Geography GIS and Crime Analysis
Lin Liu
  • LAST MODIFIED: 28 July 2021
  • DOI: 10.1093/obo/9780199874002-0233


Spatial analysis of crime has gained increasing attention during the past thirty years, coupled with the growth of geographic information systems (GIS). Most crime analysis tasks are either carried out in a GIS environment or supported by a GIS. GIS is typically used as a tool for data management, data processing, data visualization, and data analysis for crime studies. Crime analysis normally involves the following elements: uncovering spatio-temporal patterns of crime distribution, such as crime hotspots; explaining these patterns and discerning major contributing factors based on multivariate regression modeling; predicting future crime patterns using machine learning and other predictive methods; developing crime prevention approaches based on historical and future crime patterns; and evaluating the effectiveness of crime prevention, to find out if crime is reduced in the targeted area and whether the nearby areas are affected by the intervention. It should be noted that crime analysis is inherently multidisciplinary, including but not limited to geography, criminology, computer science, statistics, urban planning, and sociology. Therefore, an effective crime analyst should be well trained in multiple disciplinary approaches. Any crime analysis that leads to real-world impact must rely on sound theories and effective methodologies. Many of the theories covered in this article are related to geography, criminology, and sociology. The methods are mostly influenced by GIS, spatial statistics, and artificial intelligence. Crime analysis also involves multiple stakeholders, including at least government agencies, universities, and private companies. Universities conduct basic and applied research, private companies convert the research to products, and government agencies provide funding for research and implement crime prevention strategies. In addition, crime analysis needs to pay close attention to potential issues related to ethics, privacy, confidentiality, and discrimination.

GIS for Crime Analysis

Crime data and related data are often managed in geographic information systems (GIS) databases. Maintaining data in a spatial database management system helps ensure data integrity, facilitates efficient query and processing, and supports spatial analysis.

Crime Data and Environmental Data

Two types of data are involved in crime analysis. The first is data pertaining to crime itself, the second is data that help explain crime.

Crime Data

Crime data come from a variety of sources, including official crime data from police departments, data derived from court records, and data derived from social media such as news media, blogging, tweets, etc. These data describe the crime incident, the offender, and the victim. The most often used official crime data come from the crime database, calls for service, and the arrestee database (Hill and Paynich 2013). In the United States, the Federal Bureau of Investigation (FBI) compiles a Uniform Crime Report (UCR) that integrates crime data reported from individual police departments. The “more serious” Part 1 offenses in UCR include murder and non-negligent homicide, rape, robbery, aggravated assault, burglary, motor vehicle theft, larceny-theft, and arson. The first four are termed violent crimes and the latter four property crimes. The “less serious” Part 2 offenses include simple assault, curfew offenses and loitering, embezzlement, forgery and counterfeiting, disorderly conduct, driving under the influence, drug offenses, fraud, gambling, liquor offenses, offenses against the family, prostitution, public drunkenness, runaways, sex offenses, stolen property, vandalism, vagrancy, and weapons offenses. The calls-for-service database contains assignments to police, fire department, and emergency medical services as a result of the calls made by the public via emergency telephone services, such as 911 in America, 999 in United Kingdom, and 112 across Europe. While many countries use a single emergency call number, others use different numbers for different types of emergency services. For example, 110 is the number for police, 119 for firefighters, and 120 for the ambulance in China. Crime studies mostly use calls about the police. Calls for service are often used to represent the need for policing resources. It should be noted that the classifications of the calls do not necessarily match those of the crime database. Thus, linking crime databases to calls for service can be challenging. The arrestee database contains information about the arrested offenders, including their home addresses. It is possible to link an arrestee to the corresponding crime incident.

Social and Built Environment Data

Data for explaining crime can be broadly grouped into two categories: social environments and built environments. Social environments pertain to economical status, racial composition, population structure, housing, neighborhood stability, etc. Built environments, on the other hand, cover transportation network, land use, facility that attracts or generates crime, natural edges or transitional areas between zones, etc. The most dominant source of data related to social environments is the census. The census can derive hundreds of variables on population and economy, such as income, housing, race and ethnicity, etc. However, one of the main drawbacks is that the census is typically updated once every ten years. The relatively low spatial resolution is another drawback. For example, the smallest unit of the US census is a block. In cities, a census block corresponds to a city block bounded by four streets. With the emergence of geotagged social media data, cell phone data, and GPS tracking data, new variables with high spatial and temporal resolutions become available. The integration of these new data and the traditional census data has shown great promise. Built environment data come from a variety of sources. Land use may be interpreted from remote sensing images. Streets and points of interest (POIs) are available from mapping and navigation companies such as Google Map. POIs have been used to represent crime attractors and generators. Satellite nightlight data can help capture the city space of human activities. LiDAR—Light Detection and Ranging—can create precise three-dimensional (3-D) models of cities. The 3-D models help describe micro built environments, which are critical for crime studies in high spatial resolutions. Another important source of data for environment auditing is street view images.

Geocoding and Address Matching

In most cases, every crime incident, victim, and offender are associated with an address. The very first step of crime analysis is geocoding or address matching, a process that converts an address to a location on the map. There are two dominant approaches for geocoding: one is the linking of the address to the corresponding parcel, the other is through linear interpolation of the address based on address ranges of the street segment. The former is available in some commercial online maps such as Google Map, in which every address is represented as a point of interest, with a precise location. Therefore, this approach leads to highly accurate results. A typical GIS adopts the latter approach. Both OpenStreetMap and commercial street maps can be used for geocoding. Because many of the earlier computer-aided dispatch systems relied on outdated maps and address databases, geocoding in GIS cannot match all the addresses. For a typical American city, 92–95 percent of the addresses can be matched automatically. This matching rate drops significantly in rural areas. The matching rate also varies significantly across countries. For example, the matching rate is below 80 percent in Chinese cities. Furthermore, the locational accuracy of geocoding, measuring the closeness between the actual location and the geocoded location, also varies between cities and rural areas, and among various countries. Both matching rates and locational accuracies are important in crime studies. The minimum of 85 percent matching rate has been well accepted following the work of Ratcliffe 2004. However, Briz-Redón, et al. 2020 recently reveals that the 85 percent rate may be too low, while Andresen, et al. 2020 shows that the needed matching rates are lower than those found in previous research. These conflicting findings further underscore the complexity of the issue. A low matching rate impairs the representativeness of the geocoded incidents. Similarly, a low locational accuracy may result in misrepresentation of the spatial distribution of crime. This problem is more pronounced in micro studies of crime in small units such as individual addresses or street intersections. For example, if one is interested in studying robberies within fifty meters of a bus station, a location accuracy lower than fifty meters makes the geocoded crime map unusable.

  • Andresen, Martin A., Nick Malleson, Wouter Steenbeek, Michael Townsley, and Christophe Vandeviver. “Minimum Geocoding Match Rates: An International Study of the Impact of Data and Areal Unit Sizes.” International Journal of Geographical Information Science 34.7 (2020): 1306–1322.

    DOI: 10.1080/13658816.2020.1725015Save Citation »Export Citation » Share Citation »

    In an effort to revisiting the issue of minimum acceptable geocoding match rate, this international study finds that the match rate depends on the number of points and the number of areal units under analysis, but generally shows that the needed match rates are lower than those found in previous studies.

    Find this resource:

  • Briz-Redón, Álvaro, Francisco Martinez-Ruiz, and Francisco Montes. “Reestimating a Minimum Acceptable Geocoding Hit Rate for Conducting a Spatial Analysis.” International Journal of Geographical Information Science 34.7 (2020): 1283–1305.

    DOI: 10.1080/13658816.2019.1703994Save Citation »Export Citation » Share Citation »

    This article revisits the issue of minimum acceptable geocoding hit rate. The results indicate that variations in intensity, clustering, and aggregation levels of spatial analysis lead to different acceptable rates. Therefore, the previous minimum of 85 percent needs to be raised.

    Find this resource:

  • Ratcliffe, Jerry H. “Geocoding Crime and a First Estimate of a Minimum Acceptable Hit Rate.” Journal International Journal of Geographical Information Science 18.1 (2004): 61–72.

    DOI: 10.1080/13658810310001596076Save Citation »Export Citation » Share Citation »

    This highly cited article establishes the famous minimum acceptable geocoding success rate of 85 percent for crime analysis, to ensure the needed statistical representativeness of the geocoded crime incidents.

    Find this resource:

Crime Mapping and Visualization

Once crime incidents are geocoded, they can be mapped and visualized. Geocoded crime incidents are mapped in various symbols of point, line, or area. The earliest crime maps are pin maps, with each incident mapped as a pin. In modern GIS, a crime incident is mapped in a point symbol of unique shape, such as a dot, circle, triangle, square, etc. Each shape can represent a crime type, thus displaying multiple crime types on a single map. However, when multiple incidents are accumulated in the same location, pin maps or point symbol maps are no longer effective. One way to solve this problem is to calculate density, measured by the number of incidents divided by the size of the area. Density maps are an effective representation of crime distribution in space. The simple density treats all incidents in the area equally. A variation of density maps is the kernel density map, in which a kernel dictates how incidents are weighted in the density calculation. An incident near the center of the area is weighted higher than an incident away from the center. When crime incidents are aggregated to the nearest street segments, line symbols are used to map crime counts on the street segments. The intensity of a color can be used to show the crime count on a line with a fixed width. The convention is that darker colors correspond to higher counts. The width of a line can also be made proportional to the crime count. If crime incidents are aggregated to areas or polygons, both point and area symbols can be effective for mapping crime counts within the areas. Graduated symbols such as circles make the sizes proportional to the counts. Graduated colors make the intensity proportional to the counts, resulting in what is called choropleth maps. While choropleth maps are widely used to show crime counts, they can be misleading when the size of polygons varies significantly across the space. An alternative approach is to use dot-density maps, in which dots are randomly placed inside a polygon and each dot represents a certain number of crime incidents. Chainey and Ratcliffe 2005 and Gorr, et al. 2018 are two representative books on crime mapping in a GIS environment.

Spatial Analysis

Virtually all-analytical functions in a GIS outlined in Chang 2019 and Santos 2017 can be used in crime studies. This section covers some of the most commonly used functions. Crime hotspot detection is perhaps the most commonly used function for crime analysis in GIS. The spatial distribution of crime follows the power law, or the 80–20 rule, which dictates that majority of the crime incidents are concentrated in a few small areas. Crimes are inherently clustered. Spatial autocorrelation functions such as global Moran’s I can test the degree of crime clustering. LISA (local indicators of spatial association) plots are frequently used to detect clusters of areas of high crime concentration surrounded by other areas of a high prevalence of crime. Density maps are also effective in detecting hotspots. Some scholars argue that areas with a density higher than the average are regarded as hotspots, but this approach tends to detect too many unstable hotspots that change over time. If the main interest is to locate permanent hotspots, the density threshold should be set at the 85th percentile or higher. It should be pointed out that not everywhere inside a hotspot has crime. The location with the highest density value may be inside a gap in-between high crime concentration areas. Buffering is another function frequently used in crime analysis. The phenomena of distance decay are prevalent in crime distribution. An offender tends to commit more crime near home than in distant places. More crimes tend to occur near crime attractors such as bars or crime generators such as bus terminals than distance places. These patterns can be tested by incremental buffering analysis. Buffering can also help create proper units of analysis. For example, a buffering area within fifty meters of a bus stop may be used as the unit of analysis to assess the impact of the bus stop on crime. Overlaying is also often used in crime analysis. Quite often the unit of crime analysis does not coincide with the census unit. To obtain census attribute values of the analysis unit, an overlay is performed on these two sets of units. A portion of the attribute values of the census unit is assigned to the analysis unit based on the proportion of the overlapping part over the entire census unit. This process is called the proportional assignment. While this overlay process is often performed in the vector domain, some overlay tasks are more suited in the raster domain. Risk terrain modeling, for example, adds the risk value of individual risk factors to create an overall risk map.

