Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Exploring the Potential of a School Impact on Pupil Weight Status: Exploratory Factor Analysis and Repeat Cross-Sectional Study of the National Child Measurement Programme

  • Andrew James Williams ,

    a.j.williams@ed.ac.uk

    Affiliations Farr Institute @ Scotland and Scottish Collaboration for Public Health Research and Policy, Usher Institute of Population Health Sciences and Informatics, University of Edinburgh, 20 West Richmond Street, Edinburgh EH8 9DX, United Kingdom, Institute of Health Services Research, University of Exeter Medical School (formerly Peninsula College of Medicine and Dentistry), South Cloisters, St. Luke’s Campus, Exeter EX1 2LU, United Kingdom, Children’s Health and Exercise Research Centre, Sport and Health Sciences, St. Luke’s Campus, University of Exeter, Exeter, EX1 2LU, United Kingdom

  • Katrina M. Wyatt,

    Affiliation Institute of Health Services Research, University of Exeter Medical School (formerly Peninsula College of Medicine and Dentistry), South Cloisters, St. Luke’s Campus, Exeter EX1 2LU, United Kingdom

  • Craig A. Williams,

    Affiliation Children’s Health and Exercise Research Centre, Sport and Health Sciences, St. Luke’s Campus, University of Exeter, Exeter, EX1 2LU, United Kingdom

  • Stuart Logan,

    Affiliation Institute of Health Services Research, University of Exeter Medical School (formerly Peninsula College of Medicine and Dentistry), South Cloisters, St. Luke’s Campus, Exeter EX1 2LU, United Kingdom

  • William E. Henley

    Affiliation Institute of Health Services Research, University of Exeter Medical School (formerly Peninsula College of Medicine and Dentistry), College House, St. Luke’s Campus, Exeter EX1 2LU, United Kingdom

Abstract

Schools are common sites for obesity prevention interventions. Although many theories suggest that the school context influences weight status, there has been little empirical research. The objective of this study was to explore whether features of the school context were consistently and meaningfully associated with pupil weight status (overweight or obese). Exploratory factor analysis of routinely collected data on 319 primary schools in Devon, England, was used to identify possible school-based contextual factors. Repeated cross-sectional multilevel analysis of five years (2006/07-2010/11) of data from the National Child Measurement Programme was then used to test for consistent and meaningful associations. Four school-based contextual factors were derived which ranked schools according to deprivation, location, resource and prioritisation of physical activity. None of which were meaningfully and consistently associated with pupil weight status, across the five years. The lack of consistent associations between the factors and pupil weight status suggests that the school context is not inherently obesogenic. In contrast, incorporating findings from education research indicates that schools may be equalising weight status, and obesity prevention research, policy and practice might need to address what is happening outside schools and particularly during the school holidays.

Introduction

The rising prevalence of overweight and obesity and the associated non-communicable diseases are among the leading public health concerns of the 21st century [13]. Childhood is seen as a key stage of life in which to intervene to prevent the development of overweight, obesity and the associated behaviours [1,4,5]. Schools, as an environment to which the majority of children are exposed and in which they eat, drink, exercise and learn, have been perceived as important sites for obesity prevention interventions [57]. Consequently, there have been a large number of school-based obesity interventions [4,812], however, a recent systematic review concluded that they had only been moderately effective [10]. Data from the National Child Measurement Programme (NCMP) in England have identified that the prevalence of obesity doubles during the period of primary education (ages 4–11 years), which may indicate that primary schools may be crucial to the development and prevention of obesity [1317].

The World Health Organization health promoting schools initiative recognises the importance of the school context, however, few school-based obesity prevention interventions are developed to be adapted to each schools context [18]. Bonell et al. [5] recently reported a systematic review of both quantitative and qualitative evidence on the effect of school and school environment on health. The review, concluded that the empirical evidence supported a number of the theories as to how schools affect health such as the theories of human functioning and school organisation, social capital and social development [5]. Bonell et al. [5] called for future school-based health research to be informed by theory and examine a broader range of health and education outcomes than those examined in the studies reviewed (e.g. substance misuse, violence, physical activity and healthy eating). The review highlighted the strong theoretical basis for a school effect on pupil weight status but emphasised the lack of empirical research [5]. Procter et al. [19] undertook a cross-sectional ‘value-added’ analysis of 2,367 children attending 35 primary schools in Leeds, England, seeking to address the question of a school effect on pupil weight status. Their findings led them to conclude that schools had an independent impact upon children’s weight status [19]. However, this study lacked the power to evaluate which elements of the school context might be associated with the school effect on weight status.

We previously undertook an analysis replicating the methods proposed by Procter et al. [19] across five years [20]. Although we were able to replicate their findings within any single year, each schools impact (value-added ranking) varied considerably across the five years [20]. We therefore concluded that the effect observed by Procter et al. [19] did not reflect a systematic differential school impact on pupil weight status [20]. However, the possibility remains that there are features of the school context that influence weight status but that these are not evident when studying consistency of school rankings over time because the rankings are sensitive to other factors including the characteristics of the pupils in each yearly cohort [21]. Furthermore, studying associations between school context and pupil weight status related to a systematic common school impact, is complicated by the difficulty of collecting data on a suitably large matched control sample of children who do not attend school. In this study we explored the impact of measured characteristics of the school context on pupil weight status by exploiting the differential exposure to the school context of pupils in the different year groups, measured within the NCMP. Recognising that the school context is not one dimensional and heeding the danger of over adjustment raised by Bonell et al. [5], we first used exploratory factor analysis of variables related to school context to derive a robust set of interpretable school-based contextual factors that summarise the key sources of variation between schools [2224]. Subsequently, we tested for consistent associations between pupil weight status and the school-based contextual factors, to explore the possibility that common features shared by groups of schools may have a systematic impact on pupil weight status.

