Spatial Statistics
Determining the distribution of dataset or the pattern in which they occur is very important in geographical analysis, as it can be used in decision making. Spatial statistics can be used to analyze if distribution patterns are clustered or sparse; if the feature distribution is random or not.
In this exercise, phone calls from a fire department were analyzed. I determined if distributions were clustered or not and if factors such as; the priority of phone calls and distance; had an effect on clustering of phone calls.
Description of Procedure
1. The objective in this procedure was to locate clustering of false alarm calls in the neighborhood. The area of the neighborhood was first noted, this would be used to restrict data in the study area. The nearest neighbor tool was used to calculate distance from where each false alarm phone calls were made to its nearest neighbor. The nearest neighbor index value was less than 1, from this value I was able to determine that the phone calls were clustered. Confidence level and Z-scores were also calculated in the report from running the nearest neighbor tool.
Nearest Neighbor Workflow(click to enlarge)
Average Nearest Neighbor tool result summary(click to enlarge)
Average Nearest Neighbor map(click to enlarge)
2. I also determined if calls that had high priorities were clustered. I also considered distances at which these clustering occurred the most. Using the Calculate Distance band tool, I determined where the minimum, maximum and average distance to 7 neighboring points to be considered. Using the distances of 200 – 1200 ft. The Z-scores were calculated for each distance band using the High/Low clustering tool. This was done to be sure of which distance had the highest amount of clusters. The distance band with the highest Z-score was the distance that had the highest cluster of phone calls.
High-low clustering workflow(click to enlarge)
Summary result after running the High-Low clustering tool(click to enlarge)
Map showing call priorities(click to enlarge)
3. Determining the clustering between phone calls was based on all neighboring features within range of distances and priority of these phone calls. This was done using the Multi- distance spatial cluster analysis tool. The ‘FEE’ field was used as the priority weigh input field and beginning distance set at 200ft. I also set the confidence envelope to 99 permutations. The difference between the “observed value” field and the upper limit of the confident envelope field at all distances were calculated. the distance that had the highest difference value was the distance at which I had the highest cluster
Workflow to determine clusters considering distance and call priority(click to enlarge)
Multi-Distance Spatial Cluster Analysis Map(click to enlarge)
4. The densities of calls per block and the distance at which these densities cluster were determined in this step. A grid of 200ft was overlaid with point data representing number of calls by using the spatial join tool. The number of calls exceeding ‘zero’ per grid were selected resulting in each grid having at least 1 count of phone call. The distance band was within 250 to 600ft, with and interval of 50ft. the spatial auto-correlation tool was run for each of these distances and the z-scores were recorded. The distance that had the highest Z-score value was the distance with the highest cluster.
Auto-correlation workflow(click to enlarge)
Summary Report after running the Moran's I tool(click to enlarge)
Map showing densities of call clustering(click to enlarge)
Application and Reflection;
The spatial statistic tools are very important in determining how features are distributed; it can also be used to examine different kinds of characteristics that are responsible for these distributions. The health department might what to know if a particular disease outbreak is clustered in a certain location. And one can weigh features in that location to find out if some environmental factors actually promoted the outbreak of this disease.