The Edge Effect:
Exploring High Crime Zones near Residential Neighborhoods
Justin Song, Valerie Spicer, and Patricia Brantingham
Institute of Canadian Research Studies (ICURS)
Simon Fraser University
Burnaby, BC, Canada
Abstract—Crime Pattern Theory offers a theoretical mostly occur in these awareness spaces or at the
framework for a micro level explanation of the dynamics of crime
in an urban environment. The research focuses on the novel use attractors [1], [2], [7], [9].
the concept of boundaries or edges to analyze micro level Research has also found that people in general and
differences in crime in various urban neighborhoods. Edges are offenders tend to find routine navigation routes along
identified where there is dissimilarity between adjacent areas.
Crime is more likely to occur along edges. Edges can be physical, major roads or public transit routes. Crime tends to
social, temporal and economical or a product of these dimensions. be highest along these routes or at, or near, common
This study compares crime data from the Municipality of Burnaby destination points like shopping areas, large schools,
in British Columbia, Canada with patterns in land use data. Single
family residential neighborhoods are constructed by joining parks, or clusters of working locations. In turn,
adjacently zoned single family areas. The edges of these zoning patterns for land uses associated with higher
neighborhoods are the areas where the single family zoning use tend to be placed near major travel routes, that is
changes to commercial, parks and higher density residential
zoning The results finds crime is sixty four percent higher in these areas that are easily accessible. Commercially zoned
edges than in the interior of the neighborhoods. The results are areas are near major streets or transit; industrially
discussed and future research proposed to repeat and enhance the zoned areas are easily accessible; large senior level
model.
schools serve multiple neighborhoods and frequently
Keywords: crime-pattern recognition – GIS techniques – crime are in high access locations. The high access attribute
analysis continues for drinking establishments, movie
INTRODUCTION
I. theaters, shopping malls and high use parks. With
that crime concentrates near these high access
Crime Pattern Theory [1], [2] is tested using locations [1], [2], [8], [9].
computational criminology, an emerging An important question is whether these crime
multidisciplinary field that uses criminology, concentrations expand or permeate into the
geography, computer science and mathematics to surrounding, less accessible areas. This paper
better understand patterns of criminal events, presents the results of a computational criminology
offender target choices and social networking of co- approach to explore this micro level crime spread
offenders. Analysis is frequently done through from high activity areas to surrounding ones. This is
spatial-temporal algorithms developed to describe not done through a diffusion model but through the
the city’s environment and, using these backcloth use of knowledge about human navigation in the
structures, identify features to predict where crimes urban environment and the cognitive images created
are likely to occur [3]. for urban environments [5], [6].
Crime pattern research has found that offenders The permeability of the edge or boundary of single
appear to navigate through an urban setting between family residential areas is the focus of the research.
crime attractors and along major roads. There are It is expected that crime will move from higher
basic routine travel patterns that develop between concentrations of crime along major access routes,
home, work or school, shopping and entertainment. but not far beyond the edge into the surrounding area
These locations and the routine pathways between and less towards the interior of the surrounding
these locations form an awareness space. Crimes neighborhoods. Edges from spatially distinct single
family dwellings are identified for a major suburb of
978-1-4673-6213-9/13/$31.00 ©2013 IEEE 245 ISI 2013, June 4-7, 2013, Seattle, Washington, USA
the City of Vancouver, Canada. Buffers of varying Paths: These channels designed for movement such as streets,
sizes are used to test the permeability of the single sidewalks, bike paths, or transit lines. Major paths would carry
large numbers of people whereas minor ones have lower flow
family areas to crime. of people.
The purpose of this study is to explore in some
detail the differences in the edges based on land use Edges: These are barriers between two places and vary in how
classifications. While there are many other types of penetrable they are. For instance, a river is very impenetrable
edges based on socio-economic or even architectural whereas a change in land-use is less of a physical barrier, but
may create a mental barrier.
design that could impact their behavior, the first
iteration of this type of research focuses on the Nodes: These are geographic points where activity
dynamics of land-use changes, a highly visible concentrates. Nodes can be the crossing of two paths or large
change within a city. Future studies will bring in people attractors in a city. Nodes are connected to paths and are
other dynamics into play within these spaces to generally predominant features in perceptual images of the city.
explore core attributes. Districts: These are sections of the city and each one has its
own identifying character. Cities are structured through districts
and while these districts hold similar geographical definitions
II. THEORETICAL REVIEW from a general perspective, each individual person will place
A. Crime Pattern Theory these in a different hierarchical order depending on their own
personal experience.
In 1993, Crime Pattern Theory [2] was formalized
through four assumptions about crime. In a first Landmarks: These are reference points that are physical
instance crime events are described as complex objects such as large building or mountains in the environment.
