Forecasting: Complex statistical methods and crime mapping
Chapter 6: Crime Mapping Futures

In cooperation with the Pittsburgh Police Department, Olligschlaeger (1997) employed advanced statistical methods in an attempt to identify emerging drug markets, which are relatively difficult to detect by conventional means. This is because they often are revealed only indirectly through the commission of other crimes, such as robberies or burglaries, and then only after a delay. Given the importance of drug markets as crime generators, early warning is useful. Three types of calls for service (CFS) were used to develop models: weapon-related, robbery, and assault. Commercial land use and seasonality were also included as model criteria based on evidence in the literature. A grid of cells measuring 2,150 square feet was then superimposed on the city, with the size of the cells determined by the need to have cells big enough to represent a reasonable number of CFS, but small enough to supply an adequate number to satisfy statistical modeling requirements. The grids were then used as the framework for choropleth maps.

Following rules based on gaming simulation, a statistical model estimated the consequences of the various types of CFS at certain levels. (For details of the model's architecture and specifications, see Olligschlaeger, 1997). Then forecasts from different methods were compared with actual patterns to provide a basis for evaluation.

Experience showed that a type of gaming simulation model known as the neural model did better than other types tested. A difficulty with the analysis—namely, the use of large quantities of computer resources—becomes less of a problem as the power of personal computer processors increases. Overall, the analysis suggested that advanced methods of the type tested—the neural model—could be useful tools in spatial forecasting.

Mapping change: From pins to grids

Also employing a grid mode of representation for Pittsburgh, Gorr used 750-foot cells approximating four city blocks. Based on a pin map, a grid map (figure 6.4) displays the Part I crimes from the pin map in each grid cell, suppressing cells with only one crime. The grids show the total demand for serious crime suppression, in combination with an indicator of crime-prone land uses. The latter indicator came from the PhoneDisc3 yellow pages and includes the total number of restaurants, fast-food stands, bars, drug stores, retail stores, pawn shops, jewelry stores, etc., by grid cell, suppressing cells with only one such establishment. This example shows how the reformulation of information and the introduction of a related layer (or layers) of data can provide new and more useful interpretations than the original data alone.

Figure 6.4

As shown in figure 6.5, this methodology was used to analyze change by converting the two pin maps into grids and then subtracting one from the other to get the measure of change. Grid cells have several advantages:

Figure 6.5

  • They clearly show crime intensity in places with many overlapping point markers.

  • Their data are in the form needed for time-series plots, bar charts, and statistical analyses, which are examples of crime space/time series, with one "slice" shown in figure 6.5.

  • The grids can be used to produce change maps that are more legible than pin maps.

    Making maps come to life4

    Application of Virtual Reality Modeling Language (VRML) to crime data allows the user to change a viewpoint by rotating, translating, zooming in and out, and tilting maps, providing a dynamic way of viewing crime. The images shown in figures 6.6 and 6.7 are part of an animation that depicts different crime types reported to police for various time periods in Vancouver, British Columbia, Canada. In its original context, this animation could be activated by clicking the start button shown in the images. The process of creating the images involves first rasterizing the data, then developing a color code key for the crime types, and, finally, designing a system for displaying the crime—in this case, as a histogram.

    Figure 6.6

    Figure 6.7

    In figures 6.6 and 6.7, six different types of crime are illustrated: assault (ASLT), breaking and entering (BNE), family trouble (FAMTRB), mischief (MSCHF), auto theft (TFAUTO), and theft (THEFT). The height of the stack is proportional to the total number of crimes in an area, so hot spots can be recognized as "highrises." The two figures display the same data—the same map—from different viewpoints after rotating and zooming in.

    Another approach that is easy and yet quite effective as a means of visualizing change involves animating a two-dimensional map. In figure 6.8, calls for service in Mesa, Arizona, were mapped and (in the original) animated. This map uses isoline mapping (joining points of equal value—in this case, equal CFS counts5). The animation is a rapid-sequence display of a series of maps of successive arbitrary time intervals giving the visual impression of movement, much like an animated cartoon.

    Figure 6.8

    These applications illustrate how we can expect maps to become more dynamic, more maneuverable, in the years ahead. Not only is it likely that the flexibility of maps will improve, but a more user-friendly environment will likely evolve in parallel. The average analyst will simply not have time to do the programming that Lodha and Verma (1999) did to produce their maps, and tools of this sort will, of necessity, become easier to use.

Chapter 6: Crime Mapping Futures
Previous Contents Next
Return to Home Page



Mapping Crime: Principle and Practice, by Keith Harries, Ph.D., December 1999