Distributions of participants according to individual-level and area-level characteristics are shown by level of street connectivity in Table 1
. There were noticeable and statistically significant differences in street connectivity according to family SES, neighborhood safety, amounts of litter in neighborhoods, and rundown homes in neighborhoods. For example, 40.7% of students in the lowest family SES group resided in a neighborhood in the highest street connectivity quartile as compared to 23.8% of students in the highest family affluence group (p < 0.0001). Level 2 (area-level) measures of SES were also associated with street connectivity such that higher SES areas had higher street connectivity scores. Geographic location was associated (p < 0.0001) with street connectivity; all students in the highest connectivity quartile resided in an urban core.
A description of physical activity levels in the total sample and by gender and grade is presented in Table 2
. Of the total sample, 5.5% participated in no physical activity outside of school hours, 37.0% participated in at least 4 hours per week of physical activity outside of school, and 16.4% participated in at least 7 hours per week of physical activity outside of school. A lower percentage of those who participated in no physical activity outside of school were males than females (44.6% vs.
55.4%), while a higher percentage of those who participated in at least 4 hours per week of physical activity outside (54.9% vs.
45.1%) and at least 7 hours per week of physical activity outside of school (59.4% vs.
40.6%) were males than females. Physical activity levels outside of school were slightly higher in Grade 6–8 than Grade 9–10 students. For instance, while 57.7% of those who accumulated less than 4 hours per week of physical activity outside of school were comprised of Grade 6–8 students, only 54.8% of those who accumulated 4 or more hours per week of physical activity outside of school were comprised of Grade 6–8 students.
Of the total sample, 26.9% (n = 2,296) were in street connectivity group 1 (highest connectivity), 21.7% (n = 1,851) were in street connectivity group 2, 27.8% (n = 2,374) were in street connectivity group 3, and 23.6% (n = 2,014) were in street connectivity group 4 (lowest connectivity). The percentage of the sample that were physically activity (i.e.
, 4 hours per week of physical activity outside of school) within each of the street connectivity groups ranged from a low of 30.7% in street connectivity group 1 to a high of 38.6% in street connectivity group 4 (Table 3
). Table 3
lists the bivariate relations between the three street connectivity measures and the overall street connectivity scales with physical activity. All three connectivity measures, as well as the composite street connectivity scale, were associated with physical activity in a consistent fashion. For the overall street connectivity scale, by comparison to group 1 (highest connectivity), the relative risks of being physically active outside of school hours were higher in group 2 (RR: 1.29, 95% CI: 1.16–1.42), group 3 (RR: 1.31, 95% CI: 1.18–1.44), and group 4 (RR: 1.26, 95% CI: 1.14–1.40). The three components of the street connectivity scale (connected node ratio, intersection density, and average block length) were also related to the physical activity outcome such that the relative risks were significantly increased in groups 2, 3 and 4 by comparison to group 1. Note that the relations presented in Table 3
are bivariate relations. In other words, the relative risks presented in this table were not adjusted for any of the confounding variables.
presents the associations between the covariates and the physical activity outcome. Of the individual-level (Level 1) covariates, gender, grade, family SES, perceived neighborhood safety, and perceived litter in the neighborhood were all significant independent predictors of physical activity outside of school hours (see Multivariate Model 3). While the geographic location and parks/recreational facilities area-level (Level 2) covariates were related to physical activity in the bivariate models, they were no longer significant in multivariate model 2, implying that they were not independent predictors of physical activity. Similarly, the perceived rundown homes and area-level SES variables were not related to physical activity.
The results of the multivariate model building process for the association between street connectivity and physical activity is also shown in Table 4
. Street connectivity was significantly associated with physical activity outside of school hours. This relationship was consistent between the bivariate and three multivariate models. In other words, street connectivity remained a significant predictor of physical activity outside of school hours after adjustment for salient covariates (gender, grade, family SES, perceived safety, and perceived litter). The final multivariate model (model 3) suggested that, compared to students living in the first (highest) street connectivity quartile, those in the second (RR: 1.22; 95% CI: 1.10–1.35), third (RR: 1.25; 95% CI: 1.13–1.37), and fourth (lowest; RR: 1.21; 95% CI: 1.09–1.34) street connectivity quartiles were significantly more likely to be physical activity for 4 hours per week outside of school hours.