  • Chang, Kang-Tsung. Introduction to Geographic Information Systems. 9th ed. New York: McGraw Hill Education, 2019.

    Save Citation »Export Citation » Share Citation »

    This is a popular GIS textbook. It covers all the commonly used spatial analytical functions. The labs and exercises are closely tied to ArcGIS, the most popular GIS software.

    Find this resource:

  • Santos, Rachel Boba. Crime Analysis with Crime Mapping. 4th ed. Los Angeles: SAGE, 2017.

    Save Citation »Export Citation » Share Citation »

    This book covers fundamentals of crime analysis, and various crime analysis techniques for tactical, strategic, and administrative purposes.

    Find this resource:

Elements of Crime Analysis

Crime analysis normally involves the following elements: uncovering spatio-temporal patterns of crime distribution; explaining crime patterns by contributing factors; predicting future crime patterns; developing crime prevention approaches; and evaluating the effectiveness of crime prevention.

Spatial and Temporal Patterns

It is well known that crime concentrates in small geographic areas called hotspots, as is underscored by Ackerman and Murray 2004; Johnson and Bowers 2004; Eck, et al. 2005 and Eck, et al. 2007; Hipp 2016; Steenbeek and Weisburd 2016; and Andresen, et al. 2017. Sherman, et al. 1989 found that 50 percent of the calls for service are concentrated in 3 percent of the places. Weisburd 2015 and Bernasco and Steenbeek 2017 also confirmed this law of crime concentration in space and time. Lee, et al. 2017 reviewed forty-four empirical studies and concluded that crime is indeed concentrated at a few places, and that calls for services appear more concentrated than crime. While commonality in the spatial distribution of crime exists across nations, as the principles of environmental criminology theories are generally applicable virtually everywhere, there exist notable differences among nations as well. For example, areas surrounding high schools tend to attract more crime in America, but there is less crime in China due to strict policies on management and surveillance.

  • Ackerman, W. V., and A. T. Murray. “Assessing Spatial Patterns of Crime in Lima, Ohio.” Cities 21.5 (2004): 423–437.

    DOI: 10.1016/j.cities.2004.07.008Save Citation »Export Citation » Share Citation »

    This paper presents an analytical and theoretical framework to study violent and property crime at the macro, meso, and micro levels in Lima, Ohio. Both qualitative and quantitative techniques are employed. The authors identify problem neighborhoods, problem areas within identified neighborhoods, and hot spots within those problem areas.

    Find this resource:

  • Andresen, M. A., S. J. Linning, and N. Malleson. “Crime at Places and Spatial Concentrations: Exploring the Spatial Stability of Property Crime in Vancouver BC, 2003–2013.” Journal of Quantitative Criminology 33.2 (2017): 255–275.

    DOI: 10.1007/s10940-016-9295-8Save Citation »Export Citation » Share Citation »

    This paper presents that property crime in Vancouver is highly concentrated in a small percentage of street segments and intersections. The spatial point pattern test shows that spatial stability is almost always present in the consecutive ten years when taking all street segments and intersections into consideration.

    Find this resource:

  • Bernasco, W., and W. Steenbeek. “More Places than Crimes: Implications for Evaluating the Law of Crime Concentration at Place.” Journal of Quantitative Criminology 33.3 (2017): 451–467.

    DOI: 10.1007/s10940-016-9324-7Save Citation »Export Citation » Share Citation »

    This paper proposes generalizations of the Lorenz curve and the Gini coefficient for correcting bias when crime data are sparse. By examining two types of crime in the city of The Hague, the results show that the generalizations can better represent crime concentration and improve comparisons of crime concentration in terms of sparse crime data.

    Find this resource:

  • Eck, J., S. Chainey, J. Cameron, and R. Wilson. Mapping Crime: Understanding Hotspots. National Institute of Justice, NCJ Number 209393, 2005.

    Save Citation »Export Citation » Share Citation »

    This report discusses hot spot analysis techniques and software and identifies when to use each one. The authors argue that mapping a crime pattern should be consistent with the type of hot spot and possible police action. For example, when hot spots are at specific addresses, a dot map is more appropriate than an area map, which would be too imprecise.

    Find this resource:

  • Eck, J. E., R. V. Clarke, and R. T. Guerette. “Risky Facilities: Crime Concentration in Homogeneous Sets of Establishments and Facilities.” Crime Prevention Studies 21 (2007):225.

    Save Citation »Export Citation » Share Citation »

    This paper presents that the distribution of crime across a population of similar facilities follows a J curve: a few of the facilities account for most of the crime in these facilities. The J curves of crime are an emergent macro property of the interaction of individual decisions. We should focus on sets of facilities rather than each high-crime facility.

    Find this resource:

  • Hipp, J. R. “General Theory of Spatial Crime Patterns.” Criminology 54.4 (2016): 653–679.

    DOI: 10.1111/1745-9125.12117Save Citation »Export Citation » Share Citation »

    This paper proposes a general theory for examining the spatial distribution of crime. It addresses the spatial distribution of the residences of offenders, targets, guardians, and their respective expected movement patterns across space and time. Further, it creates estimates of potential crime at various locations at various points in time.

    Find this resource:

  • Johnson, S. D., and K. J. Bowers. “The Burglary as Clue to the Future: The Beginnings of Prospective Hot-Spotting.” European Journal of Criminology 1.2 (2004): 237–255.

    DOI: 10.1177/1477370804041252Save Citation »Export Citation » Share Citation »

    This paper uses statistical techniques developed to study the transmission of disease to analyze burglary data for the county of Merseyside. It presents that residential burglary clusters in space and time, and that a residential burglary flags the elevated risk of further residential burglaries in the near future (one to two months) and in close proximity (up to three hundred to four hundred meters) to the victimized home. Preventive action should not be restricted to the burgled home.

    Find this resource:

  • Lee, YongJei, John E. Eck, SooHyun O, and Natalie Martinez. “How Concentrated Is Crime at Places? A Systematic Review from 1970 to 2015.” Crime Science 6.6 (2017).

    DOI: 10.1186/s40163-017-0069-x.2Save Citation »Export Citation » Share Citation »

    This paper presents a systematic review and meta-analysis of the evidence that crime is concentrated among places, based on forty-four empirical studies. It concludes that crime is indeed concentrated at a few places, and that calls for services appear more concentrated than crime.

    Find this resource:

  • Sherman, L. W., P. R. Gartin, and M. E. Buerger. “Hot Spots of Predatory Crime: Routine Activities and the Criminology of Place.” Criminology 27.1 (1989): 27–56.

    DOI: 10.1111/j.1745-9125.1989.tb00862.xSave Citation »Export Citation » Share Citation »

    This paper presents results with spatial data on 323,979 calls to police over all 115,000 addresses and intersections in Minneapolis over one year. It found that relatively few “hot spots” produce most calls to Police (50 percent of calls in 3 percent of places) and calls reporting predatory crimes (all robberies at 2.2 percent of places, all rapes at 1.2 percent of places, and all auto thefts at 2.7 percent of places).

    Find this resource:

  • Steenbeek, W., and D. Weisburd. “Where the Action Is in crime? An Examination of Variability of Crime across Different Spatial Units in The Hague, 2001–2009.” Journal of Quantitative Criminology 32.3 (2016): 449–469.

    DOI: 10.1007/s10940-015-9276-3Save Citation »Export Citation » Share Citation »

    This paper presents that about 58–69 percent of the variability of crime in the city of The Hague from 2001 to 2009 can be attributed to street segments (micro level), with most of the remaining variability in districts (macro level). Micro geographic units are key to understanding the crime problem. Neighborhood does not tell a lot.

    Find this resource:

  • Weisburd, D. “The Law of Crime Concentration and the Criminology of Place.” Criminology 53.2 (2015): 133–157.

    DOI: 10.1111/1745-9125.12070Save Citation »Export Citation » Share Citation »

    This paper presents strong support for a law of crime concentration. By providing the first cross-city comparison of crime concentration using a common geographic unit, the same crime type, and examining a general crime measure, it is shown that crime concentration stays within a narrow bandwidth across time, despite strong volatility in crime incidents.

    Find this resource:

Pattern Detection

Crime patterns can be detected globally or locally. Lee and Eck 2019 summarized the benefits and limitations of four global measures of concentration, including Gini, Simpson, Shannon, and Decile indices. Other global clustering detection approaches include the nearest neighbor analysis such as k-nearest neighbors algorithm (KNN) and Ripley’s K and L functions, as depicted by Rogerson and Sun 2001. Density mapping and spatial autocorrelation analysis such as Moran’s I and LISA (local indicators of spatial association) plot by Anselin, et al. 2000 are popular local measures of concentration. Nath 2006 applied data mining approaches for crime patterns. Most of the clustering methods are area-based, an example of which is Grubesic 2006. For street crimes, line-based clustering is more appropriate, to ensure clusters conforming to streets such that none of the no crime areas are identified as part of a hotspot. The temporal patterns of crimes can be assessed across multiple years, across seasons, during holidays, between weekdays and weekends, between day and night, as is documented in Haberman, et al. 2017. Each crime type may have its unique temporal patterns. Theft, associated with the concentration of people, tends to occur during the day and early evening. Robbery tends to concentrate in the evening. Aggravated assault tends to occur more in the late evening. Burglary can happen in both day and night, although the exact time of the offense is hard to confirm. Due to the change in routine activities, the temporal crime patterns are usually different between weekdays and weekends. Summer tends to have more crimes than winter. Holidays can greatly impact crime. Crimes such as burglary and robbery tend to increase during Thanksgiving and Christmas. However, crimes drop significantly in China during the Chinese New Year, when people return to their hometown for two to four weeks. The difference in culture seems to play an important role. Nelson, et al. 2001 and Rey, et al. 2012 attempted to combine spatial and temporal patterns of crime. One of the popular spatio-temporal approaches is the hotspot matrix by Ratcliffe 2004, which combines three types of spatial distributions and three types of twenty-four-hour distributions, resulting in a total of nine possible spatio-temporal patterns. Temporal patterns at different scales such as weekly and seasonal can also be combined with spatial patterns to create spatio-temporal patterns. Since crimes cluster in space and time, the phenomena of near-repeat in space and time have been extensively investigated by scholars, including Townsley, et al. 2003 and Ratcliffe and Rengert 2008. Many studies have revealed that one crime incident may lead to another crime within two hundred meters in less than two weeks. Slightly over 30 percent of crimes fall to the near-repeat pairs. Literature has suggested that gang (group) related crimes have a higher tendency of near repeat. Such findings may provide insight into crime prevention.