These can be well-known landmarks and can symbolize a
requiring various elements to occur. Second, crime is
direction.
not random due to the non-uniform spatial and
temporal distribution of motivated offenders and Couclelis, Golledge, Gale and Tobler [6], build
targets. Third, offenders and victims use the the anchor-point theory on these elements
environment as anyone else does and develop routine established by Lynch [5] to show how space is
movement patterns. Through this routine activity conceptually clustered, and guides judgment about
they develop an awareness space of limited parts of space. Individual cognitive maps evolve according to
a city. Motivated offenders are more likely to Lynch’s principles. However, where landmarks are
commit crime within the sphere of their normal both individually and collectively experienced,
activity space. Finally, motivated offenders, through anchor-points are only found in individual cognitive
daily routines and activities combined with their maps. These are places like home or work and are
social and physical interactions with the unrelated to the collective because these would be
environment, over time will develop a stable specific to the person forming the cognitive map.
offending template. Brantingham and Brantingham [2] use the
B. Structual Elements in Urban Environment geographical concepts of nodes, edges, paths and
In the Image of the City, Lynch [5] describes how anchor-points to explain, in part, the occurrence of
cognitive images of the city are created and related crime events. Offenders, in the sense of learning
to way-finding. Images are developed through an about space and developing routines, follow similar
interactive process between the individual and the functions to non-offenders. And most offenders
environment. These images of the city are individual, spend a large portion of their time in non-offending
but can resemble the image of other people who may activity. Space in these theoretical approaches is
share experiences of the environment, live in similar structured in such a way that crime can be analyzed
areas or are attracted to similar activity areas. The and predicted both at the individual and aggregate
city image is structured through constructed using level. Cognitive image-making ties into Crime
major or minor elements: paths, edges, nodes, Pattern Theory as activity spaces follow these
districts and landmarks. geographical principles [7], [8].
246
C. Edges and Crime Occurrences Research Studies (ICURS) at Simon Fraser
Edges occur between two distinct spaces defined University. PIRS contains officially reported crime
by a variety of characteristics including physical, dataset events for Royal Canadian Mounted Police
social or economical attributes. The temporal and (RCMP) jurisdictions in British Columbia. This
physical permeability of these edges also vary dataset contains about 4.4 million crime events.
depending on a number of factors. Edges also range There are 43,225 crime events from the middle of
from sharp and well defined, to diffused and 2001 to the middle of 2006 in the city of Burnaby,
progressive. the study area in this study. These data contain
For example, a river running through a city creates attributes about the crime event such as date, time,
a very defined edge between two shores. Whereas a location, offender, and specific crime type. This data
commercial strip which transitions to mixed-use does not contain attributes of residents of the
structures to multiple dwelling residential buildings locations victimized.
to single family dwellings represents a more
progressive transition. Furthermore, major pathways Land use Data: In 2001, Metro Vancouver
such as major roads, highways or mass transportation published the land use dataset used in this study. It is
networks, such as light rail systems, form edges that classified as 14 land use categories: Agricultural,
can either bisect neighborhoods or define Commercial, Commercial – Residential Mixed,
communities. Industrial, Institutional, Lakes and Water Bodies,
From a perceptual perspective, edges emanate Open and Undeveloped, Recreation and Protected
territorial cues that reduce spatial ownership, Natural Area, Residential – High Rise Apartments,
increase potential conflicts and decrease feelings of Residential – Rural, Residential – Single Family,
safety. For these reasons, crime events are more Residential – Townhouse and Low rise Apartments,
likely to occur in these areas. Furthermore, edges and Transportation, Communication and Utilities.
often contain a variation and concentration of
criminal opportunities. In this study, the Residential – Single Family class is
analyzed in detail. The area of this land use type is
D. Summary of Theoretical Framework 33.6 square kilometers and covers 33.6 % of the
Crime Pattern Theory provides an environmental study area.
context for crime events. The non-uniform
distribution of crime is connected to the physical and British Columbia Assessment Authority (BCAA)
cognitive structure of the urban domain. Physical, data: 32,375 single family dwelling addresses were
social and economical factors guide how people used to identify the exact x, y coordinates of those
interact within the urban setting. In particular, the dwellings. They were used to calculate the crime
dynamics of edges can create high crime areas. density (crimes per dwellings).