Based on the RR estimates provided for the street connectivity scale in multivariate model 3 in Table 4
, and the prevalence of the study sample in the different street connectivity groups (Table 3
), the population attributable risk for the physical activity outcome was calculated as following: [21.7%(1.22 − 1)/(1 + 21.7%(1.22 − 1)] + [27.8%(1.25 − 1)/(1 + 27.8%(1.25 − 1)] + [23.6%(1.21 − 1)/(1 + 23.6%(1.21 − 1)]. This population attributable risk calculation suggested that 15.8% (95% CI: 7.7–23.8) of the physical activity outcome in the study sample was attributable to not living in the most highly connected street connectivity group (group 1). In other words, had none (0%) of the sample been in the most highly connected street connectivity group (group 1), the prevalence of physical activity in the sample would have been 15.8% higher.
presents a sensitivity analysis conducted within a subset of the HBSC survey. This subset consisted of 2,922 English speaking grade 9 and 10 students from the province of Ontario in whom supplemental information on vehicle traffic, stoplights/stop signs, and bike lanes/sidewalks was captured. As shown in the final multivariate model (model 2), high levels of vehicle traffic (RR: 0.87, 95% CI: 0.76–0.98) and the presence of stoplights or stop signs at busy intersections (RR: 1.16, 95% CI: 1.01–1.30) were related to the physical activity outcome, albeit in opposite directions. Conversely, the availability of bicycle lanes and sidewalks was not related to the physical activity outcome (p = 0.78 from bivariate model). A comparison of the RR estimates for the street connectivity scale in multivariate model 1 and multivariate model 2 indicate that that adjustment for the vehicle traffic and stoplights/stop sign measures did not alter the associations between the street connectivity and physical activity measures.
Youth from neighbourhoods with lower street connectivity scores (i.e., quartiles 2–4) were more likely to be physically active outside of school than youth from neighbourhoods with the highest street connectivity scores (i.e., quartile 1). There appeared to be a threshold effect for street connectivity as the relative risks for physical activity, while different in the highest street connectivity quartile, were quite similar in each of the lower three quartiles. The population attributable risk estimates suggest that 15.8% of the physical activity outcome in the study sample was explained by street connectivity. Thus, from a public health perspective street connectivity has a meaningful impact on the physical activity of young people.
In addition to street connectivity, several of the covariates that were examined in this study were independently associated with physical activity. Participants reporting a high perceived safety of their neighbourhood were 1.47 (95% CI: 1.34–1.59) times more likely to be physically active outside of school hours, participants from a high family SES were 1.45 (95% CI: 1.30–1.61) time more likely to be physically active outside of school, and girls were 0.73 (95% CI: 0.86–0.77) times less likely to be physically active outside of school. Our final models controlled for the aforementioned factors.
The threshold effect we observed for the street connectivity exposure on the physical activity outcome is an important finding. Youth in the high street connectivity quartile were less likely to be physically active, and closer examination revealed that each of the schools in this quartile was located in the urban core of a large Census Metropolitan Area (e.g., Toronto, Montreal, Vancouver). While schools in the second most connected quartile were mainly from urban cores as well, these schools were located in less densely populated urban cores. Therefore, students living in the most highly populated urban cores reported considerably lower levels of physical activity outside of school hours than their peers. Based on the street connectivity illustrations shown in Figure 2
, it is clear that the 5 km buffers around the schools capture vastly heterogeneous environments, with relatively dense street networks around most schools and varying levels of reductions in the street network density at further distances from the schools. By using the street connectivity score within the 5 km buffer as a proxy for the residential neighbourhood of all students attending that school, we do appreciate the fact that misclassification of our key study exposure occurred. Thus, the associations between street connectivity and physical activity that were observed in our study were likely underestimated.