  • Anselin, L., J. Cohen, D. Cook, W. Gorr, and G. Tita. “Spatial Analyses of Crime.” Criminal Justice 4.2 (2000): 213–262.

    Save Citation »Export Citation » Share Citation »

    This paper presents some practical and accessible methods of exploratory data analysis to explore the relationship between place and crime. Those methods include hot spot modeling and analysis, exploratory spatial data analysis (point pattern analysis, distance statistics, areal analysis, etc.), LISA statistics, spatial autocorrelation, and spatial modeling.

    Find this resource:

  • Grubesic, T. H. “On the Application of Fuzzy Clustering for Crime Hot Spot Detection.” Journal of Quantitative Criminology 22.1 (2006): 77.

    DOI: 10.1007/s10940-005-9003-6Save Citation »Export Citation » Share Citation »

    This paper introduces a generalized partitioning method known as fuzzy clustering for hot-spot detection. It makes functional and visual comparisons of fuzzy clustering and two hard-clustering approaches (medoid and k-means). Using 613 crime events from the Pleasant Ridge neighborhood of Cincinnati, Ohio, in 2003, the results suggest that a fuzzy clustering approach is better to handle intermediate cases and spatial outliers.

    Find this resource:

  • Haberman, C. P., E. T. Sorg, and J. H. Ratcliffe. “Assessing the Validity of the Law of Crime Concentration Across Different Temporal Scales.” Journal of Quantitative Criminology 33 (2017): 47–567.

    DOI: 10.1007/s10940-016-9327-4Save Citation »Export Citation » Share Citation »

    The paper summarizes the cumulative percentages of Philadelphia, PA USA street blocks and intersections experiencing 25 and 50 percent of street robberies by hour of the day, days of the week, and seasons of the year.

    Find this resource:

  • Lee, YongJei, and John E. Eck. “Comparing Measures of the Concentration of Crime at Places.” Crime Prevention and Community Safety 21.4 (2019): 269–294.

    DOI: 10.1057/s41300-019-00078-2Save Citation »Export Citation » Share Citation »

    This paper compares four concentration measures, including Gini, Simpson, Shannon, and Decile indices. It summarizes the benefits and limitations of each measure and the circumstances for which each is most useful.

    Find this resource:

  • Nath, S. V. “Crime Pattern Detection Using Data Mining.” Paper read at 2006 IEEE/WIC/ACM International Conference on Web Intelligence and Intelligent Agent Technology Workshops. 2006.

    Save Citation »Export Citation » Share Citation »

    This paper introduces a data mining framework to detect crime patterns. The author enhances k-means clustering, uses a semi-supervised learning technique, and develops a weighting scheme for attributes. These techniques are applied to real crime data from a sheriff’s office and results are validated.

    Find this resource:

  • Nelson, A. L., R. D. F. Bromley, and C. J. Thomas. “Identifying Micro-spatial and Temporal Patterns of Violent Crime and Disorder in the British City Center.” Applied Geography 21.3 (2001): 249–274.

    DOI: 10.1016/S0143-6228(01)00008-XSave Citation »Export Citation » Share Citation »

    This paper presents a combination of different micro-spatial information in identifying patterns of violence in Cardiff and Worcester. Evidence shows primary clusters at night in the pub/club leisure zones, secondary clusters during the shopping day in major retail streets, subsidiary afternoon clusters near licensed premises, and a late-night confluence flashpoint at a node of pedestrian activity.

    Find this resource:

  • Ratcliffe, J. H. “The Hotspot Matrix: A Framework for the Spatio‐temporal Targeting of Crime Reduction.” Police Practice and Research 5.1 (2004):5–23.

    DOI: 10.1080/1561426042000191305Save Citation »Export Citation » Share Citation »

    This paper presents techniques for identifying the spatial and temporal components of crime hotspots. Three broad categories of temporal hotspot (diffused, focused, acute) and three broad categories of spatial hotspot (dispersed, clustered, and hotpoint) are described in the form of a Hotspot Matrix. The hotspot matrix could be used to determine an appropriate prevention or detection.

    Find this resource:

  • Ratcliffe, J. H., and G. F. Rengert. “Near-Repeat Patterns in Philadelphia Shootings.” Security Journal 21. 1 (2008):58–76.

    DOI: 10.1057/ Citation »Export Citation » Share Citation »

    This paper explores near-repeat patterns of shooting in Philadelphia. Using new tools developed to quantify the spatio-temporal patterns of near-repeats, it is found that there are elevated patterns of near-repeat shootings within two weeks and one city block of previous incidents. The elevated risk of a shooting during this period is found to be 33 percent greater than expected. Possible reasons may include coercion, retaliation, and escalation.

    Find this resource:

  • Rey, S. J., E. A. Mack, and J. Koschinsky. “Exploratory Space–Time Analysis of Burglary Patterns.” Journal of Quantitative Criminology 28.3 (2012): 509–531.

    DOI: 10.1007/s10940-011-9151-9Save Citation »Export Citation » Share Citation »

    This paper introduces two new methods, a conditional spatial Markov chain and its extension, for the analysis of the spatiotemporal dynamics of residential burglary patterns in Mesa, Arizona, from 2005 to 2009. Spatial clustering of burglary activity is present in each year and has an important influence on both the conditional and joint evolution of burglary activity across space and time.

    Find this resource:

  • Rogerson, P., and Y. Sun. “Spatial Monitoring of Geographic Patterns: An Application to Crime Analysis.” Computers, Environment and Urban Systems 25.6 (2001): 539–556.

    DOI: 10.1016/S0198-9715(00)00030-2Save Citation »Export Citation » Share Citation »

    This paper combines the nearest neighbor statistic and cumulative sum methods for detecting changes in spatial patterns of point events over time. Using 1996 arson data from Buffalo, the first two hundred arsons were used as a base pattern, the remainder was used for surveillance of the pattern. It is found that arsons occurring later in the year were more likely to cluster in space.

    Find this resource:

  • Townsley, M., R. Homeland, and J. Chaseling. “Infectious Burglaries: A Test of the Near Repeat Hypothesis.” British Journal of Criminology 43.3 (2003): 615–633.

    DOI: 10.1093/bjc/azg615Save Citation »Export Citation » Share Citation »

    This paper adopts epidemiological methods for the study of infectious diseases to investigate the phenomenon of near repeat victimization of burglaries. Proximity to a burgled dwelling increases burglary risk during the period immediately following the burglary.

    Find this resource:

Explanatory Models

Once patterns of crime in space and time are revealed, the next step is to explore what factors may help explain these patterns. This task is typically accomplished via multivariate regression analysis. There are a variety of models for crime analysis, including negative binomial models by Bernasco and Block 2011 and Nobles, et al. 2016, discrete choice models by Bernasco and Nieuwbeerta 2004, Song, et al. 2019, and Vandeviver and Bernasco. 2020, multilevel models by Sampson, et al. 1997, hierarchical linear models by Davies and Johnson 2015, and time constrained regression models by Ratcliffe 2006. Any discovered relationship is likely to be associative, not causal. In most cases, a crime incident is linked to a precise location where the incident occurred or reported and to a time when the incident occurred. Therefore, crime incident data have an extremely high spatio-temporal resolution. Explanation of individual crime incidents is virtually impossible because almost all explanatory variables have a much lower spatio-temporal resolution. Therefore, explanatory models are implemented at a reduced resolution in space and time, aiming to explain the overall crime pattern rather than individual crime incidents. Crime incidents are typically aggregated to areal units or street segments in explanatory models. If the interest is solely on the presence or absence of crime, the dependent variable is coded as 1 or 0, with 1 indicating the presence and 0 absence. Logistic regression then relates explanatory variables to crime, an example of which is Kennedy and Forde 1990. If, however, the interest is to explain the quantitative variation of crime in space, non-logistic regression models are applied, an example of which is Roncek and Maier 1991. The dependent variable is typically a crime count or crime rate calculated as a ratio of count over the population. Crime density, measured as a ratio of count over the area, is rarely used as a dependent variable. Like all count data, crime counts follow a Poisson distribution, or negative binomial distribution when an over-dispersion exists. Therefore, Poisson regression and negative binomial regression are popular choices for crime modeling. If the dependent variable has too many zeroes, zero-inflated models may be more appropriate.

  • Bernasco, W., and R. Block. “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 48.1 (2011): 33–57.

    DOI: 10.1177/0022427810384135Save Citation »Export Citation » Share Citation »

    This paper examines the effects of crime generators, crime attractors, and offender anchor points on the distribution of street robberies across the nearly twenty-five thousand census blocks of Chicago. The results of negative binomial regression models show that blocks with elevated numbers of robbery also radiate their elevated crime risk to adjacent blocks. Ecological disadvantage includes the presence of small legal commercial business, illegal markets, accessibility facilitators, and offender anchor points.

    Find this resource:

  • Bernasco, W., and P. Nieuwbeerta. “How Do Residential Burglars Select Target Areas?: A New Approach to the Analysis of Criminal Location Choice.” The British Journal of Criminology 45.3 (2004): 296–315.

    DOI: 10.1093/bjc/azh070Save Citation »Export Citation » Share Citation »

    This paper introduces the discrete spatial choice approach to the study of criminal target choice. Using data on 548 residential burglaries, committed by 290 burglars from the city of The Hague, the results show that the likelihood of a neighborhood’s being selected for burglary is increased by its ethnic heterogeneity, its percentage of single-family dwellings, and its proximity to where the offender lives.

    Find this resource:

  • Davies, T., and S. D. Johnson. “Examining the Relationship between Road Structure and Burglary Risk via Quantitative Network Analysis.” Journal of Quantitative Criminology 31.3 (2015): 481–507.

    DOI: 10.1007/s10940-014-9235-4Save Citation »Export Citation » Share Citation »

    This paper examines the effects of the configuration of the street network on the spatial distribution of residential burglary at the street segment level in Birmingham. The results of a hierarchical linear model show that betweenness was a highly significant predictor of the burglary victimization count. More linear streets were generally found to be at lower risk of victimization. More potential usage is likely to incur a higher risk of burglary.

    Find this resource:

  • Kennedy, L. W., and D. R. Forde. “Routine Activities and Crime: An Analysis of Victimization in Canada.” Criminology 28.1 (1990): 137–152.

    DOI: 10.1111/j.1745-9125.1990.tb01321.xSave Citation »Export Citation » Share Citation »

    This paper presents evidence that strong interaction effects between demographic characteristics of victims and certain routine activities also exist for violent crime. Using data from the Canadian Urban Victimization Survey, the results of logistic regression models show that personal crime is contingent on the exposure that comes from following certain lifestyles. This is particularly true for certain demographic groups, particularly young males.

