 

.. . .. t.
u 3 7V 3%.“.3
. 1 . . E.
1
_

1?. ”$1....

:1 ”Es.
mumum n
mam?“

a
r]. ..
:1:
$1.:
3?

av

Evmﬁr». .
. ugh

1&3

xiio.‘
3.1125
:3... 4 {I}:
sit... .
‘15.,
$0) 3.3).)!‘2‘! I.
5:1...rzziak
(1...;th

WW1.

 

mm“
1%
.1. bL‘V‘til‘
‘ E. . RR
.2 an Hahn... uvnm voila
p.391... ital?!“ .! V.
alviuiié.‘ ‘ ‘1‘.
3.1.5.. \
‘S.'—-I§I

.2 ..
:

3:: i.
a: .33.
1...}.
in
11.2.31

:2.

.3...

.2.

.
ll
, e“ snag... .2
«Eurasian».
3...}...
‘7...

I... 35‘
nunshnﬂiv. .

. 3-...25:
magnum”
.2.) ,

:3:
4.35!!!
L ‘y. 3
i. dun»:

I.-

 

..nL.LO-L a; .

 

3 LIBRARY
2 0 09 Michigan §tate
Univemty

This is to certify that the
dissertation entitled

NEIGHBORHOOD DISORDER, DILAPIDATED HOUSING,
AND CRIME: MULTILEVEL ANALYSIS WITHIN A MIDSIZED
MIDWESTERN CITY CONTEXT

presented by

JINSEONG CHEONG

has been accepted towards fulfillment
of the requirements for the

PhD. degree in Criminal Justice

 

 

 

 

v Major rofessor’s ng nature

8, Eli/02008

Date

 

MSU is an afﬁrmative-action, equal-opportunity employer

 

PLACE IN RETURN Box to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

1131211!
MAthzom

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 ICIProi/AocaPreleIRClDateDuojndd

NEIGHBORHOOD DISORDER, DILAPIDATED HOUSING, AND CRIME:
MULTILEVEL ANALYSIS WITHIN A MIDSIZED MIDWESTERN CITY CONTEXT

By

J inseong Cheong

A DISSERTATION

Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Criminal Justice

2008

ABSTRACT

NEIGHBORHOOD DISORDER, DILAPIDATED HOUSING, AND CRIME:
MULTILEVEL ANALYSIS WITHIN A MIDSIZED MIDWESTERN CITY CONTEXT

By

Jinseong Cheong

This dissertation had two main purposes. First, it attempted to test the broken
windows theory within a midsized Midwestem city context, using two waves of ﬁeld
observation data and other various official datasets. Second, cross-level interaction
dynamics were examined via a Hierarchical Mixed Poisson Model including a spatial lag
term. In order to make the multivariate contextual analyses more credible, this study
further attempted to clarify sensitive methodological issues that are likely to inﬂuence
study outcomes.

The Hierarchical Mixed Poisson Model revealed several notable findings. First,
the results for violent crime were different from those for property crime, which appeared
to be supportive of Clarke’s (1970) notion that more crime-speciﬁc approaches need to
be taken in research and crime prevention policy. Second, the effect of physical disorder
change on violent crime change was negatively moderated by the level of concentrated
disadvantage. In other words, the link between physical disorder and violent crime was
less valid in prosperous neighborhoods than in disadvantaged counterparts, which
implied that strategies focusing solely on reducing physical disorder might have little
effect on dropping the level of violent crime in poor neighborhoods. Third, social
disorder change and residential instability had a negative interaction effect on property

crime change. It indicated that the impact of social disorder change on property crime

change became weaker as the level of residential instability increased, which further
implied that dealing with social disorder might be less effective in reducing property
crime in non-stable communities than in stable communities.

Along with the necessity of crime-speciﬁc approach, the above cross-level
interaction effects made a suggestion that the link between disorder and crime could be
different depending on various community contexts, and thus more context-speciﬁc
approaches need to be taken. However, it was necessary for the ﬁndings to be interpreted
with a caveat due to some limitations. As an example, the important mediating factor,

collective efﬁcacy was not controlled for.

