By Christopher Bruce
Data-Driven Approaches to Crime and Traffic Safety (DDACTS) is a partnership between the National Highway Traffic Safety Administration, the National Institute of Justice, and the Bureau of Justice Assistance. It features a series of workshops at which analytical, supervisory, and executive representatives of police agencies create models for analysis and enforcement specific to their jurisdictions and traffic and crime problems. DDACTS is not itself a model, nor a single strategy, but rather a philosophy, process, and a set of guiding principles under which more specific models are developed.
These models share in common a focus on both collision and crime hot spots, and the use of highly-visible enforcement as one tactic (National Highway Traffic Safety Administration, 2009, p. i). In DDACTS, we encourage synthesis of analysis and response to both crime and traffic safety issues. In some agencies, this takes the form of enforcement at literally overlapping crime and collision hot spots; in others, it is more complicated, with considerations of remote enforcement and offender travel patterns. There are hundreds of variations of DDACTS depending on the goals of the agencies, their resources, and the nature of their jurisdictions.
The totality of DDACTS is often mistaken for the particular models developed under the DDACTS name, but we mean the term “approaches” quite literally. Effective DDACTS models can and do draw from best practices in crime analysis, hot spots policing, evidence-based policing, problem-oriented policing, predictive policing, and other progressive, data-driven models. For this reason, blanket statements such as “our agency does DDACTS” or “DDACTS works” are somewhat meaningless without a solid understanding of the specific models that agencies have chosen to develop. What we can say definitively is that “the right sort of DDACTS models ‘work,’” and “for those that don’t—well, monitoring, evaluation, and adjustment is one of the core guiding principles of DDACTS.”
Mapping in DDACTS
Where DDACTS primarily focuses on hot spots, mapping is a vital part of any model. We use DDACTS maps for four primary purposes:
1. Identify hot spots. As I discuss below, several techniques are useful for identifying hot spots for both collisions and crimes.
2. Designate target areas. Agencies combine one or more hot spots—for collisions, crimes, or both—into one or more target areas. These target area polygons serve as the basis for deeper analysis, enforcement, problem-solving, and evaluation.
3. Identify enforcement points. Within each target area, agencies identify one or more enforcement points—places where traffic enforcement, directed patrols, and other tactics are likely to have the most impact.
4. Evaluate results. A GIS helps us measure changes in the target areas, compare them to control areas, and test for displacement and diffusion of benefits.
Unique Geocoding Challenges
Most analysts have extensive experience geocoding crime locations, but geocoding collision locations can present some unique challenges. First, since they occur on the street, address-matching routines that place points on parcels or buildings, or offset them from the street centerline, create confusing and nonsensical maps for collision analysis. Second, address fields for collisions are often more complicated than are those for crimes, with indicators that include direction of travel and distance offsets. It can be difficult to incorporate these fields into an address-matching scheme despite their importance in identifying the specific location of the collision.
Precision in geocoding is arguably more important for traffic collisions than for crime, especially when attempting to identify roads and intersections that require engineering (rather than enforcement) solutions. Default address matching routines might locate all collisions at a complex “intersection” in the literal center of the two streets, ignoring the specifics of ramps, lanes, and directions—specifics that we need to truly understand what is occurring at collision hot spots. Using the distance offsets from collision data (e.g., “Main Street 300 feet west of Elm Street”) to adjust coordinates can help improve geocoding precision, but the only way to achieve true precision is through GPS coordinate collection (which is rare) or digitizing the collision location based on the officer’s diagram (which is time-consuming).
