Crime Pattern Theory

Description

Summarize crime pattern theory and the two assigned empirical studies. In doing so, be sure to discuss the findings from each study in relation to crime pattern theory. Answer must be 2.5 pages, essay format, also APA format. I have attached the articles needed to answer the question.

Unformatted Attachment Preview

Journal of Research in Crime
andhttp://jrc.sagepub.com/
Delinquency
The Variable Impacts of Public Housing Community Proximity on
Nearby Street Robberies
Cory P. Haberman, Elizabeth R. Groff and Ralph B. Taylor
Journal of Research in Crime and Delinquency 2013 50: 163 originally published
online 8 November 2011
DOI: 10.1177/0022427811426335
The online version of this article can be found at:
http://jrc.sagepub.com/content/50/2/163
Published by:
http://www.sagepublications.com
On behalf of:
John Jay College of Criminal Justice, City University of New York
Additional services and information for Journal of Research in Crime and Delinquency can
be found at:
Email Alerts: http://jrc.sagepub.com/cgi/alerts
Subscriptions: http://jrc.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav
Citations: http://jrc.sagepub.com/content/50/2/163.refs.html
>> Version of Record – May 6, 2013
OnlineFirst Version of Record – Nov 8, 2011
Downloaded from jrc.sagepub.com at University of Texas at San Antonio on January 10, 2014
What is This?
Article
The Variable
Impacts of
Public Housing
Community
Proximity on
Nearby Street
Robberies
Journal of Research in Crime and
Delinquency
50(2) 163-188
ª The Author(s) 2013
Reprints and permission:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0022427811426335
jrcd.sagepub.com
Cory P. Haberman1,
Elizabeth R. Groff1, and Ralph B. Taylor1
Abstract
Objectives: Use crime pattern theory to investigate the proximity effects
of public housing communities on robbery crime while taking into account
the presence of nearby nonresidential facilities. Method: The study uses
data describing 41 Philadelphia public housing communities and their surrounds. Surrounds are defined using two increments of street block-sized
buffers. Multilevel models (buffer areas nested around public housing communities) allowing the proximity effect to vary across communities and predicting its shape with public housing level predictors are estimated.
Results: The multilevel models show that the shape of proximity effects
varies across public housing communities and depends on community size,
even after factoring in presence of nonresidential facilities. Spatially, multiple
1
Department of Criminal Justice, Temple University, Philadelphia, PA, USA
Corresponding Author:
Cory P. Haberman, 515 Gladfelter Hall, 1115 W. Polett Walk, Philadelphia, PA, 19122, USA.
Email: [email protected]
Downloaded from jrc.sagepub.com at University of Texas at San Antonio on January 10, 2014
164
Journal of Research in Crime and Delinquency 50(2)
public housing communities close to one another have more intense
robbery patterns. Conclusions: Labeling all public housing communities
as equally criminogenic robbery exporters is unwarranted. In fact, some
communities have lower robbery counts than the areas surrounding them.
Consequently, effectively addressing robbery in and around public housing
communities will require careful consideration of where the problem is
located. Locating public housing communities more than two blocks apart
may reduce robbery.
Keywords
public housing, proximity, distance, crime pattern theory, multilevel
modeling
Public housing is a prominent feature of the urban cores of most American
cities (Holzman 1996). Originally conceptualized as a job creation plan and
temporary housing for families hoping to rebound in the post-Depression
era, American public housing policy has slowly transitioned to long-term
housing for the perpetually disadvantaged (Huth 1981). Today, television
and print news stories as well as quantitative (Holzman, Hyatt, and Kudrick
2005; Roncek, Bell, and Francik 1981)1 and qualitative social science
research (Kotlowitz 1992; Venkatesh 2002; Venkatesh 2008) have made
public housing virtually synonymous with poverty and crime (Farley
1982; Roncek et al. 1981).