Copyright by
JINSEONG CHEONG

2008

To my parents, wife, son, and daughter, who encourages me to stand up again and again

ACKNOWLEDGEMENTS

I could not have ﬁnished this dissertation without such great supports and
sacriﬁces of many people. My special gratitude goes to Dr. Christopher Maxwell, my
advisor and chair of the guidance and dissertation committee, who has always stood by
my side to encourage me to reach my full potential. He led me to the ﬁleds of broken
windows research, crime mapping, and high level of data analysis technique, which made
me more competitive in academia than I expected. Throughout the last several years of
doctoral program, he has been very generous with his time, insights, and fun.

I also owe many thanks to Dr. Edmund McGarrell, Dr. Merry Morash, and Dr.
Craig Harris for their genuine interests, kind guidance, and precious insights. Dr.
McGarrell has been enormously supportive despite his tight schedule, and his comments
have never missed the point. He will be remembered as a respectable man. Dr. Morash
has always been there whenever I need her, giving encouraging and thought provoking
words. She will remain as a role model inside me. Dr. Harris has been very cooperative,
and his insights have widened the scope of my idea. I will remember him as an
exemplary fellow.

Other faculty and staff members in and out of the School of Criminal Justice have
offered a great guidance and help. My deep gratitude goes to Dr. Ashton Shortridge for
his expertise in spatial analysis and humanity. I admire Dr. Kevin Kelly for his
exhaustive knowledge on social dynamics and change. Dr. Steve Dow’s class of Law
and Society was also quite profound. I would not have been able to survive in the

program without the incredible assistance of Dr. Vincent Hofﬁnan, Dr. Mahesh Nalla,

vi

Mary Lee VanderMoere, Melissa Christle, Peggy Donahue, Terri Bulock, and Hao Lu. I
enjoyed interacting with those wonderful people.

My friends and colleagues have lived through this with me, offering courage,
insights, and true friendship: Nickqursaro, Mengie Parker, Jason Ingram, Carol
Zimmermann, and Mohammad Ali. And I would like to thank my Korean ﬁiends for
their support and love, which is beyond any description: Eui-Gab Hwang, Dae-Hun
Kwak, Wook Kan g, Suyeon Park, Mirang Park, Seok-J in J eong, and Byung Hyun Lee.

Finally, I owe my biggest gratitude to my family. Youngsuk sacriﬁced her life so
much that I will never be able to pay back. She raised our son and daughter, Younghun
and Seohyun, so well as healthy and nice kids with my little support. Younghun and
Seohyun did not complain about the absence of their dad when they need him. My
parents, parents in law, sisters, and brothers have always been supportive and thoughtful.
When bad things happened, they did not call me if I were in the middle of doing
important works like taking a ﬁnal exam. Everyone should be so lucky to have family

like mine. Your support had me stand up again and again to get this done.

vii

TABLE OF CONTENTS

LIST OF TABLES ........................................................................ xi
LIST OF FIGURES ....................................................................... xiii

CHAPTER I: INTRODUCTION ......................................................
Neighborhood Disorder, Dilapidated Housing, and Crime .......................
Lansing Dilapidated Housing Project ...............................................
Study Purpose 1 ........................................................................
Contextual Effects .........................................................................
Study Purpose 2 .......................................................................
Methodological Issues in the Broken Windows Field .............................

HOOOQON-bt—U-I

ﬂ

Overview ...................................................................................

CHAPTER II: THEORY — MAIN CONCEPTS AND CAUSAL PATHS .. 13

Ecological Perspective in Crime Research ........................................... 14
Nature and Scope ........................................................................ 14
Social Disorganization Theory: Chicago School ................................... 16
Contemporary Ecological Theories: Resurrection of Chicago Tradition .... . 18
Environmental Design .................................................................. 20
Situational Studies ...................................................................... 21
New Social Disorganization Theory .................................................. 24
Broken Windows in Brief ............................................................ 26
Summary: Main Causal Factors ..................................................... 27

Broken Windows in Detail ............................................................. 29
Causal Paths ........................................................................... 3 1

