Week 5: Crime Hotspot Analysis

This week's lab involved creating several different Crime Hotspot Maps and comparing them in their predictive power.  We created hotspots using Local Moran's I, Kernel Density, and Grid Overlay from 2007 burglary data.  Then we compared the number of 2008 burglaries within these 2007 hotspots.

 I argue that the best crime hotspot predictive method in this scenario is using Kernel Density.  I base this on the observed highest crime density of 2008 burglaries within the 2007 hotspots. This is category (above) that I view as the best metric of predictive power because it takes into account hotspot total area, and thus would provide efficient preventative resource allocation.

It is clear that the Grid Overlay hotspot contains the most 2008 burglaries, therefore it is certainly a suitable predictive tool.  However, the total area is >65km² and may be too large to effectively patrol by police.  With unlimited resources (i.e. police officers/vehicles), it would be feasible to use this model as area to patrol. In a limited resource situation – which is what is most often the case – higher priority areas must receive preferential resource allocation.  Thus, the Kernel hotspots.

The Local Moran’s I hotspots are large, but contain the fewest amount of 2008 burglaries.  The large area appears to be a result of extreme density regions.  By visual assessment, it appears that areas between several clusters, which don’t look particularly dense themselves, are influenced by adjacency (see screenshot below).  It appears that the red area toward the top is in between two hotspots.