Andresen, M.A., & Wuschke, K. (2019). Geography. In R. Wortley, A. Sidebottom, N. Tilley, & G. Laycock (Eds.), Routledge handbook of crime science (pp. 131 – 144). New York, NY: Routledge.
Introduction
Crime science is the study of crime, quite often its events, in an effort to reduce and/or prevent criminal activity. This method of crime prevention is not focused on any particular theory, but is multidisciplinary in nature, guided by the scientific method. As such, rather than being self-fulfilling, at least in its essence, crime science has the purpose of developing new knowledge with the goal of reducing crime. This is accomplished through the identification of empirically-supported evidence for crime prevention, the integration of that empirically-supported evidence across disciplines, and the further development of that empirically-supported evidence.
One of the disciplines contributing to crime science is geography. At its essence, (human) geography is the study of people and their communities and how those people interact within and through their environment. Critical to this discipline are the concepts of space and place, which bring together an understanding of how spatial relationships and local-level investigations help us to comprehend human activities in particular places. One of those activities is crime.
Geography, or at least the knowledge of spatial variations, has been a part of criminology for nearly 200 years. It should then come as no surprise that geography is a part of contemporary crime science. In short, criminal opportunities vary across space, as do motivated offenders; as such, crime exhibits predictable spatial patterns. In this chapter we will cover some of the contributions of geography to crime science: spatial analysis, the importance of scale and place, and policy mobilities1. Though we cover topics that crime scientists may approach through the use of computer software such as geographic information systems, we try to emphasize how thinking geographically, even within numerical calculations, can help to understand spatial crime patterns. This, in turn, will help inform the development of policies and practices to reduce and/or prevent criminal activity.
Spatial analysis
The most obvious contribution of geography to crime science is the development and use of spatial analysis. In fact, the use of maps to show geographical crime patterns goes back at least as far as the work of Guerry (1832, 1833) and Quetelet (1831, 1842) in early eighteenth century France. Quite simply, spatial analysis is the examination of spatially-referenced data that considers the geographical relationships between spatial units of analysis. It is critical to consider the importance of such relationships because spatially-referenced data can be analyzed without considering spatial relationships. For example, a researcher may have criminal event and census data, all at the level of the census tract. The researcher could use a spatial analysis method because census tracts have spatial information available, but the researcher could also simply undertake classical statistical analysis using the census tracts as observations. Such approaches have been commonplace, at least up until the last couple of decades when spatial analysis methods have become more available. However, ignoring spatial relationships can lead to statistical problems, and may also prevent the discovery of important theoretical and practical insights for the study of criminal activity.
There are three fundamental spatial units of analysis: points, lines, and areas. In crime science, points (criminal events, for example) and areas (neighborhoods, for example) are the most common, but some research does consider line segments (street networks, for example). Though spatial-linear analyses were not undertaken, some research in the crime and place literature that uses line segments as the unit of analysis very clearly shows evidence for “linear clustering” of criminal events (Curman et al., 2012; Weisburd et al., 2012). When one considers that the activity patterns of individuals tend to follow major arterial roadways (Brantingham and Brantingham, 1981, 1993), it should come as no surprise that criminal activity tends to cluster along these roadways. Because of the dominance of points and areas in criminological research, we will focus on those two spatial units of analysis here.
The geographic element that is used most frequently within crime science is the hot spot map. Hot spot maps are generated using (criminal event) point data to create a surface over the entire study area representing the density of those points. Broadly speaking, the methods that can be used to create these hot spots are surface-generating techniques or spatial interpolation. The most common of these is called kernel density estimation.
The kernel density process is illustrated in Figure 1, considering theft from vehicle in Vancouver, Canada. Figure 1a is a dot map showing the locations of all the thefts from vehicle in a year. Because of the relatively low volume of this criminal event, approximately 10,000 events, a spatial pattern emerges over the city—a more common crime type may cover the entire city with dots, such that there is no discernable concentration of criminal events. The greatest density of theft from vehicle occurs in the northern peninsula of the city (within Vancouver’s central business district) and on the east side of the city; these areas represent a high density of targets, and relatively lower socio-economic status, respectively. Despite these densities, it should be clear that theft from vehicle occurs all over the city.