The association of public housing with poverty and crime has created a
‘‘not in my back yard attitude’’ among nearby market-value residents. Political reaction to communities’ fear of crime spillover (Santiago, Galster, and
Pettit 2003) has led to the geographic concentration of public housing communities and in turn exacerbated social problems associated with economic
disadvantage (Massey and Kanaiaupuni 1993). Federal programs, such as
the U.S. Department of Housing and Urban Development’s (HUD) HOPE
VI are based on the premise that relocating and rebuilding public housing
communities at lower densities will improve public housing residents’ standards of living. These programs are proceeding, however, without a solid
understanding of public housing proximity effects. Simply stated, policymakers need to know whether public housing increases crime in the surrounding area and if so, the size and geographic extent of those
proximity effects.
The current research embeds an investigation of micro-level proximity
effects within an environmental criminological crime pattern theory
Downloaded from jrc.sagepub.com at University of Texas at San Antonio on January 10, 2014
Haberman et al.
165
framework to address two specific research questions. First, after controlling
for the presence of other potentially criminogenic facilities, something not
done in previous works, do public housing communities affect street crime
levels nearby? Second, if proximity effects are present then are public housing proximity effects variable depending on the characteristics of individual
public housing communities?
Theoretical Perspective
Environmental criminology explains the geographic and temporal patterning of crime by understanding how human interaction creates criminal
opportunities (Wortley and Mazerolle 2008). The urban landscape is
viewed as a collection of places or nodes which are connected to one
another by a street network or other transportation modes (i.e., pathways;
Brantingham and Brantingham [1981] 1991, 1993, 1995). At the same time,
people have activity spaces which consist of the places they visit on a routine basis and the routes (i.e., paths) they take among those places (Horton
and Reynolds 1971). The distribution of a city’s land uses and particular
facilities influences where and how people will travel to use the city
each day (Groff, Weisburd, and Morris 2009; Horton and Reynolds 1971;
Kinney et al. 2008). Recognizing the importance of routine activities theory’s idea that crime events stem from the convergence of motivated offenders with suitable targets lacking adequate guardianship (Cohen and Felson
1979), crime pattern theory predicts that crime will cluster along the most
commonly traveled pathways and around particular nodes which create the
greatest number of offender-target convergences (Brantingham and Brantingham 1993).
Further, some facilities will generate crime by attracting many people
and other facilities by attracting many criminals (Brantingham and
Brantingham 1995). Subsequent research has empirically linked numerous
types of facilities with higher levels of crime in the surrounding area: high
schools (Roman 2005; Roncek and LoBosco 1983; Roncek and Faggiani
1985), bars and taverns (Roncek and Maier 1991), convenience stores
(Schweitzer, Kim, and Mackin 1999), public transportation stations (Block
and Block 2000; Block and Davis 1996), check-cashing stores (McCord and
Ratcliffe 2007), liquor stores, halfway houses, and homeless shelters
(McCord and Ratcliffe 2007; Rengert, Ratcliffe, and Chakravorty 2005).
Because public housing communities are themselves a people concentrating
node within the urban landscape, it is likely that crime will also cluster
around public housing communities.
Downloaded from jrc.sagepub.com at University of Texas at San Antonio on January 10, 2014
166
Journal of Research in Crime and Delinquency 50(2)
Crime around Public Housing Communities
Two studies have compared crime rates within public housing communities
to crime rates within adjoining buffer areas or beyond. Fagan and Davies
(2000) compared crime levels within 82 public housing communities in the
Bronx, NY, to three buffer zones extending on in 100 yard increments.
Violence rates were lowest in the 100 to 200 yard buffers and highest within
the 0 to 100 yard buffers, with the communities’ rates falling in between.
Additional results from two-stage least squares regression models suggested that crime was diffusing outward from the public housing communities and the authors suggested that public housing communities may
serve as the epicenter of social exchanges in areas that lack social control.
Holzman et al. (2005) results varied by crime type: robbery and property
crime was higher in a 300-meter (984.25 feet) buffer around the public communities and assault was higher within the community than in adjoining
buffers. The authors suggested commercial facilities in the buffer zones
provided more attractive robbery targets than the residential public housing
communities where high unemployment rates translated into high levels of
guardianship. At the same time, unemployed individuals probably contributed to higher rates of violence within the communities since they spent
more time at home where they could become involved in violent events,
especially domestic violence.
