I’m going to try to do a two post thing, for anything with some complexity. The main post will show the most interesting stuff, the key takeaways and so on, and then a methodology post will cover some of the gory details that might not be useful or interesting for everybody.
Data Sources / Methods
The commuting data is from the 2016 Census long form, and the dissemination area geography is from Statistics Canada. The walking distances were calculated using the Geofabrik downloads of the OpenStreetMap network with all roads and paths (including some that are not necessarily walkable, like freeways), using QGIS 3.2 and QNEAT3. I manually identified the train stations, starting with the OSM data, except for an open source TTC station file from Kevin Farrugia. I also calculated a straight line distance, just using buffers of the stations and the centroids of the DAs.
One part of cleaning the data is removing a small percentage of dissemination areas (DAs) that are not useful for analysis. One is if they have fewer than 100 commuters — this is pretty straightforward; these are based on small samples, and averages with those samples are unreliable, especially with StatsCan’s penchant for rounding to the nearest 5. The other is if they have an “unreliable” walking distance. Let me show you a DA as an example:
This DA (shown in red; the yellow boundaries mark other DAs) is in northwest Calgary, near Dalhousie LRT station (dark blue). It’s a classic Transit Proximate Development and contains a real mix of uses; the green areas are two different retail areas (Dalhousie Station shopping centre on the west, a Coop / Canadian Tire on the east) and their surrounding parking chasms, the light blue is park and ride and the transit loop. The residents in this area are in two clusters of townhouse developments (in yellow), and in four mid rise (3-10 storey) apartment buildings. A thin thread of road right of way connects the two parts.
The townhouse closest to the LRT station access bridge is within a 300m walking distance; the apartments north of Dalhousie Station shopping centre are about 600m to the front door. The eastern cluster of townhouses and apartments are about 1200m walking distance from the station. The Coop supermarket in the northeast end of this DA is approved for redevelopment including multifamily; the apartments there will be 1500m from the station. It just doesn’t make much sense to say that residents in this area are a mean of 800m walking distance from the LRT station and draw conclusions from that.
Handily, the walking distance was produced as a raster so there is a measure of how reliable it is in a given DA; there is a standard deviation. In this case, it’s 389m. I wound up using as a metric the standard deviation in walking distance divided by the ln of the DA area in square metres (which acknowledges that a larger DA will by definition have a higher standard deviation); this DA is just above the cutoff I used of 30.
In the figures, I blanked out these low quality DAs; the large ones almost certainly failed on this criteria, while some of the small ones may have failed on the lack of commuters.