By Raymond Guidetti, Lieutenant, New Jersey State Police; and James W. Morentz, PhD, Senior Homeland Security Advisor, SPADAC Inc.
eospatial statistical modeling is a growing tool in intelligence-led policing and takes a more proactive approach towards disrupting criminal activity. This case study examines how the New Jersey Regional Operations Intelligence Center (NJ ROIC) is applying geospatial statistical modeling to assist local police agencies with reducing violent crime.
The aim of intelligence-led policing is to anticipate and prevent crime and terrorism. Yet in order to design and implement a program of prevention, law enforcement leaders must have access to information that can uncover trends in the environment that relate to criminal or terrorist threats and other hazards, either natural or manmade. The analysis of these trends can further expose patterns: patterns that yield factors, and factors—when observed in other locales—that can reveal the high probability of reoccurrence, which leads to prevention. This case study will focus on how the NJ ROIC, through the use of geospatial statistical modeling technology,1 has provided the Jersey City, New Jersey, Police Department with the intelligence necessary for structuring prevention initiatives aimed at reducing violent crime (see figure 1).
Geospatial statistical modeling for law enforcement is a methodology based on identifying discernable geospatial preferences associated with a perpetrator’s conscious and unconscious activities leading up to criminal behavior, a gang action, or a terrorist threat. By applying a structured geospatial model, hundreds or even thousands of geospatial factors can be fused together to produce geospatial patterns of statistical similarity among criminal location preferences. These patterns, once understood by analysts, can be mapped to a geographic area. Displayed visually as a density pattern, the analysis of these factors can aid in defining the probability that conditions exist for the specific activity of concern to emerge and, more importantly, its location. Further, the analysis of underlying factors permits the development of action plans to counter the event.
Stated simply, where an individual chooses to be tells a lot about what that individual chooses to do. Humans are influenced by conscious and unconscious factors. Among the more important factors is the environment around a person—the physical, natural, cultural, and social environment—which both constrains and influences behavior.
Geospatial statistical modeling uses statistical algorithms to model that environment, using sometimes thousands of types of measurements. It uses that model to compare the location of past crimes and the features of the environment where those crimes occurred. It then produces a statistical characterization of the environment, which shows not only what is known, but also what is unknown. It shows the previously unknown areas of the environment that are suitably similar to where past crimes occurred so as to create a statistical likelihood of future crimes occurring in those geographic locations. It then uses that assessment to create a likelihood map of where future crimes may be located. Analysts, investigators, and law enforcement personnel can then use these likelihood maps as objective and scientific means to understand the environment for which they seek to police.Thus, the model does not produce dull statistics. Rather, it identifies the geospatial preference of perpetrators for future crimes on a map like the one shown in figure 2. Here, the darker the color, the higher the likelihood that the geospatial features of the environment are conducive to crimes like those modeled. When viewed in more detail, important decisions about crime prevention and response can be made.
Geospatial statistical modeling provides law enforcement personnel answers to questions which often times go unanswered. These questions include the following:
- Where is a crime—a shooting, a drug deal, a gang action, and so forth—most likely to occur?
- Where is the greatest officer safety risk or law enforcement opportunity?
- How should agency resources best be deployed?
- What has changed in the criminal environment since yesterday?
- What can the department expect to be different tomorrow?
- How can officers influence others’ actions to achieve strategic goals?
Geospatial statistical modeling uses statistical algorithms to produce objective assessments of the environment in which crime occurs. It then uses that assessment to identify other areas with a statistical similarity to the crime areas and creates a likelihood map of where future crimes may be located. Analysts, investigators, and commanders can apply these likelihood maps to support intelligence production, investigations, and strategic planning in an objective manner.
The New Jersey Project
The following case study traces the use of a geospatial statistical modeling tool through one validation test against real crime data from Jersey City. The results of this analysis show the strategic, operational, and tactical value of this type of analysis in helping law enforcement commanders and investigative personnel to employ data, statistics, and analytics to improve enforcement operations, investigations, and intelligence initiatives. The NJ ROIC is at the center of this case study because it has provided the Jersey City Police Department with access to analytical capacities to which it otherwise would not have had access.