CHAPTER III: LITERATURE REVIEW ........................................... 33

Structure, Disorder, and Process ....................................................... 34
Neighborhood Structure and Crime ................................................ 34
Neighborhood Disorder and Crime ................................................. 39
Methodological Issues ................................................................ 43
Neighborhood Process and Crime: Social Disorganization, Systemic

Control, Social Capital, and Collective Efﬁcacy ............................... 57

Contextual Effects ........................................................................ 63

viii

Implications ........................................................................... 65

Summary .................................................................................. 66
CHAPTER IV: THE CURRENT STUDY ........................................... 68
Study Setting: Lansing .. ................................................................ 68
Lansing Dilapidated Housing Project Revisited ....................................... 70
Study Subjects (Sample) and Unit of Analysis ..................................... 71
Non-Experimental Longitudinal Design ........................................... 74
Limitations ................................................................................ 75
A Study Model and Hypotheses 76
Summary .................................................................................. 77
CHAPTER V: METHODOLOGY ...................................................... 79
Data ......................................................................................... 79
CAD (911) ............................................................................. 79
Street Survey ........................................................................... 81
Census .................................................................................. 82
Phone Data ............................................................................. 83
Measurement .............................................................................. 84
Dependent Variable: Serious Crime ................................................ 85
Primary Independent Variables of Interest ........................................ 87
Control Variables: Neighborhood Structure (Characteristics) .................. 91
Analytic Strategies ....................................................................... 94
Spatial Analysis ....................................................................... 94

A Hierarchical Mixed Poisson Model ............................................. 98
CHAPTER VI: RESULTS ............................................................... 106
General Characteristics of Study Subjects (Street Segments & Block Groups) .. 106
Change of Disorder and Crime ...................................................... 108
Bivariate Relationships .................................................................. 110
Diagnostics of Multicollinearity .................................................... 111
Correlations between Dependent Variable and Independent Variables ....... 113
Spatial Lag and Autocorrelation of Crime ............................................ 114
The Hierarchical Mixed Poisson Model .............................................. 120
Unconditional Model (One—Way ANOVA) ....................................... 121
Cross-Sectional Analysis — Random Coefﬁcient Model ........................ 123
Cross—Sectional Analysis — Conditional Model ................................... 125

ix

Longitudinal Analysis — Random Coefficient Model ............................ 129

Longitudinal Analysis — Conditional Model ...................................... 131
Summary .................................................................................. 135
CHAPTER VII: CONCLUSION ........................................................ 137
Summary and Implications for Theory, Research, and Policy ..................... 137
Limitations ................................................................................ 146
Psychological Process ................................................................ 146
Temporal Dynamics .................................................................. 147
Spatial Dynamics ..................................................................... 148 .
Contributions to the Broken Windows Field ......................................... 149
Future Directions of the Broken Windows Research and Policy ................. 151
More Emphasis on the Role of Formal Social Control .......................... 151
Comparative Study with another Country ........................................ 153
APPENDICES .............................................................................. 156
APPENDIX A. Coding Sheet for Street and Housing Survey .................... 157
APPENDIX B. Evaluation Guidelines for Street Survey ........................... 161
APPENDIX C. Number of Serious Crime and Social Disorder for
Study Segments and Intersections ................................. 165
APPENDIX D. Multicollinearity Diagnostics of Level 2 Variables Including
Lagged Crime Rates ................................................. 166
APPENDIX E. Multicollinearity Diagnostics of Level '1 Variables (2003 and
2005) ................................................................... 167
APPENDIX F. Results for Hierarchical Linear Models .............................. 168
REFERENCES ............................................................................. 170

Table 1.

Table 2.

Table 3.

Table 4.

Table 5.

Table 6.

Table 7.

Table 8.

Table 9.

Table 10-1.

Table 10-2.

Table 1 1.

Table 12.

Table 13-1.
Table 13-2.
Table 13-3.
Table 13-4.
Table 14-1.
Table 14—2.

Table 15-1.

LIST OF TABLES

Summary of Individual Studies on the Effects of Disorder
on Crime .................................................................

Racial and EconOmic Conditions of Lansing, Michigan,
and US. ..................................................................

Description of Data ....................................................

Percentage of Crimes and Social Disorders Recorded on
Intersections ...............................................................

