4. Geographic profiling

Locating criminals constitutes a large share of the work of investigators (Canter, 1994). Geographic clues prove to be valuable for police forces during investigations, particularly in the case of repeat offences by the same individual (Rossmo, 2000). Geographic profiling used in this context may be defined as “... an information strategy for […] crime investigations that analyses crime scene information to determine the most probable area of offender residence” (Rossmo, 2000, p. 259). Some research indicates that the use of geographic profiling systems can reduce the area of investigation by 90% (Canter, Coffey, Huntley and Missen, 2000; Rossmo, 2000). Geographic profiling services provided by the police are intended for various cases: federal and provincial governments and various law enforcement agencies, particularly the RCMP, FBI and Scotland Yard (Rossmo, 2000).

According to Rossmo (2000), compiling geographic data along with information identified as useful to criminal profiling strengthens crime analysis tools. Thus, crime mapping has become a relatively common analytical practice for police services. The ability to use geographic data effectively is related to the use of geographic information systems (Rossmo, 2000).

In a survey of 2004 U.S. police departments conducted by Mamalian and La Vigne (1999), 85% of the respondents stated that crime mapping was a useful tool. The results of this survey also indicated that "crime clustering" and hot point analysis were the mapping applications most used. Information produced through mapping applications can also be subsequently compared with the information obtained, for example, through census and community members.

4.1 Theoretical framework

Geographic profiling has become an essential part of the criminal investigation process. The following sequence, proposed by Rossmo (2000), stipulates how geographic profiling fits into the investigation process:

1. Occurrence of a crime series;
2. Traditional investigative techniques;
3. Linkage analysis;
4. Criminal profile;
5. Geographic profiling; and
6. New investigative strategies.

In addition, the use of geographic profiling is based on a certain number assumptions, namely:

1. The profile must be based on multiple crime scenes (several crimes committed by the same individual or several sites linked to the same crime);
2. The crime scenes must be linked to the same offender;
3. There cannot be a great distance between the residence (or home base) of the offender committing the crimes and the area of criminal activity;
4. The crime scenes must be fairly evenly distributed around the offender’s home or anchor point; and
5. The offender cannot move anchor points or operate from multiple anchor points during his or her crime series (Bennell and Corey, 2007).

In the simplest cases, the criminal’s home is located at the centre of the crime pattern and may be discovered with the help of spatial analysis methods. However, most of the time the relationship between the crime scene and the criminal’s home is much more complex (Rossmo, 2000).

Rengert (1996) proposes four hypothetical spatial patterns that could be used to describe the geography of crime scenes; (1) a uniform pattern with no distance-decay; (2) a bull’s-eye pattern with spatial clustering, exhibiting distance-decay centered around the offender’s primary anchor point; (3) a bimodal pattern with crime clusters centered around two anchor points; and (4) a teardrop pattern with a directional tendency toward a secondary anchor point. According to Rossmo (2000), in reality, these patterns are affected by various factors such as the configuration of the road system and traffic density, type of zoning and land use. All these factors contribute to limiting the scope of geographic profiling without making it ineffective.

In addition, Felson and Clarke (1998) presented three underlying theories of the practice of geographic profiling: routine activity theory, crime pattern theory and rational choice theory.

4.1.1 Routine activity theory

This theory was first developed as an explanation for predator crime. It postulates that in order for such crimes to occur, three components need to be present at the same time and in the same space: a potential offender, a suitable victim and the absence of a guardian able to protect the victim. The risk incurred by the victim depends on four factors: target value, inertia, visibility, and ease of access. This theory assumes that the number of crimes can increase without an increase in the number of criminals if, for example, there are more potential victims or access is easier in the absence of an effective guardian. This has two implications for geographic profiling: crimes are committed particularly 1) in places corresponding to routine criminal activity, and 2) where it is relatively easy to commit crimes because of frequent time-space convergence of the three previously mentioned elements.

4.1.2 Crime pattern theory

This theory focuses on the way that persons and objects involved in criminal activities move in time and space. It links three concepts: 1) nodes (points of departure and arrival in the movement of individuals), 2) paths, and 3) edges (borders or limits between certain areas). Particular attention is given to the geographic distribution of crimes. This theory assumes that criminality is influenced by the characteristics of nodes, that it will be present more around nodes that lend themselves to crime and the paths that lead to them as well as in the proximity of certain borders that constitute sensitive spots.

4.1.3 Rational choice theory

This theory views the criminal as a rational being constantly involved in analyzing the costs and benefits of crime (Beauregard, Rossmo and Proulx, 2007). The emphasis is thus on the offender’s decision-making since the basic postulate is that crime is a purposeful behaviour undertaken for some benefit. The theory is aimed at understanding individual criminal choices in terms of motivations in a given context offering opportunities to satisfy this motivation. In terms of geographic profiling, it determines that, since the place farthest from the anchor point incurs a cost, most offenders will commit their crimes reasonably close to their home (Beauregard and Rossmo, 2007; Brantingham and Brantingham, 1990).