    Find this resource:

  • Nobles, M. R., J. T. Ward, and R. Tillyer. “The Impact of Neighborhood Context on Spatiotemporal Patterns of Burglary.” Journal of Research in Crime and Delinquency 53.5 (2016): 711–774.

    DOI: 10.1177/0022427816647991Save Citation »Export Citation » Share Citation »

    This paper examines the elative contribution of social, structural, and environmental design covariates on single and repeat/near repeat burglary counts. The results of spatially lagged negative binomial regression models show that positive and consistent association between concentrated disadvantage and racial heterogeneity and all types of burglaries was evident, while the effects for other indicators varied across classifications of single and repeat/near repeat burglaries.

    Find this resource:

  • Ratcliffe, J. H. “A Temporal Constraint Theory to Explain Opportunity-Based Spatial Offending Patterns.” Journal of Research in Crime and Delinquency 43.3 (2006): 261–291.

    DOI: 10.1177/0022427806286566Save Citation »Export Citation » Share Citation »

    This paper examines criminological implications of temporal constraints to explain some key concepts from environmental criminology after summarizing Miller’s time measurement theory. It is hypothesized that the temporal constraints of daily life are the main cause of unfamiliarity with areas beyond the offender’s immediate least-distance path. Therefore, temporal constraints are a major determinant in spatiotemporal patterns of property crime.

    Find this resource:

  • Roncek, D. W., and P. A. Maier. “Bars, Blocks, and Crimes Revisited: Linking the Theory of Routine Activities to the Empiricism of “Hot Spots”.” Criminology 29.4 (1991): 725–753.

    DOI: 10.1111/j.1745-9125.1991.tb01086.xSave Citation »Export Citation » Share Citation »

    This paper examines the effects of recreational liquor establishments (i.e., taverns and cocktail lounges) on crime from 1979 to 1981 on Cleveland’s residential city blocks. Multivariate regression results show that the number of such businesses has positive and statistically significant effects on the amount of crime. However, the effects on crime are compounded when the businesses are located in areas with more anonymity and lower guardianship.

    Find this resource:

  • Sampson, R. J., S. W. Raudenbush, and F. Earls. “Neighborhoods and Violent Crime: A Multilevel Study of Collective Efficacy.” Science 277.5328 (1997): 918.

    DOI: 10.1126/science.277.5328.918Save Citation »Export Citation » Share Citation »

    This paper examines the effects of collective efficacy on violence. Using a 1995 survey of 8782 residents of 343 neighborhoods in Chicago, multilevel analyses show that collective efficacy yields a high between-neighborhood reliability and is negatively associated with variations in violence. Collective efficacy largely mediates associations of concentrated disadvantage and residential instability with violence.

    Find this resource:

  • Song, G., W. Bernasco, L. Liu, L. Xiao, S. Zhou, and W. Liao. “Crime Feeds on Legal Activities: Daily Mobility Flows Help to Explain Thieves’ Target Location Choices.” Journal of Quantitative Criminology 35.4 (2019): 831–854.

    DOI: 10.1007/s10940-019-09406-zSave Citation »Export Citation » Share Citation »

    This paper examines whether daily mobility flows of the urban population help predict where individual thieves commit crimes by using geocoded tracks of mobile phones in a large city in China. The results from discrete choice models show that the stronger a community is connected by population flows to where the offender lives, the larger its probability of being targeted. However, mobility measure does not replace the role of distance.

    Find this resource:

  • Vandeviver, C., and W. Bernasco. “‘Location, Location, Location’: Effects of Neighborhood and House Attributes on Burglars’ Target Selection.” Journal of Quantitative Criminology 36 (2020): 779–821.

    DOI: 10.1007/s10940-019-09431-ySave Citation »Export Citation » Share Citation »

    This paper examines the simultaneous effect of neighborhood-level and residence-level attributes on residential burglars’ target choice. The results from a discrete spatial choice approach show that burglars prefer burglarizing residences in neighborhoods with lower residential density, they previously and recently targeted for burglary, and residences nearby their home. Burglars also favor burglarizing detached residences, residences in single-unit buildings, and renter-occupied residences.

    Find this resource:

Explanatory Variables and Crime Theories

The key to the success of crime models is the selection of explanatory or independent variables. Ideally, the selection must be guided by theories and literature. Several environmental criminology theories are important for crime geography and spatial analysis of crime. Brantingham and Brantingham 1981, Andresen 2020, and Wortley and Townsley 2017 are three representative books on environmental criminology. Routine activity theory (Cohen and Felson 1979) stipulates that crime occurs when a motivated offender meets a desirable target in the absence of capable guardianship. Therefore, variables of the routine activities of offenders, victims, and guardians are critical. Such data are hard to obtain in the past. Recent advances in cell phone data, GPS tracking data, and geotagged social media data have opened enormous opportunities for accurately modeling the routine activities. Social disorganization theory (Sampson and Groves 1989) considers three central elements in poverty or concentrated disadvantage, residential mobility or instability, and racial or ethnic heterogeneity. The exact variables used to represent these elements may vary across the countries. For example, while many Western countries are inhabited by multiple racial and ethnic groups, East Asian countries such as China, Japan, and Korea have one dominant ethnic group. The percent of rural to urban migrant workers over the total population is typically used as a substitute for ethnic heterogeneity in crime research in China. The rational choice theory by Cornish and Clarke 1987 assumes that offenders behave rationally, by maximizing reward, minimizing cost, and minimizing the risk of getting caught. For example, the risk for burglars can be measured by the degree of entry control in gated communities, the risk for robbers can be measured by the accessibility of escape routes out of convenience stores. Crime pattern theory by Brantingham and Brantingham 1993 explains the spatial distribution by assessing the potential overlap of activity spaces between offenders and victims. These activity spaces may be nodes or paths linking the nodes. Nodes can be further categorized into crime attractors and crime generators. Crime attractors are areas that attract potential offenders while crime generators are areas that attract potential victims. Land use data and points of interest (POIs) are often used to represent crime attractors and crime generators. Crime generators and attractors may vary across the countries. For example, bars typically attract crimes in Western countries, but not necessarily so in China. Bus stops typically generate crimes in Western cities, but not necessarily so in Chinese cities. Possible explanations are that bars are often located in upscale areas and bus stops are ubiquitous in Chinese cities, while bars tend to concentrate in the entertainment district and bus routes mainly serve people in need in Western cities.

  • Andresen, M. A. Environmental Criminology: Evolution, Theory, and Practice. 2d ed. New York: Routledge, 2020.

    Save Citation »Export Citation » Share Citation »

    This book presents a systematic treatment of environmental criminology, including its evolution, theory, and practice. Virtually all popular theories are covered, including social disorganization theory, routine activity theory, rational choice theory, crime pattern theory, and crime prevention theory.

    Find this resource:

  • Brantingham, P. L., and P. J. Brantingham. “Environment, Routine, and Situation: Toward a Pattern Theory of Crime.” In Routine Activity and Rational Choice: Advances in Criminological Theory. Vol. 5. Edited by Ronald V. Clarke and Marcus Felson, 259–294. NCJ-159998. New Brunswick, NJ: Transaction, 1993

    Save Citation »Export Citation » Share Citation »

    The crime pattern theory holds that crimes do not occur randomly or uniformly in time and space, but rather that they have distinct patterns, resulting from complex interactions.

    Find this resource:

  • Brantingham, P. J., and P. L. Brantingham, eds. Environmental Criminology. Beverley Hills, CA: SAGE, 1981.

    Save Citation »Export Citation » Share Citation »

    This is typically regarded as the earliest book on environmental criminology.

    Find this resource:

  • Cohen, L. E., and M. Felson. “Social Change and Crime Rate Trends: A Routine Activity Approach.” American Sociological Review 44.4 (1979): 588–608.

    DOI: 10.2307/2094589Save Citation »Export Citation » Share Citation »

    This is the most cited work on routine activity theory.

    Find this resource:

  • Cornish, D., and R. Clarke. “Understanding Crime Displacement: An Application of Rational Choice Theory.” Criminology 25.4 (1987): 933–947.

    DOI: 10.1111/j.1745-9125.1987.tb00826.xSave Citation »Export Citation » Share Citation »

    This is the most cited work on rational choice theory in the context of criminology.

    Find this resource:

  • Sampson, R. J., and W. B. Groves. “Community Structure and Crime: Testing Social-Disorganization Theory.” American Journal of Sociology 94.4 (1989): 774–802.

    DOI: 10.1086/229068Save Citation »Export Citation » Share Citation »

    This is the most cited work on social-disorganization theory.

    Find this resource:

  • Wortley, R., and M. Townsley, eds. Environmental Criminology. 2d ed. New York: Routledge, 2017.

    Save Citation »Export Citation » Share Citation »

    This edited volume includes chapters contributed from some of the leading scholars in environmental criminology. It is organized in three sections on crime event, crime pattern, and crime prevention.

    Find this resource:

Crime Types and Units of Analysis

Not all crimes have the same mechanism. The variables that explain one type of crime may not be effective in explaining another. Liu, et al. 2020 showed that satellite nightlight helps explain street robbery but not residential burglary. Lan, et al. 2020 demonstrated that the opening of new casino had an impact on property crime nearby, but not on violent crime. Any attempt of using a single model to explain multiple crime types is unlikely to be successful. Further, not all crime types can be explained with environmental criminology theories. For example, telemarketing scams, credit fraud, and Internet fraud are very different from theft from persons, street robbery, assault, and burglary. There exists empirical evidence that surveillance cameras may deter street crimes, but they have virtually no effect on scams. Another important consideration in crime modeling is the unit of analysis. Crime analysis is not immune to the modifiable area unit problem—the results of modeling is affected by the different aggregations of the data. Brantingham, et al. 1976 highlighted the importance of spatial resolution on crime mapping. While the unit of analysis is not necessarily tied to the resolution of the data, most studies tend to dictate the unit of analysis by the resolution of the data. Because the Census data are provided at the resolutions of tracts, block groups, and blocks, many studies use Census tracts, block groups, and blocks as the units of analysis. Ideally, the choice of the unit of analysis should be dictated by spatial distribution of the crime, the most logical units of analysis should be individual addresses, street segments, clusters of segments, and neighborhoods. Lately, more studies have adopted the multilevel approach that combines variables from different areal units, to assess the interaction between them.

  • Brantingham, P. J., D. A. Dyreson, and P. L. Brantingham. “Crime Seen Through a Cone of Resolution.” American Behavioral Scientist 20.2 (1976): 261–273.

    DOI: 10.1177/000276427602000207Save Citation »Export Citation » Share Citation »

    This paper examines the impact of resolutions on mapping crime. The scale ranges from the national level, to the state level, to the Census block group level. With the increasing resolution, more details on the spatial distribution of crime are revealed.