Factor Loadings of Physical Disorder and Serious Crime ..
Factor Loadings of Social Disorder and Serious Crime ...........
Component Loadings of Community Characteristics ..............
Descriptive Statistics ....................................................

Change of Disorder and Crime for Total and Selected
Segments ..................................................................

Correlation Matrix (Level 1) ...........................................
Correlation Matrix (Level 2) ..........................................
Tolerance and Variance Inﬂation Factors (VIF) ...................
Collinearity Diagnostics ...............................................
Unconditional Model (Violent Crime 2003) ........................
Unconditional Model (Property Crime 2003) .......................
Unconditional Model (Violent Crime 2005) ........................
Unconditional Model (Property Crime 2005) .......................

Random Coefﬁcient Model (Cross-Sectional - Violent Crime)

Random Coefﬁcient Model (Cross-Sectional - Property Crime) ..

Conditional Model (Cross-Sectional) - Violent Crime

xi

44

69

79

87

89

91

92

107

109

110

111

112

112

121

122

122

122

124

124

126

Table 15-2.

Table 16.

Table 17.

APPENDIX A.

APPENDIX A.

APPENDIX C.

APPENDIX D.

APPENDIX D.

APPENDIX E.

APPENDIX E.

APPENDIX E.

APPENDIX E.

APPENDIX F.

APPENDIX F.

APPENDIX F.

APPENDIX F.

Conditional Model (Cross-Sectional) - Property Crime ...........

Random Coefﬁcient Model (Longitudinal) .........................

Conditional Model (Longitudinal) ...................................

Table 18.

Table 19.

Table 20.

Table 21.

Table 22.

Table 23.

Table 24.

Table 25.

Table 26.

Table 27.

Table 28.

Table 29.

Table 30.

Coding Sheet for Street Survey ...........................
Coding Sheet for Housing Survey ........................

Number of Serious Crime and Social Disorder for
Study Segments and Intersections ........................

Tolerance and Variance Inﬂation Factors . . . . ..

Collinearity Diagnostics ....................................

Tolerance and Variance Inﬂation Factors (2003)

Collinearity Diagnostics (2003) ..........................

Tolerance and Variance Inﬂation Factors (2005)

Collinearity Diagnostics (2005) ..........................
Unconditional Model (Violent Crime 2003) ............
Unconditional Model (Property Crime 2003) ..........
Unconditional Model (Violent Crime 2005) ............

Unconditional Model (Property Crime 2005) ..........

xii

127

131

133

157

159

165

166

166

167

167

167

167

l68

168

169

169

Figure 1.
Figure 2.
Figure 3.
Figure 4.
Figure 5.
Figure 6.

Figure 7.

Figure 8-1.
Figure 8-2.
Figure 8-3.
Figure 8-4.
Figure 9-1.

Figure 9-2.

LIST OF FIGURES

A Renovated Housing Unit during the Intervention Period ......... 6
Neighborhood Effects ..................................................... 8
Ecological Factors and Processes ....................................... 28
Contemporary Theoretical Model Linking Disorder and Crime .. .. 32
Study Segments (n=377) .................................................. 72
Mid-Michigan and Lansing Block Groups ............................. 73
A Hierarchical Model Linking Disorder and Crime .................. 77
Quantile Maps of Crime Rates (Violent Crime 2003) ............... 114
Quantile Maps of Crime Rates (Property Crime 2003) .............. l 15
Quantile Maps of Crime Rates (Violent Crime 2005) ............... 116
Quantile Maps of Crime Rates (Property Crime 2005) .............. 117
Moran’s I and Permutation Tests for Crime Rates ................... 118
Moran’s I and Permutation Tests for Crime Rates Change ......... 118