Three other studies investigated broader scale distance effects of public
housing communities. Crime rates within census blocks were considered as
a function of distance from public housing communities by Roncek et al.
(1981). In Cleveland, an indicator of public housing communities’ proximity linked to higher census block violent crime rates but not higher property
crime rates. Further, census blocks adjacent to public housing communities
experienced two more violent crimes per year than nonadjacent blocks,
even after controlling for citywide public housing community influence
on each census block.
In Atlanta, McNulty and Holloway (2000) explored whether the census
block group level race–crime relationship could be explained by the geographic anchoring of impoverished minority populations in public housing
communities. They found that an interaction between a block group’s percentage of Black residents and distance to public housing was inversely
associated with murder, rape, assault, and public order crimes (but not robbery and property crimes), but simple slope analyses showed that the relationship between percentage African American and the dependent variables
became negligible for block groups a mile or more from the nearest public
Downloaded from jrc.sagepub.com at University of Texas at San Antonio on January 10, 2014
Haberman et al.
167
housing community. The authors concluded that the neighborhood-level
race–crime relationship is related to the geographic distribution of public
housing because that distribution spatially concentrates poverty.
Holloway and McNulty (2003) reframed the relationship between block
group distance to public housing and violent crime by arguing that this
effect would vary depending on a community’s physical design and immediate context. When estimating separate models for each of the forty-two
different public housing communities in Atlanta, and using block group
distance from public housing as a predictor, they found the public housing
distance–violent crime link varied. Distance was a significant predictor in
the expected negative direction in models for twenty public housing communities, a significant positive predictor in the models for five others, and
an insignificant predictor for the remaining seventeen. Proximity impacts
were stronger for public housing communities with more housing units and
high-rise family developments. The authors concluded that public housing
communities are not homogenous and stereotyping all public housing communities as equally criminogenic is unwarranted.
Extending Past Studies
In sum, prior studies show public housing communities adversely affect
crime in their surrounds. Crime generally decreased as distance from public
housing increased. These studies, however, suffer from three shortcomings
addressed here. First, prior works have failed to control for compositional
differences in facility patterns in the surrounding areas. As noted above,
many studies have linked the presence of certain facilities to higher levels
of crime. Failing to model the impact of other facilities may have led to less
robust models in earlier work. Second, existing studies, excluding Holloway
and McNulty (2003), implicitly assume that all public housing communities
impact crime in the surrounding areas uniformly. The crime pattern theory
framework adopted here anticipates proximity effects will vary depending
on the characteristics of the public housing communities themselves. For
example, communities with more total residents may have proximity effects
which ‘‘drop-off’’ faster as distance increases because more populous communities may create more spatially dense offender–victim encounters
within the community. Similarly, a faster drop-off in the proximity effect
might be expected moving farther away from family communities as compared to elderly communities because members of family communities may
lead more active lifestyles and thereby increase the likelihood of victim–
offender encounters within the community as socializing occurs (Kotlowitz
Downloaded from jrc.sagepub.com at University of Texas at San Antonio on January 10, 2014
168
Journal of Research in Crime and Delinquency 50(2)
1992; Venkatesh 2002; Venkatesh 2008). Third, extant research relies on
census delineated geographic units and measures proximity across large
distances. In keeping with recent research suggesting high-traffic facilities
impact nearby areas over relatively short distances (Groff 2011; Ratcliffe In
press), the varying proximity effects examined here are measured at the
micro level of street blocks. Overall, more micro-scale distances are used
while allowing proximity effects to vary while controlling for compositional
differences in facilities, a factor which has been insufficiently integrated into
works to date (Brantingham and Brantingham [1981] 1991, 1995).
Data and Method
Because buffers at different distances from the public housing communities
are nested by the public housing communities themselves, multilevel models
(MLMs) are used (Raudenbush and Bryk 2002).