Figure 1b is a hot spot map created using the same data from Figure 1a through kernel density estimation. Figure 1b definitely shows a high density of theft from vehicle in the central business district area, but a relatively low density everywhere else. Though such a representation does show the overall pattern—almost the entire central business district is covered with dots—it gives the idea that theft from vehicle is not a problem for most of the city. In fact, this just means that the rest of the city has a relatively low density of theft from vehicle (when compared to the central business district) but it may still have a high volume of theft from vehicle. Such a potential problem in inference emerges because kernel density estimation provides a picture of crime based on comparative levels of concentration.
Figure 1. Theft from vehicle, dot map and hot spot map (Vancouver, 2001).
a) Dot map |
---|
b) Hot spot map |
Figure 2 clarifies why this generalization takes place. When making a kernel density map, the entire study area is covered in a grid; the size of the grid cells can be defined by the user, but tend to be rather small. Each one of these grid cells is represented by a location (point) s, which is the centroid of the cell. And at each one of these locations in the study area, a circle is drawn with a radius, or bandwidth, that is also defined by the user—most GIS software programs have default settings determined by the size of the study area. The number of events within the circle are counted and then used to calculate the “height” of the location, called the kernel. These values can also be used to create 3D maps. Two aspects of this estimation should be emphasized. First, as the bandwidth increases in size, more events will be captured within the circle to calculate the kernel. This creates a smoothing effect for the hot spot map. Consequently, hot spots will often appear to be bigger than they really are. Second, it is quite likely (perhaps even common) that no events occur close to the location (point) s, but the location is given a positive value due to a large default bandwidth setting. As a result, a location that is crime free will have a positive value because criminal events occurred “nearby”. One could argue that this must still be representing the reality of crime in the neighbourhood. However, we recently created a kernel density map with a default bandwidth of just under 600 meters. Though this could be classified as “close” it would be far enough away to have very different characteristics – this will be further discussed in the spatial scale and the crime at places section, below. These two characteristics of kernel density estimation emphasise the importance of careful consideration of both the default settings within GIS software programs, and of the generalized results of such mapping techniques.
Another issue that is often overlooked relates to the data used to calculate hot spot maps. Figure 3a is another kernel density hot spot map created using only criminal event data. This is the most common type of hot spot map. However, it is well accepted that there will be more criminal events where there are more targets, and people are an excellent representation of those targets. More people means more direct targets for violent crime and more indirect targets for property crime because people bring property with them wherever they go.
Figure 2. Kernel density estimation
Source. Andresen (2014).
Because only one variable was used to calculate the hotspot map displayed in Figure 3a, it is called a single kernel. Figure 3b, on the other hand, is called a dual kernel hot spot map because it considers two variables: criminal event data, and residential population count data from the census. The latter variable represents the population at risk of criminal victimization and should be considered as part of the calculation. Why? Because we know, and we have known for a long time, that more crime occurs where there are more targets, particularly in central business districts such as the dark area in Figure 3a (Andresen & Jenion, 2010; Boggs, 1965; Schmid 1960a, 1960b). One can think of Figure 3b as being the result of subtracting a kernel density map of the population at risk from the original criminal event kernel density map. With a dual kernel that considers the population at risk, Figure 3b shows that the hot spot is actually not as “bad” as it looked in Figure 3a. As such, Figure 3a is best thought of as a map showing the volume of criminal events whereas Figure 3b is best thought of as a map showing the risk of being a victim of a criminal event, similar to comparing criminal event counts to crime rates. This discussion should emphasize how geography can contribute to crime science not only through the application of geographical methods to criminal event data, but also through its underlying understanding of the spatial representations used in crime science.
Figure 3. Theft from auto, Single versus dual kernel density estimation (Vancouver, 2001)
a) Single kernel |
---|
b) Dual kernel |
Source. Andresen (2014).
In Figures 1 and 3 it is clear that locations that are close to one another have similar values. This is often referred to as spatial autocorrelation—a numerical representation is shown in Figure 4. Figures 1, 3, and 4 are examples of positive spatial autocorrelation: areas are surrounded by other areas with similar values; there is also the case of negative spatial autocorrelation: areas are surrounded by other areas with dissimilar values.