xiii

CHAPTER I: INTRODUCTION

Neighborhood Disorder, Dilapidated Housing, and Crime

Crime and fear of crime are often among the top concems of urban residents.
Along with poor schools, these conCems are frequently viewed as the leading proximate
causes of urban decline across America. Criminologists have long compiled evidence
that physical landscape and the social environment in inner cities serve as critical factors
that can heighten fear of crime and crime rates (Skogan, 1990). For instance, Wilson and
Kelling’s (1982) classic essay, “Broken Windows,” has inﬂuenced two decades of public
policy by refocusing efforts away from targeting individuals towards addressing crime
generators such as persistent, lower-level problems that violate social norms of public
order. From this perspective, quality of life issues such as vandalism, panhandling, litter
and trash, grafﬁti, and abandoned housing constrain pro-social neighborhood processes
that have become known as a community’s collective efﬁcacyI (Sampson, Raudenbush,
& Earls, 1997). In turn, the decrease in collective efﬁcacy that arises out of social and
physical disorder is linked to the growth of serious street crime (Sampson, Morenoff, &
Raudenbush, 2005; Taylor, 2002). Thus, the link between neighborhood disorder and
crime appears causal, inasmuch as the growth of disorder is consistently associated with
an increase in serious crime. However, when communities address low-level disorder,
whether social or physical, they communicate to those at risk for deviant behavior that

the area is under capable guardianship and that there is a will to improve the community

 

' Sampson and his colleagues (1997) deﬁne collective efﬁcacy as “social cohesion among
neighbors combined with their willingness to intervene on behalf of the common good.” It is
proposed that the level of collective efﬁcacy is the key connecting neighborhood structural
conditions and crime rates. Refer to Chapter HI for more detailed discussion on the neighborhood
process mechanism.

and combat crime (e. g., Green, 1995). Accordingly, the “broken windows” thesis has
become a working model under which some community policing (e. g., Katz, Webb, &
Schaefer, 2001) and community prosecution (see White, F yfe, Campbell, & Goldkamp,
2003; Worrall & Zhao, 2003) initiatives operate.

Meanwhile, abandoned or dilapidated housing has received special attention as a
kind of representative measure of physical disorder. Neighborhoods are easily disﬁgured
by ugly dilapidated houses, and those buildings often serve as a source of other types of
disorderly conditions and behaviors. The inﬂuence of abandoned properties on crime
appears to be strongly supported by empirical evidence. Tiechert (1999), for example,
found that Franklin Villa, a small portion of Sacramento, California, was made up of 85%
absentee landlord rental units and suffered from the highest crime rates in the entire city.
According to Spelman (1993), Austin, Texas, experienced a dramatic 50% increase in the
property crime rate and an 89% increase in robbery rate from the mid-19805 through
1991, when 25% of houses were vacant, with the number of vacant properties increasing
each year. Sherman, Gartin, and Buerger (1989) discovered that in Minneapolis,
Minnesota, 50% of the police service calls were concentrated in only 3% of its entire area,
which were called “hot spots.” Those hot spots were revealed to be more likely to occur
in poor neighborhoods where the number of abandoned houses was also increasing (Rider,
2003)

The city of Lansing, Michigan, despite its relatively small size, appears to have
suffered from similar problems. For instance, in 2003, the Chestnut neighborhood2 that

comprises only 11.8% of Lansing’s land area contributed 20% of the city’s dilapidated

 

2 It falls neatly inside the area assigned to Team 3 of the LPD’s Team Policing. Team Three’s
policing area is bordered by Saginaw Highway, Martin Luther King Blvd, East St, and the Grand
River greenway north of Grand River and Willow Avenues (Rider, 2003).

houses and over 21% of criminal incidents reported by the Lansing Police Department
(hereinafter LPD; Rider, 2003). Serious concerns with the issue of dilapidated housing
among the residents in Lansing communities have often been expressed by the local
media. As an example, an article in the Lansing State Journal stated:

Residents say the hulking eyesores — deemed uninhabitable by the city —
make potential home buyers balk at moving in and send a bad vibe to visitors
driving Lansing’s major thoroughfares. The houses are dangerous playgrounds
for curious kids and attract homeless people and drug users. A suspicious ﬁre
gutted a red-tagged, two-story apartment house in Old Town last month.