Level-Two Units: Public Housing Communities
The level-two units in this analysis are public housing communities in Philadelphia, PA, USA. The Philadelphia Public Housing Authority (PHA),
serving over 81,000 Philadelphians, is the fourth largest in the country. The
Philadelphia Police Department (PPD) and HUD provided data describing
public housing communities in Philadelphia.
The PPD supplied geographic data in the form of an ArcGIS shapefile
containing the boundary of each public housing community (PPD, 2011).
These outlines were used to construct the buffer areas discussed in the following sections of the article. The Philadelphia PHA operated 41 public
housing developments in 2007; however, only 40 level-two units are analyzed in this study. Two developments were aggregated into one community due to their geographic proximity.2 These two communities are
separated by only a shared residential street and were deemed analytically
indistinguishable.
The HUD (2009) data set provided community level measures for the 40
public housing communities in the geographic data from the PPD. These
data were collected between July 2007 and December 2008 by local PHAs
and landlords operating federally subsidized housing using official HUD
forms, HUD-50058 and HUD-50059. Roughly 88 percent of occupied
units returned a completed self-reporting form in Philadelphia, but the
only level-two variable that would have been impacted by a low response
Downloaded from jrc.sagepub.com at University of Texas at San Antonio on January 10, 2014
Haberman et al.
169
rate used in the present analysis, residential population, was obtained from the
Philadelphia PHA’s inventory database and therefore is a 100 percent count.
Level-Two Independent Variables
Level-two independent variables, characteristics of public housing communities, were derived from the HUD (2009) data set. The size of each public
housing community was provided as the total number of residents in each
community. The resident type of each public housing community was measured using two dummy variables: family communities and senior communities, with a reference category representing communities with mixed,
family, and senior, populations. Descriptive statistics for all variables are
shown in Table 1.
Level-One Units: Buffer Areas
The level-one units in this analysis are buffer areas surrounding individual
public housing communities. Each public housing community was buffered
at 50 feet, 450 feet, and 850 feet. The 50 feet buffer includes the actual public housing development, representing the community itself and the streets
running adjacent to the public housing development. This was done because
geocoding is a slightly imprecise process (Chainey and Ratcliffe 2005:6063). Crimes that occurred within the community would actually be geocoded on the street running along its border. Street crimes in Philadelphia
are also sometimes geocoded to intersections when their precise locations
are unclear. Therefore, the 50 feet buffer would correctly attribute all
crimes occurring within the community and geocoded to the street in front
of the community to the community itself. The average length of a Philadelphia street block is 400 hundred feet (McCord and Ratcliffe 2009; Ratcliffe
and Rengert 2008), so the second and third buffers represent the area surrounding the public housing community at distances of one and two blocks
out, respectively. This distance was used to reflect research that has found
that the effects of criminogenic facilities only extend a relatively short distance from a facility (Groff 2011; Ratcliffe In press).
After the 50, 450, and 850 feet buffers were created for each public housing complex, parts of the buffers of 18 public housing communities overlapped.3 In order to retain statistical independence and avoid double
counting crime incidents in the overlapped buffers, each overlapping buffer
area was assigned to just one community using the following systematic
process. First, the buffers were decomposed into 212 independent buffer
Downloaded from jrc.sagepub.com at University of Texas at San Antonio on January 10, 2014
170
Journal of Research in Crime and Delinquency 50(2)
Table 1. Descriptive Statistics of Dependent and Independent Variables.