Though negative spatial autocorrelation does emerge in the social sciences and, consequently, crime science, positive spatial autocorrelation is the most common. Interesting in its own right, (positive) spatial autocorrelation is problematic for statistical analysis. Many statistical techniques require the independence of observations, so if you can predict the value for one spatial unit of analysis when you know its neighbour’s value, independence is violated.
Figure 4. An example of positive spatial autocorrelation
Source. Andresen (2014).
This issue is most manifest in the context of regression analysis. In the presence of positive spatial autocorrelation, ordinary least squares regression underestimates the standard errors for coefficients. Consequently, some explanatory variables may appear to be more significant than they really are. In purely academic circles, the consequences of retaining theoretically informed variables in a regression model are not particularly severe: theoreticians may think something matters more than it really does, if it matters at all. However, in the context of crime science, this could lead to the belief that a relationship exists when it does not and may lead to inappropriate crime prevention interventions. This undermines the credibility of crime science and wastes crime prevention resources. Spatial statistical methods that address this spatial autocorrelation are instructive here because they can filter out the problematic spatial autocorrelation and provide more reliable statistical results for the application of crime science. Two such methods include spatial lag models and spatial error models.
Though extremely useful and appropriate for the analysis of spatial data, techniques such as spatial lag and spatial error regression are “global” statistics, representing the entire study area. Fundamental to geography is the recognition that relationships are specific to places and vary across space. This does not mean that geographers do not believe in global processes, just that they must be interpreted with caution. It should therefore not come as a surprise that the local has become particularly important in the field of geography in recent decades (Fotheringham, 1997). There are many forms of local statistical analysis that could be utilized in crime science, but the statistics that are most commonly used in geography of crime studies relate to forms of spatial clustering (the recognition of positive and negative spatial autocorrelation occurring within the same study area) and geographically varying regression coefficients (also being positive in some places and negative in others).
The most commonly used spatial statistic for local spatial clustering is local Moran’s I, developed by Anselin (1995). Local Moran’s I is part of a subset of local spatial statistics called local indicators of spatial association (LISA) because it is related to a global statistic. Moran’s I, the global statistic, measures the presence of spatial autocorrelation (positive, negative, or neither) across an entire study area. Though instructive, and important for residuals in the context of a regression using spatial data, global patterns most often hide local patterns. Local spatial statistics produce a statistic for each spatial unit under analysis; as such, local Moran’s I generates a statistic that measures the degree of spatial autocorrelation for each unit as it relates to every other spatial unit of analysis. With a set of local statistics this information can be mapped to search for local patterns, such as clustering.
A map of local Moran’s I is presented in Figure 5, showing clusters of crime in the central business district of Vancouver: High-High refers to places with high crime surrounded by other areas with high crime, low-high refers to places with low crime surrounded by areas with high crime, and so on. Recalling from above (Figures 1 and 3), this is an area of high crime. However, this map shows that within this area of high crime there are pockets of low crime that do not emerge when using more conventional spatial analysis techniques, particularly kernel density estimation hot spot maps. The largest of these low crime areas, located close to the center of the map, is a relatively recent expensive and trendy development that has been able to repel crime within while being surrounded by older areas that historically have high levels of crime. It has been found that the areas exhibiting positive spatial autocorrelation (High-High and Low-Low) can be predicted from social disorganization theory (Sampson & Roves, 1989) and routine activity theory (Cohen & Felson, 1979), whereas the areas exhibiting negative spatial autocorrelation (Low-High and High-Low) tend to not follow these expectations (Andresen, 2011).
Figure 5. Theft from Auto, Local Moran’s I (Vancouver central business district, 2001).
Source. Andresen (2011, 2014).
The last spatial analysis method to be discussed here, though there are many more, is a locally-based regression technique called geographically weighted regression. As with local Moran’s I, geographically weighted regression recognizes that regression parameter estimates may vary across space. As such, with geographically weighted regression a map of parameter estimates is generated with each spatial unit of analysis having its own value (see Brunsdon et al., 1996 and Fotheringham et al., 2002). This local spatial analytic technique is particularly interesting because regression analysis is a common statistical method in crime science. However, geography tells us that while there may be global tendencies, there are almost always local variations that may prove to be instructive.