Therefore the continued exploration of edges is an
important contribution to the theoretical GIS Innovation data: The 2006 road network data
understanding of these spaces. from the GIS Innovation was used to geocode crime
event locations. This road network data was also
III. METHODS used to visualize the output results.
A. Data B. Geocoding
Four data sources are integrated in this study: Crime events records retrieved from PIRS have an
Police Information Retrieval System (PIRS), Land address field that includes the street address of the
use data, British Columbia Assessment Authority crime location. The street addresses are converted to
(BCAA) data, and GIS Innovation data. geographic coordinates through Geocoding. For
example, a street address, 1020 North Road, could be
PIRS: Crime Data-Warehouse (CDW) is a collection geocoded somewhere at the road segment with the
of datasets located at the Institute of Canadian Urban starting and ending house numbers of 1000 and 1098
247
respectively. The exact location is therefore interior areas as light blue with the edges as white
determined through the interpolation of house bands surrounding these residential interiors.
number and street address. The side offset option in
geocoding was set to five meters.
C. Identifying the Edge and the Interior
Land use type classified as “Residential - Single
Family” was extracted and defined as either interior
or edge. The edge was identified within each
Residential – Single Family zone and was a buffer
area with a width of 0 to 100 meters inward from the
outer boundary of the Residential – Single Family
zones. The interior was the complementary area not
part of the edge buffer.
Once the interior and the edge were identified, the Figure 2: Study Area with Edges and Interior (100 meters Buffer)
edge was defined more thoroughly to determine the
relationship between the edge type and adjacent land D. Crime Density Calculation
use type. Edge – adjacent to Commercial zone, Edge The BCAA data was used to calculate the crime
– adjacent to Residential Multiple Family density. BCAA data is point data representing each
(Residential – Townhouse and Low rise lot. Once the buffer zone was established, including
Apartments), Edge – adjacent to Institutional, Edge the interior and exterior boundaries, the housing unit
– adjacent to Natural area (Open and Undeveloped, density count for BCAA data was calculated. Crime
Recreation and Protected Natural Area), and Edge – count data was then divided by this BCAA data in
adjacent to Industrial are defined by their adjacent order to establish data crime density value.
land use zone. For example in figure 1, the Edge –
E. Iterative Process – Re-defining the Buffer Zone
adjacent to Commercial was identified in relation to
the residential interior area and the adjacent The experiment was run to clarify crime patterns in
commercial zone. residential neighborhoods, and more specifically
compare crime density in the edge of single dwelling
neighborhood compared to the interior aspect of
these areas. Therefore, defining the edge and interior
is a critical aspect of this experiment.
To further identify the characteristics of the edge, a
crime density decay graph was used. First, the
shortest distance between the crime location and the
boundary of the Residential Single Family zone was
calculated for each crime occurred in the Residential
Single Family zone of the study area.
The distance between a point (c) and a boundary set
Figure 1: Identifying Interior and Edge (B) is the infimum of the distances between the point
and the polygon that forms the boundary of the set.
This process was replicated over the entire study area
with both the interior residential areas identified and d(c, B) = inf {di(c,b)}, where b אB
the edges to each of these interior spaces defined.
The map of the study area in figure 2 shows the A polygon is a collection of finite polylines;
therefore, the distance between a point (c) and a
248
polygon could be written as the minimum of the calculated for the number of crimes per the number
distances between a point (c) and a collection of of dwellings and plotted as shown in figure 5.
finite polylines.
d(c, B) = min {di(c, b)}, where b אB
If the perpendicular line is within the polyline, as
shown on the left in the figure 3, the shortest distance
between the point c(x1, y1) and the polyline is the
perpendicular distance. If a perpendicular line is
outside of the polyline, as shown on the right in the
figure 3, the shortest distance is the distance between
the point c(x1, y1) and the closest vertex. [4]
Figure 5 Crime Decay
ȋ1ǡ1Ȍ ȋ1ǡ1Ȍ The peak value of crime density was found around
the 10 meters from the outer buffer boundary and
drops rapidly at 30 meters. Also, the density value
found steady after 30 meters. The results of this
experiment are taken from within the 30 meters
buffer zone.
RESULTS
IV.
Figure 3: Distance Between a Point and a Polyline The crime density in the homogenous interior and the
30 meter edge were calculated by crime/dwelling
After the distances between the crime point and the density. Table 1 shows how the crime density in the
boundary was calculated, crime density was edge is higher than that of homogenous interior with
calculated every five meters in the buffer zone from a 64% difference.
the outer boundary of the Residential - Single Family
Table 1: Interior and Edge Comparison
zone inward. The circles in figure 4 show the location Residential Crime Density
Dwellings Crimes Count
of dwellings and the red circles indicate the crime (Single Family) (crimes / dwellings)
locations. Interior 26,110 30,931 1.18
Edge 6,264 12,173 1.94
Further analysis was conducted to determine how the
interior area interacts with different types of land use.