Materials and Methods

The NCMP was introduced in England in 2005 in order to monitor progress towards a 2004 Public Service Agreement to reduce the prevalence of obesity among children less than 11 years old to the level reported in 2000 [25,26]. Within the programme, each year the children in the first (Reception, 4–5 year olds) and last (Year 6, 10–11 year olds) years of primary education at state-maintained primary schools are weighed and measured by health professionals. The data are collected in order that body mass index standard deviation scores (BMI-SDSs) relative to the UK 1990 growth reference can be calculated [27,28]. Subsequently each individual is categorised by weight status (underweight, healthy weight, overweight or obese) according to the epidemiological definition for monitoring purposes [28].

The data in this study relate to the Local Authority of Devon, a large county in the South West of England, with a mix of rural and urban, isolated and dense dwellings, including the city of Exeter. Although the urban areas of Plymouth and Torbay are within the county of Devon they are governed by separate unitary authorities and hence not included in this study. Devon has two coastlines, low levels of deprivation and low ethnic diversity; summary characteristics of the study sample, Devon and England as a whole are presented in Table 1 [29]. Due to low numbers some of the ethnicity categories were combined; Chinese with Asian or Asian British and Black or Black British with Any other ethnic group [30]. Reports on the NCMP note that the South West has a statistically lower than average prevalence of obese children, however, there is a statistically higher than average prevalence of overweight children in Reception, but not Year 6 across the region (Table 1) [1417].

thumbnail
Table 1. Devon schools compared with national schools data 2008/09 to 2010/11a.

https://doi.org/10.1371/journal.pone.0145128.t001

Individual weight status data from the NCMP

Following approval from the institution’s ethics committee and the Caldicott guardian, five years (2006/07 to 2010/11) of anonymised cross-sectional data from the NCMP were received from the primary care trust (NHS Devon) to form the individual data for this study. Across the five years there were 62,554 participants with valid BMI-SDS which had been checked internally, as well as according to the national validity checks [30]. Two binary outcome variables were generated indicating those categorised as overweight (1.04≤BMI-SDS<1.64) or obese (BMI-SDS≥1.64). Within Devon, children and parents are required actively to opt out of the NCMP and no data are collected on non-participants. However, NHS Devon were able to provide data on the proportions of each school year group which did not participate. Across the five years analysed participation in Devon was lowest in 2006/07 (82%) and rose to 91% in 2010/11 [1317].

The anonymised NCMP dataset contained the following demographic data on each participant: age in months at time of measurement, gender and ethnicity. The location of participant’s home was indicated by their Lower Super Output Area (LSOA) [31]. The LSOA data were used to link individuals and schools to the following indices of area deprivation: Index of Multiple Deprivation (IMD), Income Deprivation Affecting Children Index (IDACI) and Child Wellbeing Index (CWI) [32,33]. Multiple indices of deprivation were examined due to the previously identified association between weight status and socioeconomic status [26]. In order that the indices were comparable they were rescaled nationally prior to linking with the dataset, so that 1 was the most deprived and 0 the least deprived LSOA in England.

School contextual data

A reference number for each participant’s school and the school’s LSOA were provided as part of the NCMP data extract. A wide variety of school data were sought relating to demographics, socioeconomic status, built environment, physical activity, diet and ethos from multiple sources. The selection of school data was informed by previous research [24,3439]. The acquired data were combined into a school dataset as outlined in S1 File, however, due to the widely recognised difficulties in collecting dietary data, no routinely collected sources of such data were identified. Only a few variables (e.g. school building age, catchment area and SATs pass rate) contained large quantities of missing data. Therefore, it was felt prudent not to include these three variables in the exploratory factor analysis, but otherwise to undertake a complete case analysis.

The school variables collected reflected the compositional (e.g. ethnic mix), collective (e.g. Healthy School award), as well as contextual (e.g. location) aspects of the school as discussed by Macintyre et al. [40]. As the school composition varies from year to year, and Year 6 pupils would have been exposed to the school environment for a number of years, each of the compositional variables was averaged across the six years of primary education. This assumed that children remained at the same school for their entire primary education and the effect of each year was equal, which although unlikely to be absolutely true was a useful preliminary assumption. Reception pupils' values for the compositional variables were those of the year of measurement. Summary statistics and histograms of each school variable were examined and highly skewed variables were categorised as shown in S1 File. The school and individual datasets were merged using the school reference number.

Analysis

In line with the study purpose there were two phases to the analysis; firstly deriving the school-based contextual factors and, secondly testing the impact of the school-based contextual factors on pupil weight status. All data preparation and analysis was undertaken in Stata IC 11 [41].

Phase 1: derivation of school-based contextual factors.

Any impact of school context on pupil weight status would be expected to be more evident among those with more exposure to the school context (Year 6 compared with Reception). Hence, only the school variables related to Year 6 pupils were used to derive the school-based contextual factors. Furthermore, as the intention was to test the consistency of the associations between pupil weight status and school context, the school-based contextual factors were developed using the 2010/11 data as, within the dataset this was the year with the highest NCMP participation, resulting in one observation per school (n = 289). With the majority of school variables being categorical the factor analysis was based on the polychoric correlation matrix [42]. This required that all the categorical variables were either ordinal or binary and therefore school governance and denomination were transformed into binary variables (community or other and not-religious or other respectively). As well as excluding school building age, catchment area and SATs pass rate from the factor analysis three further variables were excluded. Governance and denomination were almost identical variables and therefore only governance was included. By 2010/11 all schools were required to have a travel plan and therefore this variable was excluded. Non-participation was excluded from the factor analysis as it related to the NCMP and not the school context, but to assess how non-participation may have biased the results, we adjusted for non-participation in all models in phase 2. Promax rotation was used to calculate the school-based contextual factors as it does not rely on the variables being uncorrelated (non-orthogonal). Examination of the variables with highest loadings in each school-based contextual factor and data from the excluded variables was then used to define a label for each school-based contextual factor.