In an analysis of violent crime in Portland, Oregon, Cahill and Mulligan (2007) compared ordinary least squares results with geographically weighted regression results, arriving at some interesting findings. In short, these authors found that regression parameters do vary across space, in some cases rather extremely. Moreover, the use of geographically weighted regression was able to explain one counter-intuitive finding within the ordinary least squares results: affluent areas have higher levels of violence. When considering the geographically weighted regression results, these authors found that most of the city had either statistically non-significant relationships or negative relationships, as expected, but a small number (10 percent) of affluent areas had greater than expected levels of violent crime. These areas dominated the ordinary least squares regression, making the estimated parameter positive at a global level. Consequently, not only did the use of geographically weighted regression show that local parameter estimates can be statistically insignificant, negative, and positive within the same study space, it was also able to explain why counter-intuitive results emerged in a global statistical technique. This underscores why such a geographical (local) perspective should be considered important within crime science when aiming to understand spatial patterns and prevent criminal activity.
The importance of scale, its complications, and the understanding of place
It is almost tautological to state that one of the contributions of geography to crime science is the importance of scale. Scale, when considered as defining the nature of the data: points, lines, and areas, has implications for the method of (spatial) analysis and any subsequent interpretations. Though arguments can be made for which is the most ideal form of (spatial) data, usually micro-level data are better because they can be aggregated: address level data can be aggregated to neighbourhoods that can be aggregated into cities, depending on research needs. However, we are often circumscribed by the data provided to us by various agencies. As such, we need to be cognizant of the challenges that go along with spatial data, and of the corresponding limitations in any interpretations. These are, primarily, the ecological/atomistic fallacy and the modifiable areal unit problem (MAUP)—both of which are related to spatial heterogeneity.
The ecological fallacy can occur when research is undertaken that uses aggregate data such as neighborhoods, municipalities, states/provinces, or countries. The fallacy actually occurs when the researcher (or someone interpreting the research) makes assumptions about individual units (persons living within an area, for example) based on the statistical characteristics of their aggregated areas. In essence, the fallacy occurs when someone assumes that what is true of the whole is also true of its parts--the atomistic fallacy operates in the opposite direction. Technically, what is true of the whole is true, on average, for its parts, but to say that any one individual is the average can be quite a leap. The ecological fallacy is a fallacy, of course, because such a relationship cannot be assumed.
The ecological fallacy was first formally discussed by Robinson (1950) in the context of the census and the individuals on whom it was conducted.. More than three decades later, Openshaw (1984a) investigated the relationships, if any, between individual-level correlations and area-level correlations. Generally speaking, Openshaw (1984a) found that the severity of any problems with interpretation that would emerge depended on the particular method of analysis. Further, he found that the differences in the individual-level and area-level correlations cannot be known before the analysis, and any subsequent comparisons are undertaken. In a study of two census units of analysis in each of Vancouver, Canada and Leeds, England, Andresen and Malleson (2013) most often found that there was a small set of geographically smaller units driving the change for the geographically larger units. This confirmed that what is true of the whole is not necessarily true of all of its parts. However, they found in a number of cases that (statistically speaking) none of the geographically smaller units of analysis exhibited the same change as the geographically larger units of analysis.
Though the ecological fallacy is easy to avoid, in theory (never make inferences at a level different from your unit of analysis), it is a very easy fallacy to commit when interpreting data. This highlights the importance of understanding scale when undertaking (spatial) analysis.
The modifiable areal unit problem (MAUP), though formally defined and discussed by Openshaw (1984b), was first identified by Gehlke and Biehl (1934). The MAUP is an issue that emerges in the analysis of spatial data because of the arbitrary nature in which micro-level (individuals or individual households) data are often aggregated into areas. This is most common in the analysis of census data in census tracts, block groups, and output areas. The MAUP has two common forms: the scale problem and the zoning problem. The scale problem manifests itself when the same analysis is undertaken at two different scales: census tracts and neighborhoods, for example. The zoning problem manifests itself when the size and shape of the spatial units of analysis are the same, but they are placed differently on the study area.
