The neighborhood social environment and body mass index among youth: a mediation analysis

Child and adolescent overweight and obesity represents a significant global public health burden, with a high and increasing prevalence in many developed countries, serious health consequences both during childhood and adolescence and in adulthood, and increased likelihood of remaining overweight and obese as adults [1, 2]. Prevention of overweight in childhood and adolescence is therefore critical and a thorough understanding of influences on the development of overweight is needed. Ecological models suggest that childhood overweight is the result of a complex array of factors operating at multiple levels, from individual to family to school, neighborhood, community and policy influences [3]. Recently, the importance of neighborhood environments for promoting or hindering youth physical activity and active transport has received considerable attention [4] and many studies have also examined the influence of neighborhood environments on youth overweight [5–7]. However, most of those studies focus on the built environment. Studies of social factors within neighborhoods and youth overweight are less common and have generally been limited to an examination of perceived safety and crime [7]. Constructs such as neighborhood social cohesion and social networks are rarely studied. However, it has been suggested that socially cohesive neighborhoods may influence youth physical activity by facilitative enforcement of healthy norms, community awareness of programs and facilities, collective action to improve the local area and limiting crime and disorder [8].

Associations between social capital, trust or cohesion and physical activity have been observed cross-sectionally [9–11] and longitudinally [8]. Positive associations between social ties or networks (i.e. area has lots of children, friends live nearby, know neighbors) have also been reported for walking and cycling cross-sectionally [12, 13] and longitudinally [14]. However, few studies have examined such social constructs in relation to overweight and all are cross-sectional [10, 15–17]. Those studies found inverse associations between the social environment and overweight, but each examined different aspects of the social environment, including social trust [10], informal social control [16] and collective efficacy [17].

Although there is some evidence that neighborhood social environments are associated with overweight in children, this evidence is based mainly on cross-sectional samples, with little consistency in measures of the social environment. In addition, no studies have considered pathways through which the social environment may operate to influence childhood obesity. Physical activity, time outdoors, sedentary time (objectively assessed time spent sedentary), and screen-time (watching TV, using the computer and playing electronic games) are likely to be on the causal pathway between the social environment and overweight as these behaviors have been shown to be associated with body mass index (BMI) in youth [18, 19] and are also likely to be influenced by the social environment [20]. Two key aspects within the social environment are social networks (i.e. how well neighbours know each other and whether there are other children around to play with) and social trust and cohesion between neighbours. Therefore, this study aimed to examine: 1) cross-sectional and prospective associations between neighborhood social networks and social trust/cohesion and BMI among youth; and 2) to determine whether these associations were mediated by the time spent in a range of physical activity behaviors, screen-time and total sedentary time. It was hypothesized that children who lived in neighborhoods with strong social networks and cohesion would spend less time sedentary and in front of screens and more time in moderate- to-vigorous physical activity (MVPA), more time outdoors, and would take more walking and cycling trips in their neighborhood, which would reduce the likelihood of overweight and obesity.

The data in this study were drawn from the first and second follow-up data collection waves of the 'Children Living in Active Neighborhoods (CLAN)' study [21, 22]. At baseline (2001), 1210 families of 5-6 year old and 10-12 year old children were recruited from 19 state schools in high (n = 10) and low (n = 9) socio-economic areas of metropolitan Melbourne (44% response rate from those individuals invited). Schools were selected using stratified random sampling proportionate to school size, based on the Australian Bureau of Statistics' Socioeconomic Indices for Areas [23]. Recruitment of the baseline sample has been reported previously [21, 22]. Families who indicated that they were interested in participating in further research were re-contacted in 2004 (n = 587). The sample comprised 189 children aged 8-9 years and 398 children aged 13-15 years. This process was repeated in 2006 (n = 487). The present study includes both the younger and older cohorts of children and examines data from 2004 and 2006. Only participants with complete outcome variable (BMI) data in 2004 (n = 544) and 2006 (n = 430) respectively were included in the cross-sectional and prospective analyses. Participants included in the present study did not differ significantly from those who did not continue past baseline or who had missing BMI data with regards to the sex of the child or the age of the parent completing the survey. There was, however, a significant difference with regard to BMI. Those included in the analyses had a lower BMI than those who did not continue past baseline or with missing BMI data (18.3 vs 19.4 kg/m², p < 0.05).