Phase 2: testing the impact of school context on pupil weight status.

A series of logistic regression models were developed to explore the associations between the derived school-based contextual factors and pupil weight status (overweight or obese). As the data were hierarchical, the primary analysis involved fitting multilevel logistic regression models with a random school effect (intercept). Non-hierarchical (single level) models were also estimated in order to provide some insight into the impact of the various multilevel structures evaluated. Having hypothesised that the school context would have greater impact on Year 6 than Reception pupils, the two year groups were examined separately in two-level models (pupils within schools). Three-level models (pupils within year groups within schools) were also developed in order to be able to explore the differences within schools (between year groups, akin to ‘value-added’) as well as between schools and the interactions between year group and the school-based contextual factors. Neighbourhood has been considered an important context in the child’s life. However, our previous research, and that of Procter et al. [19], found cross-classification by neighbourhood to have little impact upon the results, and hence cross-classification was not undertaken in the present study [20].

In line with our aim to examine not only significance but also consistency, each year (2006/07 to 2010/11) was analysed separately. Initially empty null models were examined to identify the intraclass correlation coefficients (ICCs). Due to the underlying logistic distribution the individual variation in the null models was assumed to be π2/3 [37,43]. Subsequently, the derived school-based contextual factors were examined individually in models which were adjusted for individual gender, ethnicity and socioeconomic status (based on previous analyses IMD was chosen as the sole individual measure of socioeconomic status [39]) and NCMP non-participation. All analyses were two-tailed with α = 0.05. Restricted maximum likelihood estimation and numerical integration with seven quadrature points were used for the multilevel analyses. The normality and homoscedasticity of the residuals from the school and year group levels of each model were assessed for any violation of assumptions using graphical methods.

Caterpillar plots were constructed from the fitted models to quantify each school’s residual effect upon pupil weight status after adjustment for the fitted variables. The width of the confidence intervals was set at ±1.4 times the standard error in order that 5% significance was maintained, given that the ratio between the majority of pairs of standard errors did not exceed 1:2 [43,44]. Plots of the school-level residuals from the two-level and three-level models, illustrated the differences between schools. However, caterpillar plots of the difference in year group (Year 6-Reception) residuals from the three-level models indicated the differences between year groups within schools (‘value-added’).

Results

Descriptive characteristics of the individual data can be found in Table 2 and school data in S1 File. The mean age of the children in each year group decreased slightly over the five years, while the gender balance remained fairly constant with slightly more males than females. Around 95% of the sample were White—British and the levels of deprivation were low. The BMI-SDS and weight categories reflect the national pattern of steady increase in weight status; however, in 2008/09 there was a larger increase in BMI-SDS, overweight and obesity. This coincides with a change in the NCMP legislation to enable letters be sent to parents informing them of their child's weight status [45]. Whilst it might have been expected that parents of overweight or obese children would not have wanted to know their child's weight status and therefore prevented the child participating, this would have resulted in a drop in prevalence of obesity; furthermore, the proportion of pupils participating in the NCMP continued to rise across this time period [1317]. Therefore the jump in prevalence requires further exploration beyond the current study. Additionally, contained in Table 2 is information on the distribution of pupil weight status within schools. These values demonstrate that there is marked variation in pupil weight status between schools, with the prevalence of overweight and obesity varying by more than 0.1 (10%).

Phase 1: derivation of school-based contextual factors

We initially examined the extent to which the factors identified changed depending on which combination of school variables were included in the factor analysis, (to assess the robustness of the factors). Furthermore, in recognition that deprivation is a recognised determinant of weight status, we wanted to see whether deprivation as a factor could be isolated from other aspects of the school context. The experimentation primarily consisted of varying the number of measures of school socioeconomic status included in the factor analysis, while also comparing the results of the promax rotation with those from varimax rotation. We found little variation in the nature of the factors identified with the inclusion and exclusion of variables and between types of rotation and therefore settled on the pragmatic approach of including all the deprivation variables (results not included but available from the corresponding author on request). Subsequently, four school-based contextual factors were identified with eigenvalues greater than 1 and the Kaiser-Meyer-Olkin measure of sampling adequacy was 0.55 indicating that factor analysis was appropriate (Bartlett’s test p<0.0001). Table 3 lists the composition and definition of the four factors, where loading values of less than 0.3 have been suppressed. In order to test the stability of each factor, the scores for a school with a high, medium and low score in 2010/11 were compared across the five years and between year groups. These results can be found in S2 File, alongside the polychoric correlation matrix from which the factors were derived.

thumbnail
Table 3. School-based contextual factor composition and definition.

https://doi.org/10.1371/journal.pone.0145128.t003

Factor 1.

The variables which loaded most strongly in factor 1 were the measures of school socioeconomic status and proportion of pupils with special educational needs. Profiling schools with high and low factor scores identified that the schools with lowest scores were large schools with low deprivation, prevalence of special education needs and use of active travel (walking or cycling) to school, which tended to be oversubscribed. Whereas, schools with high scores for factor 1 were high deprivation, smaller schools with more special educational needs and higher tendency to active travel. Henceforth factor 1 will be referred to as Deprivation.

Factor 2.

The variables which loaded most strongly in factor 2 were: school capacity, proportion not White—British, proportion eligible for free school meals, school location, proportion walking or cycling to school and proportion for whom English was an additional language. Schools with low scores for factor 2 were small rural schools with older buildings, large catchment areas, classes containing pupils from different year groups and little to no ethnic diversity. The high score schools were large urban schools with modern buildings, and more ethnic diversity. Factor 2 will be referred to as Location.

Factor 3.

