177 477
(0.2463) (0.3425) (0.3666) (0.3813)
Mathematics grades 0.0783 0.4609 0.1855 −0.4333 477
(0.3874) (0.4394) (0.4727) (0.4656)
Science grades −0.0781 0.1356 0.1753 −0.5142** 476
(0.3135) (0.4532) (0.3842) (0.2555)
Language grades 0.2310 0.3740 0.7349* −0.3110 476
(0.3622) (0.4739) (0.4455) (0.4041)
Panel B. Inactive pupils
GPA 0.2654 0.2756 0.2353 0.2880 400
(0.2601) (0.3087) (0.3481) (0.3850)
Examination GPA 0.1045 0.2236 −0.0038 0.0332 399
(0.2713) (0.3303) (0.3816) (0.3903)
Assessment GPA 0.3163 0.2629 0.3482 0.3719 398
(0.2582) (0.3142) (0.3404) (0.3759)
Reading grades 0.2379 0.3416 0.0497 0.3055 400
(0.2694) (0.3021) (0.3201) (0.4806)
Mathematics grades 0.1716 −0.0282 0.4219 0.1973 400
(0.3564) (0.4572) (0.4930) (0.4286)
Science grades 0.2126 0.1331 0.3700 0.1445 398
(0.3367) (0.4059) (0.4267) (0.4333)
Language grades 0.9717*** 1.3398*** 0.6009 0.7938* 397
(0.3372) (0.4188) (0.4292) (0.4775)
Notes: The unstandardised regression coefficients present adjusted difference-in-differences regression controlled for parental
income, education and ethnicity. Each row represents separate regressions. Robust standard errors corrected for school cluster
are in parentheses. Panel A presents the estimated effect of the intervention for active pupils captured by baseline fitness above
average. Panel B presents the estimated effect of the interventions for inactive pupils captured by baseline fitness at or below
average.
*, **, ***indicate that the coefficients are statistically significant at 10, 5, and 1% levels, respectively. GPA = grade point average. I =
a dummy equal to one if the school was selected for any intervention. C = a dummy equal to one if the school was selected for
the control. HIT = a dummy equal to one if the school was selected for the intervention “high-intensity training.” PHT = a dummy
equal to one if the school was selected for the intervention “physical test.” IOS = a dummy equal to one if the school was selected
for the intervention “including organised sports.”

interventions will be a partly self-selected subset of the pupils who were originally selected for the
intervention. Thus, a comparison between those selected for an intervention and the control
group can be biased, due to variation in the take-up rate.
Following Angrist and Pischke (2009), we have addressed issues regarding take-up rate by using
random allocation into the interventions as an instrumental variable for improvement in fitness.
Unfortunately, the first-stage results were not significant. Thus, lacking better estimates on interven-
tion completion, we are not able to distinguish between those who did and did not participate among
those who were selected for the intervention.
Another method of approaching the take-up issue is to consider the effect of the intervention only
on the pupils improving their physical fitness during the experiment. To investigate this issue, esti-
mation can be made from a difference-in-differences model within the group of those who improved
their fitness and within the group whose fitness did not improve. This strategy is based on the
assumption that improvement in physical fitness in the intervention groups is a good indicator of
full participation in the RCT. As stated above, we cannot be absolutely certain about this. As the
numbers of pupils who did and did not improve their fitness are 406 and 471, respectively, this
assumption appears to be too strong. Thus, variation in take-up rate is only one explanation.
SCANDINAVIAN JOURNAL OF EDUCATIONAL RESEARCH 495

Other factors may explain the lack of increase in fitness, which cannot be fully measured with the
available data.
Table 6 analyses variation in take-up. Overall, the Table does not reveal many significant results,
but some interesting differences are seen. The intervention HIT did have a positive effect on the
language grades within the increased-fitness group, which increased by an estimated 0.87 grade
points. Nearly all other HIT coefficients are positive in both groups, but none of them are significant.
These results indicate that school performance can be improved by intensive training.
In contrast, the PHT intervention seems to improve academic achievement on languages, even if
the intervention does not improve the pupils’ physical fitness. Grade point average in languages
improves by 0.95 grade points. Bearing in mind that PHT is the most academically oriented of
the interventions, these results could possibly be explained by a cognitive spill-over effect; skills
acquired in one arena could affect learning in another.
Most disappointing are the results for the intervention IOS. For pupils who participated in the
intervention and improved their fitness level, examination GPA and mathematics grades were
reduced by 0.78 and 0.80 grade points, respectively, while all the coefficients within the “decreased
fitness” group were not significant. Coleman’s (1961) “zero-sum thesis” could be a possible

Table 6. Difference-in-differences. Take-up rate.