More female carers of children with complete BMI data also reported to have a high educational level at baseline compared with females who did not participate in follow-up data collections (41.2% vs 30.5% had high level of education, p < 0.05). Written consent was required from parents as well as from the older cohort of children. The Deakin University Human Research Ethics Committee, the Victorian Department of Education and the Catholic Education Office provided ethical approval.

Statistical analyses

To examine the cross-sectional (2004 only) and prospective associations between the neighborhood social environment and BMI, and to examine the mediating effects of sedentary and physical activity behaviors on this association, a series of multi-level linear regression analyses (using MLwin version 2.02) were conducted. In the prospective analyses, BMI values measured in 2006 were used as the outcome variable, while the values measured in 2004 were used for the predictor and mediator variables. Two levels were defined in the multi-level regression: 1) student, and 2) school. We checked for possible confounding by sex, socioeconomic status (SES) and age. To test whether SES was a confounder, two indicators of SES were included separately in the main regression analyses: 1) maternal education; and 2) area level SES (based on the score for the Socio-Economic Index for Areas (SEIFA) Index of Relative Socio-Economic Advantage/Disadvantage score for each child's residential postal code). As there was no evidence of confounding, neither of the SES indicators were included in the final models.

The cross-sectional and prospective associations between the neighborhood social environment and BMI was calculated (c-coefficient/total effect), by regressing the social environmental variable (i.e. social network or social trust/cohesion) and confounders on the outcome variable (i.e. BMI 2004, or BMI 2006) (see Figure 1). The mediated effect of screen-time, sedentary time, MVPA time, time spent outdoors, and the number of walking/cycling trips on the association between neighborhood social environment and BMI, was then calculated using the products of coefficient method [35]. First, the associations between the neighborhood social environment and the potential mediators were calculated using separate regression models (a-coefficient). Second, the associations between the potential mediators and BMI, controlling for the neighborhood social environment variables were calculated using separate regression models (b-coefficient). This model also provided the direct effect of the neighborhood social environment on BMI, adjusting for potential mediators (c'-coefficient). The mediated effect is the product of the a and b coefficient (a*b) and provides an estimate of the relative strength of the mediation effect. The Sobel test was used to assess the statistical significance and 95% confidence intervals of a mediating effect by dividing the products-of coefficients (a*b) by its standard error SE_ab = √((a²*SE_b²) + (b²*SE_a²)). In this equation, 'a' stands for the a-coefficient and 'b' stands for the b-coefficient.

Table 1 shows the characteristics of the sample and the distribution of all variables. The average age of children in 2004 was 12.6 years (47% boys), and the average age of the parent completing the survey in 2004 was 42.8 years (88% female). Average BMI of children was 20.5 in 2004 and 21.4 in 2006.

This study examined cross-sectional and prospective associations between aspects of the neighborhood social environment and BMI, and examined whether this association was mediated by a range of physical activity and sedentary behaviors. Consistent with previous cross-sectional studies [10, 16, 17], we found negative associations between neighborhood social networks and social trust/cohesion and BMI, with stronger associations cross-sectionally between social networks and BMI. Interestingly, inverse associations were also found prospectively two years later. This indicates that the more positive the neighborhood social networks and social trust/cohesion in 2004, the lower the children's BMI at that time point and two years later. These findings suggest that these aspects of the neighborhood social environment may be important determinants of overweight and obesity in youth. Maternal and area-level SES did not confound these associations.

In order to explain associations between neighborhood social networks and social trust/cohesion and overweight it is important to examine the associations between social networks and social trust/cohesion and the potential mediators (a-coefficient) and the associations between the mediators and BMI (b-coefficient). Our 2004 cross-sectional findings suggest that children who live in a neighborhood with strong perceived social networks tend to spend less time in screen-time, more time in MVPA outside-school hours, and more time outdoors. These results confirm other studies [9, 12, 13, 15] including a study by Franzini et al. that found that the neighborhood social environment was positively associated with self-reported physical activity, which was negatively associated with child obesity [15]. In the current study, no associations were found between social networks and the number of walking/cycling trips or percentage of sedentary time outside-school hours. Social trust/cohesion was not associated with physical activity, screen-time or sedentary time. This indicates that how well people get along with and trust their neighbors may be of less importance to these behaviors than factors relating to social networks, such as children having other children nearby to play with. This is consistent with our previous cross-sectional research that highlighted the importance of social networks to children's outdoor play in the neighborhood [36, 37]. Given the inconsistent findings across the measures of the social environment, further research using different measures of aspects of the neighborhood social environment may be warranted.

