Title: Hot Spot Analysis of 911 Calls
Map Extent: Metropolitan area
Map Scale: 1:35,000 (for all four data frames)
Page Size (w x h): 24 x 34 inches (ANSI D-size Portrait)
Map Category: Analysis Map
Key to Icons Opens in new window
|Add a picture of a geoprocessing model to a page layout|
|Create multiple data frames showing the same extent|
|Create a custom legend for raster data|
This map was created to show how a typical analysis of point data might be done using the Hot Spot Analysis tools that were introduced in ArcGIS 9.1.
This map needed to show the various stages of analysis and—because this information is not easy to understand in just one form—some of the ways to present this kind of information.
This particular map illustrates an analysis of 911 call data. To perform this analysis, we created a model that integrates several spatial statistic tools. Our mission was to evaluate the spatial pattern of these 911 calls and look for hot spots. That is, we wanted to see where the calls were clustering together in space. Then, ultimately using the results of our hot spot analysis, we wanted to compare clustering that we found to the location of emergency response stations.
In the upper left, each of the points represents a single call into a 911 dispatch center. Notice that it would be difficult to find whether and where clustering exists. Certainly, some areas appear to be clustered, but even with a truly random pattern, we expect to see some pockets of local clustering. The first thing our model did was collect the incident data into weighted points. This yielded single features for locations with counts of calls (upper right map).
The weighted points show some clustering more clearly, such as near the center. The question remains as to whether this cluster is statistically significant or not. To answer that question we used Getis-Ord G* statistics. The idea was to learn where we were getting more calls than expecedt and where we were getting fewer calls than expected. The Hot Spot Analysis tool evaluates the call data by comparing the local mean to the global mean and then determining whether the difference between them is statistically significant—in other words, how likely it is that we would see a pattern like this one if the underlying processes are random.
The results are presented as standard deviation z-scores (lower left). Once these were calculated, we could immediately see the 911 call clustering. The red indicates a lot of calls. The blue shows that, given the overall region, areas that don’t get very many calls. To get a clearer idea, these points were interpolated into a surface (lower right).
Results: Locations of emergency response stations. Two stations are located in hot spots. The third may not be ideally located.