By Charles Anyinam, PhD


The importance of place in crime causation and crime prevention has been emphasized over the years (Weisburd, 2008). Current environmental criminological theories of crimes such as rational choice (Clarke & Felson, 1993); routine activity theory (Cohen & Felson, 1979); and crime pattern theory (Brantingham & Brantingham, 1993) give weight to the importance of place for understanding crime. As McCord and Ratcliffe have summed up, “together these three theories state that specific types of land uses and facilities generate crime due to the daily activities associated with them and the number and types of people they attract” (McCord & Ratcliffe, 2009, p. 18). With increasing effectiveness, GIS and spatial analysis are being used to more rigorously examine the effect of “place” on crime. In the last few years, one approach that has been advanced to understand the environmental context of crime and the role that sites, locations or places in urban areas play in attracting criminal activities is Risk Terrain Modeling (RTM) (Caplan & Kennedy, 2010a; Caplan, Kennedy, & Miller, 2011).

Risk Terrain Modeling is an analytic technique based on the idea that crime offenders, crime victims, and crime targets operate in space and time and that the risk of a crime event occurring at a specific location is determined by a combination of social, cultural, economic, and physical environmental risk factors (Caplan & Kennedy 2010; Caplan & Kennedy, 2011; Caplan, Kennedy & Piza 2012). When these risk factors, which may be “generators” and/or “attractors” of crime, intersect in space and time, they have the greatest potential to yield a particular crime outcome. When such factors are identified, they can be used to create separate map layers to represent their presence, absence, or intensity (Caplan & Kennedy, 2010). When these risk map layers are further combined, a composite “risk terrain” map can be produced. Risk terrain modeling of crimes, thus, results in maps that show places with the greatest risk or likelihood of becoming places for crimes to occur in the future. The result is the production of a geospatial hot spot map of areas with the highest probability of crimes occurring within a certain time period. This analytic technique provides crime analysts the opportunity to describe, demonstrate, analyze and explain the socio-cultural, economic, and physical/geographical environmental context within which certain types of crime are concentrated in urban areas. In a nutshell, the RTM technique examines the features of places that contribute to crime concentration.

This report briefly illustrates how the risk terrain modeling (RTM) technique was used to identify the environmental context within which violent crimes are most likely to occur in the city of New Haven, Connecticut. The report also tests the predictive value of the composite “risk terrain” map to determine the empirical credibility and degree of confidence in using the model for future forecasts of violent crimes in New Haven.


New Haven, Connecticut is the second-largest city in Connecticut with a population of 129,779 people in 2010. The city has a total area of 20.1 square miles (52.1 km2), of which 18.7 square miles (48.4 km2) is land and 1.4 square miles (3.7 km2), or 6.67%, is water. Several risk factors that have been found to relate to violent crimes by existing criminological theories and empirical studies are also found to be relevant in the city of New Haven. Based on personal knowledge of criminal activities in New Haven as well as past literature (for example, Anderson, 1999; Bernasco & Block, 2011; Brantingham & Brantingham, 1982, 1995; Caplan et al., 2011, Caplan, Kennedy, & Baughman, 2012; Fass & Francis, 2004; McCord, Ratcliffe, Garcia, & Taylor, 2007), 27 potential risk factors were selected as “attractors” and/or “generators” of violent criminal activities in New Haven. For convenience, these are grouped into three main categories:

A. Commercial Infrastructural Services

Banks, bars/clubs/restaurants, beauty salons/barber shops, check-cashing stores, convenience stores, entertainment facilities, fast foods, gas stations, hotels/motels/inns, package stores, pawn shops, recreational facilities, strip malls

B. Municipal Infrastructural Services

Apartment complexes, bus stops, cemeteries, elderly housing, halfway houses, parks, public housing, schools, Section 8 housing

C. Potential Offenders/Perpetrators of Crime

Drug arrestees’ home addresses, public drug complaint calls for service, parolees, released prisoners, probationers

The Risk Terrain Modeling process tests a variety of factors that are thought to be geographically related to crime incidents. Valid factors are selected and then weighted to produce a final model that basically paints a picture of places where crime is statistically most likely to occur (Caplan, et. al 2011). In 2013, Rutgers Center for Public Security released the RTMDx Utility software, a tool that facilitates the empirical method of risk terrain map production. Prior to the development of this tool, the operationalization of risk factors was undertaken manually using the 10 steps described by Caplan and Kennedy (2010) with the help of an ArcGIS toolbox designed by Caplan and colleagues for use in ArcMap.