Through our examination of associations between the potential mediators and BMI (b-coefficient) we also identified that children who spent more time in front of screens had a higher BMI both cross-sectionally and two years later. Surprisingly, neither total sedentary time outside-school hours, as assessed by accelerometry, or any of the physical activity behaviors were associated with BMI in any analyses. Previous cross-sectional studies have shown inverse associations between children's objectively measured physical activity and BMI [38], and positive associations between children's screen-time and overweight and obesity [19]. Epstein et al. (2008) demonstrated that changes in screen-time caused changes in adiposity in young children, and this was the result of changes in energy intake rather than physical activity [39]. In addition, some studies suggest displacement of physical activity with screen-time, although these data are equivocal [19]. A small number of studies have examined the relation between objectively-assessed sedentary time and children's overweight and obesity. A cross-sectional study involving over 5,000 12 year old child in the UK found for every hour per day spent sedentary (operationalized as < 200 accelerometer cpm) after adjusting for sex, social factors, sleep, TV time and pubertal status, children were 32% more likely to be obese (body fat was measured by DEXA) [40]. Interestingly, the relationship between sedentary time and obesity was no longer significant once analyses adjusted for MVPA. Consistent with findings from the current study, few other cross-sectional or prospective studies have shown that objectively measured sedentary time is associated with obesity in children and adolescents [41–43].

While it is intuitive that positive social environments are likely to influence weight by providing environments in which youth feel safe enough to walk, cycle, or be active, the findings from this study showed no evidence that the physical activity or sedentary behaviors included in our model mediated the association between neighborhood social networks and social trust/cohesion and BMI. The lack of mediated effect indicates that the social environmental determinants did not exert their effects on BMI via the sedentary and physical activity behaviors assessed. The working mechanisms through which neighborhood social networks and social trust/cohesion exert their effect on BMI are therefore still unclear. Although, the measures of SES available in this study did not confound the associations between the social neighborhood variables and BMI, it is possible that neighborhoods that score high on social environment are neighborhoods with greater levels of socio-economic advantage. It may also be possible that children living in areas with high perceived social environment may have healthier eating patterns, which may explain mediating pathways with BMI. Future research could incorporate children's eating behaviors in addition to physical activity and sedentary behaviors.

The study had some limitations. The neighborhood social environment was limited to the examination of social networks and social trust/cohesion as reported by parents and children may have perceived their social environment differently to their parents. The inclusion of other aspects of the social environment and different ways of measuring social networks and social trust may have also resulted in alternative findings. In addition, parents reported screen-time and walking and cycling trips in the neighborhood for the children in the younger cohort, whereas these behaviors were self-reported by the older cohort. There were also significant differences in BMI and maternal education between those included in the present study and those who did not continue past baseline or who had missing BMI data. Further, due to the sample size it was not possible to analyze data separately by age and sex. It may be that social networks and social cohesion/trust exert a greater influence on physical activity levels of older children who are more likely to be able to walk and cycle in their neighborhood without adult supervision [22]. In addition, other factors known to influence obesity and that might confound the relationship between the neighborhood social environment and obesity (e.g. parental obesity, cultural background/ethnicity, diet and sleep) were not assessed. Future studies may benefit from including dietary intake data as a potential mediating variable and including additional aspects of the social environment.

Despite these limitations, this research is unique in examining prospective associations between the participants' perceived neighborhood social environment and BMI two years later in youth. In addition, to our knowledge, this is the first study to have examined the mediating mechanisms through which the social environment may operate to influence BMI in youth. Including objectively measured behaviors (i.e. MVPA and sedentary time) is a further strength.

In conclusion, the study findings suggest that neighborhood social networks and social trust/cohesion may be important in preventing overweight and obesity in youth. Considering that there was no evidence for a mediating effect of a range of physical activity behaviors, screen-time or sedentary time, future research investigating the working mechanisms through which the neighborhood social environment exerts its effect on BMI is needed. These studies should include measures of dietary intake and more precise measures of physical activity and sedentary behavior.