Brantingham and Brantingham (1990) were also involved in geographic profiling using environmental criminology, which focuses more on the study of the crime context than criminal motivation. Thus, environmental criminology is based on the principle that a crime is the result of the confluence of 1) at least one criminal, 2) a victim (or a target of some sort), 3) laws in place and, 4) a given place and time. This means that crime analysis has four dimensions: a legal dimension, a criminal dimension, a victim (or target) dimension and a spatial-temporal dimension. Brantingham and Brantingham (1990) also add that these dimensions must be interpreted in historical, social, economic and political contexts, as well as on the basis of the biological and physical characteristics of the environment, since each of these characteristics contributes to creation of the context of the crime. For example, Tita and Ridgeway (2007) showed how the formation of street gangs influenced the prevalence and distribution of criminal activity.

Brantingham and Brantingham (1990) proposed that crimes are often committed within the context of daily life events and daily paths of individuals. Thus, individuals generally commit their crimes within close proximity of places where they spend most of their time, such as their home, place of work, school, businesses and recreational areas that they frequent. Similarly, individuals who are victims of criminal acts are generally in places they frequent, or along the paths that link these various places. This means not only that criminal events can be understood and predicted from a knowledge of the place where a criminal lives and frequents, but also that the crime can generally be understood and predicted through the analysis of the urban structure (arrangement of urban areas, road configurations, and transport system configuration).

Another environmental criminology theory maintains that crime scene location is determined through research and a structured decision-making process influenced by the criminal’s perception of environmental clues that distinguish between a "good criminal opportunity" and a "bad criminal risk." These authors also maintain that spatial perception varies with the criminal’s age.

Brantingham and Brantingham (1990) also pointed out the necessity of taking ecological labels into consideration. These are described as "reputations popularly appended to particular places or neighbourhoods" (p. 4) with an effect on crime. These labels especially influence the type of individuals (and socioeconomic group) who will be attracted or repulsed by a certain area and the perception that the police, social workers and other individuals have of it. They also affect criminals’ perception of whether the area is an adequate one to commit a crime. These observations are consistent with the results of Dunham, Alpert, Stroshine and Bennett (2005) and Ingram (2007).

Based on urban structure and knowledge of perception and cognition in the criminal context, Brantingham and Brantingham (1990) came to the following conclusions:

1. Older cities, often characterized by a concentric shape with a dense core, have a crime distribution pattern with increased criminality in the proximity of the centre;
2. younger cities in the form of a mosaic have a more disparate c riminal pattern;
3. younger cities with widely dispersed business districts have a higher property crime rate;
4. the development of major transport arteries leads to a concentration of criminal events close to highways, particularly close to major intersections;
5. areas developed using grids generally lend themselves to crime more than "organic street layouts";
6. older cities where low-income housing is dispersed throughout the area are likely to have a concentration of crimes close to the heart of the city and close to various low-income housing areas;
7. relocation of workplaces from the city core to the outskirts tends to increase the suburban crime rate;
8. major recreational complexes such as arenas are likely to increase the localized crime rate. If these complexes are located close to a residential area harbouring several potential criminals, the crime rate increases drastically;
9. Cities with a red light district in their core are likely to have a higher concentration of crime in that area. However, forcing the dispersal of typical red light activities will not necessarily lead to a decrease in the total number of crimes, but will modify their spatial distribution.

4.2 Application

Various methods have been developed with a view to applying geography to the field of criminal profiling. The potential for the application of geographic information systems (GIS) in serial murder investigations has been recognized for a few years. Given that a GIS allows for the combination of geographic attributes and spatial data with other relevant data, it is a useful tool for organizing information that might at first glance seem disparate (Oatley, Ewart and Zeleznikow, 2006; Rossmo, 2000). According to Rogers, Craig and Anderson (cited in Rossmo, 2000), geographically coded information can be used to detect crime trends and recurrent patterns to confirm the presence of certain individuals in certain sectors and mark off areas where patrols should be concentrated. According to the same authors, GIS use could contribute to the identification of a serial killer through the retrospective analysis of known cases and could, by that very fact, prove useful in solving unsolved murder cases.