(MacDonald, 5/19/2003)

In sum, disorderly behaviors and conditions, including dilapidated housing, seem
to be signiﬁcantly associated with civil concerns, fear of crime, and actual concentration
of criminal incidents in urban neighborhoods. In fact, as mentioned above, the link
between disorder and crime is well conceptualized and contains very appealing policy
implications. As a result, many public agencies eagerly have adopted initiatives that
purport to reduce crime by dealing with disorderly street activities and conditions.3
Responding to the civil concerns about dilapidated residential properties and crime, since
2003, the city of Lansing also has initiated some measures to improve physical
environments across the city. The efforts provided a nice opportunity for empirical
research. In response to that opportunity, the School of Criminal Justice at Michigan

State University and the city of Lansing developed a cooperative initiative, Lansing

Dilapidated Housing Project (hereinafter LDHP), to examine the effectiveness of the

 

3 A typical example is the SMART (Specialized Multi-Agency Response Team) program in
Oakland, California (Green, 1995). The key aspect of the program was constructing cooperative
relationships among diverse public agencies (e.g., police, housing inspectors, public workers, gas
and electric agencies), citizens, landlords, and business owners. The proactive enforcement of
various city codes and civil laws (e.g., housing code, health and safety code, drug nuisance
statement law, etc.) was effective in reducing disorder and drug problems in targeted areas. Even
a small net effect of spatial diffusion of the beneﬁts was detected.

city’s efforts to deal with crime and related problems. A brief description of the project

is presented below.

Lansing Dilapidated Housing Project

When residential units are poorly maintained and unsafe, consistent with the
city’s housing code, ofﬁcers from the Lansing C ode Compliance Ofﬁce (hereinafter
LCCO) issue and attach a red tag to the front door. In June, 2003, there were about 479
red-tagged houses located in 386 street segments across the city.4 Those houses, as
mentioned above, raised serious residential concerns about decreasing property values,
high crime rates, and urban decline.

In response to these concerns, the project city’s government and council have
taken several ofﬁcial actions to improve the conditions of dilapidated housing properties
with an aim of reducing crime. As an example, an amended state law was passed in 2003
that allowed cities to raze red-tagged houses with fewer constraints than existed
previously. As a result of the amendment, they did not have to wait for six months of
vacancy nor did they have to wait until the property value surpassed the State Equalized
Valuation. Meanwhile, the LCCO hired a caseworker. Her key actions included
cooperation with the LPD to identify neighborhoods with the most crime and dilapidation
issues, cooperation with code inspectors to identify the neighborhoods with the most
abandoned housing, working with those speciﬁc neighborhoods, locating homeowners

and determining potential actions, arranging for boarding and securing of houses, and

 

4 In other words, some segments had more than one red-tagged house. For instance, two
segments contained as many as four tagged houses. Additionally, the roads of the project city
consist of 5,953 segments.

connecting homeowners to diverse agencies, such as bankers, contractors, realtors, and
property demolition experts (Rider, 2003).5

Those efforts provided an opportunity for empirical research on the theoretical
link between physical disorder (represented by abandoned housing and its surrounding
conditions), social disorder, and crime in the city’s neighborhoods. The initial research
plan was to perform a randomized ﬁeld experiment, which was agreed upon by the city
government. Researchers assigned 260 red-tagged houses into a treatment group and 217
into a control group." The LCCO inspectors and caseworker were asked to work with
owners and other interested parties (e. g., banks, community groups, home owners, etc.)
only for the 260 treatment houses. To provide a pretest, two graduate students from the
School of Criminal Justice conducted a street survey on the tagged houses and their
segments during the spring and summer of 2003.7

Unfortunately, however, due to budget constraints, the original plan came to an
early end without accomplishing the goal. The caseworker had to quit her job after
treating 50 houses in 40 segments. This early end of the experimental approach left the
researchers with only one option: the researchers decided to take a non-experimental
approach, which was to observe and model the natural change of physical conditions,

social disorder, and crime across the study segments. With the help of the Michigan

 

5 The community-oriented policing approach of the LPD deserves attention in terms of its
structure. They divided the city into several team areas where permanent patrol ofﬁcers and
leadership were assigned. Also, every two team areas shared investigators so that they could
become familiar with the people and other general conditions and situations in that area.
However, much less organized efforts to enhance collective efﬁcacy through proactive
interactions with residents appear to be present in Lansing as compared to other well established
COP programs (Rider, 2003). For this reason as well as data unavailability, this dissertation does
not include the COP of Lansing as another dimension of the ofﬁcial actions.

