This study examined mid-way and post-intervention effects of a 20 month school-based obesity prevention intervention upon a wide range of determinants of PA and SB. In addition moderating effects of gender, weight status and parental education were assessed and the influence of perceived intervention dose received by the participants on the determinants was explored.
For the whole sample favorable effects on both psychological and social-environmental determinants of PA were found mid-way. However, post-intervention the effect was only sustained for social support from teachers, whereas an unexpected negative effect on self-efficacy for PA was revealed. The intervention did not affect any of the SB determinants. Moderation effects of weight status and parental education were observed, and subgroup analyses showed that the intervention did not work equally well in all subgroups. In addition, analyses of intervention dose received indicated that the effect on the determinants was influenced by the adolescents’ reported degree of exposure to and participation in the intervention.
Most of the intervention components that targeted the adolescents emphasized promoting enjoyment of PA, possibly explaining the overall positive mid-way effect on enjoyment. Our finding are consistent with the results of a 12 week long intervention among younger girls  and would seem encouraging given that enjoyment of PA has been shown to be of great importance for activity initiation and continued interest . Moreover, enjoyment has been identified as a mediator of PA change in adolescent girls . However, in accordance with results from a longer lasting intervention with a similar age group as the current one  no overall favorable effect on enjoyment was seen post-intervention, while a clear, reduction in enjoyment was detected among the overweight. The latter result would seem troublesome, and might reflect that various intervention activities have not met with the needs of those being overweight. .
Even though there was an overall marginally positive mid-way effect for self-efficacy, the subgroup analyses revealed a positive effect among the normal weight only, while there was a tendency towards a negative effect on self-efficacy among the overweight. Despite focusing on low threshold intervention activities, the negative mid-way trend for self-efficacy together with the reduction of enjoyment seen post-intervention among the overweight could well reflect that the overweight group has not felt at ease with these activities provided over time. Indeed, a sense of competence and feeling efficacious has been shown to be a key factor for enjoying PA [37, 38]. Alternatively, social comparison processes with those being normal weight might have led to unfavorable self-perceptions and enjoyment among the overweight .
As to self-efficacy post-intervention, the results showed an effect in the undesired direction for the whole sample. However, this type of unexpected result has also been seen in other studies [40, 41]. Due to the comprehensive nature of our intervention we cannot draw conclusion to which intervention components the effect can be attributed. However, it could well be that participants were more unaware of barriers to PA change in the first school-year of the intervention, but as the intervention moved along they might have become more aware of and realistic about barriers for PA.
While no overall effects on social support from friends or parents were seen mid-way or post-intervention, the negative post-intervention change for perceived social support from parents among the adolescent with medium level of parental education is not readily explainable. While one could assume that this result was due to baseline differences between control and intervention group this was, however, not the case (Table 5).
To our knowledge, this is the first study to report a positive effect on perceived social support from teachers and this was observed both mid-way and post-intervention. These results are encouraging because teachers are in the position to reach most adolescents and hold the role as change facilitators in most school-based interventions. The post-intervention subgroup differences for parental education level revealed that the effect on social support from teachers was predominantly seen among adolescents with lower and higher parental education background. Most importantly, these results indicate that teachers also seem to be able to reach children with lower socio-economic status when it comes to providing support for PA, and that teacher support may be a source of social influence that holds the potential to influence the social gradient that seems to exist concerning PA among adolescents . The yearly kick-off meetings for the teachers targeting the whole teacher team at each school to support the intervention might have contributed to a sense of enhanced support from teachers. At the same time low baseline values means that there was greater room for improvement in social support from teachers compared to many of the other determinants.
The intervention did not have an impact on determinants for SB post-intervention (perceived regulation of parental TV-viewing and computer/game-use), even though mid-way effects on TV-viewing and computer/game-use among girls have been documented previously in the HEIA study . One explanation could be that the intervention targeting SB (in 7th grade only) was not extensive enough to influence the determinants of SB, since the components included only one computer tailoring session and one fact sheet to parents. No effect on perceived social inclusion at school was found either. However, all these determinants showed quite high baseline values (range 3.53-4.34, Table 3). Hence, a possible ceiling effect might also explain these post-intervention results.