    Find this resource:

  • Lan, M., L. Liu, and J. E. Eck. “A Spatial Analytical Approach to Assess the Impact of a Casino on Crime: An Example of JACK Casino in downtown Cincinnati.” Cities (2020).

    DOI: 10.1016/j.cities.2020.103003[class:journalArticleSave Citation »Export Citation » Share Citation »

    This paper uses the weighted displacement quotient and a series of negative binomial models to assess the change of crime patterns before and after the opening of the JACK casino. Results show that the casino had differing effects on property and violent crime. While property crime increased near the casino, violent crime did not change much.

    Find this resource:

  • Liu, L., H. Zhou, M. Lan, and Z. Wang. “Linking Luojia 1-01 Nightlight Imagery to Urban Crime.” Applied Geography 125 (2020): 102267.

    DOI: 10.1016/j.apgeog.2020.102267Save Citation »Export Citation » Share Citation »

    This paper uses the composite edge derived from the Luojia 1–01 satellite nightlight imageries to explain robbery and burglary at the Census block level. Results show that the composite edge has a significant effect on robbery, but not on burglary.

    Find this resource:

Model Fitness and Multicollinearity

The fitness of a model may be measured by R2, pseudo R2, AIC (Akaike information criterion), or BIC (Bayesian information criterion). A larger R2 value indicates a better fit, while a smaller AIC or BIC value indicates a better fit. Whether a variable plays a statistically significant role in explaining crime is dictated by the p-value of its coefficient. A p-value less then 0.05 is said to be significant. A positive coefficient contributes to crime, while a negative coefficient deters crime. Non-standardized coefficients in a model cannot be compared directly. Once standardized, a variable with a larger standardized coefficient is said to have a greater influence on crime. The test of multicollinearity among explanatory variables is also important in multivariate regression. The variance inflation factor (VIF), the quotient of the variance in a model with multiple variables by the variance of a model with one variable alone, is typically calculated for this purpose. While there is no universally accepted standard, most agree that a VIF value above 4 indicates collinearity being too high for the model to be reliable. An overwhelming majority of crime models would not have a VIF higher than 4. A more stringent requirement is that a VIF should be no more than 2.5.

Prediction Models

Police departments are interested in finding out what would happen in the future, such as the next day, the next week, or the next month. Risk terrain models are usually not time-sensitive, despite some equating risk terrain models as predictive models. True predictive models must have an explicit time component. Chandra, et al. 2008 used a time series approach for crime prediction. Short, et al. 2010 emphasized that the underlying assumption of predictive models is that past trends are likely to be repeated in the future. Spatial and temporal autocorrelations are the theoretical foundation in Mohler 2014. The patterns of distance decay and temporal decay are well accepted in environmental criminology and crime geography. Near (in space) and recent (in time) incidents are weighted higher in the models. Virtually all crime prediction models are raster-based.

Density Mapping and Machine Learning Algorithms

Liu and Brown 2003 and Gerber 2014 used crime density maps for prediction. Density maps that consider spatial and temporal autocorrelation tend to perform better than simple density maps that treat all previous crime incidents equally. Spatio-temporal interpolation methods such as Kriging and co-Kriging have also been used to predict future crime risks by Yang, et al. 2020. Polynomial models calibrated by the location and time of the past crime incidents can also be used for predictions. Various machine learning algorithms have been applied to crime prediction by Castelli, et al. 2017. These algorithms include artificial neural networks (ANN) by Lin, et al. 2018, support vector machine (SVM) by Lin, et al. 2018, random forest by Alves, et al. 2018, Long short-term memory (LSTM) by Zhang, et al. 2020, Deep Learning by Kang and Kang 2017, etc. Variables representing spatial and temporal autocorrelation can be explicit input to the model, or the algorithms are capable of learning various degrees of autocorrelation. Variables for crime prediction include data about historical crime, built environment, social environment, seasonal variation, weather, etc. Historical crime is the most important input. The addition of the other variables may improve prediction accuracy, as is confirmed by Yang, et al. 2020 and Zhang, et al. 2020. The specification of the machine learning models is beyond the scope of this paper. In general, models that are more capable of depicting various forms of spatio-temporal autocorrelation tend to perform better. Machine learning models are normally considered black-box models. While effective in prediction, they may not be used for explanation. Recently efforts have been made to add explanatory power to machine learning models. One such example is the calculation of variable importance in random forest models. A variable with a higher value of variable importance makes a larger contribution to the model.

  • Alves, L. G. A., H. V. Ribeiro, and F. A. Rodrigues. “Crime Prediction through Urban Metrics and Statistical Learning.” Physica A: Statistical Mechanics and Its Applications 505 (2018): 435–443.

    DOI: 10.1016/j.physa.2018.03.084Save Citation »Export Citation » Share Citation »

    This paper uses a random forest regressor to predict crime and quantify the influence of urban indicators on homicides. A grid-search algorithm with the stratified k-fold cross-validation (with k = 10) is used to find the best combination of trees and the maximum depth of the trees. This approach can have up to 97 percent of accuracy on crime prediction. Among urban indicators, unemployment and illiteracy are the most important variables for describing homicides in Brazilian cities.

    Find this resource:

  • Castelli, M., R. Sormani, L. Trujillo, and A. Popovič. “Predicting per Capita Violent Crimes in Urban Areas: An Artificial Intelligence Approach.” Journal of Ambient Intelligence and Humanized Computing 8.1 (2017): 29–36.

    DOI: 10.1007/s12652-015-0334-3Save Citation »Export Citation » Share Citation »

    This paper proposes an artificial intelligence system for predicting per capita violent crimes in urban areas starting from socioeconomic data, law-enforcement data and other crime-related data. The proposed framework incorporates genetic programming that uses the concept of semantics during the search process with a local search method. The results show that the proposed method produces a lower error with respect to the existing techniques. It is particularly suitable for analyzing large amounts of data.

    Find this resource:

  • Chainey, S., L. Tompson, and S. Uhlig. “The Utility of Hotspot Mapping for Predicting Spatial Patterns of Crime.” Security Journal 21 (2008): 4–28.

    DOI: 10.1057/ Citation »Export Citation » Share Citation »

    This paper uses history crime data for a period to generate hotspot maps and test their accuracy for predicting where crimes will occur next. Hotspot mapping accuracy is compared with other mapping techniques (thematic mapping of Census Output Areas, spatial ellipses, grid thematic mapping, and kernel density estimation) and by crime type. The results show that kernel density estimation consistently outperformed the other techniques, while street crime hotspot maps were consistently better at predicting where future street crime would occur.

    Find this resource:

  • Chandra, B., M. Gupta, and M. P. Gupta. “A Multivariate Time Series Clustering Approach for Crime Trends Prediction.” Paper read at 2008 IEEE International Conference on Systems, Man and Cybernetics, 12–15 October 2008.

    Save Citation »Export Citation » Share Citation »

    This paper introduces a novel approach based on dynamic time wrapping for crime prediction. A parametric Minkowski model is proposed to find similar crime trends among various crime sequences at different locations. Subsequently, this information is used for future crime trends prediction. Analysis on Indian crime records show that the proposed technique generally outperforms the existing techniques in clustering of such multivariate time series data.

    Find this resource:

  • Gerber, M. S. “Predicting Crime Using Twitter and Kernel Density Estimation.” Decision Support Systems 61.1 (2014):115–125.

    DOI: 10.1016/j.dss.2014.02.003Save Citation »Export Citation » Share Citation »

    This paper investigates the use of spatiotemporally tagged tweets for crime prediction. Twitter-specific linguistic analysis and statistical topic modeling are used to automatically identify discussion topics in Chicago. These topics are then incorporated into a crime prediction model. The results show that, for nineteen of the twenty-five crime types, the addition of Twitter data improves crime prediction performance versus a standard approach based on kernel density estimation.

    Find this resource:

  • Kang, H. W., and H. B. Kang. “Prediction of Crime Occurrence from Multi-Modal Data Using Deep Learning.” PLOS ONE 12.4 (2017): e0176244.

    DOI: 10.1371/journal.pone.0176244Save Citation »Export Citation » Share Citation »

    This paper proposes a deep neural network (DNN) for crime prediction. Various online databases of crime statistics, demographic and meteorological data, and images in Chicago are used for analysis. DDN consists of four kinds of layers: spatial, temporal, environmental context, and joint feature representation layers. The results show that DNN model is more accurate in predicting crime occurrence and statistically analyzes data redundancy.

    Find this resource:

  • Lin, Y. L., M. F. Yen, and L. C. Yu. “Grid-Based Crime Prediction Using Geographical Features.” ISPRS International Journal of Geo-Information 7.8 (2018):298.

    DOI: 10.3390/ijgi7080298Save Citation »Export Citation » Share Citation »

    This paper incorporates the concept of a criminal environment in grid-based crime prediction modeling, and establishes a range of spatial-temporal features based on eighty-four types of geographic information by applying the Google Places API to theft data for Taoyuan City, Taiwan. The best model was found to be deep neural networks, which outperforms the popular Random Decision Forest, Support Vector Machine, and K-Near Neighbor algorithms.

    Find this resource:

  • Liu, H., and D. E. Brown. “Criminal Incident Prediction Using a Point-Pattern-Based Density Model.” International Journal of Forecasting 19.4 (2003):603–622.

    DOI: 10.1016/S0169-2070(03)00094-3Save Citation »Export Citation » Share Citation »

    This paper introduces a point-pattern-based transition density model for space–time event prediction. It relies on criminal preference discovery as observed in the features chosen for past crimes. This model is evaluated on breaking and entering burglary point data from Richmond, Virginia. The results show that it is more effective than the best of current “hot spot” methods.

    Find this resource:

  • Mohler, G. “Marked Point Process Hotspot Maps for Homicide and Gun Crime Prediction in Chicago.” International Journal of Forecasting 30.3 (2014): 491–497.

    DOI: 10.1016/j.ijforecast.2014.01.004Save Citation »Export Citation » Share Citation »

    This paper presents how point process models of crime can be extended to include leading indicator crime types. They capture both short-term and long-term patterns of risk. Several years of data and many different crime types in Chicago are systematically combined to yield accurate hotspot maps. These maps can be used for the purpose of predictive policing of gun-related crime.

    Find this resource:

  • Short, M. B., P. J. Brantingham, A. L. Bertozzi, and G. E. Tita. “Dissipation and Displacement of Hotspots in Reaction-Diffusion Models of Crime.” Proceedings of the National Academy of Sciences 107.9 (2010): 3961–3965.

    DOI: 10.1073/pnas.0910921107Save Citation »Export Citation » Share Citation »

    This paper presents a mathematical framework based on reaction-diffusion partial differential equations to study the dynamics of crime hotspots. It is shown that subcritical crime hotspots may be permanently eradicated with police suppression, whereas supercritical hotspots are displaced following a characteristic spatial pattern.

    Find this resource:

  • Yang, Bo, L. Liu, M. Lan, Z. Wang, H. Zhou, and H. Yu. “A Spatio-Temporal Method for Crime Prediction Using Historical Crime Data and Transitional Zones Identified from Nightlight Imagery.” International Journal of Geographical Information Science 34.9 (2020): 1740–1764.