The edge where there was the highest crime was the
Utility area with crimes being 187.9% higher in the
edge that in the interior. Conversely, the edge with
the lowest impact is Multiple Family edge which was
only 20% higher than the interior. The Commercial
edge contained 45.6% more crime than the interior,
the Institutional edge, 105.6%, the Industrial edge
Figure 4 Residential Dwellings and Crime Locations contained 127.7% and the Natural Area edge 69.2%
more crime than the interior. Table 2 displays the
In order to calculate crime decay, the crime density crime distribution in these areas.
for each successive five meter buffer zone was
Table 2: Crime in Edges
249
Crimes Crime Density be generalized to all jurisdictions. However, this
Edge Dwellings
Count (crimes / dwellings)
research does explore the intricacies at play within
Edge - Commercial
3,410 5,676 1.66 edges with strong results encouraging future research
Edge - Multiple
Family 5,010 7,885 1.57 in other jurisdiction to verify whether land-use
Edge - Institutional impacts edges in a similar manner.
5,197 8,464 1.63
Research on edges is important to civic agencies
Edge - Natural Area
6,215 8,492 1.37 engaged in crime prevention. While calibration
Edge - Industrial
1,199 2,485 2.07 would be necessary in a different city, should the
Edge - Utility
pattern be repeated in a similar form, it would outline
1,016 1,716 1.69
the importance of considering and focusing physical
DISCUSSION
V. crime prevention in the possibly narrow high crime
buffer zones surrounding residential neighborhoods.
The results of this experiment support the Crime
Resources could be maximized in these edges to
Pattern Theory in a number of ways. First, crime
reduce crime opportunities while maximizing
events are complex as these relate to urban dynamics.
offender apprehension. Future land use decisions
Second, the crime events explored here are not
could also be calibrated to incorporate these concepts
uniformly distributed in the edges. Some edges have
and reduce high crime edges through alternate
significantly higher levels of crime. This could be
development decisions. Future research will
attributed to both the distribution of targets and the
concentrate of delving further into these edges to
specific nature of the edge. Third, the variation in
discover the intricacies of the edge effect.
crime distribution may be attributed to the movement
patterns within these edges. Finally, the daily REFERENCES
routines that occur in these edges may be extremely [1] P.L. Brantingham and P.J. Brantingham, Notes on the geometry of crime,
varied both physically and temporally. In P.J. Brantingham and P.L. Brantingham (Eds.), Environmental
Criminology, Sage Publications, Beverly Hills, CA, (1981).
This research further supports the edge theory, [2] P.L. Brantingham and P.J. Brantingham, Environment, routine, and
with crime higher on the border/edge than the situation: Toward a pattern theory of crime, In R.V. Clarke and M. Felson
(Eds.), Routine Activity and Rational Choice: Advances in
interior. The results show that there is 64% higher Criminological Theory, (5th Ed.), Transaction Publishers, New
crime in the overall edge area compared to the Brunswick, NJ, (1993).
[3] P. L. Brantingham, M. Ester, R. Frank, U. Glässer, and M. A. Tayebi. Co-
interior. Furthermore, the variation of the edge offending network mining. Counterterrorism and Open Source
comes to the forefront. In this research, the highest Intelligence, (2011), 73-102.
crime edge is the Industrial one and the lowest is the [4] ESRI, ArcGIS 10.1 Help, http://resources.arcgis.com/en/help/main/10.1
/index.html#//000800000048000000.
Multiple Family edge. This initial research combined [5] K. Lynch, The Image of the City, The MIT Press, Cambridge,
crime types. Future research into edges will Massachusetts (1960).
[6] H. Couclelis, R.G. Golledge, N. Gale and W. Tobler, Exploring the
concentrate on the variation by crime type that occurs Anchor-Point Hypothesis of Spatial Cognition. Journal of Environmental
in these spaces (eg: property crime, violent offenses Psychology, 7(1) (1987), 99-122.
or public disorder). Similarly this initial study [7] W. Bernasco and F. Luykx, Effects of Attractiveness, Opportunity and
Accessibility to Burglars on Residential Burglary Rates of Urban
focused on the difference in edges by land use and Neighborhoods, Criminology, 41 (2003), 981–1001.
attributes available in the data sets. Future analysis [8] A. Malms, Environmental Criminology in Francis T. Cullen & Pamela
Wilcox (eds) Encyclopedia of Criminological Theory. Sage Publications,
will include, with the acquisition of the population Beverly Hills, CA (2010).
data, even greater distinctions including factors such [9] A. Reid, R. Frank, N. Iwanski, V. Dabbaghian, and P. Brantingham,
Uncovering the Spatial Patterning of Crimes: A Criminal Movement
as income, education levels and ethnicity. Model (CriMM), Journal of Research in Crime and Delinquency, (2013).
This study is limited to one jurisdiction in
metropolitan Vancouver as such these results cannot
250