Hot Spot Methodologies
As other sources (e.g., Eck, Chainey, Cameron, Leitner, & Wilson, 2005) have covered, there are many ways to conceptualize “hot spots,” including:
- Aggregation by point, line, or polygon
- Techniques that identify clusters of points based on distance measures
- Interpolation techniques, such as kernel density estimation
Kernel density estimation (KDE) is perhaps the most popular hot spot routine; it results in an attractive map, and many commercial applications do it. But analysts need to understand the weaknesses of the technique: the nature of the hot spots it identifies may vary dramatically depending on settings; the resulting calculations are very difficult to explain to an audience; the routine will happily identify a “hot spot” at the highest-volume locations regardless of how many actual incidents occurred; and, perhaps most important, the very logic of the routine—calculating risk for every point on the map based on locations where incidents actually occurred—is simply invalid for many crime types and certainly for most collision types. That is to say that risk of collisions, for almost all causes, does not transmit in a two-dimensional manner from the collision location. At best (as in the case of speed-related collisions), it transmits in a linear manner along the street segment, but in many cases (as in red-light collisions at intersections), no risk transmits at all.
For these reasons, I tend to favor aggregation by line segment (calculated as a rate per X linear feet) as the primary mechanism for identifying hot spots (or, more accurately, “hot lines”) for collisions, and clustering techniques for identifying hot spots for crimes. In particular, I favor CrimeStat’s Nearest Neighbor Hierarchical Spatial Clustering (NNH) routine, using either random or fixed distances depending on the nature of the likely response (see National Institute of Justice, 2013). I grudgingly admit that KDE is an acceptable alternative for crime, when the analyst knows what he or she is doing with the settings.
Of course, what to map is an equally important consideration as how to map. The choice of crimes and collisions to include in the assessment of hot spots and target areas makes an enormous difference in both the resulting maps and the likelihood of success with the DDACTS model.
The choice of crimes must make sense given the nature of the chosen solution. If the agency’s model relies heavily on visible enforcement, for instance, it makes the most sense to include in the hot spot assessment those crimes that will respond to the physical presence of a police officer on the street. These include street robberies, residential and commercial burglaries, thefts from vehicles, auto theft, and (in agencies with high enough volume) some street violence, like shootings and gang assaults. It is not, conversely, likely to affect aggravated assaults or sexual assaults in general (the majority of which take place behind closed doors between people with a prior relationship). Shoplifting is also a poor choice for identifying hot spots, particularly given inconsistencies in reporting.
With collisions, similarly, analysts must be intelligent about what they include in their hot spots maps. Many agencies find that their top collision hot spots are shopping mall parking lots or freeway on- and off-ramps where the sheer amount of congestion causes many cars to occasionally collide—but not in a way that the presence of a police officer could have prevented. Agencies often find it more valuable to map collisions with particular causes—particularly speeding, light and sign violations, and drunk driving—to identify hot spots where their enforcement efforts are likely to have the most significant result. When an agency does not have these causal factors readily available in the data, mapping injury-only collisions often serves as a good proxy.
DDACTS is a useful program that gets police agencies excited about the possibilities inherent in mapping and data analysis. Agencies that have never used analysis in the past have enjoyed significant successes with operational models—both basic and advanced—that direct resources to hot spots. Those with significant experience with crime analysis and mapping have still benefited from the DDACTS program’s emphasis on the various guiding principles, including the setting of explicit goals, the establishment of stronger partnerships, more systematic use of analysis and mapping, more rigorous evaluation, and an analytical approach to traffic collisions specifically. Though it fuses some of the best current techniques of crime analysis and progressive policing models, there is still plenty of room for innovation. I look forward to more advanced articles on DDACTS techniques in upcoming issues of Crime Mapping & Analysis News.
Eck, J. E., Chainey, S., Cameron, J. G., Leitner, M., & Wilson, R. E. (2005). Mapping crime: Understanding hot spots. Washington, DC: National Institute of Justice.
National Highway Traffic Safety Administration. (2009). Data-driven approaches to crime and traffic safety (DDACTS): Operational guidelines. Washington, DC: Author.
National Institute of Justice. (2013). CrimeStat materials. Retrieved April 1, 2014, from http://www.nij.gov/topics/technology/maps/pages/crimestat-downloads.aspx