Health Reports
A pan-Canadian measure of active living environments using open data

by Thomas Herrmann, William Gleckner, Rania A. Wasfi, Benoît Thierry, Yan Kestens and Nancy A. Ross

Release date: May 15, 2019

DOI: https://www.doi.org/10.25318/82-003-x201900500002-eng

Modifiable elements of neighbourhood environments (e.g., number of sidewalks, proximity to commercial services, population density) can increase rates of active transportation (walking and cycling for the purpose of transportation, and using public transportation).Note 1Note 2Note 3Note 4Note 5 Public health and urban planning researchers often measure three characteristics of communities that support active travel: higher street connectivity (e.g., intersection density, route directness), higher density (e.g., population density, dwelling density), and greater numbers and diversity of nearby destinations.Note 4Note 6Note 7 Canadian research suggests that exposure to these favourable “active living environments” is associated with more optimal markers of health, including more optimal systolic blood pressure,Note 8 decreased obesity, overweight and diabetes prevalence,Note 9 and improved body mass trajectories among men.Note 10

A national metric of active living environments is desirable to facilitate the direct comparison of communities, national surveillance of population health and data linkage with existing Canadian national health surveys (e.g., Canadian Community Health Survey, Canadian Health Measures Survey) and investigator-led cohort studies. Currently, very few pan-Canadian measures of the active living environment exist and those that do are not readily accessible or free to use.Note 11

The purpose of this paper is to describe the development of the Canadian Active Living Environments (Can-ALE) dataset: a Canada-wide set of four individual and four summary measures that characterize the favourability of active living environments in Canadian communities at the dissemination-area (DA) level (Figure 1). This study reports on analyses which guided the selection of measures and derivation data sources to select for the dataset. The objective was to produce a national database entirely from open data and to evaluate the performance of open data compared with traditional or proprietary sources. A 2006 version and a 2016 version of Can-ALE are available to download online (https://nancyrossresearchgroup.ca/research/can-ale/).

Data and methods

Study design

Multiple candidate measures of active living environments were created to represent connectivity, density and access to destinations for 56,589 DAs (2016 data) in Canada using a Geographic Information System (GIS). The candidate measures were derived using ArcMap v10.5, and the measures included in the final Can-ALE dataset were derived using PostGIS v2.3.3, a GIS extension for the PostgreSQL object-relational database management system (coding available from authors upon request). DAs are geographic units defined by Statistics Canada. They are the smallest standard geographic area for which complete census data are disseminated across Canada and have populations of 400 to 700 people.Note 12 A correlation analysis was used to inform decisions on the measures used in the final pan-Canadian dataset. These decisions were guided by principles according to which the selected measures were (1) associated with walking rates, (2) suitable to a variety of built environments (e.g., urban, suburban, rural), and (3) openly available to researchers and the public health community. A k-medians cluster analysis—an iterative process used to assign observations (in this case DAs) to a group with the closest within-group median values and maximal between-group median values—was conducted to guide the creation of a categorical variable characterizing the favourability of the active living environment of Canadian communities for both research and public health surveillance.

Unit of analysis

The principal unit of analysis was a circular, one-kilometre buffer around the centroid of the DA. Previous studies have shown differences in associations between measures and walking behaviour or health outcomes according to the type of geographic unit used to derive them.Note 13Note 14Note 15 As a result, buffer choices are often debated by researchers. To determine the most appropriate buffer shape and size for the national dataset, 12 active-living measures were derived using 4 types of buffers, which varied according to shape (circular versus street-network-based) and radius (500 metres versus 1 kilometre) for a subset of DAs on the island of Montréal. Using walking-for-transportation rates derived from the 2013 Montréal Origin-Destination Survey, 10 of the 12 measures had the highest correlations with walking rates when derived using the circular, one-kilometre buffer. In general, measures derived from the 500-metre network buffers were the least associated with walking rates (results available from the authors). Network buffers, by design, are a type of connectivity measureNote 16Note 17 and their use essentially causes connectivity to be counted twice in summary measures. The strength of the associations with walking rates and the confounding influence of network buffers led to the adoption of one-kilometre, circular buffers. Circular buffers are also favourable for computing resources relative to network buffers.

