Data sources and study sample

This study was based on the cross-sectional 2009/10 Canadian Health Behaviour in School-Aged Children Survey (HBSC), which consists of two components: (1) a classroom-based health survey conducted on a representative sample of children in grades 6 to 10, and (2) GIS measures of the environment in the neighborhoods surrounding the HBSC participants homes.

The 2009/10 HBSC is a cross-national survey conducted in affiliation with the World Health Organization. This study was limited to the Canadian sample. The survey covered several aspects of health, health behaviors, and physical and social determinants of health. The Canadian sample was designed according to the international HBSC protocol [12]. The sampling strategy followed a systematic multi-stage cluster technique, whereby individual students are nested in school classes that are nested within schools and school boards. The 2009/10 Canadian HBSC included 26,078 students from 436 schools, with distributions reflecting the distribution of Canadians in grades 6 to 10 (approximate age range 11 to 15-years-old). All provinces and territories in Canada participated with the exception of Prince Edward Island and New Brunswick. Students enrolled in private, special needs, or home schools, as well as incarcerated youth, were excluded; combined they contribute to <10% of the 11 to 15-year-old population. Consent was obtained and provided by school boards, individual schools, participants, and their parents/guardians. Ethics approval was obtained from the General Research Ethics Board of Queen’s University and Health Canada’s ethics board.

Given the objective of this study, an a priori decision was made to examine children and not adolescents. Therefore, 6,667 grade 9 and 10 high school students were excluded from the original sample of 26,078 grade 6 to 10 students. Because postal codes were used as proxies of the home address when obtaining the neighborhood environment measures (explained below), of the 19,411 grade 6 to 8 children, 8,032 with missing or invalid postal code data were excluded. Next, we excluded 2,351 students from rural areas as postal codes cover a large geographic area in rural areas; in Canadian urban settings postal codes cover a small geographic area (i.e., 1 or 2 blocks) [13]. An additional 3,524 participants were excluded because Google Earth street view images were not available for their neighborhood, and these images were needed to measure the undeveloped green spaces. Finally, 366 participants were excluded because they did not report their physical activity or covariate data in the questionnaire. The final sample consisted of 5,138 grade 6 to 8 students who were approximately 11 to 13-years-old.

Undeveloped green space in the home neighborhood (exposure)

A 1 km radius circular buffer around participants’ homes, as measured from the geographic center of their postal code, was used to define their home neighborhoods. A 1 km distance represents a 10 to 15 minute walking time [14]. Previous research supports this as an appropriate buffer size for children. The majority of children are not allowed to not travel >1 km from home unsupervised [15]. Measures of street connectivity obtained for a 1 km buffer were more consistently related to physical activity in youth than similar measures obtained at larger buffer sizes [16]. Furthermore, in the Canadian HBSC fast food restaurant measures obtained for a 1 km buffer are more strongly related to eating behaviors than measures obtained for smaller and larger buffer sizes [14].

The percentage of land area comprised of undeveloped and publicly accessible green space was determined for each neighborhood. The undeveloped green space consisted of meadows and treed areas. A meadow is defined here as a field vegetated primarily (>50%) by grass and other non-woody plants. A treed area is defined here as a field vegetated primarily (>50%) by trees and shrubs. Green areas that were not publicly accessible (e.g., fenced off, industrial property, yards at peoples’ homes), that were covered by water (e.g., lakes, wetlands), and that consisted of developed green space (e.g., parks with amenities such as a playground, sports fields, school grounds) were excluded. The exclusion of these areas essentially left meadows and treed areas as the remaining green spaces.

To process for measuring undeveloped green space started in ArcGIS version 10.2 software (Esri, Redlands, CA). First, we created 1 km circular radius buffers around the central point of each participant’s postal code to define their neighborhood. Next, for each of the participating schools we created a polygon that covered all of the neighborhood buffers for that school’s participating students. These polygon shapes were exported into Google Earth (Google, Mountain View, CA) where they were added as a layer on top of Google Earth satellite images. A layer of developed park spaces was also exported from ArcGIS into Google Earth to help distinguish between developed and undeveloped green space. Once these layers were superimposed on the satellite images in Google Earth, potential undeveloped meadows were identified by carefully inspecting the satellite images. These potential undeveloped meadows were then inspected at the ground level using the Google Earth street view tool. Areas that were confirmed as meadows were then outlined with a polygon shape on the satellite images using the Google Earth editing tools. This process was repeated until all of the meadows in the larger polygon were identified, and then repeated to identify the treed areas. Once completed, the newly developed layers that contained the smaller polygons for the meadows and treed areas were exported into ArcGIS. ArcGIS was then used to calculate the land coverage area of these undeveloped green spaces and the total land area (not including water and wetlands) separately for each participant’s neighborhood buffer. We then calculated the percentage of total land area that was made up of the undeveloped green spaces.

All of the undeveloped green space measures were obtained by a single rater. To determine the intra-rater reliability of the undeveloped green space measures, that rater completed the measures a second time, several weeks after obtaining the initial measures, for 1,404 participants from 33 schools. There was excellent agreement between the land coverage values for the first and second measures with correlation coefficients of r = 0.99 for meadows and treed areas.

Physical activity in free-time outside of school (outcome)

Responses to the question “Outside school hours: how often do you usually exercise in your free time so much that you get out of breath or sweat?” were used to measure physical activity in free-time outside of school hours. Ordinal responses to this question were: “Never”, “Less than once a month”, “once a month”, “once a week”, “2 to 3 times a week”, “4 to 6 times a week”, and “Every day”. This question has been a mandatory item in the international HBSC survey dating back to the 1989/1990 cycle.

Confounding variables

Variables considered as potential confounders as self-reported in the HBSC were gender, grade, race (white or other), number of parents in the household (single or dual), perceived neighborhood safety (assessed with the question “It is safe for younger children to play outside during the day” with 5 ordinal responses), and perceived family wealth (assessed with the question: “How well off do you think your family is?” with five ordinal responses). GIS measured confounders included average income in the neighborhood as obtained from the 2006 Census of Population, the number of recreational facilities in the neighborhood, the proportion of neighborhood land area made up of developed park and playground space, and the percentage of the total road distance within the buffer that was comprised of low speed roads (i.e., speed limit ≤50 km/h). These GIS measures were obtained in the 1 km neighborhood buffers and are described elsewhere [8]. Finally, we considered whether the survey was administered in the winter (December-March), fall (September-November), or spring (April-June).

Statistical analysis

Analyses were performed in SAS version 9.3 (SAS Inc., Carry, NC) and accounted for the sample weights and clustered nature of the data. Distributions of key variables were characterized using conventional descriptive statistics. Spearman correlations were used to explore the relationships between the physical environment measures. Bivariate ordinal logistic regression models were initially used to describe the relationships between the undeveloped green space exposures and confounders with the physical activity outcome. This was followed by multivariate ordinal logistic regression models that included the primary exposure variables and all of the confounders that were related (p < 0.1) to the outcome in the bivariate models. With the exception of gender, race, parents in household, and winter season, all of the exposure and confounding variables were entered into the ordinal logistic regression models as continuous variables. The odds ratios and 95% confidence intervals from these models are expressed as follows: per 5% change in the proportion of land coverage for the undeveloped green space and developed park space variables, per each grade, per one unit change in ordinal responses for the perceived family wealth and perceived neighborhood safety variables, per $10,000 for average household income, per each additional neighborhood recreational facility, and per 10% change in the proportion of neighborhood road distance made up of low speed roads. Analyses indicated that there were no age or gender interactions; therefore, all participants were included in the same regression models.