Though there is no intrinsic problem with changing the scale or zoning in a spatial analysis, research has consistently shown that when the spatial units of analysis are changed in some manner, the statistical relationships between the variables under analysis change. Typically, as found by Gehlke and Biehl (1934), as the geographical size of the units of analysis increases so does the strength of the correlation between those variables. However, more troubling is the research of Fotheringham and Wong (1991) who found in a regression context almost any result could be generated using different aggregations of data. This is highly problematic because most research only considers one spatial unit of analysis. Is any given piece of research using the “correct” geographical aggregation, by chance, or it is using one of the many others producing spurious results? Basic probabilities do not bode well here.
However, in the context of geographical criminology research, Wooldredge (2002) has found that the geographic size of the spatial units of analysis matters very little, generating very similar results considering multiple scales of analysis. The limitation with Wooldredge (2002), however, is that his analyses were only a comparison of a few aggregations in one location. Perhaps all census-based and neighbourhood-based statistical results are similar, but all incorrect. Geography has a lot to add to crime science with regard to the unit of analysis when investigating crime patterns.
Recognizing the importance of scale in understanding spatial crime patterns, it is no surprise that concern over the appropriate spatial unit of analysis goes back almost 200 years to the work of Adolphe Quetelet and André-Michel Guerry. This concern is rooted in the finding that when geographically smaller units are analyzed, researchers find spatial heterogeneity: a province or state may have a high crime rate relative to the rest of the country, but the municipalities within that province or state vary significantly. This phenomenon has been illustrated indirectly through a series of different research studies analyzing different spatial units of analysis, and directly within one research study conducted by Brantingham and colleagues (1976). In their study, the authors analyzed crime patterns at the national, state, municipality, census tract, and block group levels, each time finding significant variations in crime rates within the geographically smaller units of analysis. Because of this spatial heterogeneity, the trajectory over the past 200 or so years has been toward increasingly smaller spatial units of analysis (Weisburd et al., 2009a).
The more recent crime and place literature emphasizes the importance of spatial scale and the understanding of place. Crime and place research considers the micro-spatial unit of analysis, often street segments, intersections, and specific addresses. Weisburd and colleagues (2012) have argued that the street segment—both sides of the street between two intersections—is optimal for understanding geographical crime patterns because it is small enough to avoid significant concerns regarding spatial heterogeneity and large enough to gather data on for subsequent analysis. The first citywide crime and place study, conducted by Sherman and colleagues (1989), found that approximately 5 percent of street segments account for approximately 50 percent of crime, or calls for police service. This statistic has been replicated in a number of other contexts in different countries around the world (Andresen & Linning, 2012; Andresen & Malleson, 2011; Curman et al., 2015; Groff et al., 2010; Melo et al., 2015; Weisburd, 2015; Weisburd et al., 2004, 2009b, 2012).
One criticism of this statistic is that when one investigates crime patterns at the street segment level for an entire municipality, one will almost certainly find concentrations. This is because there will be thousands (even tens of thousands) of street segments and likely fewer criminal events. If a researcher is analyzing 1000 criminal events on 10,000 street segments, at the very least (assuming no clustering of crime) 90 percent of the municipality will be free from these criminal events. Though this must be kept in mind when interpreting any such statistics, it is still meaningful because the researcher can know that very few places actually have crime. However, research that has considered only those street segments that experience crime still finds high degrees of concentration within that subset of street segments (Andresen & Linning, 2012; Andresen & Malleson, 2011; Melo et al., 2015).
Perhaps most interesting is that some of this research has found place to be critical for understanding crime patterns. For example, Groff and colleagues (2010) found that the trajectories of street segments2 can vary considerably from street segment to street segment. Moreover, from a place perspective, the street segment may even be considered too large to really understand the underlying crime patterns. Why would two contiguous street segments have completely different trajectories when one of the long-standing facts within the geography of crime research is that crime clusters in space? The short answer is that it depends what is on each of those street segments. One street segment with nothing but residential housing and another that has a convenience store or a drinking establishment will have very different crime patterns, whether they’re geographically close or not. In fact, most of the street segment containing the convenience store or drinking establishment will likely be “crime free”—barring any spillover effects—and, therefore, similar to its neighbouring street segment. As such, understanding the geography of crime makes understanding the importance of place and local context critical.