Three GIS models are currently in use for the geographic targeting of criminals (Paulsen, 2006). The first was developed based on research conducted by Simon Fraser University and the Vancouver Police Department. It was called the Criminal Geographical Targeting (CGT) model and was based on the theoretical model of Brantingham and Brantingham (1990). According to this model, the geographic coordinates of crime scenes are analyzed with the help of a distance decay function allowing for the production of a two- or three-dimensional spatial representation of the probabilities of locating the criminal’s place of residence (Paulsen, 2006; Rossmo, 2000). According to Rossmo (2000), the CGT model proved valid and reliable when strict application criteria were respected, particularly for decreasing process subjectivity. For example, only the crime scenes recognized as such should be considered valid data as opposed to, for example, the place where the victim was seen for the last time. According to Rossmo (2000), the main limitation in the application of the CGT model is that it is less useful when the criminal has travelled long distances to commit his crime. However, it was pointed out by Rossmo (2000) that when a crime has been committed far from the criminal ’s place of residence, it is often possible to identify the criminal’s workplace, a former residence or the residence of a relative or friend of the criminal in close proximity to the crime scene. This model uses two software programs: RIGEL and RIGEL Analyst (a simplified version of RIGEL).

The two other models are Canter’s, which uses Dragnet software, and Levine’s, which is based on Crimestat software (Paulsen, 2006). Unlike Rossmo’s and Canter’s models, Levine’s model is based on the Journey to Crime (JTC) type. The distinction between the two approaches is that the JTC model is strictly statistical and not based on a theoretical framework of geographic profiling, as is, for example, Brantingham and Brantingham’s model (Paulsen, 2006).

Nevertheless, Rossmo (2000) and Ainsworth (2001) note that it is important to keep in mind the fact that although the model produces very accurate results, it does not take the police right to the criminal’s door. Geographic profiling software must be seen as information management systems with a geographic perspective useful for the investigation process rather than a panacea. They could help prioritize certain suspects and focus patrol staff in a limited area and thus possibly shorten the time between the moment the crime was committed and the moment the criminal is apprehended (Rossmo, 2000).

4.3 Empirical support

4.3.1 Reduction of the search area

One of the advantages attributed to geographic profiling is the reduction of the surface area to be explored before localizing an offender’s base of operations. According to the research available, geographic profiling could reduce the territory of investigation by 90% (Bennell, Snook, Taylor, Corey and Keyton, 2007). Research conducted by Beauregard and Rossmo (2007) is even more optimistic. According to them, the proportion of surface area that should be examined by police would be 7.1% for robbery, 5.1% for homicides, 4.7% for sexual assault and 2.2% for arson. These results are the same in terms of size as those obtained by Laukkanen and Santtila (2006), who obtained a median research area equivalent to 4.69% of the area covered by crimes. Canter and Larkin (1993) observed a median research area of 1.07% when the paths of offenders seem to correspond to the circle heuristic. However, when the offender travelled to the crime scene, predictions were much more inaccurate, with the median research area covering 24.06% of the total surface area.

4.3.2 Accuracy

The effectiveness and accuracy of geographic profiling can vary according to the type of crime. To locate a terrorist base of operation, for example, Bennell and Corey (2007) obtained rather inconclusive results with the Dragnet software. They pointed out the difficulty of profiling crimes where the perpetrators covered long distances. It seems that terrorism does not lend itself very well to geographic profiling since it is often the work of organizations with several bases of operation, and a series of attempts could involve more than one perpetrator.

In the case of burglary, Bennell, Snook, Taylor, Corey and Keyton (2007) observed a reduction in the margin of error in the prediction of the offender’s place of residence by police officers with minimal training in the use of simple heuristics (circle heuristic and decay heuristic). Their performance after training surpassed that of a control group that had not learned to handle heuristics. This reduction in the error of prediction was more significant when the number of crimes profiled increased from three to five, then from five to seven. However, the true accuracy of the methods is difficult to evaluate since the results are provided solely in millimetres of deviation on a geographic map without conversion to kilometres (on a real scale).

Paulsen (2006) points out that the accuracy is generally modest: "Importantly, the results also seem to indicate that none of the strategies are very accurate, with the average error distance across all strategies being 4.45 miles, a significant distance in an urban area" (p. 316). In his comparison of seven computerized algorithms and three heuristic methods, Paulsen (2006) also noted that the geographic profiles obtained using heuristic methods were, on average, more often correct than those provided by the seven software profiles, the place of residence being included in the profile obtained by the heuristic method 27.6% of the time versus 11.4% for computerized methods. This better performance would not be obtained at the expense of a loss of accuracy: "Importantly, these strategies also have a substantially smaller average top profile area than all but one probability distance strategy (negative exponential) indicating that their accuracy is not necessarily due to overly large average top profile areas" (Paulsen 2006, p. 316). The success rate of profiling, included as a percentage of profiles effectively containing the criminal’s home, would be relatively modest according to these results (from 20% to 30% in the best cases).

