JCLambert GIS Enterprise: M1 Crime Analysis

For this week’s lab assignment, we performed crime analysis for the Washington, D.C., and Chicago areas. In the first part of the lab, we created choropleth maps using data already provided to us. Performing spatial joins between feature classes gave us results for high burglary rates in the D.C. area. The latter part of the lab consisted of three parts using three different methods of identifying and visualizing hotspots in the city of Chicago. In this part of the lab, we are comparing the techniques in analyzing crime data; Grid Based, Kernel Density, and Local Moran's I. Through completing each of these steps we had to perform a spatial join, select by attributes, create layers from selection, and other geoprocessing steps to get the final output.

After setting up the environment for Chicago, I performed different geoprocessing operations to get the data for the number of homicides occurring in specific areas. The first analysis involved adding the clipped grid cells and the 2017 homicide layers and then spatially joining them. The Grid Based analysis is simple with the results based on a grid overlay of an area of .50 mile cells that are clipped to the Chicago boundary. To find areas with the highest rate of homicide I used the select by attributes to get rates greater than zero and manually counted the top 20%, the top quintile equal to 62. I added a field for these, exported the layer, and used the Dissolve tool for a smooth service area.

The Kernel Density Analysis uses search distance from the radius outward from a point of the highest value. This analysis covers a large area and accounts for outliers. The Kernel Density toolset was used in the analysis of this part. The same data from the Grid Based analysis were used. The toolset parameters needed to be set up with input features, distance, and because rasters are used in this analysis cell size and an output raster needed to be designated. After reclassifying the data for two break values, calculated by multiplying 3 times the mean, and converting the raster to a polygon, I selected attributes having a value of two giving me the final output. The output numbers were different from the Grid Based.

The Local Moran’s I identifies significant clusters and outliers in data. Using the same data for this analysis the output numbers were different from the previous two. Local Moran’s I used weighted features to get results, the parameters for this were the number of homicides per 1000 housing units for the Chicago area. The geoprocessing tool is Cluster and Outlier Analysis (Anselin Local Moran's I) for calculating spatial analysis statistics for the number of homicides. The output of the analysis gave me four classifications for homicides per household, the high-high cluster result is the only one used for the Dissolve tool setup.

The differences between these after calculating the results are noted in the table below for the numbers calculated. Which is better? I think Kernel Density given that a search from the point with the highest value radiating outward. The calculated result is the magnitude per unit of the cell within the study area so a per unit (cell) from a specified point of high value will give more accurate results.