Other factors, such as vehicle traffic, may have influenced the lower levels of physical activity reported by students living in the most highly connected neighbourhoods. Increased traffic in highly populated and connected neighbourhoods could lead to parent and youth concerns of outdoor safety and subsequently to a decrease in youth physical activity participation. While perceived vehicle traffic was related to physical activity levels in the subset of the study sample in whom these vehicle traffic measures were obtained (Table 5
), adjustment for vehicle traffic in the multivariate model did not alter the affect estimates for street connectivity. Therefore, perceived vehicle traffic did not mediate or account for the relationship between street connectivity and physical activity. Furthermore, the perceived availability of bike lanes and sidewalks was not related to physical activity and did not mediate the relations between street connectivity and physical activity. This suggests that the relations between street connectivity and physical activity were not related to the active transportation component of physical activity.
Another potential explanation of the observed disparity is the lack of outdoor play space in neighbourhoods with highly connected streets. Homes in neighbourhoods with the highest connectivity are packed very tightly together (Figure 2
), leaving little room for yards and driveways for young people to use for physical activity, which may lead to a decrease in outdoor activity. Also, the short blocks and lack of cul-de-sacs may make it difficult to play on the street. Poorly connected neighbourhoods with many cul-de-sacs present a space for youth to play, in a relatively safe and low traffic environment [11
]. Future studies should consider the concept of outdoor space as a determinant of physical activity for youth residing in highly connected neighbourhoods, and attempt to characterize areas in which young people most often play outdoors.
Findings of this study are disparate from some of the five previous studies that examined the relationship between street connectivity and moderate-to-vigorous physical activity in youth. Mota et al.
] reported no association, Norman et al.
] and Boone-Heinonen and Gordon-Larsen [16
] reported higher levels of physical activity in less connected areas for girls but not boys, and Kligerman et al.
] and Leung et al.
] reported higher levels of physical activity in more connected areas for both genders. The lack of consensus in these studies may be explained by their use of varying measures of connectivity, their study of highly specific geographic areas, and their comparatively small sample sizes. The size (n = 8,535), heterogeneity, nationally-representative nature, and use of a comprehensive and objective measures of street connectivity in our study is a methodological improvement on past research, and may explain why this study identified a different relationship.
The presence of lower levels of physical activity in highly connected neighborhoods is important as public health interventions that target youth physical activity levels have the potential to greatly impact population health. A challenge public health officials and urban planners will face when developing strategies for optimizing street connectivity is that the relations between street connectivity and physical activity for youth, as reported here, are in opposite direction to those previously reported for active transportation in adults [4
]. Thus, the current public health and urbanist movement to create highly connected neighbourhoods [27
], with the goal of increasing active transportation, may have a negative effect on the physical activity patterns of our youth. In light of these observations, consideration should be given to neighbourhoods that have a low street connectivity, but contain networks of pedestrian paths to increase their overall connectivity [28
]. Development of neighbourhoods that are conducive to physical activity in all age groups, while challenging, have the potential to substantially ameliorate the health of the population.
Limitations of this study merit consideration. First, the cross-sectional nature of the HBSC data makes this research limited in its ability to determine the temporality of any observed relationship. With that being said, it is unlikely that active youth would be able to influence their family to move to less connected neighbourhoods to promote their physical activity, supporting the implied temporal sequence. Second, there is potential for area-level associations to be residually confounded by variables not captured in this research. An example of this is parental influences. Street connectivity influences adult physical activity [4
], and parental physical activity is known to be a determinant of childhood physical activity [29
]. Ethnicity has also been shown to affect youth physical activity levels [29
], and not accounting for this variable in the analyses may also have resulted in residual confounding. Third, the 95% confidence intervals surrounding the relative risk estimates, as well as population attributable risk estimates derived from these relative risk values, may be biased slightly relative to more directly measured estimates of relative risk and associated population attributable risk estimates. Fourth, there is the possibility of misclassification on the basis of the street connectivity exposure. Since this research used a 5 km radius around the schools as a proxy for home neighbourhoods, this may have resulted in area-level characteristics being ascribed to students who in fact do not live within this radius. As no standard method exists for the measurement of neighbourhood environments, it was unclear what type of buffer should be used (radial versus
street network buffers) and what radial distance around schools would be appropriate as a proxy to capture the participants’ home neighbourhood environment. It would be ideal to use a smaller buffer around participants’ homes; however, the Health Behaviour in School-Aged Children survey did not collect the necessary personal information on student addresses as it was an anonymous survey.