The variables which loaded most strongly in factor 3 were: an indicator of whether the school was split over more than one site (e.g. separate playing field), grass play area and site area per pupil. Profiling schools identified that factor 3 ranked schools from (low scores) large rural community governed schools that have received recent investment, to (high scores) older small urban (town centre) voluntary (church) governed schools with little access to grass play area. Notably a number of the schools with low scores have subsequently become academies. Subsequently, factor 3 has been labelled Resource.

Factor 4.

The variables which loaded most strongly in factor 4 were: an indicator of whether the school catchment area included a coastline, school achievement of an Active Lifestyle or PEDPASS (Physical Education, Daily Physical Activity and School Sports) award and subscription (whether the number of pupils was markedly higher (oversubscribed) or lower (undersubscribed) than the school capacity). Factor 4 ranked schools similarly to factor 3 from large catchment area rural schools to small urban schools with increased active travel. There also appeared to be some association with school inspections (Office for Standards in Education (Ofsted) reports) with lower scoring schools receiving consistent Good grades, whereas those with higher scores were improving or in need of improvement. Schools with lower scores, compared to those with higher scores, had more quickly achieved Healthy School status particularly focussing on pupil emotional health and wellbeing but had not received the more physical activity specific awards (Active Lifestyle and PEDPASS) [46,47]. Consequently, factor 4 was labelled Prioritisation of physical activity.

Notably, none of the measures of school socioeconomic status loaded heavily into factors 3 or 4.

Phase 2: testing the impact of school context on pupil weight status

The final (complete) analysis included 55,826 pupils (Table 4). The ICCs of the null models are listed in Table 4 with less than 3% of the variation in pupil overweight status and less than 9% of pupil obesity status attributed to differences between schools or year groups. Contrary to expectations, in the two level-models the school ICCs tended to be larger for Reception than Year 6 pupils. In the three-level models the school ICC was generally <0.1%, indicating that any clustering was due to differences between year groups rather than between schools. Overall, the school and year group ICCs varied from year to year.

The results of the models fitted to test the significance and consistency of the associations between school based contextual factors and pupil weight status are presented in Table 5. Given the large number of statistical tests involved, the results have been presented as effect sizes (Cohen’s d values converted from odds ratios [48]), placing the emphasis on meaningful rather than solely statistical significance. Due to a sparse matrix in some models it was necessary to estimate the models in R [49] using lme4 [50] and then transfer the residuals back into Stata.

None of the school-based contextual factors demonstrated consistent associations with pupil overweight or obesity status, with effect sizes varying from model to model and year to year, all the effect sizes were less than |0.35|. Of the school-based contextual factors, Deprivation was most often meaningfully (d>|0.2|) associated with pupil weight status (increased odds of obesity). Location and Prioritisation of physical activity were never meaningfully associated with pupil overweight or obesity status. Resource was only once meaningfully associated with reduced odds of overweight. The expectation that school context would be more consistently and meaningfully associated with Year 6 pupil weight status than Reception pupil was not evident in the results. When the interactions between year group and factor were tested in the three-level models, the direction of the association of the main (Reception) and interaction (Year 6) effects were almost always in opposition (Table 5). For example in the 2007/08 Deprivation three-level interaction model of obese status the effect size in Reception (the main effect) is 0.235, while in Year 6 it is 0.124 (= 0.235–0.111), with the interaction effect flattening the main effect (Table 5). This indicates that any associations with contextual factors were stronger during Reception than during Year 6. The full details of the models can be found in S3 File alongside the sensitivity analysis undertaken using overweight and obesity categorised according to the International Obesity Task Force classification [51] and continuous BMI-SDS as outcomes. The sensitivity analysis supports the lack of consistent and meaningfully significant associations.

S4 File contains all the calculated caterpillar plots, which were smoother than those calculated in our previous study [20], demonstrating the effect of the fitted variables. The caterpillar plots differed very little between factors, which is coherent with the minimal impact of the factors demonstrated in Table 5. For brevity in Table 6 the plots have been categorised by the amount of uncertainty in each school or year group difference residual; little to none (−), small (~), medium (≈), large (≋). There were two slight variants of these four plot types labelled positive skew (+) and ‘outlier’ (°), although no true outliers were identified and none were consistent. Generally there was more variation in residuals from models of obesity than overweight, and the positive skew variant only occurred from obesity models, which likely reflects the increasing prevalence of obesity during primary school. The caterpillar plots confirm the finding from the ICCs (Table 4) that in the three-level models there is no variation between schools. Introducing the year group interaction revealed a small quantity of between school variation in some models, but the variation between year groups within schools is more marked. The caterpillar plots support our previous finding that there is no systematic differential school impact on pupil weight status, as well as indicating that there are unlikely to be unmeasured features of groups of schools that influence weight status [20]. Any remaining systematic common school impact is confounded with age-period-cohort effects which we cannot assess.

Table 6 also lists the amount of variation explained comparing the fitted models with the null models (Table 4), which alongside the ICCs (Table 4) demonstrate that the variation components are unstable, varying between year and model structure. Consistent with the findings of Miyazaki and Stack [52], the percentages of variation explained was sometimes negative as a result of chance fluctuations. This suggests an underlying complex variation structure, invalidating the use of simple methods to explore school impact on pupil weight status.

Discussion

The overarching objective of this study has been to explore the potential for measurable characteristics of the school context to have an impact on pupil weight status. Based on previous research, data were collected on a variety of school variables reflecting the contextual, compositional and collective attributes of the schools [24,3440]. Exploratory factor analysis identified four school-based contextual factors from the school variables representing; Deprivation, Location, Resource and Prioritisation of physical activity. Given that the school context is complex, drawing conclusions from studying individual school variables in isolation is inappropriate, whereas these school-based contextual factors reflect the multiple inter-related components of the school context. However, no meaningfully significant and consistent associations between any of the derived school-based contextual factors and pupil overweight or obesity status were identified. The Location and Prioritisation of physical activity factors were consistently not meaningfully associated with pupil weight status, while Deprivation was most consistently associated with obesity. Three additional findings from the second analytical phase provide further insights into the impact of school context on pupil weight status:

  • The relative and absolute proportion of variation in pupil weight status attributable to differences between schools decreased from Reception to Year 6.
  • When differences between year groups within schools had been accounted for, there was little to no variation in pupil weight status attributable to differences between schools.
  • When exploring the interactions between year group and school-based contextual factors, the associations weakened between Reception and Year 6.