Policy mobilities and crime prevention
The reduction and/or prevention of crime is the fundamental goal of crime science, and much of what we know regarding practical crime prevention today is rooted explicitly or implicitly in situational crime prevention (Clarke 1980, 1983, 2012). Situational crime prevention recognizes the importance of the situation at hand when designing a specific crime prevention application. In geography, the equivalent terminology would be understanding the “local”, emphasizing the importance of local spatial analysis, as discussed above.
Situational crime prevention applies to the crime types that one is trying to prevent or reduce, the location(s) where crime is occurring, when those crimes are occurring, and so on. Situational crime prevention is not without its critiques (Wortley, 2010), but is an integral component of crime science due to its focus on prevention (see Clarke 1997, 2012) and its recognition of the importance of the local when implementing initiatives. What works in one place may not work in another, and what works at one time may not work at another. Quite simply, different crime types, locations, and times have different opportunities and will attract different offenders. Consequently, without a focus on the local/situational factors, crime prevention initiatives are unlikely to be successful. But is this focus on the local restricted to the crime prevention initiatives themselves? We would argue no.
Over the past decade in human geography, a new branch of literature has developed that recognizes the importance of the local when developing policy. This literature, urban policy mobility, has shown that (urban) policy has a number of factors that affect its successful implementation: spatial scales, communities, and institutions (McCann, 2008, 2011; Peck, 2011; Temenos and McCann, 2012, 2013). In other words, urban policy is highly situational. This means that any policy aimed at the reduction of crime is a geographical process because of the varying social, political, and crime cultures in a place (McCann, 2013; McCann and Temenos, 2015; Robinson, 2011; Ward, 2006); (crime prevention) policy itself, therefore, moves through and is shaped by the local. From a geographical perspective, understanding urban local policy mobility becomes critical for knowing if and when any global crime prevention policy can be applied in a local context.
The urban policy mobility literature shows that the situational component of situational crime prevention not only operates for crime prevention initiatives, but at the level where the crime prevention policies are developed. We need to conduct investigations into local conditions to not only design successful crime prevention initiatives, but also to design the crime prevention policies considering local conditions so that they may be applied in their respective locations with success. Though such a statement may appear trite, its importance and relevance becomes clear when we recognize that crime prevention policies are often set at the national, state/provincial, and municipal levels. However, recall from the discussion of scale and place above that crime can vary from street block to street block. When success is measured against nationally-defined performance measures and crime patterns change on the next block, sustained support for crime prevention activities in the context of crime science, despite its empirical support, may lose traction. As such, crime prevention policy is empty without considering the local and the ability of the policy to be mobile to particular places.
Conclusion
Geography has informed criminological research for nearly two centuries, emphasizing the importance of considering space when investigating crime. This spatial perspective is particularly relevant within crime science. Geography has offered a number of concepts and approaches to the field of crime science; within this chapter we have focused on three such contributions.
Crime science has adopted a variety of spatial analytical approaches for representing and understanding crime events. From hotspot mapping to spatial statistics, geography has emphasized the need for special consideration of space and place, particularly when studying human behaviour. One must pay careful attention to the spatial units of analysis (points, lines and areas) as well as to scale in order to limit the likelihood of spurious analytical results. Such attention is particularly important when considering the policy implications, where results translate to actionable crime reduction approaches. Along this vein, policy mobilities literature guides the successful implementation of crime reduction policy by re-emphasizing the importance of adopting a local focus at all stages of investigation – from the spatial crime analysis, to the policy implementation, and into the development of crime prevention initiatives.
References
Andresen, M. A. (2011). Estimating the probability of local crime clusters: The
impact of immediate spatial neighbors. Journal of Criminal Justice, 39(5), 394
- 404
.Andresen, M. A. (2014). Environmental criminology: Evolution, theory, and practice.
New York, NY: Routledge.
Andresen, M. A. and Jenion, G. W. (2010). Ambient populations and the calculation of
crime rates and risk. Security Journal, 23(2), 114 - 133.
Andresen, M. A., & Linning, S. J. (2012). The (in)appropriateness of aggregating
across crime types. Applied Geography, 35(1-2), 275 - 282.