Variable
N
Dependent variables
Street robberies
120
Exposure variable
Area square miles
120
Level-two independent variables
Total residents
40
Building resident typea
Family
19
Senior
12
Family and Senior
9
Level-one independent variables
Buffer distance
50 feet buffer
40
50-450 feet buffer
40
450-850 feet buffer
40
Buffer intersections
120
Beer establishments
120
Check-cashing businesses 120
Drug treatment centers
120
Halfway houses
120
High schools
120
Homeless shelters
120
Parks
120
Pawn brokers
120
Subway stations
120
Min
Max
M
SD
Skewness
0
26
6.26
6.04
1.67
0.003
0.133
0.06
0.04
0.28
17
1,696
430.40
412.03
1.30
—
—
—
—
—
—
—
—
—
—
—
—
—
—
—
—
—
—
0
0
0
0
0
0
0
0
0
0
—
—
—
4
1
2
2
2
2
3
3
1
2
—
—
—
0.63
0.08
0.08
0.12
0.08
0.24
0.09
0.44
0.03
0.04
—
—
—
1.04
0.26
0.32
0.41
0.31
0.50
0.39
0.67
0.18
0.24
—
—
—
1.82
3.27
4.66
3.70
3.91
2.00
5.23
1.40
5.27
6.36
Source: Department of Housing and Urban Development (2009); Philadelphia Liquor Control
Board (2009); Philadelphia Police Department (2007); Philadelphia Police Department (2008)
Level-two units public housing communities (n ¼ 40). Level-one units buffer areas (n ¼ 120).
a
Building resident type consists of 2 dummy variables. For the family variable, communities
with only family units were coded as 1 and all other buildings were encompassed by the reference category. For the senior variable, communities housing only elderly residents were
coded as 1 and all other building types were included in the reference category.
slivers covering a total of 6.82 square miles. Of the 212 total buffer slivers,
129 slivers covering 6.25 square miles, or 91 percent of total area under
study, were associated with only one community from the start. The remaining eighty-nine slivers were areas where the buffers of two or more communities overlapped. These eighty-nine slivers contained only 0.57 square
miles or roughly 9 percent of the total area under study. For all overlapping
buffers, there was never an instance where the entire buffer of a community
was completely subsumed by the buffer of another community.
Downloaded from jrc.sagepub.com at University of Texas at San Antonio on January 10, 2014
Haberman et al.
171
Allocation rules were then established to assign each of the eighty-nine
overlapped buffer slivers to only one public housing community. First, an
attempt was made to assign the overlapped buffer areas to the most proximate public housing community. For example, if the 450 feet buffer of
development X overlapped with the 850 feet buffer of development Y, then
the area would be designated to development X’s 450 feet buffer area. A
total of sixty-one overlapped buffer slivers were assigned using this first
rule. The remaining twenty-eight overlapping slivers were then allocated
through random assignment. In other words, these overlapping buffer slivers were theoretically attributed to the ‘‘wrong’’ community in roughly
half of the allocations and made finding a statistical relationship more difficult. After each overlapped area was assigned to only one community, the
slivers were reaggregated to create a single buffer at each of the three distances for each community. In total, 120 level-one public housing buffer
areas were created; one buffer at each of the three distances for each of the
forty communities.
Level-One Independent Variables
Level-one variables include measures of the individual buffer areas. First,
because the area of the level-one buffer areas varies across public housing
communities, an exposure variable of area in square miles was used in our
models to control for those differences. The first predictor at level-one is the
proximity effect. The proximity effect variable was coded with the first 50
feet buffer, including the public housing community, as 0, the next 450 feet
buffer away as 1, and the second 450 feet buffer as 2. The second predictor
controls for the buffer overlap among different communities.4 This variable
is simply a count of the number of different community buffers a buffer
intersected with. Third, counts of nine different nodes within each buffer
were entered at level one as well. The relationship between alcohol serving
establishments and crime is well established (Roncek and Maier 1991
among many others). Other studies have found a positive relationship
between check-cashing businesses (McCord and Ratcliffe 2007), drug
treatment centers (Taniguchi and Salvatore in press), halfway houses
(Rengert et al. 2005), high schools (Roman 2005; Roncek and Faggiani
1985; Roncek and LoBosco 1983), homeless shelters (McCord and
Ratcliffe 2007), parks (Groff and McCord In press), pawn brokers (McCord
and Ratcliffe 2007), and subway stations (Block and Block 2000; Block and
Davis 1996; McCord and Ratcliffe 2009)5. All data except beer establishments were provided by the PPD’s crime analysis unit (PPD 2008). Beer
Downloaded from jrc.sagepub.com at University of Texas at San Antonio on January 10, 2014
172
Journal of Research in Crime and Delinquency 50(2)
establishments were extracted from Philadelphia Liquor Control Board’s
(PLCB) liquor license database using license classifications for small markets, delis, sandwich shops, and taverns permitted to sell alcohol for on-site
and off-site consumption.
