Methodology

Data

Through UBC’s Department of Geography we had access to all the base geographic files of Vancouver necessary for this project: the shape files of parks, point data for schools, libraries and community centres, major roads, city land use types, the base layer of Vancouver itself, and the average residential monthly rent prices by dissemination area (9). In addition, we obtained data from Vancouver's Open Data Catalogue.

Methods

The goal of our project was to apply the geographic principals of Megan’s Law (and its variations) to Vancouver, to discover what the "optimal" residential "no-live" zones as a percentage of the entire city would be. As discussed in much of the literature reviewed, one of the issues with these laws residential restrictions is that they create exclusion zones around high-density children sites which often cause much of urban areas to be unlivable for sex offenders (8). Consequently, they have the potential to force sex offenders into specific, geographically-restricted locations which constitute a very small percentage of the total city area (8). If applied "successfully" according to the requirements of these laws, the entirety of an urban area’s sex offender population could thus be forced to reside within a few city blocks. In this project we were interested in finding the optimal balance between these two factors in our own city: requiring sex offenders to reside a certain distance from sites highly frequented by children, but also avoiding forcing sex offenders into too small an area within the city.

To these ends we primarily used the ArcGIS ModelBuilder function to perform our analyses. Based on research we decided to include the following as our high-density children sites: schools, parks, libraries, and community centres. In ModelBuilder we performed multiple ring buffers around each site at fifty metre increments from fifty to five hundred metres (ten in total). These distances reflected the variation in scope we found through research (from 500 to 2500 feet) and would allow us a broad range from which to choose our "optimal" buffer size (see Figure 2). We then clipped the resultant buffers to the Vancouver area (as some of them were shape files for the entire Metro Vancouver area) and unioned all the buffers of the same size for each buffer distance. We also intersected this remaining available area with residential land use, so the final map would just include areas in Vancouver zoned for residential housing.

We now had to make a decision: did we want to stop running the buffers once a minimum acceptable area for residency had been achieved (as a percentage of Vancouver’s total residential area), or did we want to continue running the model until all of the ten buffer distances had been performed? In theory we built the model to stop once a certain percentage of available residential area remained (see Figure 2), but in reality we ran into numerous difficulties once we tried actually running the model (as inherent in any GIS venture!). We determined that twenty-five percent of remaining residentially-zoned area would be an appropriate proportion to aim for. We built the model to erase the buffered areas from the total city area, once the twenty-five percent had been achieved, and then we manually intersected that area with the average monthly rent prices by Dissemination Area in Vancouver. This gave us the rent prices in the remaining residentially-zoned areas where sex offenders were ‘allowed’ to live by buffer distance (as one of the issues discussed in our research was that much of the housing actually available to sex offenders would be largely unaffordable for offenders just released from prison [5]).  

The main problem we had while trying to run ModelBuilder was an error that occurred indicating that the "geometry is not M-aware" and prohibited us from successfully running our model. Due to this and other scripting complications, we ended up doing most of the iterations manually (with the same final result as if the model we built had worked). Once we had run the iterations for all ten buffer distances, we decided to include five in our final results: 100m, 200m, 300m, 400m, and 500m. This gave us a broad cross-section of different buffer distances to consider and from which to select our "optimal" minimum "no-live" zone, without being an overload of information.