I HIT PHT IOS N
Panel A. Increased fitness
GPA −0.0385 0.0933 0.0716 −0.5382 406
(0.2460) (0.2588) (0.3720) (0.4117)
Examination GPA −0.1837 0.0305 −0.1424 −0.7820** 405
(0.2385) (0.2841) (0.3171) (0.3548)
Assessment GPA 0.0268 0.1164 0.1767 −0.4292 406
(0.2603) (0.2613) (0.4228) (0.4372)
Reading grades −0.0615 0.0690 −0.0810 −0.3565 406
(0.2436) (0.2886) (0.3435) (0.4544)
Mathematics grades 0.0179 0.2466 0.1793 −0.8020** 406
(0.2947) (0.3179) (0.4087) (0.3878)
Science grades 0.0072 0.1569 0.1804 −0.6343 406
(0.3523) (0.4257) (0.4701) (0.4228)
Language grades 0.5194 0.8723** 0.2863 −0.0460 404
(0.3264) (0.3624) (0.4394) (0.3636)
Panel B. Decreased fitness
GPA 0.0840 0.0551 0.2382 −0.0100 471
(0.2502) (0.2921) (0.3846) (0.3001)
Examination GPA 0.0844 0.1343 0.1573 −0.0185 471
(0.2426) (0.2934) (0.3995) (0.3028)
Assessment GPA 0.0665 −0.0104 0.2862 −0.0328 469
(0.2680) (0.3195) (0.3972) (0.3062)
Reading grades 0.1095 0.1771 0.1862 −0.0125 471
(0.2544) (0.2978) (0.3561) (0.3919)
Mathematics grades 0.1621 0.0302 0.2926 0.1811 471
(0.4220) (0.5308) (0.5459) (0.4956)
Science grades 0.1136 0.0291 0.3427 0.0130 468
(0.2391) (0.3406) (0.3170) (0.2321)
Language grades 0.6405* 0.7783 0.9473** 0.2765 469
(0.3507) (0.5459) (0.4557) (0.4204)
Notes: The unstandardised regression coefficients present adjusted difference-in-differences regression controlled for parental
income, education and ethnicity. Each row represents separate regressions. Robust standard errors corrected for school cluster
are in parentheses. ‘Increased fitness’ is a dummy equal to one if the pupil had improved their physical fitness from 2008 till 2010
and equal to zero if not. ‘Decreased fitness’ is a dummy equal to one if the pupil did not improved their physical fitness from 2008
till 2010 and equal to zero if they did.
*, ** indicate that the coefficients are statistically significant at 10 and 5% levels, respectively. GPA = grade point average. I = a
dummy equal to one if the school was selected for any intervention. C = a dummy equal to one if the school was selected for the
control. HIT = a dummy equal to one if the school was selected for the intervention “high-intensity training.” PHT = a dummy
equal to one if the school was selected for the intervention “physical test.” IOS = a dummy equal to one if the school was selected
for the intervention “including organised sports.”
496 A. QUINTO ROMANI AND T. B. KLAUSEN

explanation for this, as the intervention might increase pupils’ participation in sports or PE, but the
increasing physical activity might take away time and energy available for academic purposes.
Crowding out arises if the increased physical activity in schools causes a decrease in leisure-time
physical activity. Thus, the success of school-based interventions essentially depends on the extent to
which the increased in-school public supply of physical activity replaces the existing after-school pri-
vate supply. If the expansion in in-school public supply of physical activity simply induces already
active pupils to switch from private to public provision, it may have no impact on their school
performance.
Table 7 explores to what extent the crowding out thesis is a plausible explanation for the disap-
pointing results in this paper. The Table shows two statistically significant results: There was an
increase in seated activities within the HIT intervention with a coefficient of 0.08 (p < 0.05) and
within the IOS intervention the amount of intensively active decreased by 0.12 (p < 0.1). Further,
it is worth mentioning that also within HIT there is a negative coefficient for intensively active, con-
tributing to a statistically significant coefficient for the intervention-groups combined of approxi-
mately −0.06. Thus, the results confirm that increasing in-school physical activity seems to be
replacing after-school physical exercise.