Consequently, through novel, theory driven, analysis it has not been possible to identify common and measurable features of the school context that have an obesogenic impact.

This study has made extensive use of routinely collected data. The use of cross-sectional data meant that the consideration of causal links and age-period-cohort effects relevant to this study was more difficult; at present an equally large longitudinal dataset is not available. Recent changes to the NCMP mean that from around 2020 pupil Reception and Year 6 data will be linkable enabling more refined analyses [53]. Using routinely collected school data, has meant that only the measureable objective aspects of the school context have been captured, a particular weakness being the inability to include dietary variables. We note that from September 2015, as part of the School Food Plan, school inspections will include inspections of the school food and food environment [54]. The subjective school context may not be represented, although some of the derived school-based contextual factors reflected less tangible aspects of the school context (e.g. prioritisation of physical activity). But, using routinely collected data and thorough analyses have permitted the undertaking of a cost-efficient large-scale study to test and subsequently contest a common and important assumption that schools are contributing to the obesity epidemic. In calculating the compositional variables for Year 6 pupils, we made the assumption that each child remained at the same school for the entirety of their primary education and that the influence of each year was equal. These simplifying assumptions were necessary because of the lack of information on which school each Year 6 pupil had attended in previous years and are likely to attenuate the effects of the contextual variables. However, in the absence of any significant findings no analysis of the sensitivity of this assumption has been undertaken. In order to overcome some of the issues with multiple testing the results have been presented as effect sizes (Cohen’s d values) which also placed the emphasis on meaningful, rather than just statistically significant results. We note that the Kaiser-Meyer-Olkin measure of sampling adequacy was low but considered adequate to proceed within this study.

Being conducted in Devon, England, precluded this study from exploring possible ethnic differences, which have been the focus of previous studies in this area [55,56] and limited the scope for examining associations with socioeconomic status (Table 1). However, our finding that deprivation was more markedly associated with pupil obese than overweight status is consistent with Conrad and Capewell [57]. Previous studies have reported school-based contextual variables which they had found to be associated with pupil weight status, diet or physical activity [24,3438]. In this study we strove to examine multiple aspects of the school context simultaneously, as advocated by Bonell et al. [5], in the form of school-based contextual factors, which may explain the absence of previous associations among the findings. Although the school ICCs observed within this study were small, they are consistent with previously reported values, as was the finding that school ICCs were larger among Reception rather than Year 6 pupils [26,56,58,59]. Pallan et al. [56] acknowledge that this finding is counterintuitive, but posit two explanations; that the early primary school years are more conducive to physical activity whereas the later years focus more on academic attainment, and/ or that across the school years the influence of non-school environments swamp the school environment. Our results are consistent with their second explanation, but when considering their first explanation we argue that nine months or less exposure (for Reception pupils) is unlikely to impact significantly on weight status in comparison with five years of exposure to the home environment. Moreover, unlike previous studies the additional year group level in the three-level models made it possible to examine both between and within school effects. The finding that with the additional level the school ICCs in the three-level models effectively becomes zero indicates that there are no observed differences between schools, only differences between year groups within schools. Subsequently, the observed doubling in the prevalence of obesity during primary education does not appear to differ significantly between schools [1317].

It has been a commonly held assumption that the school environment has been contributing to the obesity epidemic through prolonged sedentary behaviour, less healthy foods and insufficient physical activity [57]. However, within this study no evidence of such an effect has been detected. Schools have been targeted as they offer a structured environment within which the majority of children can be reached, but as Downey et al. [60] emphasise, even during the half of the year that children attend school the average child spends more than 60% of their waking time outside school. Studies with repeated measures of pupil weight status have found that pupil weight status increases more markedly during the school holidays than during term time [6164]. Within the education literature this phenomenon is observed in regards to educational attainment and known as the summer learning gap [60,6567]. Specifically, the evidence demonstrates that children tend to learn at the same rate during term time, but during the summer holiday the drop-off in learning increases with deprivation to such an extent that the summer learning gap is considered to be responsible for the socioeconomic inequalities in educational attainment, with schools considered the ‘great equaliser’ [59,6567]. Gershenson [68] studied the differences in summer time-use by socioeconomic status and found that television viewing was the time-use most related to socioeconomic status. Given that sedentary behaviour is associated with weight status, it seems highly plausible that this impacts on weight status as well as educational attainment [9,24]. Our finding that the impact of school context seemed to lessen from Reception to Year 6 would be consistent with increasing disparities due to either increasing biological variation, or unequal opportunities outside the school, or a combination thereof. Furthermore if, as in the findings from those studies with repeated measures, growth in weight status slows or even decreases during term time, the school context may be stabilising pupil weight status and functioning as the great equaliser, holding-back weight gain [6164]. The flattening of the association between school-based contextual factor and pupil weight status identified when exploring the interaction with year group may reflect this phenomenon. We are not aware of any longitudinal study with sufficiently frequent (at the beginning and end of each school year or term) anthropometric measures to test this hypothesis. However, Anderson et al. [69] using the endogeneity in school starting age found that pupils starting school earlier (as the youngest in the class) had better weight outcomes than those starting later (as the oldest in the class) supporting this hypothesis. Therefore, it would appear that non-school time and school holidays in particular are a promising focus for obesity prevention interventions, especially as such interventions have the potential to influence the adult, as well as the child population [70].