Researchers have recently questioned whether certain types of facilities
need to be present or is it just simply that ‘‘busy places,’’ a mix of nonresidential facilities, create crime opportunities and determine spatial crime
patterns (Wilcox and Eck 2011). Here we do not distinguish between facilities which increase criminal opportunity at a place and those that simply
increase the number of people; both are consistent with crime pattern theory. In other words, from the former perspective, our facility variables
account for the crime opportunities that specific types of facilities might
create. From the latter perspective, they serve as a proxy for foot traffic density in the buffer areas. Observational counts of pedestrian traffic and nonresidential land uses at the street block level have found that pedestrian
counts load strongly on a nonresidential versus residential street block composition component (see Taylor, Shumaker, and Gottfredson 1985). Such a
pattern suggests pedestrian counts are indistinguishable at the street block
level from nonresidential land use patterns.
Dependent Variable
The dependent variable in the present analysis is the number of street robberies occurring in each individual buffer area during calendar year 2007.6
Incident level data for street robbery were obtained from the PPD. A single
type of violent crime, street robbery, is used in order to reduce the heterogeneity inherent in aggregate crime categories (Clarke 2008; Smith, Frazee,
and Davison 2000). Moreover, the violent nature and frequency of street
robbery makes it a major concern in Philadelphia (Ramsey 2008). In
2007, the robbery rate in Philadelphia reached 714.58 robberies per
100,000 citizens compared to the national robbery rate of 147.6 robberies
per 100,000 residents (Federal Bureau of Investigation 2007). Additionally,
because Americans rightly perceive the likelihood of robbery victimization
is much greater than most other crimes (Ferraro 1995:47), understanding
the dynamics of street robbery might lead to policies that impact a wider
range of Americans. Finally, street robbery matches the theoretical foundation of the present analysis as robbery is the quintessential predatory crime,
and ethnographic research with active robbers has demonstrated that opportunity concentrating places figure prominently in robbers’ searches for suitable targets (St. Jean 2007; Wright and Decker 1997). The mean of the
Downloaded from jrc.sagepub.com at University of Texas at San Antonio on January 10, 2014
Haberman et al.
173
dependent variable is 6.26 with a variance of 36.48, indicating a Poisson
distribution with over dispersion (Long and Freese 2006:350).
Results
The series of multilevel count models estimated included an initial null
model confirming significant (p < .001) variation in expected robbery counts across public housing communities; an analysis of covariance MLM examining the fixed effect of buffer proximity while controlling for nearby facilities (first half of Table 2) and confirming significant robbery differences between communities remained (p < .001) even after controlling for facility differences across buffers; and a full model where a significantly varying proximity impact (p < .05) was predicted using public housing community features (second half of Table 2). The level two total residents’ variable in the full model was grand mean centered. Analysis of Covariance (ANCOVA) Model The ANCOVA model shows expected robbery counts decrease 43 percent for each additional buffer farther from a public housing community, holding the presence of other nonresidential facilities constant.7 This finding agrees with previous work (Fagan and Davies 2000; Holzman et al. 2005; Roncek et al. 1981), but extends it in two important ways: proximity effects appear even when using short distances and controlling for variation across the buffers in nonresidential facilities. Additionally, the ANCOVA model shows that certain facilities are associated with increases in robbery counts and others with decreases. Specifically, the presence of a high school increases the expected robbery count in a buffer by 50 percent compared to 63 percent for a homeless shelter, 55 percent for a pawn broker, and 30 percent for a subway station. On the other hand, the presence of a drug treatment center or halfway house decreases the expected robbery count by 24 percent and 29 percent, respectively. Full Model: Average and Varying Proximity Effects In the full-model allowing proximity impacts to vary by public housing community, each additional 400 feet traveled from a public housing community, on average, results in a significant 34 percent decrease in expected robbery counts when all other variables in the model are held constant, an impact about a quarter smaller than seen in the ANCOVA model. Again, Downloaded from jrc.sagepub.com at University of Texas at San Antonio on January 10, 2014 174 Downloaded from jrc.sagepub.com at University of Texas at San Antonio on January 10, 2014 5.31 — — — -0.55 — — — 0.05 0.38 0.22 0.27 0.34 0.41 0.49 0.007 0.44 0.26 Variance Component 0.73 — 122.65*** — 0.16 — — — 0.06 — — — 0.13 0.26 0.15 0.11 0.12 0.09 0.08 0.09 0.23 0.16 Chi-square SE 39 — 202.95 — — — 0.57 — — — 0.95 1.47 1.25 0.76 0.71 1.50 1.63 1.01 1.55 1.30 df Event Rate Ratio 33.61*** — — — -8.63*** — — — -0.41 1.50 1.51 2.43** 2.94** 4.56*** 5.84*** 0.074 1.93* 1.70* t-ratio 4.92 0.0007 0.17 0.52 0.42 0.0005 0.23 0.22 0.11 0.54 0.09 0.08 0.55 0.42 0.54 0.05 0.64 0.36 Variance Component 0.83 0.13 Beta Coefficient 116.81*** 54.03** 0.32 0.0004 0.33 0.46 0.15 0.0002 0.15 0.20 0.06 0.19 0.13 0.12 0.16 0.13 0.06 0.10 0.26 0.13 Chi-square SE 36 36 136.35 1.0007 1.19 1.68 0.66 0.999 0.80 0.80 1.12 1.71 0.92 0.93 0.57 1.52 1.71 1.06 1.90 1.44 df Event Rate Ratio Full Model 15.50*** 1.84* 0.53 1.12 2.90** 2.91** 1.51 1.08 1.74* 2.83*** 0.67 0.62 3.45*** 3.328*** 8.48** 0.60 2.44** 2.71*** t-Ratio Note: Level-two units are public housing communities (n ¼ 40). Level-one units are buffer areas (n ¼ 120). Dependent variable specified as a Poisson distribution with overdispersion and an exposure variable of area (square miles). The distance variable is a fixed effect in the ANCOVA model and specified with a random slope in the full model. Level-two variable total residents is grand mean centered. *p < .1. **p < .05. ***p < .01. Intercept Buffer distance Intercept Total residents Family community Senior community Buffer distance Total residents Family community Senior community Buffer intersections Beer establishments Check-cashing businesses Drug treatment centers Halfway houses High schools Homeless shelters Parks Pawn brokers Subway stations Random effect Beta Coefficient ANCOVA Model Table 2. Multilevel Model Results for Robbery Count Dependent Variable. Haberman et al. 175 public housing proximity effects are found using short distances and while controlling for potentially criminogenic facilities. On the other hand, there is significant variation (p < .05) around this average proximity impact in Philadelphia. Even at this micro level, all public housing communities were not equally criminogenic for their surrounding communities. This finding both confirms and extends earlier work. Holloway and McNulty’s (2003) differential proximity impact is extended because said impact appears even though much shorter distances are used and the variation in nearby facilities is taken into account. Thus, the findings from this more micro-level study offers additional evidence countering the stereotyped notion that all public housing communities are equally criminogenic. The variation in proximity effects aligns as hypothesized by crime pattern theory with a community feature. More total residents in the public housing community linked negatively to the proximity slope, in other words, each additional 100 residents in a public housing community over the average-sized community made the proximity impact 1 percent steeper. This differential impact fits within a crime pattern theory framework; more populated communities, as compared to less populated ones, will create more convergences of motivated offenders and suitable targets within the community. Examining the local context of Philadelphia suggests community sizelinked design features might have contributed to variations in proximity effects. Only six communities had above-average populations. Three consisted of vast campuses of mid-rise apartment buildings, two were campuses with multiple high-rise buildings, and the remaining community was an extensive neighborhood of row homes. In short, Philadelphia’s more populated communities are also the communities with designs that would facilitate robbery in the community (see Newman 1973; Newman and Franc