Conclusion and Discussion


Taking different theoretical perspectives on physical activity and school performance as a point of
departure, this paper has investigated the relationship between physical activity and school perform-
ance, using data from a randomised school-based intervention study.
Most results in this paper reveal no significant association between physically demanding inter-
ventions and school performance, that is, the interventions did not have an overall effect on the
pupils’ performance.
The only positive overall result was that HIT did improve language grades by 0.75 points. This
approximately resembles a difference on the European Credit Transfer System grading scale of
just below half a grade, which is quite a substantial difference.
When making separate analyses for boys and girls, there were no big differences between the gen-
ders within the HIT intervention. Language grades for girls and boys improved by 0.77 and 0.72
points, respectively. However, it was only statistical significant for girls. In contrast, the IOS inter-
vention had a negative impact on school performance among boys, but not girls. More specifically,
examination grades and reading grades were reduced by 0.61 points among boys.
We also found some variation in intervention effects across the baseline activity distribution.
High-intensity training did mainly have an effect on school performance among the pupils that

Table 7. Pre and post differences by control and intervention school. Crowding out.
I C HIT PHT IOS
Sedentary activity 0.0270 0.0054 0.0804** −0.0292 0.0122
(0.0181) (0.0209) (0.0227) (0.0328) (0.0300)
Intensively active −0.0579* 0.0673 −0.0805 0.0297 −0.1175*
(0.0321) (0.0444) (0.0486) (0.0554) (0.0550)
N 554 371 223 169 162
Notes: The unstandardised regression coefficients present adjusted simple differences regression controlled for parental income,
education, and ethnicity. Each row represents separate regressions. Robust standard errors corrected for school cluster are in
parentheses. “Intensively active” is a dummy equal to one if the pupil practices sports vigorously often during the week and
equal to zero if not. “Sedentary activity” is a dummy equal to one if the pupil spends most of their leisure time watching tele-
vision, playing computer games, listening to music or in other seated activities and equal to zero if not.
*, **, indicate that the coefficients are statistically significant at 10 and 5% levels, respectively. GPA = grade point average. I = a
dummy equal to one if the school was selected for any intervention. C = a dummy equal to one if the school was selected for the
control. HIT = a dummy equal to one if the school was selected for the intervention “high-intensity training.” PHT = a dummy
equal to one if the school was selected for the intervention “physical test.” IOS = a dummy equal to one if the school was selected
for the intervention “including organised sports.”
SCANDINAVIAN JOURNAL OF EDUCATIONAL RESEARCH 497

were inactive at the beginning of the experimental period. Language grades increased by 1.34 points
within this sub-group. Including organised sports increased language grades by 0.79 points among
the pupils who were inactive at baseline. Physical test, however, had a positive effect among the
pupils who were active at the beginning of the experimental period; their language grades improved
by 0.73 points.
Further analyses show that variation in take-up and crowding out might be plausible explanations
of the mostly disappointing results. When considering variation in take-up, our results reveal that
among pupils who improved their fitness, HIT had a positive effect in that language grades improved
by 0.87 points. Including organised sports had a negative effect, however, as examination GPA and
mathematics grades decreased by 0.78 and 0.80 points, respectively. In contrast, when considering
pupils who did not improve their fitness, PHT had a beneficial impact in that their language grades
improved by 0.95 points. Regarding crowding out, the results revealed that the HIT increased the
likelihood of pupils being seated in their leisure time and IOS reduced the likelihood of pupils
being intensely active in their leisure time. Thus, there seems to be a tendency towards substitution
rather than accumulation of physically demanding activities. This finding should attract the atten-
tion of researchers wanting to study the impact of in-school physical activity.
It is difficult to make a direct comparison of our results with the results from earlier studies within
this research area as differences in design, duration, and so on will certainly produce different results.
However, a more indirect comparison is possible. Recent empirical work has found that physical
activity is positive in relation to school performance (Dwyer et al., 2001; Field et al., 2001; Nelson &
Gordon-Larsen, 2006). In line with Rees and Sabia (2010), our analyses on take-up rate and crowd-
ing out indicate that these positive correlations might partly be driven by unmeasured heterogeneity
and not purely by causal relations. Thus, generally speaking, our modest results seem to be more in
accordance with earlier research using econometric methods than with cross-sectional studies and
RCT. This suggests that the positive academic spill-over associated with physical activity might be
overstated, implying that political interventions designed to encourage physical activity may not
have the expected positive effect on the pupils’ school performance.
This comes as no surprise, as we argued in favour of method dependency in the results of earlier
research in our introduction. Thus, we recommend the use of econometric methods within RCT fra-
meworks (such as our own) in order to solve not only the selection bias problem, but also to address
imperfect randomisation and possible omitted variable bias.

Limitations
Although we find little evidence to support the claim that physical activity is positively related to
school performance in a causal sense, our results do not entirely rule out that differences in intensity,
frequency, and duration of physical activity could be possible explanations as to why our results
diverge from other RCT that find a positive causal relation between physical activity and school per-
formance (Donnelley et al., 2009, 2013; Ericsson & Karlsson, 2014; Shephard et al., 1984). Compared
to Ericsson and Karlsson’s (2014) nine-year intervention study, our 24-month-long RCT was short.
Even with a far shorter experimental period of three years, Donnelley et al.’s (2009) study was sub-
stantially longer than ours. The fact that with an experimental period of about the same length as
ours, Sallis et al. (1999) did not reveal any positive effect strengthens the argument that duration
might be an issue when comparing results from different experimental studies. Thus, it seems likely
that duration might be a part of the explanation of the diverging results.
As regards intensity and frequency, we believe that the quality of our RCT compares successfully
with that of other research. The HIT-group received high-intensity training for 20 minutes twice a
week during the intervention and the other two interventions were fairly frequent too. The fact that
we include an intervention of both fairly high frequency and high intensity enables specification of
the effect of this in a more precise manner than has been the case in earlier research; in several cases,
498 A. QUINTO ROMANI AND T. B. KLAUSEN

frequency has been captured by adding a number of PE lessons to the intervention group, but the
intensity of these lessons has not been specified.
When it comes to the content of our interventions, they could have been better designed. The
interventions used in this RCT were originally designed to investigate the association between phys-
ical activity and childhood obesity. It might be argued that despite its policy relevance, such an
approach does not necessarily contribute with sufficient knowledge on the exact mechanisms linking
physical activities and school performance. This may be accomplished in several ways, but using rig-
orously designed interventions capturing the different pathways through which physical activity can
affect school performance is crucial.
Finally, the gap between the completion of the RCT and collection of GPA might explain the
rather disappointing effect of physical exercise on school performance. It could be argued that a fairly
long and intensive RCT does have long-term effects. However, the ideal solution would have been to
collect information about school performance immediately after the end of the RCT and compare
with grades from the registers at the end of form 9.
Regardless of the limitations, our findings add new evidence to the debate about the role of
schools as a promoter of physical fitness in order to improve school performance. One major goal
of the Danish 2014 school reform was to establish an integrated school day, meaning that a large
share of physical activity could be transferred from the private to the public sphere. This has resulted
in longer school days of 35 lessons or more per week, where two or three PE lessons per week are
supplemented by 45 minutes of daily exercise as part of the integrated school day. Even if some
of our results suggest that differentiation of physical education might have a beneficial impact on
school performance, attention should be paid to the possible crowding out effect the school initiat-
ives might have on leisure time activity, and the possible negative impact it might have on both phys-
ical fitness and school performance. However, among pupils not participating in organised sports,
the reform might be levelling the playing field by benefitting the more inactive pupils in particular.

Acknowledgements
The Danish Gymnastics and Sports Associations in Northern Jutland and the Danish Municipality of Aalborg have
contributed with data collection.

Disclosure statement
No potential conflict of interest was reported by the author.

Funding information
The article was funded by a grant for the research project “Structural Factors in Relation to Youth Behaviour, Lifestyle,
and Health” from the Danish Council for Independent Research [grant number 12-127741].
Project 3A, the randomized intervention study, was funded by Aalborg University, the Danish Health Insurance
Foundation [grant number 2008A003], the Danish TrygFonden Foundation [grant number 5556-07], and the Danish
Obel Family Foundation [grant number 12664]. The Danish Gymnastics and Sports Associations (DGI) in Northern
Jutland and the Danish Municipality of Aalborg.

References
Andersen, L. B., Henckel, P. & Saltin, B. (1987). Maximal oxygen uptake in Danish adolescents 16–19 years of age.
European Journal of Applied Physiology, 56, 74–82.
Angrist, J. D., & Pischke, J. (2009). Mostly harmless econometrics. An empiricist’s companion. Princeton, NJ: Princeton
University Press.
Åstrand, P. O. (1952). Experimental studies of physical working in relation to sex and age. Copenhagen: Munksgaard.
Bandura, A. (1997). Self-efficacy. The exercise of control. New York: Freeman.
SCANDINAVIAN JOURNAL OF EDUCATIONAL RESEARCH 499

Cawley, J., Frisvold, D., & Meyerhoefer, C. (2013). The impact of physical education on obesity among elementary
school children. Journal of Health Economics, 32, 743–755.
Coe, D. P., Pivarnik, J. M., Womack, C. J., Reeves, M. J., & Malina, R. M. (2006). Effect of physical education and
activity levels on academic achievement in children. Medicine & Science in Sports & Exercise, 38, 1515–1519.
Coleman, J. (1961). The adolescent society: The social life of the teenager and its impact on education. New York, NY:
The Free Press.
Cratty, B. (1997). Coordination problems among learning disabled. In B. Cratty & R. Goldman (Eds.), Learning dis-
abilities, contemporary viewpoints (pp. 143–183). Amsterdam: Harwood Academic Publishers.
Daley, A. J., & Ryan, J. (2000). Academic performance and participation in physical activity by secondary adolescents.
Perceptual and Motor Skills, 91, 531–534.
Danish Evaluation Institute. (2004). PE in schools. A subject in motion [Idræt i folkeskolen. Et fag med bevægelse].
Copenhagen: Danish Evaluation Institute.
Dills, A. K., Morgan, H. N. & Rotthoff, K. W. (2011). Recess, physical education, and elementary school student out-
come. Economics of Education Review, 30, 889–900.
Dollman, J., Boshoff, K. & Dodd, G. (2006). The relationship between curriculum time for physical education and lit-
eracy and numeracy standards in South Australian primary schools. European Physical Education Review, 12, 151–
163.
Donnelley, J. E., Greene, J. L., Gibson, C. A., Smith, B. K., Washburn, R. A., Sullivan, D. K., … . Williams, S. L. (2009).
Physical activity across the curriculum (PAAC): A randomized controlled trail to promote physical activity and
diminish overweight and obesity in elementary school children. Preventive Medicine, 49, 336–341.
Donnelley, J. E., Greene, J. L., Gibson, C. A., Sullivan, D. K., Hansen, D. M., Hillman, C. H., … Washburn, R. A. (2013).
Physical activity and academic achievement across the curriculum (A + PAAC): rationale and design of a 3-year,
cluster-randomized trial. BMC Public Health, 13, 1–8.
Dwyer, T., Sallis, J. F., Blizzard, L., Lazarus, R., & Dean, K. (2001). Relation of academic performance to physical
activity and fitness in children. Paediatric Exercise Science, 13, 225–238.
Ericsson, I. & Karlsson, M. K. (2014). Daily physical education improves motor skills and school performance. A nine-
year prospective intervention study. Scandinavian Journal of Medicine & Science in Sports, 24, 273–278.
Field, T., Diego, M., & Sanders, C. E. (2001). Exercise is positively related to adolescents’ relationships and academics.
Adolescence, 36, 105–110.
Gjesing, G. (1997). Kropumulige Unger. Köpenhamn: DHLs förlag.
Hanks, M. & Eckland, B. (1976). Athletics and social participation in the educational attainment process. Sociology of
Education, 49, 271–294.
Heckman, J. J. & Carneiro, P. (2003). Human capital policy. In J. J. Heckman, A. B. Krueger, & B. M. Friedman (Eds.),
Inequality in America: What role for human capital policies? (pp. 77–239). Cambridge, MA: MIT Press.
Heckman, J. J., Jeffrey, A. & Smith, J. A. (1995). Assessing the case for social experiments. Journal of Economic
Perspectives, 9, 85–110.
Holland, A., & Andre, T. (1987). The effect of participation in extracurricular activities in secondary school: What is
known, what needs to be known? Review of Education Research, 57, 437–466.
Jensen, E. (1998). Teaching with the brain in mind. Alexandria, VA: Association for Supervision and Curriculum
Development.
Laub, T. B., & Pilgaard, M. (2013). Danskernes motions- og sportsvaner 2011. (Danes’ exercise and sports habits 2011).
Copenhagen, Denmark: Idrættens Analyseinstitut.
Nelson, M. C., & Gordon-Larsen, P. (2006). Physical activity and sedentary behavior patterns are associated with
selected adolescent health risk behaviors. Pediatrics, 117, 1281–1290.
OECD. (2013). PISA 2012 Results in focus. What 15-year-olds know and what they can do with what they know. Paris:
OECD.
Ploughman, M. (2008). Exercise is brain food: The effects of physical activity on cognitive function. Developmental
Neurore habilitation, 11, 236–240.
Quinto Romani, A. (2014). Physical activity and child health: Can school-based intervention make a difference? Nordic
Journal of Health Economics. 2, 41–58.
Rees, D. I., & Sabia, J. J. (2010). Sports participation and academic performance: Evidence from the National
Longitudinal Study of Adolescent Health. Economics of Education Review, 29, 751–759.
Sallis, J., McKenzie, T., Kolody, B., Lewis, M., Marshall, S. & Rosengard, P. (1999). Effects of health-related physical
education on academic achievement: Project SPARK. Research Quarterly for Exercise & Sport, 70, 127–134.
Shephard, R. (1997). Curricular physical activity and academic performance. Paediatric Exercise Science, 9, 113–126.
Shephard, R. J., Volle, M., Lavallée, H., LaBarre, R., Jéquier, J. C., & Rajic, M. (1984). Required physical activity and
academic grades: A controlled longitudinal study. In J. Ilmarinen, & I. Valimaki (Eds.), Children and sport (pp.
58–63). Berlin: Springer Verlag.
Steptoe, A., & Butler, N. (1996). Sports participation and emotional wellbeing in adolescents. Lancet, 347, 1789–1792.
Tomporowski, P. D. (2003). Cognitive and behavioral response to acute exercise in youths: A review. Paediatric
Exercise Science, 15, 348–359.
500 A. QUINTO ROMANI AND T. B. KLAUSEN

Tremblay, M. S., Inman, J. W. & Willms, J. D. (2000). The relationship between physical activity, self-esteem, and aca-
demic achievement in 12-year-old children. Paediatric Exercise Science, 12, 312–324.
Trudeau, F. & Shephard, R. J. (2008). Physical education, school physical activity, school sports and academic perform-
ance. International Journal of Behavioural Nutrition and Physical Activity, 5, 1–12.
Wedderkopp, N., Froberg, K., Hansen, H. S. & Andersen, L. B. (2004). Secular trends in physical fitness and obesity in
Danish 9-year-old girls and boys: Odense School Child Study and Danish substudy of the European Youth Heart
Study. Scandinavian Journal of Medicine and Science in Sport, 14, 150–155.
Appendix A

Table A1. Articles addressing child physical activity (PA) and school performance.
Study Population Design Exogenous variation Regression Results
Cross-sectional studies
Nelson and Gordon- 11,957 pupils Pooled data Controlling confounding variables Poisson regression across clusters Physical activity improves grades
Larsen (2006) NLSAH
7th–12th forms
Field et al. (2001) 89 upper-secondary Cross-sectional Mean for high and low PA Physical activity improves grades
children
Dwyer et al. (2001) 7,961 pupils aged 7–15 Cross-sectional Weighted linear regression Physical activity improves grades
Tremblay et al. 6,923 pupils in 6th form Cross-sectional Controlling confounding variables Odds ratios regression Physical activity has a negative impact on
(2000) grades
Daley and Ryan 232 pupils Cross-sectional Correlations Physical activity has a negative impact on
(2000) grades
Dollman et al. (2006) 117 pupils attending 3rd, Cross-sectional Controlling confounding variables Multiple regression Physical activity has no impact on grades
5th, and 7th form
Experimental studies

SCANDINAVIAN JOURNAL OF EDUCATIONAL RESEARCH


Shephard et al. 546 pupils Quasi randomised Additional PA 260 instead of 40 Increased physical activity has no impact
(1984) controlled trial min per week grades
Ericsson and 251 pupils,1st–3rd forms Randomised Additional PA 225 min per week, Mean change in school performance. Increased physical activity improves grades
Karlsson (2014) controlled trial 9th year Logistic regression
Donnelley et al. 1,527 pupils Cluster randomised Additional PA 90 min per week, 3 Mean change in school performance Increased physical activity improves grades
(2009) controlled trial years controlling for confounding variables
Sallis et al. (1999) 655 pupils 5th and 6th Cluster randomised Additional PA 27–42 min per week, Mean change in school performance Increased physical activity has no impact on
forms controlled trial 2 years percentiles. ANOVA grades
Coe et al. (2006) 214 pupils 6th form Quasi randomised Additional PA 55 min per day, one Mean and standard division Increased physical activity has no impact on
controlled trial semester grades
Econometric methods
Rees and Sabia 11,261 pupils NLSAH Pooled data Econometric methods FE and IV Physical activity has no impact on grades
(2010) 7th–12th forms
Dills et al. (2011) 1–5th grade Natural experiment States and year variation in PE IV Physical activity has no offsetting impact on
ECLS-K 1998-1999 Pooled data mandates (increased school days) academic score
Cawley et al. (2013) 43,430 pupils Natural experiment. States and year variation in PE IV Physical activity reduces obesity in 5th grade
1st–5th forms Pooled data mandates (increased school days) and has no offsetting impact on academic
ECLS-K 1998-2004 score
Notes: ECLS-K = Early Childhood Longitudinal Study, Kindergarten Cohort. FE = Fixed effect and IV = instrument variable. NLSAH = National Longitudinal Study of Adolescent Health.

501
502 A. QUINTO ROMANI AND T. B. KLAUSEN

Appendix B
Table B1. Difference-in-differences, different baselines.
I HIT PHT IOS N
Panel A. Adjusted difference-in-differences baseline 2009
GPA 0.3714* 0.3768 0.7809*** −0.0605 934
(0.2070) (0.2402) (0.2314) (0.2722)
Examination GPA 0.2620 0.3681 0.6337** −0.2722 931
(0.2390) (0.3051) (0.2600) (0.2886)
Assessment GPA 0.4087** 0.3642* 0.8609*** 0.0032 932
(0.2015) (0.2228) (0.2205) (0.2897)
Reading grades 0.3132 0.2436 0.8606*** −0.1584 933
(0.2251) (0.2773) (0.2180) (0.3023)
Mathematics grades 0.0269 0.0495 0.3923 −0.3826 934
(0.2509) (0.3037) (0.2923) (0.2815)
Science grades 0.4106* 0.4879* 0.6503** 0.0591 931
(0.2372) (0.2954) (0.3016) (0.2408)
Language grades 0.9703*** 1.2455** 1.2760*** 0.2754 930
(0.3178) (0.3762) (0.3552) (0.3908)
Panel B. Adjusted difference-in-differences baseline 2010 N
GPA 0.2417 0.2728 0.2966 0.1455 905
(0.1904) (0.2277) (0.2586) (0.3068)
Examination GPA 0.1660 0.1923 0.0783 0.2258 904
(0.2006) (0.2703) (0.2645) (0.3427)
Assessment GPA 0.2728 0.3086 0.4312 0.0626 903
(0.1999) (0.2312) (0.2631) (0.2948)
Reading grades 0.1281 0.1417 0.2348 −0.0010 905
(0.2031) (0.1771) (0.3129) (0.3936)
Mathematics grades 0.1487 0.2142 0.1046 0.1141 905
(0.2653) (0.3998) (0.3663) (0.3747)
Science grades 0.3600 0.2631 0.5710** 0.2600 902
(0.2878) (0.4751) (0.2675) (0.3326)
Language grades 0.8042*** 1.0104*** 0.8051*** 0.5406* 901
(0.2543) (0.2868) (0.2991) (0.3222)
Notes: The unstandardised regression coefficients present adjusted difference-in-differences regression controlled for parental
income, education and ethnicity. Each row represents separate regressions. Panel A presents the estimated effect of the inter-
vention using 2009 as a baseline. Panel B presents the estimated effect of the interventions using 2010 as baseline. Robust stan-
dard errors corrected for school cluster are in parentheses.
*, **, and ***indicate that the coefficients are statistically significant at 10, 5, and 1% levels, respectively. GPA = grade point aver-
age. I = a dummy equal to one if the school was selected for any intervention. C = a dummy equal to one if the school was
selected for the control. HIT = a dummy equal to one if the school was selected for the intervention “high-intensity training.”
PHT = a dummy equal to one if the school was selected for the intervention “physical test.” IOS = a dummy equal to one if
the school was selected for the intervention “including organised sports.”
Copyright of Scandinavian Journal of Educational Research is the property of Routledge and
its content may not be copied or emailed to multiple sites or posted to a listserv without the
copyright holder's express written permission. However, users may print, download, or email
articles for individual use.

You might also like