    DOI: 10.1080/13658816.2020.1737701Save Citation »Export Citation » Share Citation »

    This paper presents a spatio-temporal co-kriging method for crime prediction. The results show that the model with a covariate derived from nightlight imagery outperforms the one without.

    Find this resource:

  • Zhang, X., L. Liu, L. Xiao, and J. Ji. “Comparison of Machine Learning for Predicting Crime Hotspots.” IEEE Access 8 (2020): 181302–181310.

    DOI: 10.1109/ACCESS.2020.3028420Save Citation »Export Citation » Share Citation »

    This paper compares several machine learning algorithms for crime prediction and concludes that LSTM model outperformed KNN, random forest, support vector machine, naive Bayes, and convolutional neural networks.

    Find this resource:

Accuracy Assessment

The performance of these models is affected by input variables, the model specification, and the spatio-temporal resolution of the output raster map. On the temporal side, the higher the resolution, or the shorter the time interval, the lower the accuracy. Hourly prediction is far more challenging than the weekly prediction, which in turn is more challenging than the monthly prediction, mainly due to increasing sparsities of crime incidents in the shorter time windows. On the spatial side, since crimes tend to cluster in a small number of places, the majority of cells on the map are crime-free. Cell-by-cell based correlation and mean square error measures can greatly inflate the prediction accuracy. A prediction of zero crime everywhere may still have decent accuracies by these two measures. The two most popular accuracy measures are prediction accuracy index (PAI) by Chainey, et al. 2008 and the prediction efficiency index (PEI) by Hunt 2016. PAI is the percentage of predicted crime count in the total crime count divided by the percentage of the number of predicted crime cells in the total number of cells on the map. The higher the total number of cells, the higher the PAI. Thus, PAI favors high spatial resolutions. PEI is the ratio of predicted crime count over the maximum obtainable crime count in the number of predicted crime cells. For example, if the number of predicted crime cells is six, the maximum obtainable crime count is the summation of crime counts in the six cells with the largest crime counts. PEI favors low spatial resolutions. Larger cell size leads to higher PEI. No prediction can maximize both PAI and PEI at the same time. Therefore, these two indices leave room for improvement. Zhang, et al. 2020 recently proposed two indices for prediction accuracy: the grid hit rate and the case hit rate. The grid hit rate is the ratio between the number of correctly predicted hotspot grids and the total number of actual hotspot grids. The case hit rate is the ratio between the number of cases (crime incidences) in the correctly predicted hotspot grids and the total number of cases. Still, these indices have room for improvement. An ideal index would reward correct prediction, but penalize false positive prediction (the non-crime area being predicted to have crime) and false-negative prediction (crime area being predicted to have no crime). In theory, an index of reward minus penalty makes a more realistic assessment of crime prediction accuracy.

  • Chainey, S., L. Tompson, and S. Uhlig. “The Utility of Hotspot Mapping for Predicting Spatial Patterns of Crime.” Security Journal 21 (2008): 4.

    DOI: 10.1057/ Citation »Export Citation » Share Citation »

    This paper uses history crime data for a period to generate hotspot maps and test their accuracy for predicting where crimes will occur next. It also proposed a crime prediction accuracy measure called prediction accuracy index (PAI).

    Find this resource:

  • Hunt, J. “Do Crime Hot Spots Move? Exploring the Effects of the Modifiable Areal Unit Problem and Modifiable Temporal Unit Problem on Crime Hot Spot Stability.” MA Theses. American University, 2016.

    Save Citation »Export Citation » Share Citation »

    This paper explores the effects of the modifiable areal unit problem and modifiable temporal unit problem on the stability of crime hot spots. It also proposed a crime prediction accuracy measure called prediction efficiency index (PEI).

    Find this resource:

  • Zhang, X., L. Liu, L. Xiao, and J. Ji. “Comparison of Machine Learning for Predicting Crime Hotspots.” IEEE Access 8 (2020): 181302–181310.

    DOI: 10.1109/ACCESS.2020.3028420Save Citation »Export Citation » Share Citation »

    This paper compares several machine learning algorithms for crime prediction and concludes that LSTM model outperformed KNN, random forest, support vector machine, naive Bayes, and convolutional neural networks. It also proposes two new indices for prediction accuracy: the grid hit rate and the case hit rate.

    Find this resource:

Policing and Crime Prevention

Crime prediction provides insight into where policing resources should be directed. Crime prevention is carried out in different forms of policing. The baseline is random policing, in which police officers do not target any specific places. Weisburd and Eck 2004 outlined other enhanced forms of policing including community policing, hotspot policing, and problem-oriented policing. Community policing promotes outreach to the community, to improve police-citizen relationship and collaboration. According to Sorg, et al. 2013, hotspot policing focuses resources on hotspot areas. Problem-oriented policing centers on particular problems instead of specific places. Empirical research suggests that these three enhanced approaches are often more effective than random policing. Chen, et al. 2017 developed an online cooperative police patrol routing strategy for more effective crime prevention. There have been multiple experiments with hotspot policing. Police resources are directed to the places in greater needs, often revealed by hotspot analysis and crime prediction. In addition to adjusting the routine patrol routes, Koper 1995 discovered that stopping at individual hotspots for eleven to fifteen minutes had shown most effectiveness. While the hottest spots rarely change, some of the moderate hotspots may go through cycles of presence and absence. Therefore, the patrol routes and stops should be periodically adjusted to “chase” these shifting hotspots. Verification of policing efforts is an important part of policing, as not all officers go by the ordered prevention tactics. Resisting changes is part of human nature, so any change on routine patrols may face resistance. Not increasing the workload, either real or perceived, helps encourage the full participation of patrol officers. Jeffery 1971 argued that crime prevention may also be achieved through environmental design. Newman 1972 coined the concept of defensible space for crime prevention through environment design (CPTED). The micro built environment, affecting people’s perception of safety, can impact crime. Cozens and Love 2015 provided a comprehensive review of CPTED and found that CPTED had gained popularity and was supported by many governments in the world. Clarke 1980 proposed situational crime prevention, by reducing physical opportunities for crime, and increasing the risks of being caught. Clarke 1997 later summarized successful case studies of situational crime prevention in a book.

  • Chen, H., T. Cheng, and S. Wise. “Developing an Online Cooperative Police Patrol Routing Strategy.” Computers, Environment and Urban Systems 62 (2017):19–29.

    DOI: 10.1016/j.compenvurbsys.2016.10.013Save Citation »Export Citation » Share Citation »

    This paper proposes a set of guidelines for patrol routing strategies to meet the challenges of police patrol. These guidelines include efficiency, flexibility, scalability, unpredictability, and robustness. The authors develop an innovative heuristic-based and Bayesian-inspired real-time strategy for cooperative routing police patrols. Using two real-world cases in London and Chicago and a benchmark patrol strategy, an online agent-based simulation is implemented to test the proposed strategy.

    Find this resource:

  • Clarke, R. V. G. “Situational Crime Prevention: Theory and Practice.” British Journal of Criminology 2 (1980):136–147.

    DOI: 10.1093/oxfordjournals.bjc.a047153Save Citation »Export Citation » Share Citation »

    This paper presents that an alternative theoretical emphasis on choices and decisions made by the offenders contribute to a broader approach to crime prevention. Two suggestions out of the situational research can be divided into two measures: reducing physical opportunities for crime, and increasing the risks of being caught. Examples of reductions are given in relation to offenses of theft and vandalism.

    Find this resource:

  • Clarke, R. V. G. Situational Crime Prevention: Successful Case Studies. Monsey, NY: Criminal Justice Press, 1997.

    Save Citation »Export Citation » Share Citation »

    This book argues that situational prevention comprises opportunity-reducing measures that are directed at highly specific forms of crime, involve the management, design, or manipulation of the immediate environment, and make crime more difficult and riskier. Many successful situational preventions involve such measures as surveillance cameras for subway systems and parking facilities, defensible space architecture in public housing, target hardening of apartment blocks and individual residences, alcohol controls at festivals and sporting fixtures, etc.

    Find this resource:

  • Cozens, P., and T. Love. “A Review and Current Status of Crime Prevention through Environmental Design (CPTED).” Journal of Planning Literature 30.4 (2015): 393–412.

    DOI: 10.1177/0885412215595440Save Citation »Export Citation » Share Citation »

    This paper reviews the current status of the concept of crime prevention through environmental design (CPTED). It provides an overview of its history and origins and defines how it is commonly understood and conceptualized. Globally, CPTED is an increasingly popular crime prevention strategy supported by governments all over Europe, North America, Australia, New Zealand, Asia, and South Africa. This review inspects some of the evidence associated with CPTED and provides a detailed overview of the main criticisms facing this field.

    Find this resource:

  • Jeffery, C. R. Crime Prevention through Environmental Design. Beverley Hills, CA: SAGE, 1971.

    Save Citation »Export Citation » Share Citation »

    This book draws on social, behavioral, political, psychological, and biological explanations for the occurrence of crime. The author calls for a new school of thought in the field of environmental criminology. Urban planning and design, social planning, systems analysis and decision theory, governmental policies, and training in environmental criminology are discussed in the development of concept of crime prevention through environmental design (CPTED). The environment, not the criminal justice system, is what must be reformed.

    Find this resource:

  • Koper, C. S. “Just Enough Police Presence: Reducing Crime and Disorderly Behavior by Optimizing Patrol Time in Crime Hot Spots.” Justice Quarterly 12.4 (1995): 649–672.

    DOI: 10.1080/07418829500096231Save Citation »Export Citation » Share Citation »

    This paper employs survival models to test the effects of police patrol presence on crime prevention. The results show that patrol stops must reach a threshold dosage of about ten minutes in order to generate significantly longer survival times without disorder. The optimal length for patrol stops appears to be eleven to fifteen minutes. After that point, continued police presence brings diminishing returns.

    Find this resource:

  • Newman, O. Defensible Space: Crime Prevention through Urban Design. New York: Macmillan, 1972.

    Save Citation »Export Citation » Share Citation »

    This book concerns how the forms of urban residential areas, especially low-income housing, contribute to the victimization of residents. The proposed defensible-space model is conducive to discouraging crime opportunities by limiting the space that incurs criminal activities and by increasing the surveillance opportunities for residents. This book serves to enhance an understanding of social and psychological constraints associated with mass housing.

    Find this resource:

  • Sorg, E. T., C. P. Haberman, J. H. Ratcliffe, and E. R. Groff. “Foot Patrol in Violent Crime Hot Spots: The Longitudinal Impact of Deterrence and Posttreatment Effects of Displacement.” Criminology 51.1 (2013): 65–101.

    DOI: 10.1111/j.1745-9125.2012.00290.xSave Citation »Export Citation » Share Citation »

    This paper revisits the Philadelphia Foot Patrol Experiment and explores the longitudinal deterrent effects of foot patrol on violent crime hot spots. Sherman’s (1990) concepts of initial and residual deterrence decay are used as a theoretical framework. Multilevel growth curve models reveal that beats staffed for twenty-two weeks had a decaying deterrent effect during the course of the experiment, whereas those staffed for twelve weeks did not. The displacement uncovered had decayed during the three months after the experiment.

