Emerging Hotspot Analysis
​
For the emerging hotspot analysis final outputs, any cells that were placed in any category besides ‘no pattern detected’ possessed statically significant changes through time in the method outlined by that category’s definition, and these confidence interval surfaces for all property crime, residential and commercial break-and-enters, as well as vehicle thefts have been included within the appendix (Fig A1; Fig A2; Fig A3).
​
For the maps of all three categories, there were large ‘persistent hot spots’ in the downtown core of Vancouver, and break-and-enters and all property crime also possessed persistent hot spots around False Creek (Fig 2; Fig 3, Fig 4). The all property crime map also identified ‘intensifying hot spots’ around the downtown eastside and port regions of downtown, as well as a large intensifying hot spot around Mount Pleasant, with large regions of ‘sporadic hot spots’ further east (Fig 4).
​
The south and southwest regions of the city largely possessed cold spots of various categories for all three property crime maps. The all property crime and break-and-enter maps largely identified similar regions of ‘intensifying cold spots’ and ‘persistent cold spots’ in Point Grey, the Dunbar-Southlands, and in Shaughnessy, while the vehicle theft map identified much more of these regions as intensifying cold spots (Fig 2; Fig 3; Fig 4). Intensifying and ‘sporadic cold spots’ were also found eastward along Marine drive (Fig 2; Fig 3; Fig 4).
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
Fig 2. Map of emerging hotspot analysis for thefts of vehicles in the city of Vancouver, from the years 2007-2017.
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
Fig 3. Map of emerging hotspot analysis for residential and commercial break-and-enters in the city of Vancouver for the years 2007-2017.
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
Fig 4. Map of emerging hotspot analysis for all property crime (residential/commercial break-and-enters, thefts of vehicles, thefts from vehicles, thefts of bicycles and 'other' theft) in the city of Vancouver for the years 2007-2017.
​
Maximum Entropy Crime Suitability
​
The results of the Maxent model suggest that the variables and training data have significant statistical power. The average AUC value for each of the crime types was 0.796, with a highest value recorded at 0.873 and a low at 0.707. A random prediction represents an AUC value of 0.5, meaning that the model performed consistently at rates significantly better than random. Table 1 contains the AUC results sourced from the Maxent summary data.
​
​
​
​
​
​
​
​
​
Table 1: Table of AUC results for MaxEnt suitability models for 2017 Vancouver property crime.
The importance of the contributing variables also varied across the factors analyzed. ‘Distance from shelter’ had the highest average permutation importance across all of the point data sets analyzed at 51.32, with ‘distance from park’ having the lowest average permutation importance at 6.77. The permutation importance of ‘distance from shelter’ varied significantly across the different sample sets however, and ranged from 33.6 (in ‘other theft’) to 71.9. ‘Distance from shelter’ also lost power across the board when the path the heuristic took was factored out of the analysis.
​
​
​
​
​
​
​
​
​
​
​
Table 2: Table of average permutation importance for MaxEnt suitability models for 2017 Vancouver property crime.
​
​
‘Distance from school’, the variable with the second highest permutation importance score, was the single most important variable when isolated for all of the data sets via a jackknife test of variable importance, except for ‘other theft’ and ‘commercial break-and-enter’, where ‘distance from shelter’ predominated. Table 2 describes the resultant average permutation importance for each of the variables, while Table 3 describes the result of the jackknife test of permutation importance, with figures A4(2) through A4(9) displaying the visual results of the jackknife tests.
​
​
​
​
​
​
​
​
​
​
Table 3: Table of Jackknife test results, showing which environmental layer the analyses found to be of greatest import.
​
​
The visual output of the MaxEnt model across the crime datasets analyzed aligns with the expectations developed through the emerging hotspot analysis. The six resultant maps can be found in the Appendix in Figures A4 through A9. Areas of high suitability can be found downtown and in Mount Pleasant, radiating outward along major transit routes and streets into the suburbs. These suburbs represent areas of lesser suitability, likely influenced by the lesser influence of access to transit (via the ‘distance to transit’ layer). The ‘other theft’ category represents the most concentrated and constrained suitability model, with most of the ‘suitable’ terrain located in downtown. Contrarily, the ‘residential break and enter’ category has the most widely diffuse suitability surface, with most of the study area identified as being susceptible. Both of these findings align with the nature of the crime datasets - the suburbs represent large residential areas, whereas ‘other theft’ represents an aggregated category of the theft of personal items like cell phones that is likely tied to the density of people and therefore accessibility of them to criminal activity.
​
​