Data

Active living environment measures

The active living environment measures were derived from three primary sources: Statistics Canada (2016 Road Network File, population and dwelling counts), 2017 DMTI Spatial (Enhanced Points of Interest [EPOI] file) and OpenStreetMap (OSM) (road and footpath network, points of interest [POIs]). The 2016 Statistics Canada Road Network File is a digital representation of Canada’s national road network.Note 18 The 2017 DMTI EPOI file contains approximately 1.7 million geocoded records of businesses and institutions (e.g., schools, hospitals) and other selected features.Note 19 DMTI Spatial is a proprietary data source available to Canadian university researchers through special research licensing and is the mainstay of destination-based measures. OSM is a digital world map predicated on open data licensing and user-generated data.Note 20 OSM contains spatial data for transportation infrastructure (e.g., streets, footpaths, rail stations), natural and topological features, POIs (e.g., shops, schools), and administrative boundaries. While most researchers in Canada use proprietary datasets to derive active living environment measures, OSM data have been validated with government and proprietary datasets and used for research in CanadaNote 21 and other Western countries.Note 22Note 23OSM data were obtained from the Geofabrik data extract server in October 2017.Note 24

Two connectivity measures (three-way intersection density and four-way intersection density) were calculated three times, each time using a different methodological approach or derivation source. First, intersection density was derived using road intersections from the 2016 Statistics Canada Road Network File. Second, intersection density was calculated using OSM road features, which are identical, in principle, to the Statistics Canada file. Third, intersection density was calculated from OSM intersections of roads, footpaths and recreational trails. Limited-access roads (e.g., highways and freeways) were removed from each of the road files before calculating intersection density.

Four density measures were derived, and they varied in terms of underlying data (population count versus dwelling count) and calculation method (gross density versus weighted density). Gross population density and gross dwelling densitywere derived by dividing the population or dwelling count by the area of the DA. Weighted population density and weighted dwelling densitywere calculated by aggregating the population or dwelling count of each DA within the buffer and dividing by the area of the buffer. If a DA fell entirely within the buffer, its entire population and its dwellings were added to the count for the buffer. If a DA fell partially within the buffer, the population or dwelling counts of the DA were adjusted according to the proportion of the DA within the buffer. For example, if a DA with 1,000 inhabitants was only 25% within the buffer area, the buffer was assigned 250 of its inhabitants. Major coastal water bodies and DAs with no data were excluded from this calculation. The full and adjusted values were summed to determine an approximate population and dwelling count for the buffer. Population and dwelling counts were obtained from the 2016 Census conducted by Statistics Canada.

The POI measures differed only by underlying data source. The DMTI EPOI dataset contains the geographic coordinates of businesses and government institutions in Canada, as well as a few additional special POIs (e.g., provincial and national parks, border crossings). OSM POIs consist of a wide variety of mapped features (e.g., schools, shops, parks, benches, ATMs, soccer fields) obtained from administrative data sources and user contributions. Both measures were calculated by counting the number of POIs within the one-kilometre circular buffer. Additionally, a public transportation stop measure was derived for DAs within census metropolitan areas (CMAs)—urban areas with a population of 100,000 or more. This measure reflects the number of transit stops or stations (e.g., bus stops, light rail stations, subway stations) within the DA buffer. The coordinates of each transit stop were obtained in December 2017 from datasets on municipal or transit agency websites that conform to the General Transit Feed Specification (GTFS) data standard. Areas outside of CMAs were omitted because of inconsistent availability of GTFS data. The transit stop measure was derived for 35,338 DAs (97.1% of DAs within CMAs). Spatial data on transit stop locations were not found for some smaller CMAs: Belleville, Ontario; Peterborough, Ontario; Saguenay, Quebec; and Trois-Rivières, Quebec.