    Find this resource:

  • Weisburd, D., and J. E. Eck. “What Can Police Do to Reduce Crime, Disorder, and Fear?” The ANNALS of the American Academy of Political and Social Science 593.1 (2004): 42–65.

    DOI: 10.1177/0002716203262548Save Citation »Export Citation » Share Citation »

    This paper argues that the strongest evidence of police effectiveness in reducing crime and disorder is found in geographically focused police practices, such as hot-spots policing. Community policing does not consistently affect either crime or disorder, without models of problem-oriented policing.

    Find this resource:

Impact Assessment

Place-based crime prevention experiments have led to various and contrasting results. Most find a reduction of crime in the target (treatment, or experiment) areas, while others end with inconclusive evidence on crime reduction. In addition to crime reduction in the treatment areas, effects in their surrounding areas with no targeted policing are possible. One scenario is that the surrounding areas also experience crime reduction. This effect is called “diffusion of crime prevention benefits” or “diffusion of benefits” in short, as is presented in Clarke and Weisburd 1994. In contrast, targeted policing may cause crime and disorder to shift away from the targeted areas, thus increasing crime in the surrounding areas. This effect is called “displacement of crime” or crime displacement. In addition to spatial displacement, crime can be displaced in crime type, time, target, offender, and tactic. Guerette and Bowers 2009 examined 102 evaluations of interventions for possible displacement of crime.

  • Clarke, R. V., and D. Weisburd. “Diffusion of Crime Control Benefits: Observations on the Reverse of Displacement.” Crime Prevention Studies 2 (1994): 165–184.

    Save Citation »Export Citation » Share Citation »

    This paper presents the phenomenon of “diffusion of benefits,” the unexpected reduction of crimes not directly targeted by the preventive action. Two processes underlying diffusion are identified: offenders’ uncertainty about the extent of the increased risk, and their exaggerated perception that the rewards of particular crimes are no longer commensurate with the effort. These processes can be labeled deterrence and discouragement respectively.

    Find this resource:

  • Guerette, R. T., and K. J. Bowers. “Assessing the Extent of Crime Displacement and Diffusion of Benefits: A Review of Situational Crime Prevention Evaluations.” Criminology 47.4 (2009): 1331–1368.

    DOI: 10.1111/j.1745-9125.2009.00177.xSave Citation »Export Citation » Share Citation »

    This paper examines 102 evaluations of crime-prevention projects to assess crime displacement. The results indicate that displacement was observed in 26 percent of observations, while the opposite of displacement, diffusion of benefit, was observed in 27 percent of the observations. Moreover, the analysis of thirteen studies reveals that when spatial displacement did occur, it tended to be less than the treatment effect. This means the intervention was still beneficial.

    Find this resource:

Displacement Quotients

A simple comparison between crime counts in the target area before and after the intervention (treatment, or experiment) may not be sufficient to conclude the effectiveness. This comparison has to be controlled by what happens in areas with similar characteristics but with no treatment. The control areas can be identified by a process of propensity score matching. A thorough assessment typically involves three types of areas: the target areas, the control areas, and the surrounding areas. Both target areas and surrounding areas are compared to the control areas, to examine if the intervention is effective in the target areas, and if the intervention leads to diffusion of benefits or displacement of crime in the target areas. Bowers and Johnson 2003 developed a weighted displacement quotient (WDQ) for directly comparing the effects in the surrounding areas over the target areas, with the consideration of control areas. Positive WDQ values indicate an effect of diffusion of benefits, while negative WDQ values indicate an effect of crime displacement. Similar to the use of control areas, historical crime data of the target areas and surrounding areas can be used as controls. Hall and Liu 2009 extended the WDQ index to a spatio-temporal weighted displacement quotient (STWDQ). STWDQ can be used to compare the effectiveness of changing policing tactics during a period. Lately, Wheeler and Ratcliffe 2018 proposed a simple weighted displacement difference test (WDD), capable of providing a measure of the standard error.

  • Bowers, Kate J., and Shane D. Johnson. “Measuring the Geographical Displacement and Diffusion of Benefit Effects of Crime Prevention Activity.” Journal of Quantitative Criminology 19.3 (2003): 275–301.

    DOI: 10.1023/A:1024909009240Save Citation »Export Citation » Share Citation »

    This paper introduces a weighted displacement quotient (WDQ) for assessing potential geographic displacement of crime and diffusion of benefits as a result of an intervention. It is based on comparison of the target area, a buffer area for possible displace, and a control area.

    Find this resource:

  • Hall, Davin, and Lin Liu. “Cops and Robbers in Cincinnati: A Spatial Modeling Approach for Examining the Effects of Aggressive Policing.” Annals of GIS 15.1 (2009): 49–59.

    DOI: 10.1080/19475680903271158Save Citation »Export Citation » Share Citation »

    This paper develops a spatio-temporal weighted displacement quotient (STWDQ) for assessing potential displacement of crime and diffusion of benefits as a result of an intervention. Instead of using a spatial area without invention for control, this index uses a pair of past time periods as the control.

    Find this resource:

  • Wheeler, Andrew P., and Jerry H. Ratcliffe. “A Simple Weighted Displacement Difference Test to Evaluate Place Based Crime Interventions.” Crime Science 7 (2018): 11.

    DOI: 10.1186/s40163-018-0085-5Save Citation »Export Citation » Share Citation »

    This paper proposes a weighted displacement difference (WDD) to test potential spatial displacement or diffusion of benefits. In addition to measuring the effectiveness of a crime reducing intervention, it also provides a measure of standard error.

    Find this resource:

Randomized Controlled Trial and Regression Modeling

The randomized controlled trial is a popular approach to ensure any effectiveness of intervention is statistically sound, as is presented by Ratcliffe, et al. 2011; Taylor, et al. 2011; and Garvin, et al. 2013. The difference in differences (DID) regression can test the effects and show statistical significance. It compares the average change over time in crime counts or rates in the treatment areas to the average change over time in the control areas. DID has been applied to test the impact of surveillance cameras on crime reduction. Liu, et al. 2020 (cited under the Spatio-temporal Mismatch Problem) applied a variation of the DID to test the impact of the relocation of bus stops on street crime, with the conclusion that adding a stop would increase street crime nearby but removing a stop has no significant impact. Another approach of controlling for possible confounding effect is propensity score matching. It aims to reduce potential bias caused by covariates in assessing the effectiveness of an intervention. Vito, et al. 2017 used propensity score matching for assessing the effectiveness of parole supervision, and Piza 2018 for assessing the crime prevention effect of CCTV in public places. It should be pointed out that the quality of intervention should be considered in the assessment. For example, the uneven performance of the police officers may lead to variation in the intervention effectiveness. Further, any detected effects may decay in time, therefore, intervention strategies should be periodically adjusted based on the feedback of the assessment.

  • Garvin, E. C., C. C. Cannuscio, and C. C. Branas. “Greening Vacant Lots to Reduce Violent Crime: A Randomized Controlled Trial.” Injury Prevention 19.3 (2013): 198.

    DOI: 10.1136/injuryprev-2012-040439Save Citation »Export Citation » Share Citation »

    The authors performed a randomized controlled trial of vacant lot greening to test the impact of this intervention on police reported crime and residents’ perceptions of safety and disorder. Two vacant lot clusters were randomly allocated to the greening intervention or to the control status. The results of unadjusted difference-in-differences estimates showed a non-significant decrease in the number of total crimes. People around the intervention vacant lots reported feeling significantly safer after greening.

    Find this resource:

  • Piza, E. L. “The Crime Prevention Effect of CCTV in Public Places: A Propensity Score Analysis.” Journal of Crime and Justice 41.1 (2018):14–30.

    DOI: 10.1080/0735648X.2016.1226931Save Citation »Export Citation » Share Citation »

    This paper measures the effect of CCTV in Newark, New Jersey, on auto theft, theft from auto, and violent crime respectively. CCTV viewsheds were units of analysis. Treatment cases were matched with control cases via propensity score matching (PSM) to ensure statistical equivalency between groups. Findings offer modest support for CCTV as a deterrent against auto theft while demonstrating no effect on the other crime types.

    Find this resource:

  • Ratcliffe, J. H., T. Taniguchi, E. R. Groff, and J. D. Wood. “The Philadelphia Foot Patrol Experiment: A Randomized Controlled Trial of Police Patrol Effectiveness in Violent Crime Hotspots.” Criminology 49.3 (2011):795–831.

    DOI: 10.1111/j.1745-9125.2011.00240.xSave Citation »Export Citation » Share Citation »

    This paper assesses the prevention effects of more than two hundred foot-patrol officers during the summer of 2009 in Philadelphia, through a randomized controlled trial of police effectiveness across sixty violent crime hotspots. The results demonstrate a significant reduction in the level of violent crime in treatment area after twelve weeks.

    Find this resource:

  • Taylor, B., C. S. Koper, and D. J. Woods. “A Randomized Controlled Trial of Different Policing Strategies at Hot Spots of Violent Crime.” Journal of Experimental Criminology 7.2 (2011):149–181.

    DOI: 10.1007/s11292-010-9120-6Save Citation »Export Citation » Share Citation »

    The paper randomly assigns eighty-three hot spots of violence in Jacksonville, Florida, to receive either directed-saturation patrol or a control condition for ninety days. The results show that the use of directed-saturation patrol is associated with a 33 percent reduction in “street violence” during the ninety days following the intervention. No statistically significant crime reductions are detected for the control group.

    Find this resource:

  • Vito, G. F., G. E. Higgins, and R. Tewksbury. “The Effectiveness of Parole Supervision: Use of Propensity Score Matching to Analyze Reincarceration Rates in Kentucky.” Criminal Justice Policy Review 28.7 (2017): 627–640.

    DOI: 10.1177/0887403415609717Save Citation »Export Citation » Share Citation »

    This paper examines whether a cohort of offenders released to the community in Kentucky either under parole supervision or at the expiration of their sentences are more likely to be reincarcerated within a five-year period. The participants of each cohort were constructed into two groups using propensity score matching to control for differences between them. The results from weighted logistic regression show that individuals on parole were less likely to be reincarcerated than members of the maxed-out group.

    Find this resource:

Simulation Based Virtual Experiments

Intervention in the real world is regarded as a natural experiment or quasi-experiment. A similar experiment can also be implemented in a virtual environment. Crime simulation models based on cellular automata and agent-based modeling have shown promising results in assessing the effectiveness of various policing tactics, as is presented by Liu and Eck 2008; Malleson, et al. 2010; and Weisburd, et al. 2017. According to Eck and Liu 2008, the main advantage of the virtual over the natural ones is the flexibility of configuring various conditions of the experiments. The drawback is that the theory-driven virtual models are difficult to verify, leading to suspicion on their validity by some.