Andresen, M. A., & Malleson, N. (2011). Testing the stability of crime patterns:
Implications for theory and policy. Journal of Research in Crime and
Delinquency, 48(1), 58 - 82.
Andresen, M. A., & Malleson, N. (2013). Spatial heterogeneity in crime analysis. In
M. Leitner (ed), Crime modeling and mapping using geospatial technologies
(pp. 3 – 23). New York, NY: Springer.
Analysis, 27(2), 93 – 115.
Boggs, S. L. (1965). Urban crime patterns. American Sociological Review, 30(6), 899 –
908.
Brantingham, P. L. and Brantingham, P. J. (1981). Notes on the geometry of crime. In
P. J. Brantingham and P. L. Brantingham (eds), Environmental criminology
(pp. 27 – 54). Prospect Heights IL, Waveland Press.
Brantingham, P. L., and Brantingham, P. J. (1993). Nodes, paths and edges:
Considerations on the complexity of crime and the physical environment.
Journal of Environmental Psychology, 13(1), 3 – 28.
Brantingham, P. J., Dyreson, D. A., & Brantingham, P. L. (1976). Crime seen through
a cone of resolution. American Behavioral Scientist, 20(2), 261 – 273.
Brunsdon, C. F., Fotheringham, A. S., & Charlton, M. E. (1996). Geographically
weighted regression: A method for exploring spatial non-stationarity.
Geographical Analysis, 28(4), 281 - 298.
Cahill, M., & Mulligan, G. (2007). Using geographically weighted regression to
explore local crime patterns. Social Science Computer Review, 25(2), 174 –
193.
Clarke, R. V. G. (1980). Situational crime prevention: Theory and practice. British
Journal of Criminology, 20(2), 136 – 147.
Clarke, R. V. (1983). Situational crime prevention: Its theoretical basis and practical
scope. Crime and Justice: An Annual Review of Research, 4, 225 – 256.
Clarke, R. V. (1997). Situational Crime Prevention: Successful case studies (2nd ed.).
Monsey, NY: Criminal Justice Press.
Clarke, R. V. (2012). Opportunity makes the thief. Really? And so what? Crime
Science, 1, Article 3.
Cohen, L. E., & Felson, M. (1979). Social change and crime rate trends: A routine activity approach. American sociological review, 588-608.
Curman, A.S.N., Andresen, M.A., & Brantingham, P.J. (2015). Crime and place: a
longitudinal examination of street segment patterns in Vancouver, BC.
Journal of Quantitative Criminology, 31(1), 127 - 147.
Fotheringham, A. S. (1997). Trends in quantitative methods I: Stressing the local.
Progress in Human Geography, 21(1), 88 – 96.
Fotheringham, A. S., Brundson, C., & Charlton, M. (2002). Geographically weighted
regression: The analysis of spatially varying relationships. Chichester, UK: John
Wiley & Sons.
Fotheringham, A. S., & Wong, D. W. S. (1991). The modifiable areal unit problem in
multivariate statistical analysis. Environment and Planning A, 23(7), 1025 – 1044.
Gehlke, C. E. and Biehl, H. (1934). Certain effects of grouping upon the size of the
correlation coefficient in census tract material. Journal of the American
Statistical Association, Supplement, 29(185), 169 – 170.
Groff, E. R., Weisburd, D., & Yang, S-M. (2010). Is it important to examine crime
trends at a local “micro” level? A longitudinal analysis of street to street
variability in crime trajectories. Journal of Quantitative Criminology, 26(1), 7
– 32.
Guerry, A-M. (1832). La statistique compare de l’état de l’instruction et du nombre
des crimes. Revue Encyclopédique, 55, 414 – 424.
Guerry, A-M. (1833). Essai sur la statistique morale de la France. Paris: Crochard.
McCann, E. (2008). Expertise, truth, and urban policy mobilities: Global circuits of
knowledge in the development of Vancouver, Canada’s “four pillar” drug strategy.
Environment and Planning A, 40(4), 885 – 904.
McCann, E. (2011). Urban policy mobilities and global circuits of knowledge: Toward a
research agenda. Annals of the Association of American Geographers, 101(1), 107 –
130.
McCann, E. (2013). Policy boosterism, policy mobilities, and the extrospective city. Urban
Geography, 34(1), 5 – 29.