4.3.3 Heuristics versus software

The preceding results emphasize that geographic profiling performance does not seem to depend on sophisticated software. Paulsen (2006) maintained that:

These findings cast serious doubt as to whether a law enforcement agency needing to conduct a geographic profile would find any significant benefit in using a probability strategy over a simple [heuristic] spatial distribution strategy. This is all the more important given the cost to an agency, both financial and time wise, in acquiring and learning how to use a probability strategy over a simple to use and employ spatial distribution strategy (p. 327).

This position was also defended by Snook, Canter and Bennell (2002) as well as by Bennell, Snook, Taylor, Corey and Keyton (2007), who note that "…  brief training on either the circle or decay heuristic was sufficient to increase officers’ predictive accuracy. […] Both groups achieved an average accuracy that was better that the accuracy of computationally expensive methods [CrimeStat]" (p. 128). Snook, Taylor and Bennell had already arrived at such a conclusion about the use of the circle heuristic in research published in 2004.

4.3.4 Limits

The success of geographic profiling seems to depend on certain factors. For example, a significant distance between crimes and the criminal’s place of residence as well as a wide dispersion of crimes seem to make profiling more difficult (Paulsen, 2006). If the number of serial crimes is too low (less than three) or too high (more than seven), it can also complicate the investigator’s task. In addition, certain crimes such as car theft and burglaries seem to lend themselves better to geographic profiling.

Finally, geographic profiling assumes a series of crimes or crime scenes that can be attributed to the same offender (Paulsen, 2006). The reviewed studies do not convincingly demonstrate profilers’ ability to associate a series of crimes to a given criminal. Some studies seem to suggest the feasibility of this association (Santtila, Fritzon and Tamelander, 2004; Santtila, Junkkila and Sandnabba, 2005). In other words, the basic premise in linkage analysis can apparently be demonstrated, but these studies are fraught with major methodological deficiencies. The validity of the premises in which (1) one offender demonstrates a certain stability of criminal behaviours, and (2) different offenders adopt practices that enable differentiation between them is somewhat supported in the case of burglaries (Bennell and Jones, 2005; Woodhams and Toye, 2007). However, it appears that the distance between crimes provides a better basis for associating these crimes among them than any other behavioural factor. In the absence of sound empirical evidence that crimes may be related on the basis of behavioural factors, taking into account matching crimes and offenders (linkage analysis), an integral part of investigation work, could lower the observed success rate of geographic profiling.

4.4 Conclusions

4.4.1 Summary

It seems that the potential effectiveness of geographic profiling, particularly in terms of reduction of the search area, has been empirically demonstrated several times. In addition, this effectiveness is not dependent upon the use of software. Police knowledge of a limited number of simple heuristics seems to lead to results equivalent to those obtained by Crimestat, Dragnet, RIGEL and RIGEL Analyst software. Geographic profiling is based on a certain number of theoretical referentials among which some axioms have been empirically supported.

However, the practice of geographic profiling actually consists of two stages: 1) attribution of a series of crimes to the same offender, and 2) establishment of a geographic profile defining the research area with the greatest probability of containing the offender’s residence based on the locations of various crime scenes. Although it has been empirically demonstrated that the second stage may be accomplished relatively efficiently, research is not able to judge the ability of investigators to complete the first stage. The effectiveness and usefulness of geographic profiling is based on the ability to accomplish both tasks in a reasonably accurate manner. Attribution of responsibility for a series of crimes to the same individual brings us back to geographic profiling, which has not been empirically proven. Certain studies suggest some temporal stability in the modus operandi and, especially, signature, particularly in the case of very personal crime (such as rape and homicide). In addition, these elements often have particular aspects that are distinctive enough to at least calculate the possibility of associating a suspect with a series of interpersonal crimes. However, the performance of profilers who do so does not seem to have been documented.

4.4.2 Limits

The main limit of the research reviewed was mentioned previously, specifically providing analysts with a series of crimes correctly attributed beforehand to the same offender. In reality, this first stage constitutes a challenge and a significant part of the investigation work. Empirical support of geographic profiling can thus only be considered partial. In addition, several research studies have been conducted by the writers of geographic profiling software themselves (Rossmo, Canter), which leaves some doubt as to their independence. Independent research clearly suggests that software use is not indispensable to the success of geographic profiling.

4.4.3 Recommendations

[R4] – The way in which geographic profiling coordinates are selected and entered should be standardized (for example, if an altercation starts in a bar, continues outside and ends in a homicide a few blocks away, which coordinate(s) mark the crime scene?).

[R5] –Research should be undertaken to evaluate the performance of analysts in the first stage of geographic profiling (attribution of crimes to the same suspect).

Previous Page Table of Contents Next Page