  • Eck, J. E., and L. Liu. “Contrasting Simulated and Empirical Experiments in Crime Prevention. Journal of Experimental Criminology 4.3 (2008):195–213.

    DOI: 10.1007/s11292-008-9059-zSave Citation »Export Citation » Share Citation »

    The paper compares simulated with empirical experiments on nine criteria: mechanism, data, implementation, manipulation, construct validity, statistical validity, internal validity, external validity, and application. It argues that simulations have strengths that empirical methods lack, but they also have important relative weaknesses. It concludes by listing ways simulations can be used to improve empirical experiments and discussing the differing operating assumption of empirical and simulation experimentalists.

    Find this resource:

  • Liu, L., and J. Eck, eds. Artificial Crime Analysis Systems: Using Computer Simulations and Geographic Information Systems. Hershey, NY: IGI Global, 2008.

    Save Citation »Export Citation » Share Citation »

    This edited volume includes a number of researches that utilize agent-based simulation and cellular automaton to simulation crime. The simulation software is used as a virtual laboratory for conducting experiments, by adjusting input and parameters. The interplay of related theories can be implemented and new hypotheses can be generated in the virtual experiments.

    Find this resource:

  • Malleson, N., A. Heppenstall, and L. See. “Crime Reduction through Simulation: An Agent-Based Model of Burglary. Computers, Environment and Urban Systems 34.3 (2010):236–250.

    DOI: 10.1016/j.compenvurbsys.2009.10.005Save Citation »Export Citation » Share Citation »

    This paper presents the construction and application of an agent-based model (ABM) for simulating occurrences of residential burglary. An artificial city, loosely based on the real city of Leeds, England, and an artificial population was constructed. The results of simulation experiments demonstrate the potential of this approach for both understanding processes behind crime and improving policies and developing effective crime prevention strategies.

    Find this resource:

  • Weisburd, D., A. A. Braga, E. R. Groff, and A. Wooditch. “Can Hot Spots Policing Reduce Crime in Urban Areas? An Agent-Based Simulation.” Criminology 55.1 (2017):137–173.

    DOI: 10.1111/1745-9125.12131Save Citation »Export Citation » Share Citation »

    This paper uses an agent-based model to estimate area-wide impacts of hot spots policing on street robbery. Two implementations of hot spots policing are tested in a simulated borough of a city, and are compared with two control conditions (constant random patrol and no police officers). The results show statistically significant effects for hot spots policing beyond both a random patrol model and a landscape without police.

    Find this resource:

Big Data for Crime Analysis

The Spatio-temporal Mismatch Problem

Crime incidents have precise information on location and time. Data related to the built environment and social environment, used for explaining crime, are typically updated once in several years. There exists an inherent mismatch between the high spatio-temporal resolution of the crime variable and the low resolutions of the explanatory variables. This spatio-temporal mismatch problem exists not only in crime analysis, but also in public health research and any other research that utilizes event-based data with precise time and location. Traditional approaches would typically reduce the higher resolutions to match the lower resolution. For example, individual crime events would be aggregated to a spatial unit of census tract in a time period of one year, to match those of the census variables. With the advent of big data, more explanatory variables in high spatio-temporal resolutions can be derived for crime analysis. One major source of big data is remote sensing data. Satellite nightlight data, obtained as often as once every two weeks, help derive comprehensive activity space in a city. Liu, et al. 2020 found that edges derived from the nightlight image help explain robbery at the Census block level. LiDAR data, with a spatial resolution at centimeter-level, capture the detailed morphology of the city. Micro characteristics of the built environment derived from LiDAR can potentially contribute to crime prevention through environmental design (CPTED).

  • Liu, L., H. Zhou, M. Lan, and Z. Wang. “Linking Luojia 1-01 Nightlight Imagery to Urban Crime.” Applied Geography 125 (2020): 102267.

    DOI: 10.1016/j.apgeog.2020.102267Save Citation »Export Citation » Share Citation »

    This paper uses the composite edge derived from the Luojia 1-01 satellite nightlight imageries to explain robbery and burglary at the Census block level. Results show that the composite edge has a significant effect on robbery, but not on burglary.

    Find this resource:

Big Data Solutions to the Spatio-temporal Mismatch Problem

Cell phone data record the interaction of a cell phone with the signal towers, which can infer the major activity nodes of a cell phone (user). The dominant activity node at night is likely to be the home location, while the dominant activity node during the day is likely to be the work location. The remaining activity nodes depict the rest of the activity space of the cell phone user. The temporal sequencing of the activity nodes creates trajectories. These data can derive the ambient population, representing the dynamic distribution of people. Song, et al. 2018 revealed that a cell phone-based ambient population can better explain theft than census population. Since the census attributes can be attached to the cell phone at its home location, the characteristics of individual members of the ambient population can contribute to the refined analysis, opening a completely new horizon for crime analysis. Geotagged tweets are somewhat similar to cell phone data. Lan, et al. 2019 underscored the effectiveness of tweets being used as ambient population for crime analysis. An advantage of tweets is that the content of tweets can reveal positive or negative emotions, which may, in turn, affect crime. The natural language process could also retrieve crime-related data out of the contents of tweets. GPS tracking data such as ridership of taxis are useful ambient population measurements as well. Besides, the origin and destination of each trip help reveal the movement of the population throughout a day. Such movement data can greatly enhance the traditional crime analysis. Since GPS tracking data are obtained outdoor, they can be more powerful in explaining street crimes than cell phone data and geotagged tweets, which are combinations of indoor and outdoor activities. Some bus passes also record the origin and destination of each trip. Such data can be used in a similar way as the taxi ridership data. While both realized and potential benefits of big data are enormous, there are various issues with big data. One issue is the representativeness of the big data. Virtually all big data are biased. Andresen 2015 pointed out that social media data favor young people. Lan, et al. 2019 also showed that the highest concentrations of tweets are college campuses and their surrounding areas. Taxi ridership data favors a more affluent population. Even cell phone data do not cover the entire population.

  • Andresen, M. A. “The Impact of Using Social Media Data in Crime Rate Calculations: Shifting Hot Spots and Changing Spatial Patterns.” Cartography & Geographic Information Science 42.2 (2015): 112–121.

    Save Citation »Export Citation » Share Citation »

    This paper uses social media data in Leeds, England, to measure the population at risk for violent crime. Using two local spatial statistics (Getis-Ord GI* and the Geographical Analysis Machine) and visualization, criminal event hot spots shift spatially when the volume of social media messages, instead of residential population, is used as a proxy for the population at risk.

    Find this resource:

  • Lan, M., L. Liu, A. Hernandez, W. Liu, H. Zhou, and Z. Wang. “The Spillover Effect of Geotagged Tweets as a Measure of Ambient Population for Theft Crime.” Sustainability 11.23 (2019): 6748.

    DOI: 10.3390/su11236748Save Citation »Export Citation » Share Citation »

    This paper examines the degree in which geotagged tweets can explain crime, and the impact of tweets and its possible spillover effect on theft crimes. The results from negative binomial regression models show that tweet count is a feasible measurement of ambient population and have a spillover effect on theft crimes.

    Find this resource:

  • Song, G., L. Liu, W. Bernasco, L. Xiao, S. Zhou, and W. Liao. “Testing Indicators of Risk Populations for Theft from the Person across Space and Time: The Significance of Mobility and Outdoor Activity.” Annals of the American Association of Geographers 108.5 (2018):1–19.

    DOI: 10.1080/24694452.2017.1414580Save Citation »Export Citation » Share Citation »

    This paper measures crime risk population and examines its relationship with theft from the person in a large city in China across space and time. Indicators of the risk population include residential population, subway ridership, taxi ridership, and mobile phone users. The results of negative binomial regression models show that on both weekdays and weekends, the best indicators of risk population and predictor of theft from the person vary over the course of the day.

    Find this resource:

Ethic Issues in Crime Analysis

Ethical issues exist in various elements of crime analysis. Many location data are either obtained or used without the explicit consent of the individuals. Some cell phone apps would not function without granting the permission of location sharing. Location data collected by apps are channeled to data companies that sell packaged data for profit. Data on individuals and their daily activities are highly sensitive. Therefore, any research related to the use of individual data must be approved by the institutional review board (IRB). Normally IRB would approve studies that only present aggregated results instead of individual results. Data-driven analysis, even in black-box nature for most, is not without ethical issues. Lo Piano 2020 outlined points of friction across ethical principles in machine learning and artificial intelligence, using risk assessment in the criminal justice system as an example. The issue of confidentiality extends to crime data and results of crime analysis (Wolfgang 1981). Data policies vary across the nations. America is somewhat unique in its open sharing of crime data. Many European countries have stricter control over access to crime data. Many East Asian countries consider crime data confidential. Some would even prohibit the publication of crime density maps. Therefore, many publications use a code name for the study area and remove boundary and other identifiable information on the crime maps. There are also challenges to geographic profiling (Rossmo 2000), video surveillance, and intelligence-oriented policing, due to the concerns of violating people’s privacy and ethnic discrimination. Kutnowski 2017 highlighted the potential ethical dangers of predictive policing, while accepting its merits. Therefore, researchers need to pay attention to these issues, especially if the research is carried out in unfamiliar territories. Special attention is needed when policies and regulations of the study site are different from those of the home country.

  • Kutnowski, M. “The Ethical Dangers and Merits of Predictive Policing.” Journal of Community Safety and Well-Being 2.1 (2017): 13.

    DOI: 10.35502/jcswb.36Save Citation »Export Citation » Share Citation »

    While accepting the merits of predictive policing, the paper highlights its potential ethical dangers.

    Find this resource:

  • Lo Piano, S. “Ethical Principles in Machine Learning and Artificial Intelligence: Cases from the Field and Possible Ways Forward.” Humanities & Social Sciences Communications 7 (2020): 9.

    DOI: 10.1057/s41599-020-0501-9Save Citation »Export Citation » Share Citation »

    This paper outlines points of friction across ethical principles in machine learning and artificial intelligence, using risk assessment in the criminal justice system and autonomous vehicles as examples.

    Find this resource:

  • Rossmo, D. K. Geographic Profiling. Boca Raton, FL: CRC Press, 2000.

    Save Citation »Export Citation » Share Citation »

    This book is an authority on geographic profiling, which aims to determining the most probable area of an offender’s base of activities through an analysis of past crime locations. Investigations of serial crimes may be benefited from geographic profiling. With the insight gained from geographic profiling, law enforcement can focus the limited resources on a small number of high-risk areas.

    Find this resource:

  • Wolfgang, M. E. “Confidentiality in Criminological Research and Other Ethical Issues.” Journal of Criminal Law and Criminology 72.1 (1981): 345–361.

    DOI: 10.2307/1142913Save Citation »Export Citation » Share Citation »

    This paper highlighted the importance of confidentiality and other ethical issue in criminology research. It also touched on the classical issues relate to protection of human subjects, invasion of privacy, confidentiality of records and interviews, accessibility to data, and immunity of researchers from prosecution.

    Find this resource:

back to top