McCann, E., & Temenos, C. (2015). Mobilizing drug consumption rooms: Inter-place
connections and the politics of harm reduction drug policy. Health & Place, 31(1),
216 – 223.
Melo, S. N., Matias, L. F., & Andresen, M. A. (2015). Crime concentrations and
similarities in spatial crime patterns in a Brazilian context. Applied
Geography, 62, 314 - 324.
Openshaw, S. (1984a). Ecological fallacies and the analysis of areal census data.
Environment and Planning A, 16(1), 17 – 31.
Openshaw, S. (1984b). The modifiable areal unit problem. CATMOG (Concepts and
Techniques in Modern Geography) 38. Norwich: Geo Books.
Peck, J. (2011). Geographies of policy: From transfer-diffusion to mobility-mutation.
Progress in Human Geography, 35(6), 773 – 797.
Quetelet, L. A. J. ([1831] 1984). Research on the propensity for crime at different ages
(translated by S. F. Sylvester). Cincinnati, OH: Anderson Publishing.
Quetelet, L. A. J. (1842). A Treatise on man and the development of his faculties.
Edinburgh: W. and R. Chambers.
Robinson, W. S. (1950). Ecological correlations and the behavior of individuals.
American Sociological Review, 15(3), 351 – 357.
Robinson, J. (2011). 2010 Urban Geography Plenary Lecture—The travels of urban
neoliberalism: Taking stock of the internationalization of urban theory. Urban
Geography, 32(8), 1087 – 1109.
Sampson, R. J., & Groves, W. B. (1989). Community structure and crime: Testing social-disorganization theory. American journal of sociology, 94(4), 774-802.
Schmid, C. F. (1960a). Urban crime areas: Part I. American Sociological Review, 25(4),
527 – 542.
Schmid, C. F. (1960b). Urban crime areas: Part II. American Sociological Review,
25(5), 655 – 678.
Sherman, L. W., Gartin, P., & Buerger, M. E. (1989). Hot spots of predatory crime:
Routine activities and the criminology of place. Criminology, 27(1), 27 – 55.
Temenos, C., & McCann, E. (2012). The local politics of policy mobility: The education of
attention in developing a municipal sustainability fix. Environment and Planning A,
44(6), 1389 – 1406.
Temenos, C., & McCann, E. (2013). Geographies of policy mobilities. Geography Compass,
7(5), 344 – 357.
Ward, K. (2006). Policies in motion, urban management and state restructuring: The trans-
local expansion of business improvement districts. International Journal of Urban
and Regional Research, 30(1), 54 – 75.
Weisburd, D. (2015). The law of crime concentration and the criminology of place.
Criminology, 53(2), 133 – 157.
Weisburd, D., Bushway, S., Lum, C., & Yang, S-M. (2004). Trajectories of crime at
places: A longitudinal study of street segments in the City of Seattle.
Criminology, 42(2), 283 – 321.
Weisburd, D., Bruinsma, G. J. N., & Bernasco, W. (2009a). Units of analysis in
geographic criminology: Historical development, critical issues, and open
questions. In D. Weisburd, W. Bernasco and G. J. N. Bruinsma (eds), Putting
crime in its place: Units of analysis in geographic criminology (pp. 3 – 31). New
York, NY: Springer.
Weisburd, D., Morris, N. A., & Groff, E. R. (2009b). Hot spots of juvenile crime: A
longitudinal study of arrest incidents at street segments in Seattle,
Washington. Journal of Quantitative Criminology, 25(4), 443 – 467.
Weisburd, D., Groff, E. R., & Yang, S-M. (2012). The criminology of place: Street
segments and our understanding of the crime problem. New York, NY: Oxford
University Press.
Wooldredge, J. (2002). Examining the (ir)relevance of aggregation bias for
multilevel studies of neighborhoods and crime with an example of comparing
census tracts to official neighborhoods in Cincinnati. Criminology, 40(3), 681
– 709.
Wortley, R. (2010). Critiques of situational crime prevention. In B. Fisher & S. Lab
(eds), Encyclopedia of victimology and crime prevention (pp. 885 – 888).
Thousand Oaks, CA: Sage Publications.