Acknowledgments

We would like to acknowledge the assistance and support of the staff at Devon County Council including the former NHS Devon. Namely Dr Virginia Pearson, Ian Tearle, Jane Batten, Steve Kibble, Theresa Lawless, Simon Chant, Kirsty Priestly and Ray Hennessy. Neither would the project have been possible without the work of Matthew Domminey whose Master’s project demonstrated the feasibility of this project. Finally we acknowledge Lindsay Paterson and the reviewers, whose comments, suggestions and ideas helped improve this paper.

Author Contributions

Conceived and designed the experiments: AJW KMW CAW SL WEH. Performed the experiments: AJW WEH. Analyzed the data: AJW WEH. Contributed reagents/materials/analysis tools: AJW WEH. Wrote the paper: AJW KMW CAW SL WEH.

References

  1. 1. Gortmaker SL, Swinburn BA, Levy D, Carter R, Mabry PL, Finegood DT, et al. (2011) Changing the future of obesity: science, policy, and action. Lancet 378: 838–847. pmid:21872752
  2. 2. Finucane MM, Stevens GA, Cowan MJ, Danaei G, Lin JK, Paciorek CJ, et al. (2011) National, regional, and global trends in body-mass index since 1980: systematic analysis of health examination surveys and epidemiological studies with 960 country-years and 9.1 million participants. Lancet 337: 557–567.
  3. 3. Beaglehole R, Bonita R, Horton R, Adams C, Alleyne G, Asaria P, et al. (2011) Priority actions for the non-communicable disease crisis. Lancet 377: 1438–1447. pmid:21474174
  4. 4. Brown T, Summerbell C (2009) Systematic review of school-based interventions that focus on changing dietary intake and physical activity levels to prevent childhood obesity: an update to the obesity guidance produced by the National Institute for Health and Clinical Excellence. Obesity Reviews 10: 110–141. pmid:18673306
  5. 5. Bonell C, Jamal F, Harden A, Wells H, Parry W, Fletcher A, et al. (2013) Systematic review of the effects of schools and school environment interventions on health: evidence mapping and synthesis. Public Health Research 1.
  6. 6. Swinburn B, Egger G, Raza F (1999) Dissecting obesogenic environments: the development and application of a framework for identifying and prioritizing environmental interventions for obesity. Preventive Medicine 29: 563–570. pmid:10600438
  7. 7. Clarke J, Fletcher B, Lancashire E, Pallan M, Adab P (2013) The views of stakeholders on the role of the primary school in preventing childhood obesity: a qualitative systematic review. Obesity Reviews 14: 975–988. pmid:23848939
  8. 8. Zenzen W, Kridli S (2009) Integrative review of school-based childhood obesity prevention programs. Journal of Pediatric Health Care 23: 242–258. pmid:19559992
  9. 9. Kropski JA, Keckley PH, Jensen GL (2008) School-based obesity prevention programs: an evidence-based review. Obesity (Silver Spring) 16: 1009–1018.
  10. 10. Khambalia AZ, Dickinson S, Hardy LL, Gill T, Baur LA (2012) A synthesis of existing systematic reviews and meta-analyses of school-based behavioural interventions for controlling and preventing obesity. Obesity Reviews 13: 214–233. pmid:22070186
  11. 11. Sharma M (2006) School-based interventions for childhood and adolescent obesity. Obesity Reviews 7: 261–269. pmid:16866974
  12. 12. Sharma M (2007) International school-based interventions for preventing obesity in children. Obesity Reviews 8: 155–167. pmid:17300280
  13. 13. The NHS Information Centre (2008) National Child Measurement Programme: 2006/07 school year, headline results. London: The NHS Information Centre. Available https://catalogue.ic.nhs.uk/publications/public-health/obesity/nati-chil-meas-prog-resu-2006-2007/nati-chil-meas-prog-resu-2006-2007-rep.pdf. Accessed 11 November 2015.
  14. 14. The NHS Information Centre (2008) National Child Measurement Programme: 2007/08 school year, headline results. London: The NHS Information Centre. Available https://catalogue.ic.nhs.uk/publications/public-health/obesity/nati-chil-meas-prog-resu-2007-2008/nati-chil-meas-prog-resu-2007-2008-rep.pdf. Accessed 11 November 2015.
  15. 15. The NHS Information Centre (2009) National Child Measurement Programme: England, 2008/09 school year. London: The NHS Information Centre. Available https://catalogue.ic.nhs.uk/publications/public-health/obesity/nati-chil-meas-prog-eng-2008-2009/nati-chil-meas-prog-eng-2008-2009-rep.pdf. Accessed 11 November 2015.
  16. 16. The NHS Information Centre (2010) National Child Measurement Programme: England, 2009/10 school year. London: The NHS Information Centre. Available https://catalogue.ic.nhs.uk/publications/public-health/obesity/nati-chil-meas-prog-eng-2009-2010/nati-chil-meas-prog-eng-2009-2010-rep.pdf. Accessed 11 November 2015.
  17. 17. The NHS Information Centre (2011) National Child Measurement Programme: England, 2010/11 school year. London: The NHS Information Centre. Available https://catalogue.ic.nhs.uk/publications/public-health/obesity/nati-chil-meas-prog-eng-2010-2011/nati-chil-meas-prog-eng-2010-2011-rep1.pdf. Accessed 11 November 2015.
  18. 18. World Health Organization (2013) School and youth health: what is a health promoting school? Geneva: World Health Organization. Available http://www.who.int/school_youth_health/gshi/hps/en/. Accessed 11 November 2015.
  19. 19. Procter KL, Rudolf MC, Feltbower RG, Levine R, Connor A, Robinson M, et al. (2008) Measuring the school impact on child obesity. Social Science & Medicine 67: 341–349.
  20. 20. Williams AJ, Wyatt KM, Williams CA, Logan S, Henley WE (2014) A repeated cross-sectional study examining the school impact on child weight status. Preventive Medicine 64: 103–107. pmid:24732718