Active transportation rates

Two active transportation rates were calculated from Question 43 a) of the 2016 Census long-form questionnaire.Note 25 This question asks members of the labour force aged 15 or older with a fixed workplace address how they get to work. These rates were calculated by aggregating the number of pedestrians, cyclists and public transportation users for all DAs intersecting the circular, one-kilometre buffer of the DA centroid. If a DA fell only partially within the buffer, a smaller number of commuters proportional to the area of the DA within the buffer was aggregated. For example, if a DA was 25% within the buffer area and reported 40 pedestrian commuters, it was estimated that only 10 pedestrian commuters lived within the DA buffer. The walking-to-work rate reflects the proportion of this population that reports walking as their primary mode of transportation to work. The active-transportation-to-work rate reflects the proportion of the same population that walks, cycles or uses public transportation to get to work. Public transportation use was included as active transportation, as public transportation has been shown to generate physical activity via walking to and from transit stops.Note 26Note 27Note 28

Statistical methods

Pearson correlation coefficients were calculated to assess the association between walking-to-work rates, active-transportation-to-work rates and the 13 active living environment candidate measures (Table 1). To assess whether there was a regional bias in the coverage of user-contributed OSM data, the proportion of OSM POIs to DMTI POIs in five geographical regions (Atlantic, Quebec, Ontario, Prairies and British Columbia) was calculated. It is important to note that OSM and DMTI POI datasets contain different types of records. The DMTI EPOI file mainly consists of records of businesses, while OSM POIs contain certain commercial businesses (e.g., grocery stores, restaurants, clothing stores) as well as features of public spaces and streets (e.g., benches, picnic tables, tennis courts, food stalls, ATMs, postal boxes). Accordingly, the number of points in each dataset cannot be directly compared, and instead, the proportion of OSM to DMTI points was examined for different regions of the country. The strength of the correlations in this analysis and the spatial and regional distribution of the input datasets were the primary findings that informed the selection of Can-ALE measures.

K-medians clustering was performed to classify Canadian DAs into five categories that characterize the favourability of the active living environment. K-medians clustering is a partition clustering method where the user specifies the number of clusters (k), then an iterative process is used to assign observations (in this case DAs) to a group with the closest median values. The cluster analysis was based on the three pan-Canadian DA-level measures selected for Can-ALE:Note 1 three-way intersection density of roads and footpaths derived from OSM,Note 2 weighted dwelling density derived from Statistics Canada dwelling counts, andNote 3 POIs derived from OSM. The k-medians clustering method was used, as the right skewness of the active living environment measures made the k-means approach unsuitable. A descriptive analysis compared walking-to-work rates; active-transportation-to-work rates; and average connectivity, density and POI values by cluster group.

Results

Correlation analysis

Most measures were modestly (R = 0.20 to 0.39) to moderately (R = 0.40 to 0.59) correlated with walking-to-work rates and moderately to strongly (R ⋝ 0.60) correlated with active-transportation-to-work rates (Table 2), regardless of derivation source. Of the connectivity measures, the four-way intersection density of roads and footpaths derived from OSM was the most strongly associated with walking-to-work rates (R = 0.42). The three-way intersection density of roads and footpaths derived from OSM was one of two connectivity measures that were most strongly associated with active-transportation-to-work rates (R = 0.68). However, four-way intersections were sparsely distributed throughout Canada (especially in rural and certain suburban areas) and this made the four-way intersection measure less desirable than the three-way intersection measure for a pan-Canadian dataset. For instance, 55.4% of rural DA buffers in Canada (i.e., those outside CMAs or census areas) had no four-way intersections, whereas only 33.8% of rural DA buffers had no three-way intersections.

Dwelling density measures were more strongly associated with walking-to-work rates than population density measures, and the weighted density derivation method was more strongly associated with both walking-to-work and active-transportation-to-work rates. Gross population density (R = 0.23) and gross dwelling density (R = 0.29) were both modestly associated with walking-to-work rates, while both weighted population density (R = 0.82) and weighted dwelling density (R = 0.82) were strongly associated with active-transportation-to-work rates.

Both of the DMTI (R = 0.55) and OSM (R = 0.53) POI measures were similarly associated with walking-to-work rates, and the OSM POIs were somewhat more strongly associated with active-transportation-to-work rates (R = 0.66) than the DMTI POIs (R = 0.57). A summary measure of the favourability of the active living environment measures or “ALE Index” (for definition, see Figure 1) was moderately associated with walking-to-work rates (R = 0.47) and strongly associated with active-transportation-to-work rates (R = 0.78). Within the subset of urban DAs (n = 35,319), the transit stop measure was correlated with both walking-to-work (R = 0.45) and active-transportation-to-work rates (R = 0.70). OSM data coverage varied minimally across five different regions of Canada (Atlantic, Quebec, Ontario, Prairies and British Columbia). The proportion of OSM POIs to DMTI POIs nationally is 17.9%, which varies regionally from a low of 14.1% in Quebec to a high of 20.7% in Ontario.