The 27 factors identified above were operationalized using the Risk Terrain Modeling Diagnostics (RTMDx) software that automates various steps required to determine the spatial influence and significance testing of risk factors (Caplan et al., 2013). The RTMDx tool accepts up to 30 risk factors as inputs for analysis. The 27 identified risk factors were operationalized by inputting a number of parameters: shape of study area, block length, raster cell size (i.e. the dimensions on the ground of a single cell in a raster, measured in map units), type of model, outcome event, risk factors and operationalization of the spatial influence (Caplan et al., 2013). The following are inputs set up before running the RTMDx software:

  1. Study area: New Haven city boundary
  2. Cell Size: the average block length in New Haven is 446 feet and the raster cell size was set to half-block (223 ft). There were 11,335 raster cells used in the analysis of which 1984 cells contained violent crime events.
  3. Model Type menu allows one to select either aggravating or protective model type. An aggravating model assumes that the risk factors input into the RTMDx utility correlate with the locations of outcome events and it tests for positive spatial relationships. A protective model type assumes that the risk factors correlate with the absence of outcome events and it tests for negative spatial relationships. One example of an aggravating factor might be ATMs for the crime of robbery. An example of a protective factor could be a police sub-station or community garden (Caplan et al. (2013, p.18). Aggravating model was input in as the model type for this study.
  4. Outcome Event Data: The outcome events for the study are violent crimes (principally murders, aggravated assaults and robberies) that occurred in New Haven between 2009 and 2013.
  5. Risk Factors and Operationalization of Spatial Influence: 27 risk factors were individually added, each of which was spatial influence-based and tailored to the study area: density, proximity, and both proximity and density. In this study, the maximum spatial influence was set to three blocks for each risk factor with spatial increments set to half-block. Caplan, et al. (2011) have demonstrated that some risk factors are more a function of distance from the closest feature while others are a function of the presence or absence of the feature (i.e. density of the feature). In this study, drug complaints data for 2009-2013, public housing data, drug arrestees’ home addresses, prisoners released to the city of New Haven in 2013, Section 8 houses, apartment complexes, and all 2013 parolees and probationers were operationalized as a function of density. The rest were operationalized as a function of proximity.

The RTMDx tool was run to generate a model that represented the risk factors for 6,671 violent crime events. Based on the operationalization of the 27 risk factors, 162 variables were created and tested for significance. The software builds from a null model and uses bidirectional stepwise regression process to build the “best model”. RTMDx operates based on building the optimal model, which is reflected in the Bayesian Information Criteria (BIC) value. The “Best Model Specification” section of the report generated after running the RTMDx tool provides details about the risk factors included in the risk terrain model, their optimal spatial influences and operationalization methods and their relative weights (Caplan et al., 2013).

The RTMDx tool determined that the best risk terrain model was a Negative Binomial type II model with 16 of the 27 risk factors and a BIC score of 14,067. The model also included an intercept term that represents the background rate of events and an intercept term that represents over-dispersion of the event counts. The 16 risk factors selected after running the software make statistically significant contributions to the outcome variable. Thus, all areas in New Haven with a high concentration of drug complaints, public housing, convenience stores, gas stations, bus stops, drug arrestees’ home addresses, residences of released prisoners, banks, bars, restaurants and cafes, schools, section 8 houses, apartment complexes, probationers’ residence, beauty salons and barber shops, package stores, and parks are associated with a higher concentration of violent crimes in the city (Fig. 1).

Fig. 1 shows results of the “best model” specified after running RTMDx. The relative risk value column shows the weighted values of the selected risk factors. Ten of the risk factors were operationalized based on proximity function of half-block (223 feet). The top five relative risk values are associated with drug complaints, public housing, and location of convenience stores, gas stations and bus stops. Using the “best model” for creating high-risk areas, the RTMDx creates a composite risk terrain map.

FIG. 1

Type Name Operationalization Spatial Influence Coefficient Relative Risk Value
Rate DrugCompliants2009 2013 Density 223 1.3281 3.7739
Rate PublicHousing Density 223 0.9946 2.7037
Rate ConvenienceStores Proximity 1338 0.9262 2.5249
Rate GasStations Proximity 223 0.7288 2.0726
Rate BusStops Proximity 223 0.6148 1.8492
Rate DrugArrestesHomeLocation Density 1115 0.5965 1.8157
Rate PrisonersReleased2013 Density 223 0.5274 1.6945
Rate Banks Proximity 223 0.5207 1.6831
Rate BarsClubsRestaurantsCafes Proximity 223 0.4914 1.6347
Rate Schools Proximity 223 0.4201 1.5221
Rate Section8Housing Density 223 0.4180 1.5189
Rate ApartmentComplex Density 223 0.4143 1.5133
Rate Probationers2013 Density 1338 0.4072 1.5026
Rate BeautySalonBarberShop Proximity 892 0.2726 1.3134
Rate Package Stores Proximity 1115 0.2422 1.2740
Rate Parks Proximity 446 0.2274 1.2554
Rate Intercept -2.6195
Overdispersion Intercept 0.7132