The NJ ROIC is an “all-crimes, all-threats, all-hazards” fusion center that supports law enforcement and homeland security agencies across New Jersey. Part of its mission is to assist law enforcement agencies with carrying out the governor’s Strategy for Safe Streets and Neighborhoods.2 The governor’s plan is aimed at channeling law enforcement resources at gangs, guns, and repeat violent offenders present throughout the state. In response, the NJ ROIC designed its analytical initiative, titled Project Watchtower, comprised of three core elements: NJ POP (Pins on Paper) focuses on gun violence; NJ TAG (Targeting the Activities of Gangs) focuses on targeting the activities of criminal street gangs; and NJ Trace focuses on tracing crime-related guns entering New Jersey.
Since March 2009, the Jersey City Police Department has hosted a biweekly information sharing and coordination meeting with its allied partners who have an interest in addressing the violent crime problem occurring in that jurisdiction. At these meetings, the NJ ROIC provides police commanders and investigators with intelligence derived from its Project Watchtower initiative, which is specific to Jersey City. This intelligence aids the police with developing enforcement and investigative operations.
Interpreting Jersey City’s Shooting Environment
The NJ ROIC’s highly developed NJ POP is an effective, web-based data collection and analytical application that focuses on shooting incidents in which a victim is struck with a projectile. Originally deployed to a few high-crime jurisdictions, the application is now operating throughout the state, collecting data on shootings within hours of their occurrence. The shooting data for Jersey City was available for analysis starting in December 2008.
Shooting incidents are one of those criminal acts about which it is easy to develop a mental “mind map.” Shootings occur in high-crime areas. Shootings occur where there are obvious criminal elements and where there are obvious criminal attractors, such as drugs, money, alcohol, and pornography. All of these things and more would lead law enforcement investigators and patrol officers to predict from their experience that future shootings would occur.There is a history of shootings along a narrow corridor in central Jersey City, shown in figure 3. Experience says that the next shootings will be along that corridor. After all, that is where shootings have occurred throughout December into March. But the goal of intelligence-led policing is to use this information to anticipate the future—and act on it.
The geospatial statistical modeling application has helped to anticipate the future by identifying areas in which future criminal acts are most likely to occur based on similar past criminal acts. The analysis uses as much geospatial information that is available, including census data; locations of buildings, parks, bus stops, and churches; and social facilities such as bowling alleys, food marts, bars, restaurants, and basketball courts; and economic data. The objective is to paint a picture of the community that can be statistically characterized, identify where previous crimes took place, and then fuse this information to create a statistical portrait of the preferred locations for those crimes. Finally, this leads to the ability to identify other places that have a likelihood of being preferred for those crimes. This will help neighborhoods and police understand what might happen there in the future and to develop plans to counteract criminal activities. Figure 4 shows the results of such an analysis. More than 500 different geospatial factors in the environment were included in this analysis.
The shooting incidents in Jersey City from December through March provided the locations of the criminal acts. The red color indicates areas which are geospatially similar to locations where shootings occurred historically. The darker red indicates the higher the likelihood that if a shooting occurs, it will be in a geospatially similar area. This density map of Jersey City shows the environment most conducive to future shootings.
Testing the Model
The NJ ROIC has become a test bed for the transfer of a geospatial statistical modeling tool, once limited to the military and intelligence community, into the domestic law enforcement community. As a result, armed with the December 2008 through March 2009 shooting assessments from NJ POP, the test of the model’s accuracy was the next objective.
Using the forecast of future shootings from the December through March assessment, the April and May shootings (green and blue dots) were plotted, as shown in figure 5. Two findings emerge from this assessment.
The first important finding is that all but one of the April and May shootings fell within the high-likelihood area. The one April shooting that was not anticipated by the model occurred within two or three blocks of the forecast shooting areas. However, this shooting becomes part of a special analysis to determine how it was different and, if relevant, will be used to improve the model.
The more homogeneous analysis events are, the better the model narrows the area of likelihood. If drug-related shootings, gang retribution shootings, domestic violence, and armed robbery shootings are lumped together, the results are less accurate in forecasting future locations of events, than if events are analyzed separately. This is part of the special methodology learned in the military and intelligence community that can be transitioned to domestic law enforcement.
The second important finding is that the shootings appear to have moved west. This raises strategic questions that can be addressed by the Jersey City police, such as the following:
- Were there more patrols along the high-crime corridor that pushed shootings into new areas?
- Were there social or cultural activities along the high-crime corridor that pushed shootings to the west?
- Were there opportunities to the west, such as street fairs, markets, and drug dealing, which took violence to new streets?