Cluster analysis

Five cluster groups had the lowest variation of median values for each component measure within the cluster groups and the most variation across groups. The cluster groups were ordered according to the favourability of the active living environment: DAs in Group 1 represent the least favourable active living environments and those in Group 5 represent the most favourable active living environments in Canada. A cluster group was not assigned to the 500 DAs where dwelling density could not be derived (i.e., areas where Statistics Canada does not disseminate data on dwelling counts).

The majority of Canadian DAs (64.3%) were in the lowest two cluster groups (groups 1 and 2). This reflects the right skewness of the active living environment measures on which the cluster groups are based. Rural areas and the outermost areas of cities and towns tended to be in Group 1. DAs in Group 5 tended to be in the central business districts of Canada’s largest cities. Groups 2 to 4 tended to be located in residential neighbourhoods of urban areas (Figure 2).

Active-transportation-to-work rates were positively graded by the cluster group and increased by similar magnitude by cluster group (Table 3). In contrast, the walking-to-work rates were similar and below the national average (6.0%) for groups 1 to 3 (Group 1: 5.3%, Group 2: 4.8%, Group 3: 5.3%).

Discussion

This study informed the development of the first pan-Canadian database of active living environment measures derived entirely from non-proprietary sources. Database content was guided by principles according to which the measures selected needed to be associated with walking rates, suitable to a variety of built environments (e.g., urban, suburban, rural), and openly available to researchers and the public health community. Four individual measures (three fully pan-Canadian; transit stop proximity was available only for a subset of DAs in urban areas) were ultimately selected for the database, which also contains composite measures. Three-way intersection density using the OSM road and footpath features was selected based on its higher correlation with active-transportation-to-work rates, relative to the measures with roads alone. Three-way intersection density was deemed more appropriate for the national database, as four-way intersections are concentrated in cities with a grid-like street pattern and are less common in rural and certain suburban areas of Canada, which may lead to the connectivity in these communities being underestimated. Weighted dwelling density was selected because of its higher association with walking-to-work rates relative to population density measures and the much higher association between the weighted density methods and active-transportation-to-work rates. OSM POIs were selected because of their stronger association with active-transportation-to-work rates relative to DMTI POIs and good evidence of regional similarities in feature coverage. Additionally, the transit stop measure was selected and derived for DAs within CMAs and was strongly associated with both walking-to-work and active-transportation-to-work rates.

The Canadian active living environment measures derived from open data are similarly associated with active transportation behaviour as measures derived from traditional sources, or more strongly associated. For the connectivity measures, this may be attributable to the presence of footpath features in OSM, a feature that is not well-mapped in other Canadian datasets (Statistics Canada, DMTI Streetfiles). In 2010, even before the road network was considered complete in Germany, Zielstra and Zipf noted that OSM was more optimized for pedestrian routing than an authoritative GPS dataset in medium- and large-sized German cities.Note 22 Today, the OSM road network features are considered complete in Canada and most Western countries, and the coverage of footpaths and trails continues to grow.Note 29 Another advantage of open datasets, such as OSM, is the emerging international comparison work that is facilitated by using replicable methods and data that are available internationally. For example, these methods have been replicated to produce a comparable active living environment measure for Wales.Note 30

No research has been conducted to assess the completeness of OSM POI data in Canada or elsewhere. Unlike the DMTI EPOI dataset, which primarily contains records of businesses, OSM POIs contain many features for which no publicly available, exhaustive dataset exists (e.g., beaches, playgrounds, fountains). However, research on OSM’s POI features suggests that the features in OSM are spatially and descriptively accurate and that the number of POIs is growing exponentially.Note 23Note 31Note 32 Furthermore, the small-scale, on-street features mapped by OSM are conceptually appealing given their similarity to micro-scale features of active living environments (e.g., benches, park features), which are associated with increased physical activity, but are traditionally difficult to map without conducting field audits.Note 33 Although the incompleteness of OSM data may present a barrier to research for those studying particular attribute types (e.g., liquor stores, transit stops),Note 34 the OSM POI measure is distributed similarly to the DMTI POIs by region, and both measures are similarly associated with walking-to-work and active-transportation-to-work rates.