The selected risk terrain model assigned relative risk scores to cells ranging from 1 for the lowest risk cell to 575.9 for the highest risk cell. These scores allow cells to be easily compared. For instance, a cell with a score of 575.9 has an expected rate of crime that is 575.9 times higher than a cell with a score of 1. The RTMDx utility software (professional version) generates a GeoTiff (i.e. raster image) that contains only values for the cells that were included in the “best model” specifications. The resulting composite map with GeoTiff format was further converted to create a new grid raster layer in order to undertake further analysis of the results.

To determine the effectiveness of the model for forecasting future crime events, the relative risk scores generated by the RTMDx utility tool was used to create top risky places with high probability for violent crimes to occur in the future. The raster grid layer was converted to vector shapefile (Map 1), which was further symbolized using another spatial and visualization technique. For the purposes of operational policing, the Local Indicators of Spatial Autocorrelation (LISA) method was used to identify local spatial clusters of similar values (whether high or low values). LISA was applied to the relative risk values derived from the vector shapefile. This process was undertaken with the construction of a spatial weights matrix, which imposes a neighborhood structure on the data to access the extent of similarity between locations and values. Two basic categories of defining neighborhood spatial relation are contiguity and distance. Contiguity-based weights matrices include rook and queen. Areas are neighbors under the rook criterion if they share borders (e.g., on a grid, only the cells to the North-South and East-West are neighbors). Under the queen criterion, areas are neighbors if they share either a border or point (e.g., on a grid, in addition to the four cells included under rook, the four cells sharing a corner with the central location are also counted as neighbors) (Anselin et al. 2008). The queen contiguity was used to identify spatial relationship. The local Moran’s I LISA approach indicated areas of statistically significant clusters where violent crimes are higher or lower than would be expected if the incidents were randomly distributed in New Haven. The advantage of using LISA is that it can distinguish between statistically significant clusters of high values surrounded by high values (HH), low values surrounded by low values (LL), high values surrounded by low values (HL), and low values surrounded by high values (LH) (Kennedy et. al 2011). Map 2 shows results of the LISA cluster analysis that identified only statistically significant clusters of High – High places. These top high-high risk areas make up only 6.09% of the entire area of New Haven.

Anyinam CM 9 Anyinam CM 8


To what extent does the derived composite risk terrain map in the form of LISA cluster map accurately forecast violent crime locations in New Haven? To find out, the high-high risk areas were tested with new crime data. During the first seven months of 2014, the New Haven police recorded 396 incidents of murder, robbery and aggravated assault (including non-fatal shootings). Specifically, 41% of non-fatal shootings, 57% of murders and 39% of robberies occurred in these high-risk areas, which make up only 6.09% of the city of New Haven.

These findings support the validity of RTM methodology for forecasting. The results of the study demonstrated that the co-existence of the 16 risk factors does have strong effects on the locations of violent crimes in New Haven. The results of the study were discussed with supervisors in charge of the ten policing districts. They were made aware of the high-risk areas and the different risk factors that form the “backdrop” of these areas. The need to formulate appropriate proactive and preventive strategies to address violent crimes in these areas was also discussed. Identifying these high risk areas for violent crimes adds considerable value to not only understanding the nature of violent crime problems in the city as a whole but also in prioritizing and designing appropriate police and community interventions to address the underlying risks associated with them. The risk terrain model highlights the need to devise, implement and take actions that will deter potential offenders, harden crime-prone targets, and reduce violent crime incidents in the city.


In this study, we identified 16 statistically significant risk factors that underlie violent crime occurrences in the city of New Haven, using the RTMDx software recently released by Rutgers Center for Public Security. The results point to the fact that the Risk Terrain Modeling technique has significant operational value. The fact that a high percentage of violent crimes that occurred during the first seven months of 2014 happened in areas identified by the composite RTM map lends support to the need for “place-based” strategies in dealing with violent crimes in New Haven. The results reinforce the relevance and validity of the risk terrain model in identifying geospatial hotspots or clusters of high-risk areas for violent crimes. The high-risk areas identified have the potential to make crime prevention interventions more efficient and successful in terms of resource allocations and short- and long-term planning for crime control and reduction. Risk terrain modeling is a “placed-based” forecasting technique that has the potential to assist police in not only prioritizing areas of focus for action, but also to help patrol officers find more meaningful ways to modify and/or change the geographic characteristics and dynamics that promote, encourage, and cause violence in these risk areas. The RTM approach and tool provides for a better understanding of the need and urgency to address the relatively few high-risk places and to prevent violent crimes in the city of New Haven and elsewhere.



