The study area includes eight neighbourhood areas with contrasting income levels and built environments across suburban municipalities of Metro Vancouver. Census tracts are small and relatively stable geographic units with an average population of 2500 to 8000 and were used as a base to select neighbourhood areas . The median family income of census tracts was obtained from the 2001 Census of Canada. Census tract residential density was calculated as population per hectares of residential land. Residential land was obtained from Greater Vancouver Land Use Data which assigns a land use code to all parcels . Neighbourhood areas were created by joining three to four census tracts to achieve a population between 11,000 and 17,000. Potential areas were selected based on deciles of income and residential density. Areas with the highest and lowest deciles of income were excluded as well as the lowest deciles of residential density which represented rural areas. Because our study areas are suburban and characterized by relatively low densities, the highest decile of residential density was not excluded. Among eligible census tract clusters we selected eight neighbourhood areas. Four neighbourhood areas were selected with higher residential density, two had higher median family income ($53,000-$77,000 CDN) and two had lower median family income ($32,000-$44,000). Four neighbourhood areas with lower residential density were selected, two had higher median family income ($53,000-$77,000 CDN) and two had lower median family income ($32,000-$44,000). Overall, we selected eight neighbourhood areas, two of each income/density classification (e.g. higher income, lower density). Further information on neighbourhood selection is available elsewhere .
A telephone survey was conducted by a contracted firm to obtain individual data for respondents in the selected neighbourhoods. Households were selected using Random Digit Dialling (RDD) based on a sampling frame obtained from a local telephone provider. Invalid and ineligible numbers were removed. Once a household was selected a minimum of five call-backs were attempted to minimize non-response bias. Interviews were conducted in English by experienced telephone interviewers using Computer Assisted Telephone Interviewing. The survey was conducted in February 2006, following a pilot survey in January 2006, and achieved a response rate of 29%. Data was obtained for 1935 adults aged 19 and over but 333 were excluded due to an invalid/missing postal code (n = 43) or item non-response (n = 290) resulting in a final sample of 1602. Ethics approval for data collection and analysis was sought and granted by the Office of Research Services at Simon Fraser University (application approval #38955).
Survey questions assessed participation in various types of walking and moderate physical activity. Dichotomous categories were constructed to create indicators of low physical activity and moderate or greater physical activity. Walking to work/school was assessed using the item: "In a typical week in the past 3 months how many hours did you usually spend walking to and from work or school?" Walking for errands was assessed using the item: "In a typical week in the past few months how many hours did you spend walking from home to grocery stores, banks, or to do other errands?" The responses to these questions (none, less than 1 hour, from 1 to 5 hours, from 6 to 10 hours, from 11 to 20 hours, greater than 20) were dichotomized for analysis purposes (less than one hour, one hour or greater). Moderate physical activity was assessed using the item: "In a typical week in the past 3 months how many days did you do at least 30 minutes of moderate physical activity such as brisk walking running swimming or team sports? never, 1 day, 2 or 3 days, 4 or 5 days, 6 or more". Responses were dichotomized for analysis (one day or less, two days or more). Walking for leisure was assessed from the question: "On a typical day in the past 3 months, how much time did you spend walking for leisure? 0 minutes, 15 minutes or less, 16-30 minutes, 31 minutes to one hour, over an hour." Responses were dichotomized for the analysis (15 minutes or less, greater than 15 minutes).
Individual level predictor variables include gender, age, household income, marital status, chronic conditions and obesity. Three categories of household income were constructed based on respondent self-report: low income (less than $40,000 CDN) middle income ($40,000 to $80,000 CDN) and high income ($80,000 CDN and over). Marital status of respondents was categorized as single, married/common law and divorced/widowed. Because chronic conditions may limit respondent's ability to engage in some physical activities, a variable indicating presence of a self-reported chronic condition was included. Obesity may be associated with lower levels of physical activity and to account for this, individuals with a Body Mass Index (BMI, weight (kg)/height (m) ) ≥ 30 were categorized as obese based on international standards . BMI was based on self-reported heights and weights.
Land use and neighbourhood income data
Line-based road network buffers were used to construct measures of land use based on prior work demonstrating that they offer a better representation than circular or "crow-fly" buffers of the neighbourhood that is accessible by walking . By being constrained to the road network, as actual pedestrians are, network buffers provide a more accurate assessment of the built environment as experienced by a resident walking through each neighbourhood. This is especially true in suburban areas which typically have lower street connectivity than urban areas. Respondents were geocoded using the Statistics Canada Postal Code Conversion File which assigns a latitude and longitude co-ordinate to each respondent's self-reported postal code . The British Columbia Road Network file was used to construct a one-km buffer around each postal code constrained to the road network. A 50-metre buffer was then placed around the line-based buffer to create a final buffer that was one-kilometre along the road and 50-metres on either side of the road. A detailed description of the construction of the buffer construction is available elsewhere .
Land use measures were constructed for each respondent's network buffer using Greater Vancouver Land Use Data. While land use codes differed across municipalities of the study region, a simplified layer has been constructed from more detailed land use codes to facilitate analysis across the region and a full description is available elsewhere . Four land use categories are employed for the present analysis:
Recreational and park land includes parks, play grounds, fields, and trails/wooded areas.
Residential land includes all private and rental dwellings such as high rises, low rises, garden/town homes, and single detached homes.
Commercial land includes businesses with retail sales and services and professional offices.
Institutional land includes public offices, hospitals, libraries, community centres, schools, city hall, and correction facilities.
For each respondent's line-based road network buffer, the proportion of land for each of the four land use categories was calculated. For each category, the proportions across all respondents were divided into three tertiles (low, mid, high) and the middle category was used as the reference for analysis. A fifth standard measure of land use mix was constructed by calculating the distribution of the four land uses and the measure was divided into thirds .
Because neighbourhoods were selected on the basis of income, dichotomous variables were included to indicate if a neighbourhood was in the higher or lower category. This strategy has been used in other studies . While density was also used as a selection variable, it was not included in analyses due to multicollinearity with the land use measures.
The main analytical strategy used was logistic regression to predict the influence of individual and land use characteristics on the four physical activity measures. Four sets of models are presented. Each set assesses the influence of the five land use measures on a single physical activity outcome while controlling for individual predictors and neighbourhood income. All statistical analyses were conducted using SPSS version 15.0.