Geospatial statistical modeling does not answer all these questions. Rather, it provokes additional analyses by investigators and police leadership to look at the strategic impact of the statistical analysis of the crime environment to see if there are actions that the community can control that are making a difference in crime.
With both the strategic and tactical findings, the forecast modeling of likely areas of future shootings based on past shootings seems to be relevant and accurate.
The test of the December through March data did a good job of identifying the most likely areas for shootings in April and May. However, it also identified some areas of special interest, as shown in figure 5. The arrow and circle show several blocks where, in the previous half year, there have been no shootings. Yet the geospatial environment is within the top 15 percent of most likely areas to be conducive to future shootings.
Potential Impact Elsewhere
As discussed before, geospatial statistical modeling is not conducted in isolation. Rather, it is an integrated part of the entire investigative process—it is one input that investigators and managers should include in their analyses of how best to carry out law enforcement efforts.
This finding may suggest several courses of action. One might ask if there is something special about that area that repels shootings, such as a strong youth organization or church activity. One might ask whether there is special police activity in the area that repels crime. One might ask whether the area is simply lucky. But no matter what questions remain, the potential for geospatial statistical modeling to open a new dialogue, provide new facts, create new assessments, and raise new questions is the essence of intelligence-led policing.
The Unfortunate Success
The geospatial statistical modeling assessment described here was conducted in early June 2009. An important part of that project is determining how geospatial statistical modeling can become part of the law enforcement investigation concept of operations. The first part of the project examined the existing concept of operations and identified places in investigations where analytics could be inserted to make a contribution. In New Jersey, the excellent working relationship with high-crime cities offered a way to obtain crucial investigation information, convert it into geospatial analysis (thus eliminating personal information and avoiding any privacy concerns—all geospatial statistical modeling needs to know to make an assessment is the location and the crime), and then insert the geospatial analysis into the investigation process in a way that materially helps the investigation.That process is starting to work and expand, but in June 2009, it was just beginning.
On the night of June 8, 2009, a shooting occurred in one of the areas where no shootings had occurred in the previous eight months, yet it was identified only days before as a high-likelihood shooting area (see figure 6). This was an unfortunate but meaningful way to demonstrate the value of geospatial statistical modeling.
Influencing Decision Makers
The inclusion of geospatial statistical modeling enables intelligence enterprises, such as the NJ ROIC, not only to better interpret their environments and uncover patterns, but also to strengthen their capacity to deliver visually descriptive and relevant intelligence products needed to influence police commanders and line personnel.
With regard to impacting public safety, Jersey City Police commanders, after receiving a map (see figure 7) showing the likelihood shooting assessment overlaid on police precincts and districts, were able to tailor their enforcement and investigative efforts toward high-probability violent crime areas. This simple method of increasing awareness of the environment is an example of how geospatial statistical modeling can make a difference in a commander’s arsenal. By merging incident data with geospatial environmental data to identify the likelihood that specific areas will be affected by crime in the future, commanders can allocate their finite resources efficiently and effectively.
At New Jersey’s fusion center, the integration of geospatial statistical modeling into the analyst’s toolbox is enabling individuals to graphically display high-probability violent crime areas. This has increased police officers’ abilities to uncover crime trends in the Jersey City criminal environment. This technology, originally and operationally used by the military, can transfer easily to analyze terrorist threats and other domestic environmental hazards. Essentially, geospatial statistical modeling is strengthening the fusion center’s ability to carry out intelligence-led policing. ■
|This case study is based on the testing performed at the NJ ROIC of the Signature Analyst geospatial predictive analytic software produced by SPADAC Inc. as part of an ongoing project by the Department of Homeland Security, Directorate of Science and Technology.|
1Through a project funded by the Department of Homeland Security, Directorate of Science and Technology, at the NJ ROIC, SPADAC Inc. has been working with fusion center analysts to determine the viability of transferring technologies and associated methodologies from the military and intelligence community to domestic law enforcement.
2Jon S. Corzine, A Strategy for Safe Streets and Neighborhoods (New Jersey, October 2007), http://www.nj.gov/oag/crimeplan/safe-exec-summ-complete.pdf (accessed June 25, 2010).
Please cite as:
Raymond Guidetti and James W. Morentz, "Geospatial Statistical Modeling for Intelligence-Led Policing," The Police Chief 77 (August 2010): 72-76, http://www.nxtbook.com/nxtbooks/naylor/CPIM0810/index.php#/72-76 (insert access date).