In line with one other study among children , no effect modification by gender was found on the potential determinants. Results indicate that possible working mechanisms for PA change do not differ by gender. It also corresponds with findings for change in PA itself in children and adolescents. Van Sluijs et al. 2007  concluded that most intervention studies observed no differential response by gender in PA change, and in a recent review by Cragg et al. 2011  there was no consistent evidence of an association between gender and PA change among 10–13 year olds. Moderation effects of weights status on PA have been found in a previous cross-sectional study  and on mid-way effect on SB in a prospective study from the HEIA study . However, no other studies have, to our knowledge, explored moderating effects of weight status and parental education on change in determinants for PA and SB among adolescents.
Our conflicting results on some of the determinants, especially among the overweight group, point to the importance of studying subgroup differences in the response to the intervention. Change in the expected direction in a determinant (the hypothesized mediator) is supposed to precede a desired change in behavior . Hence, the no-effect and negative effect detected in determinants in some of the subgroups could work against desirable behavioral effects (i.e. in PA or SB).
Perceived intervention dose received
The marked decrease in the proportion reporting a high intervention dose received from the mid-way (55.5%) to the post-intervention assessment (31.0%) could be one explanation for why effects were detected on several of the determinants mid-way, but not post-intervention. Furthermore, the adolescent reporting a high intervention dose received mid-way had significantly higher values on three out of four determinants than those reporting a low one. Post-intervention there were significant differences in favor of those with a high intervention dose received for enjoyment, social support from friends, perceived environmental opportunities for PA and perceived social inclusion at school (Table 6). This indicates that the intervention had an effect on change in these determinants among those most exposed to the intervention. However, the adolescents reporting a high intervention dose received mid-way showed significantly higher baseline values on self-efficacy and perceived social support from teachers compared to those receiving a low dose. Similar differences were found for enjoyment and perceived environmental opportunities for those adolescents receiving a high dose post-intervention (Table 6). Accordingly, for these determinants it seems as if the intervention increased the differences already present at baseline.
No differences in effects between the high and low intervention dose groups were seen for perceived social support from parents and perceived parental regulation of TV-viewing and computer/game-us (Table 6). Change in these parental related determinants was primarily targeted through the fact-sheets to the parents, and parental reported degree of exposure to the fact sheets would possibly be a better indication of the influence of implementation on these determinants.
However, overall the results from examining the intervention dose received suggest that the results revealing no effects in some of the outcomes in the main analyses might be due to an insufficient implementation of the intervention rather than insufficient intervention strategies. In support for this supposition, mid-way results from teacher reports of degree of implementation indicate that the overall degree of implementation was moderate . It could be that the short e-mail reminders to the teachers to prompt the implementation of the various components were not sufficient to ensure a high degree of implementation over the course of the intervention.
There are both strengths and limitations to this study. The strengths include the high quality design and the theoretically based intervention in a large, long term study in a sample drawn from a region within a European country. Effects on potential determinants for both PA and SB were examined at two time points with high response rates. The analyses of moderating effects and corresponding subgroup differences added knowledge about intervention effectiveness across subgroups. As called for, the influence of perceived exposure to and participation in the intervention on the outcomes was explored. The limitations include the power analyses which were based on detecting change in PA and BMI, and not in the determinants . However, the sample size of the study is larger than many previous studies including effect analyses on determinants [5, 6, 11]. The determinants assessed showed acceptable internal reliability at all time-points and test-retest reliability , but they might not have been sensitive enough for detecting change. The intervention was also extended to include an additional component (the computer tailoring program) in the last part. Therefore it is not possible to tease out whether the post-intervention results are related to this addition or to the intervention duration in itself. The wordings of the specific items measuring the determinants were directed towards PA and SB in general. It might have made it easier to detect intervention related changes in the determinants if they were phrased to match the intervention components more precisely since the different components were partly tailored to influence the behaviors in specific context. However, this was not possible in order to keep the questionnaire at reasonable length. While social desirability could have influenced the outcomes especially in the intervention group, the changes in the undesired direction for some of the outcomes go against such a line of reasoning. The seasonal difference between baseline (fall) and the two other data collection (spring) could also have affected the results. However, the weather conditions in Norway are quite similar for the two seasons in question and seasonal differences might be more pronounced in the actual behavior. The generalization of our findings might be somewhat weakened because a higher proportion than expected of the adolescents and parents declined to give consent. There might also be a possible attrition bias present due to the somewhat higher proportion of overweight adolescents and the lower values for perceived social inclusion found among those who only participated at baseline compared to several time-points. However, no differences between the control and intervention group were found among the non-responders at baseline (data not shown).