6 Information On two houses could not be found.

7 Coding sheets for the street survey are attached (Appendix A). More discussions on the ﬁeld
survey are presented in the methods chapter.

State University Land Policy Institute, the author carried out the ﬁeld survey for the same
housing units and segments during the fall of 2005. A total of 444 housing units and 378
segments were observed again.8 Thirty-ﬁve houses (in 8 segments) of the initial sample
appeared to have been demolished completely during the intervention periods. Further,
the researcher could witness that some of the dilapidated houses have been completely

renovated. Figure 1 shows an example.

Figure 1. A Renovated Housing Unit during the Intervention Period

 

Study Purpose 1: In short, the LDHP faced an unexpected shift of its original plan
of evaluating the project city’s comprehensive efforts via a controlled experiment. The
halt of treatment by the caseworker did not mean the end of all civil efforts, however.
Although a detailed account of the course of remaining actions and any evaluation
attempt of the efforts are beyond the scope of this dissertation, analyzing the changes in

physical disorder, social disorder, and crime would be a valuable contribution to the

 

B One segment (325 W. Sheridan Road) was newly added because it had not been surveyed in
2003. But the author did not include it in the analysis. Thus, the analysis is based on 377
segments. For more information, refer to the sample in the methods chapter.

broken windows theory. Thus, using the two waves of ﬁeld observation data and other
various ofﬁcial datasets such as CAD (computer aided dispatch) 2003 and 2005, US.
Census 2000, and Select Phone USA 2000, the author tests a multivariate model linking

disorder and crime, which is the main purpose of this dissertation.

Contextual Effects

Neighborhood research on crime begins with the assumption that a neighborhood is
not just a simple aggregation of individuals. Although little challenge has been made
against this postulation, there still remains another critical methodological problem: how do
we know that the differential outcomes for neighborhoods (e. g., crime rates) are rooted in
neighborhood characteristics rather than the self-selection of individuals to live in the
neighborhoods? This question has been one serious criticism of social disorganization
theory. Just as the self-selected people in a treatment program tend to produce more
desirable outcomes, individuals’ choice of a certain neighborhood based on their income,
education, race, and other circumstances produces a compositional effect that is clearly
distinct from contextual effects. Thus, it is crucial to control for individual characteristics to
sort out the contextual effects (Harcourt & Ludwig, 2006; Kubrin & Weitzer, 2003b;
Sampson, Morenoff, & Gannon-Rowley, 2002).

Contextual effects take two forms: pure or direct neighborhood effects and
interaction effects (Figure 2). A typical statistical solution is a contextual analysis that
controls for individual sources as confounding or moderating factors. Treating individual
factors as confounding variables involves sorting out the direct effects of neighborhoods
(see Simcha-Fagan & Schwartz, 1986), and treating them as moderating factors involves

examining the interaction effects of individual factors with neighborhood characteristics (see

Rountree, Land, & Miethe, 1994; Silver, 2000; Wikstrom & Loeber, 2000). Emphasizing
the importance of understanding contextual effects, particularly through research on
interaction effects, Miethe and McDowall (1993) argued, “[f]ai1ure to consider that the
impact of individual-level factors depends on the wider social context is a form of
speciﬁcation error that would dramatically alter substantive conclusions” (p. 752). This
growing interest in contextual effects is substantially related to the development of
Hierarchical Linear Model (hereinafter HLM) that allows for the multilevel contextual
analysis (Raudenbush & Bryk, 2002). Further, it has stimulated more systematic collection

of individual data in neighborhood research (Kubrin & Weitzer, 2003b).