Anderson, E. (1999). Code of the street: Decency, violence and the moral life of the inner city. New York, NY: Norton

Anselin L., Griffiths, E., & Tita G. (2008). Crime mapping and hot spot analysis. In: Wortley R., Mazerolle, L. (eds) Environmental criminology and crime analysis. Willan Publishing, Cullompton, Devon, pp 97–116.

Bernasco, W., & Block, R. (2011). Robberies in Chicago: A Block-Level Analysis of the Influence of Crime Generators, Crime Attractors, and Offender Anchor Points. Journal of Research in Crime and Delinquency 48 (I), 33-57.

Brantingham, P. & Brantingham, P. (1993). Environment, routine, and situation: Toward a pattern theory of crime. In R.V. Clarke and M. Felson (eds), Routine Activity and Rational Choice (vol. 5). New Brunswick: Transaction publishers, pp. 259 – 294.

Brantingham, P. & Brantingham, P. (1995). “Criminality of Places: Crime Generators and Crime Attractors”. European Journal on Criminal Policy and Research 3:1-26.

Brantingham, P. L. & Brantingham, P. J. (1982). Mobility, notoriety and crime: A study of crime patterns in urban nodal points. Journal of Environmental Systems, 11 89-99.

Caplan, J.M., Kennedy, L.W. & Baugham, J. (2012). Kansas City’s Violent Crime Initiative: A Place- Based Evaluation of Location-Specific Intervention Activities During a Fixed Time Period. Crime Mapping: A Journal of Research and Practice, 4(2), 9-38.

Caplan, J. M., Kennedy, L. & Piza, E. (2012). A Primer on the Spatial Dynamics of Crime Emergence and Persistence. Newark, NJ: Rutgers Center on Public Security.

Caplan, J.M, Kennedy, L.W., & Piza, E.L. (2013). Risk Terrain Modeling Diagnostics Utility User Manual (version 1). Newark, NJ: Rutgers Center on Public Security.

Caplan, J. M., Kennedy, L. W., & Miller, J. (2011). Risk Terrain Modeling: Brokering Criminological Theory and GIS Methods for Crime Forecasting. Justice Quarterly, 360-381.

Caplan, J.M, & Kennedy, L.W. (2010). Risk Terrain Modeling Manual: Theoretical Framework and Technical Steps of Spatial Risk Assessment. Rutgers Center on Public Security, Newark, NJ 6/29/2010.

Caplan, J.M, & Kennedy, L.W. (2011). (Edited) Risk Terrain Modeling Compendium for Crime Analysis. Newark NJ. Rutgers Center on Public Security.

Clarke, R.V. & Felson, M. (1993). Introduction: Criminology, routine activity and rational choice. In R.V. Clarke & M. Felson (eds), Routine Activity and Rational Choice vol. 5, New Brunswick: Transaction publishers.

Cohen, L.E. & Felson, M. (1979). Social change and crime rate trends: A routine activity approach. American Sociological Review, 44: 588-608.

Fass, S. M., & Francis, J. (2004). Where have all the hot goods gone? The role of pawn-shops. Journal of Research in Crime and Delinquency, 41, 156-179.

Kennedy, L.W., Caplan, J.M. & Piza. E. (2011) “Risk Clusters, Hotspots, and Spatial Intelligence: Risk Terrain Modeling as an Algorithm for Police Resource Allocation Strategies,” Journal of Quantitative Criminology 27: 339–362.

McCord, E. S., & Ratcliffe, J. H. (2009). Intensity value analysis and the criminogenic effects of land use features on local crime patterns. Crime Patterns and Analysis, 2(1), 17-30.

McCord, E. S., Ratcliffe, J. H., Garcia, R. M., & Taylor, R. B. (2007). Non-residential crime attractors and generators elevate perceived neighborhood crime and incivilities. Journal of Research in Crime and Delinquency, 44, 295-320.

Weisburd, D. (2008). Place-Based Policing. Ideas in American Policing. Police Foundation, No. 9 January.

Charles Anyinam

Charles Anyinam holds a PhD (Geography) from Queen's University, Kingston, Ontario, Canada and a Graduate Diploma (GIS) from York College of Information Technologies, Toronto. He taught at a number of universities in Canada including University of Toronto and York University, North York, Ontario, Canada. He is currently the Supervisor of the Crime Analysis Unit, New Haven Police Department, New Haven, Connecticut. His research interests include spatial-temporal analysis, predictive analytics, and use of a variety of techniques in crime mapping and analysis.

You must be logged in to post a comment