  21. 21. Merlo J, Chaix B, Yang M, Lynch J, Rastam L (2005) A brief conceptual tutorial of multilevel analysis in social epidemiology: linking the statistical concept of clustering to the idea of contextual phenomenon. Journal of Epidemiology and Community Health 59: 443–449. pmid:15911637
  22. 22. Durand CP, Andalib M, Dunton GF, Wolch J, Pentz MA (2011) A systematic review of built environment factors related to physical activity and obesity risk: implications for smart growth urban planning. Obesity Reviews 12: e173–182. pmid:21348918
  23. 23. Miller DP (2011) Associations between the home and school environments and child body mass index. Social Science & Medicine 72: 677–684.
  24. 24. Butland B, Jebb S, Kopelman P, McPherson K, Thomas S, Mardell J, et al. (2007) Tackling obesities: future choices—project report. 2nd ed. London: Department of Innovation, Universities and Skills.
  25. 25. South East England Public Health Observatory (2005) Choosing health in the South East: obesity. Oxford: South East England Public Health Observatory. Available http://www.sepho.org.uk/Download/Public/9783/1/SEPHO%20obesity%20report%20Nov%2005.pdf. Accessed 11 November 2015.
  26. 26. Townsend N, Rutter H, Foster C (2012) Age differences in the association of childhood obesity with area-level and school-level deprivation: cross-classified multilevel analysis of cross-sectional data. International Journal of Obesity 36: 45–52. pmid:22005718
  27. 27. Cole TJ, Freeman JV, Preece MA (1995) Body mass index reference curves for the UK, 1990. Archives of Disease in Childhood 73: 25–29. pmid:7639544
  28. 28. National Obesity Observatory (2010) National Child Measurement Programme 2008/09: guidance for analysis by Public Health Observatories and Primary Care Trusts. Oxford: National Obesity Observatory. Available http://www.noo.org.uk/uploads/doc/vid_4752_0809_NCMP_Analysis_Guidance.pdf. Accessed 11 November 2015.
  29. 29. Department for Education (n.d.) Statistics at DfE. Runcorn: GOV.UK. Available https://www.gov.uk/government/organisations/department-for-education/about/statistics. Accessed 11 November 2015.
  30. 30. Department of Health (2009) NHS Information Centre validation process for NCMP data. London: Department of Health. Available https://catalogue.ic.nhs.uk/publications/public-health/obesity/nati-chil-meas-prog-eng-2008-2009/nati-chil-meas-prog-eng-2008-2009-data.pdf. Accessed 11 November 2015.
  31. 31. Office for National Statistics (n.d.) Super Output Areas (SOAs). Newport: Office for National Statistics. Available http://www.ons.gov.uk/ons/guide-method/geography/beginner-s-guide/census/super-output-areas--soas-/index.html. Accessed 11 November 2015.
  32. 32. Department for Communities and Local Government (2011) The English indices of deprivation 2010. London: Department for Communities and Local Government. Available http://www.communities.gov.uk/documents/statistics/pdf/1871208.pdf. Accessed 11 November 2015.
  33. 33. Department for Communities and Local Government (2009) Local index of child well-being: summary report. London: Department for Communities and Local Government. Available http://webarchive.nationalarchives.gov.uk/20120919132719/www.communities.gov.uk/documents/communities/pdf/1126232.pdf. Accessed 11 November 2015.
  34. 34. Swinburn B, Egger G (2002) Preventive strategies against weight gain and obesity. Obesity Reviews 3: 289–301. pmid:12458974
  35. 35. Williams AJ, Henley WE, Williams CA, Hurst AJ, Logan S, Wyatt KM (2013) Systematic review and meta-analysis of the association between childhood overweight and obesity and primary school diet and physical activity policies. International Journal of Behavioral Nutrition and Physical Activity 10: 101. pmid:23965018
  36. 36. Williams AJ, Wyatt KM, Hurst AJ, Williams CA (2012) A systematic review of associations between the primary school built environment and childhood overweight and obesity. Health & Place 18: 504–514.
  37. 37. Sellström E, Bremberg S (2006) Is there a "school effect" on pupil outcomes? A review of multilevel studies. Journal of Epidemiology and Community Health 60: 149–155. pmid:16415266
  38. 38. West P, Sweeting H, Leyland A (2004) School effects on pupils' health behaviours: evidence in support of the health promoting school. Research Papers in Education 19: 261–291.
  39. 39. Williams AJ (2013) Determination of school-based contextual factors and their associations with the prevalence of overweight and obese children: Peninsula College of Medicine and Dentistry.
  40. 40. Macintyre S, Ellaway A, Cummins S (2002) Place effects on health: how can we conceptualise, operationalise and measure them? Social Science & Medicine 55: 125–139.
  41. 41. StataCorp (2009) Stata statistical software: release 11. College Station, TX: StataCorp LP
  42. 42. Statistical Consulting Group (2014) Stata FAQ: how to perform a factor analysis with categorical (or categorical and continuous) variables? UCLA: Statistical Consulting Group. Available http://www.ats.ucla.edu/stat/stata/faq/efa_categorical.htm. Accessed 11 November 2015.
  43. 43. Goldstein H (2003) Multilevel statistical models. 3rd ed. London: Hodder Arnold.
  44. 44. Goldstein H, Healy MJR (1995) The graphical presentation of a collection of means. Journal of the Royal Statistical Society Series A, (Statistics in Society) 158: 175–177.
  45. 45. Department of Health (2007) Legislative changes to the National Child Measurement Programme (NCMP). Richmond: The National Archives. Available http://webarchive.nationalarchives.gov.uk/+/www.dh.gov.uk/en/Publichealth/Healthimprovement/Healthyliving/DH_080606. Accessed 11 November 2015.
  46. 46. Department of Health (n.d.) National Healthy Schools programme: a guide for teachers. London: Department of Health. Available http://www.nice.org.uk/proxy/?sourceUrl=http%3A%2F%2Fwww.nice.org.uk%2Fnicemedia%2Fdocuments%2Fnhss_teachers-guide.pdf. Accessed 11 November 2015.
  47. 47. South West Healthy Schools Plus (2009) Handbook for schools. Bristol: South West Healthy Schools Plus Available http://www.swpho.nhs.uk/resource/item.aspx?RID=49524. Accessed 11 November 2015.
  48. 48. Chinn S (2000) A simple method for converting an odds ratio to effect size for use in meta-analysis. Statistics in Medicine 19: 3127–3131. pmid:11113947
  49. 49. R Core Team (2013) R:a language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. Available http://www.R-project.org/.
  50. 50. Bates D, Maechler M, Bolker B (2011) lme4: Linear mixed-effects models using S4 classes. R package version 0999375–42. Available http://CRAN.R-project.org/package=lme4.
  51. 51. Cole TJ, Bellizzi MC, Flegal KM, Dietz WH (2000) Establishing a standard definition for child overweight and obesity worldwide: international survey. British Medical Journal 320: 1240–1243. pmid:10797032
  52. 52. Miyazaki Y, Stack M (2015) Examining individual and school characteristics associated with child obesity using a multilevel growth model. Social Science & Medicine 128: 57–66.
  53. 53. Health and Social Care Information Centre (2013) National Child Measurement Programme: England, 2012/13 school year. Leeds: Health and Social Care Information Centre. Available http://www.hscic.gov.uk/searchcatalogue?productid=13778&q=title%3a%22national+child+measurement+programme%22&sort=Relevance&size=10&page=1#top. Accessed 11 November 2015.
  54. 54. School Food Plan (2014) The plan: Ofsted. London: School Food Plan. Available http://www.schoolfoodplan.com/ofsted/. Accessed 11 November 2015.
  55. 55. Richmond TK, Elliott MN, Franzini L, Kawachi I, Caughy MO, Gilliland MJ, et al. (2014) School programs and characteristics and their influence on student BMI: findings from Healthy Passages. PLoS ONE 9: e83254. pmid:24454697
  56. 56. Pallan MJ, Adab P, Sitch AJ, Aveyard P (2013) Are school physical activity characteristics associated with weight status in primary school children? A multilevel cross-sectional analysis of routine surveillance data. Archives of Disease in Childhood 99: 135–141. pmid:24152572
  57. 57. Conrad D, Capewell S (2012) Associations between deprivation and rates of childhood overweight and obesity in England, 2007–2010: an ecological study. BMJ Open 2: e000463. pmid:22505306
  58. 58. Johnson BA, Kremer PJ, Swinburn BA, de Silva-Sanigorski AM (2012) Multilevel analysis of the Be Active Eat Well intervention: environmental and behavioural influences on reductions in child obesity risk. International Journal of Obesity 36: 901–907. pmid:22531087
  59. 59. O'Malley PM, Johnston LD, Delva J, Bachman JG, Schulenberg JE (2007) Variation in obesity among American secondary school students by school and school characteristics. American Journal of Preventive Medicine 33: S187–194. pmid:17884567
  60. 60. Downey DB, von Hippel PT, Broh BA (2004) Are schools the great equalizer? Cognitive inequality during the summer months and the school year. American Sociological Review 69: 613–635.
  61. 61. von Hippel PT, Powell B, Downey DB, Rowland NJ (2007) The effect of school on overweight in childhood: gain in body mass index during the school year and during summer vacation. American Journal of Public Health 97: 696–702. pmid:17329660
  62. 62. Chen TA, Baranowski T, Moreno JP, O'Connor TM, Hughes SO, Baranowski J, et al. (2015) Obesity status transitions across the elementary years: use of Markov chain modelling. Pediatr Obes.
  63. 63. Moreno JP, Johnston CA, Chen TA, O'Connor TA, Hughes SO, Baranowski J, et al. (2015) Seasonal variability in weight change during elementary school. Obesity (Silver Spring) 23: 422–428.
  64. 64. Economos CD, Hyatt RR, Must A, Goldberg JP, Kuder J, Naumova EN, et al. (2013) Shape Up Somerville two-year results: a community-based environmental change intervention sustains weight reduction in children. Preventive Medicine 57: 322–327. pmid:23756187
  65. 65. Alexander KL, Entwisle DR, Olson LS (2007) Lasting consequences of the summer learning gap. American Sociological Review 72: 167–180.
  66. 66. Borman GD, Dowling NM (2006) Longitudinal achievement effects of multiyear summer school: evidence from the teach Baltimore randomized field trial. Educational Evaluation and Policy Analysis 28: 25–48.
  67. 67. Burkam DT, Ready DD, Lee VE, LoGerfo LF (2004) Social-class differences in summer learning between kindergarten and first grade: model specification and estimation. Sociology of Education 77: 1–31.
  68. 68. Gershenson S (2013) Do summer time-use gaps vary by socioeconomic status? American Educational Research Journal 50: 1219–1248.
  69. 69. Anderson PM, Butcher KF, Cascio EU, Schanzenbach DW (2011) Is being in school better? The impact of school on children's BMI when starting age is endogenous. Journal of Health Economics 30: 977–986. pmid:21733588
  70. 70. Lhachimi SK, Nusselder WJ, Lobstein TJ, Smit HA, Baili P, Bennett K, et al. (2013) Modelling obesity outcomes: reducing obesity risk in adulthood may have greater impact than reducing obesity prevalence in childhood. Obesity Reviews 14: 523–531. pmid:23601528