Active-transportation-to-work rates are graded positively by the five cluster groupings identified in the k-medians cluster analysis. This contrasts with evidence that only the most favourable active living environments (groups 4 and 5, which represent the communities of only 13.6% of Canadians) can support walking and cycling commute rates above the national average. Previous research found that public transportation users can meet recommended physical activity levels regardless of the active living friendliness of their neighbourhood.Note 26 As higher walking-to-work rates are concentrated in cluster groups 4 and 5, and as less than 2% of Canadians cycle to work,Note 35 the positive grading of active-transportation-to-work rates is largely explained by increased public transit use by cluster group. Together, these findings suggest that public transportation use may be a practical alternative to increasing physical activity levels in places where walking to work is impractical.

Cluster groups with more favourable active living environments (groups 4 and 5) had fewer DAs than those with less favourable living environments (groups 1 and 2). Because of a very high dwelling density and a high concentration of destinations in a few Canadian neighbourhoods—particularly the urban cores of major cities—measures of active living environments at the national level are highly right-skewed. Researchers have stratified communities in the past using equal interval categories (e.g., quartiles or quintiles) and have found significant relationships between these active living features and health outcomes for the highest quartile or quintile.Note 9Note 36 Future research may find more robust results if categories that reflect the right skewness of active living environment measures are used.

Limitations

One limitation of this study is the use of commuting data from the census to derive walk-to-work and active-transportation-to-work rates. Commuting data reflect only a portion of all active travel behaviour. Future researchers may wish to use measures of physical activity and health status from surveys, such as the Canadian Health Measures Survey, to analyze the association between these measures and other forms of active living. The use of aggregated transportation data for both analyses suggests an ecological association between active living environment characteristics and increased physical activity through transportation, but not necessarily a causal relationship.

Often, land use mix is included in active living environment measures instead of or in addition to destinations and POIs. While a lack of free land-use data prevented the calculation of land use mix for Can-ALE, destinations are often assessed alongside or instead of land use mix,Note 4Note 37Note 38 and estimations of land use mix using the conventional entropy formula have been criticized for overestimating the mix of uses in certain areas.Note 39

The use of open data to derive the Can-ALE database makes these measures more accessible, but also presents a challenge for reproducibility. OSM has existed for only about a decade and is always changing, which limits the exact replication of research. With the ongoing additions of new POIs by users, it may be difficult to attribute changes in OSM to changes in the built environment or simply to improvements in the data quality and quantity. Until the POIs have reached saturation at the national level (i.e., the annual number of additions or edits to the dataset is less than 3% of the total number of pointsNote 40), OSM POIs may not be a suitable source of data for studies that wish to examine changes in the built environment. However, the OSM road features are considered completeNote 29 and may be suitable for longitudinal research.

Conclusion

Addressing the prevalence of sedentary behaviour in Canada may require environmental interventions that support routine, daily physical activity. However, to date few national datasets exist to characterize active living environments and support research and public health surveillance. Can-ALE addresses this gap by providing a freely accessible dataset that is validated with national active-transportation data. The study draws attention to the methodological considerations of building measures for the national scale, and in particular, to the fact that three-way intersection density and dwelling density are suitable for national measurements. This study also expands the Canadian evidence base on the validity of using open data and suggests that certain well-mapped elements of OSM (e.g., footpaths, micro-scale features) may improve the predictive capacity of active living environment measures. As OSM reaches completion or saturation, open data may serve as an important derivation source for active living environment measures and support the creation of longitudinal datasets.

The findings also draw attention to the spatial distribution of active living environments and active travel behaviour in Canada. The non-normal distribution of active living environment measures and active travel in Canada suggests that the use of equal interval categories to stratify neighbourhoods may lead to the relationship between the built environment and health status being underestimated. In contrast, the strong grading of active transportation use by active living environment cluster group draws attention to the role public transportation may play in increasing physical activity among a larger group of Canadians.

References
Report a problem on this page

Is something not working? Is there information outdated? Can't find what you're looking for?

Please contact us and let us know how we can help you.

Privacy notice

Date modified: