The ASK study is a parallel group (intervention group vs. control group, 1:1 ratio) cluster-randomized controlled trial conducted in Sogn og Fjordane county, Norway, between August 2014 and June 2015 (Resaland et al., 2015). Randomization was performed by a neutral third part (Centre for Clinical Research, Haukeland University Hospital, Norway) and the unit of randomization was the participating schools. The procedures and methods used in the ASK trial conform to the ethical guidelines defined by the World Medical Association’s Declaration of Helsinki and its subsequent revisions (WMA, 2013). The study protocol was approved by The Regional Committees for Medical Research Ethics (REC southeast). We obtained written informed consent from each child’s parent(s)/guardian(s) prior to all testing.

We invited all 60 schools in the county that fulfilled the inclusion criteria of at least seven fifth-grade children enrolled. All schools, encompassing 1202 fifth-grade children, agreed to participate (Figure 2). Three schools (one control school and two intervention schools) from the same municipality resigned after randomization. In total, 1145 (82.1 % of the population of 10-year-olds in the county) of the 1175 invited children from 57 schools agreed to participate. The present study includes children in the lowest tertile of academic performance in numeracy at baseline. The study was registered in the Clinicaltrials.gov registry [trial registration number: NCT02132494] prior to commencement.

The ASK intervention was a part of the mandatory school curriculum for all children attending the intervention schools. It was led by the classroom teachers and consisted of three components (165 min/week); (1) physically active educational lessons, mostly performed outdoor in the school yard (3 × 30 min/week) in the subjects Norwegian, Mathematics and English, (2) physical activity breaks during classroom lessons (5 min/school-day) and (3) physical activity homework (10 min/school-day). The research group developed the intervention in collaboration with teachers at the intervention schools. Our mutual aim was to create a number of varied physical activity activities that should be carried out in small groups and which encouraged an inclusive and joyful learning environment (examples of physical activity activities¹). Two of the three intervention components, that is, the physical activity educational lessons and the physical activity homework, incorporated academic learning tasks in the physical activity (e.g., rope jumping while spelling English vocabulary words), thus adding cognitive load to the activity. Approximately 25% of the daily physical activity in the ASK intervention was intended to be of vigorous intensity. This was defined as “children would be sweating and out of breath.” Intervention schools were provided with equipment for use in the intervention activities and the children received a skipping rope and a tennis ball for physical activity homework. To ensure that the teachers were empowered, supported, and qualified to deliver the ASK intervention, we conducted three instructional seminars ahead of the intervention start for the teachers at the intervention schools. Further, we provided two regional refresher sessions during the intervention period. Finally, we provided teachers in intervention schools with e-mail- and telephone-support, as well as password-protected homepage² that supplied teachers with information, videos, and physical activity lessons.

Children from both intervention schools and control schools participated in curriculum-prescribed 90 min/week of physical education and 45 min/week of physical activity (total 135 min/week). It was specified to the control schools that they should carry out the amount of physical activity and physical education that they would have done regardless of the ASK study.

The ASK study was designed to detect an effect size (Cohen’s d) of 0.35 between the two groups for change in academic performance (main outcome). The effect size was based on findings from previous studies (Sibley and Etnier, 2003). An intra-class correlation (ICC) of 0.15 [observed clustering of academic performance during the previous school year (2013–2014)] was applied to account for the cluster-randomized design, leading to a design effect of 4.54. For further details, see Resaland et al. (2015).

All study variables were examined for distributional properties, of which TMT-b was transformed and other variables kept in theirs original form. We excluded all values exceeding five standard deviations from the mean.

Children’s characteristics are provided as means and standard deviations (SD) or frequencies. A linear mixed model including school as a random effect was used to examine differences between children in tertiles of numeracy. The descriptive analyses were conducted with SPSS software, version 23.0 (IBM SPSS Statistics for Windows, : IBM Corp., Armonk, NY, United States).

Structural equation modeling (SEM), with full information maximum likelihood estimation (FIML) was used to examine bivariate correlations and the mediation model. The analyses were implemented through the Mplus program, version 7.4 (Muthén and Muthén, 1998–2015). To estimate the mediating effects in our pretest–posttest control group design, we used the ANCOVA model as recommended and explained by Valente and MacKinnon (2017). To account for children nested within schools, the complex method with robust maximum likelihood (MLR) estimator was used in all analyses. To reduce the complexity of the mediation model, we made five parcels from the 10 items in the CBRS. The parcels were composed by a balanced approach described in Little (2013), and as we had two measurement time points we used the average loading across the time points to rank the items.

Multiple fit indices in addition to the chi-square test statistic were used to assess model fit; the Comparative Fit Index (CFI), the Root Mean Squared Error of Approximation (RMSEA), and the Standardized Root Mean Square Residual (SRMR). We used a non-significant χ² and the cutoff recommendations of CFI > 0.95, and RMSEA and SRMR < 0.05 as indications of good model fit to the data (Geiser, 2013).

Measurement invariance (metric and scalar) was tested across time and group for the latent factors of executive function, behavioral self-regulation, and school related well-being. Criteria used to test differences between nested models (configural, metric, and scalar) were ΔCFI of -0.010, ΔRMSEA of 0.015, and ΔSRMR of 0.010 (for scalar) as recommended by Chen (2007) for sample size N > 300. A p-value ≤ 0.05 was used to indicate statistical significance in all analyses.

Our results revealed that neither executive function, behavioral self-regulation, nor school related well-being mediated the effect of the intervention on academic performance in numeracy (Figure 3). Standardized β-coefficients (unstandardized p-values) for the indirect effects were 0.035 (p = 0.479) for executive function, 0.020 (p = 0.256) for behavioral self-regulation, and 0.002 (p = 0.714) for school related well-being. The mediation model had good fit [χ²(df) = 433.178 (307) (p < 0.001), RMSEA = 0.034 (90% CI, 0.026,0.041), CFI = 0.976, and SRMR = 0.049].

The effect of the intervention on executive function despite no difference in objectively measured physical activity between the intervention and control schoolchildren, suggests that the content of the intervention was of importance. The prevailing literature supports the value of the cognitive demands inherent in the physical activities in order to affect executive function (Best, 2010; Pesce, 2012; Diamond, 2015; Tomporowski et al., 2015). The ASK intervention might have enhanced the cognitive demands of the physical activity in several ways. First, the integration of academic learning tasks while being physically active likely increased the challenge on executive functions, as these functions are important for solving for example academic problem-solving tasks in mathematics (Bull and Scerif, 2001; Cantin et al., 2016). To our knowledge, no previous study have investigated the effect on executive function using physical activity integrated in mathematics lessons in a sample of children performing lower in numeracy. In general samples of elementary schoolchildren, no effects of physical activity integrated in academic lessons on executive function have been observed (Beck et al., 2016; de Greeff et al., 2016). This might indicate that children with lower math competence take unique advantage of getting involved in academic learning tasks while being physically active.

Second, the greater part of the intervention activities was organized as group activities, which require use of strategic and goal-directed behavior through social interactions in an environment that constantly change (Best, 2010). Executive function skills acquired in these activities may transfer to other executive function tasks. Supporting this line of reasoning, Schmidt et al. (2015) observed only intervention effects on executive function following a team-game intervention, and not following an aerobic exercise intervention, although both interventions increased aerobic fitness equally.

There is also a possibility that challenging features of the activities in terms of motor coordination helped make the activities more cognitively demanding. For example, all children in the intervention group received a tennis ball for the homework, which may have resulted in enhanced practice of aiming and catching skills that previously have been shown to act as a mediator in the effect of a physical activity intervention on executive function (Pesce et al., 2016). Furthermore, the daily 5 min physical activity breaks during the academic lessons in the classroom were often coordinative demanding dancing activities (e.g., Just Dance), possibly targeting the same mechanisms as studied by Budde et al. (2008), where coordinative exercise affected attention and concentration. Supporting evidence for the relevance of coordinative demands of chronic physical activity on executive functions have been observed previously (Crova et al., 2014; Koutsandreou et al., 2016).

The observed effect on executive function only in this subsample of the ASK study, could well also be attributed to the fact that the intervention optimally challenged these children in terms of the motor capabilities, as addressed by Pesce et al. (2013). Finally, the effect of the intervention on executive function might also be owing to the fact that the included children had higher body fat than the other children (however, only significantly higher than the best performing children in numeracy). Studies have suggested that overweight children benefit more from physical activity to improve executive function (Davis et al., 2011; Crova et al., 2014).

Contrary to the literature showing a consistent relationship between executive function and academic performance (Best et al., 2011; Cantin et al., 2016), we found no relationship between change in executive function and change in academic performance. However, it represents a conservative design to investigate change in the mediator on change in the outcome, as in the current case. In order to capture the effects of change in the executive function on change in numeracy, a longer follow-up period might be necessary. For example, we have previously observed a significant relation between executive function at baseline and change in numeracy (Aadland et al., 2017b).

We found no effect of the intervention on behavioral self-regulation. This stands in contrast to the effect on executive function and results from previous studies that have observed increased time on-task during or following physically active academic lessons (Bartholomew and Jowers, 2011; Riley et al., 2016) and short physical activity bouts during academic lessons with or without academic instruction (Mahar et al., 2006; Kibbe et al., 2011; Carlson et al., 2015). An explanation for this discrepant finding might be that these previous studies also increased physical activity levels, likely reflecting the importance of the physical activity dose on behavioral self-regulation. More specifically, our physical activity breaks of 5 min during academic lessons were shorter than previous studies reporting effects on on-task resulting from 10 min bouts of physical activity (Mahar et al., 2006; Kibbe et al., 2011; Carlson et al., 2015). Studies that have revealed effects of acute physical activity on arousal and attention have also used bouts of longer durations (Budde et al., 2008; Janssen et al., 2014a,b). Hence, it is possible that our intervention did not reach the physical activity dose necessary to achieve beneficial effects on arousal and attention.

Other explanations for the discrepant findings might be the measurement used for behavioral self-regulation and the timing of the measurement. For example, Riley et al. (2016) reported increased time on-task during their physically active academic lessons based on observational data. Our measure, in contrast focused on behavior in the classroom, reported by the teachers at baseline and at follow-up. Hence, our measure expressed the behavior in the classroom over a long period in which, the majority of the academic lessons did not follow or include physical activity. Thus, it is possible that our intervention children were more on-task during the intervention activities compared to their control peers, as in Riley et al. (2016), but this was not captured by our way of measurement. Furthermore, the variance in the behavioral self-regulation factor observed across the intervention and control group teachers may have delimited the construct validity thus explaining the lack of intervention effect on this mediator. The use of teacher rated behavioral self-regulation is furthermore a subject to observer bias (McClelland and Cameron, 2012). Direct observation of behavioral self-regulation was out of scope for the present study.

The relation between the change in behavioral self-regulation and change in academic performance in numeracy is an important finding, emphasizing the significant role of behavioral self-regulation for academic performance also in an older age group (fifth-grade children) than previously shown (preschoolers and first-grade children) (von Suchodoletz et al., 2013; Gestsdottir et al., 2014; Schmitt et al., 2014).

Our intervention did not improve school related well-being. Moreover, change in school related well-being was not related to change in numeracy. This is in line with the quasi-experimental study by Kall et al. (2015), who did not observe any effects of a curricular physical activity intervention on any of the Kidscreen dimensions in elementary school children. An important limitation in the Kall et al. (2015) study, however, was that they did not include objectively measurement of physical activity. Hence, it is possible that both their findings and ours revealing lack of effect on school related well-being are attributable to an insufficient dose. As observed by Norris et al. (1992), duration and intensity of physical activity may be of importance to affect emotions.

Furthermore, it could also well be that our intervention activities and the mode of delivery did not stimulate positive peer-relationships, social identity, and belonging, which all are important for well-being (Bailey, 2016; Orkibi and Ronen, 2017). Indeed, both type of activities and mode of delivery influence intervention effects (Morgan et al., 2016). In terms of delivery, the ASK study pre-intervention workshops for teachers emphasized a mastery- and autonomy supportive learning environment which has been shown to facilitate need satisfaction, quality of motivation, and psychological well-being (Ryan and Moller, 2017). However, our study included a large number of intervention teachers with variability of experience and expertise in organizing and facilitating physical activity. Hence, it is possible that the settings for the implementation of the intervention activities were different across schools with some teachers not being able to implement the physical activities such that children’s well-being were facilitated. For example, although group activities creates opportunities for cooperation, and facilitates belonging and mastery (Ryan and Moller, 2017), the different emphasis on the competition element potentially initiated by some intervention teachers may have led to various experiences of the activities among the children of which some then may have experienced reduced rather than enhanced their well-being. Indeed, teachers’ ability to create a safe learning environment, as well as promoting good social relationship and acceptance of peers, has previously been shown to be of major importance for well-being in the classroom (Holfve-Sabel, 2014). Intervention teachers may not have been able to put this into effect outside the classroom.

In terms of intervention activities, the children are different in terms of social and physical-motor competence to participate. This requires tailored activities. It is possible that the group of children in the present study joined the physical activities with less social and physical competencies to their disposal for involvement in and mastery of the group activities. Hence, the possible benefits of these activities for school related well-being might have been counteracted. Indeed, according to self-determination theory, psychological well-being rests upon satisfaction of basic needs, including not only social relatedness, but satisfaction of the need for competence as well (Orkibi and Ronen, 2017; Ryan and Moller, 2017). Future studies would do well to examine the role of these psychological prerequisites when examining well-being as a psychosocial mediating mechanism. At this point then we cannot rule out the possibility that, consistent with previous findings (Bergh et al., 2012), our intervention generated unintended effects on this special group of children.

The cluster randomized controlled design and the large sample size, together with the statistical approach used to examine our hypotheses, are strengths of the present study. In line with the recommendations by Valente and MacKinnon (2017), we estimated the mediating effect in our RCT with an ANCOVA model, which increases the precision of the treatment effect through the adjustment for baseline scores. We furthermore ran the ANCOVA model with a latent variable SEM approach, which enabled testing of the full ANCOVA model with three latent mediating factors simultaneously. The use of latent mediating factors exclude measurement error which enhance reliability and avoids underestimation of both the path between the predictor and mediator and the mediator and the outcome, and an overestimation of the direct link between the predictor and the outcome (Cole and Maxwell, 2003). Yet, it is possible that analyzing each dimension of executive function separately could have yielded other results, as previous studies have reported effects of physical activity on only one dimension of executive function and not others (Schmidt et al., 2015; Pesce et al., 2016). However, we did also analyze each dimension separately, but this approach did not change any findings. A further strength is the inclusion several mediators measured by different data sources (executive function tests, teacher reports, and self-report by the children).

The main aim of the ASK study was to determine the effect of physical activity on academic performance. The present study is based on a secondary analysis using a subsample of children, and was not a priori defined. Thus, our results should be interpreted with this limitation in mind.