Figure 2. Neighborhood Effects

 

 

Individual Pure
(Compositional) Interaction Neighborhood
Effect Effect Effect

 

 

 

 

Study Purpose 2: This dissertation does not deal with human subjects. Thus, it
was not feasible to attempt to sort out the pure and/or interaction effects of neighborhood
characteristics on crime by controlling for human factors. However, the author ﬁgured
that it would be desirable to conceptualize the link between disorder and crime in such a
way that the effect of disorder on crime at a lower level (i.e., street segment) depends on
community characteristics at a higher level (i.e., block group). To date, only a few

studies (J ang & Johnson, 2001; Miethe & McDowall, 1993; Rountree, Land, & Miethe,

1994) have examined the interaction dynamics in the area of linking neighborhood
disorder and crime. Further, concerning the cross-level interaction effect of disorder and
community characteristics on crime, only one study (Rountree et al., 1994) attempted to
test the multilevel model. Accordingly, another purpose of this dissertation is to examine
whether there are meaningful cross-level interaction effects of disorder and neighborhood
characteristics on street crimes. For this, a hierarchical model is constructed and tested.
Identifying contextual effects, particularly the interaction dynamics, has great
potential for theory elaboration and policy implication. For example, Rountree and her
colleagues (1994) found that perceived disorder and racial heterogeneity had a negative
interaction effect on burglary victimization, which suggested that the proposed link
between disorder and crime (i.e., burglary) may be less valid in heterogeneous
neighborhoods than in homogeneous counterparts. The results further implied that an
order-maintenance or control-oriented policy in the heterogeneous neighborhoods would
not be as effective as is suggested by the theory. Such conceptualization is consistent
with the ecological notion that a context-speciﬁc approach is likely to better reﬂect the

reality of crime-generating and control processes.

Methodological Issues in the Broken Windows Field

Good research and policy begin with a clear understanding of the underlying
theory(ies). Otherwise, the credibility of research and the effectiveness of policy are
open to question. The eventual by-product is ongoing controversies. The ﬁeld
concerning neighborhood disorder and crime in relation to the broken windows theory
provides a good example of such debates. Given the apparent distinction in the nature of

a community-oriented broken windows policing and a control-oriented zero tolerance

approach, it is ironic that they share the same theoretical framework. This paradoxical
situation springs either from a lack of theoretical consideration or from the partial
absorption of the broken windows idea, wittingly or unwittingly. The unfortunate aspect
is that some of the academic and political debates surrounding the New York style order-
maintenance approach have even raised questions about the theoretical value of the
broken windows theory approach (e. g., Harcourt and Ludwig, 2006; Sampson and
Raudenbush, 1999). The author opposes any hasty conclusion, whether positive or
negative, on the integrity and utility of the broken windows theory as well as other
theories. A ﬁnal judgment needs to be based on more credible, theory-driven studies,
which the researcher argues have not been carried out yet.

The bulk of sociological theories, including the broken windows theory, have
certain intrinsic limitations in their propositions and application. This is why they
necessitate empirical tests through which they evolve or are sometimes discarded.
Differential credit should be given to each study, however, depending on the level of
methodological adequacy. The adequacy criteria concerning methodological procedures
such as measurement, study design, and model speciﬁcation are suggested primarily by
theory and prior research. Accordingly, researchers with an aim of theory testing need to
go to great lengths to understand and follow the suggested rigorous methodological
standards. Unfortunately, however, for many reasons, it is often difﬁcult to fully grasp
the implied methodological criteria. One of the most critical reasons is, arguably, the
absence of a comprehensive review of prior studies. The neighborhood disorder research

is a good example. Besides theoretical confusion, the area appears to further suffer from

10

puzzlement over methodological standards, which the author believes is partly due to the
lack of a systematic review of past research."

Prior to constructing the multivariate contextual models, therefore, this
dissertation provides a comprehensive review of the broken windows theory and research
with an aim to clarify sensitive methodological issues.10 The researcher further attempts
to recommend appropriate standards for future research in the ﬁeld. To the extent
possible, the methodological approaches of this dissertation try to meet the most rigorous
criteria by paying special attention to measurement and model speciﬁcation. In this way,
this study is expected to be able to serve as a credible piece of research for future
systematic reviews. Through this effort, the author wishes that future research take a
more theory-driven and methodologically sound approach, and that eventually

conclusions about broken windows theory will be more valid.

Overview

In sum, the LDHP directly addresses the theoretical hypothesis that improving
physical conditions is an attractive alternative to reduce criminal activities, either serious
or minor, across urban neighborhoods. As several scholars have suggested (e. g., Bursik
& Grasmick, 1993; Lab, 1988), collaborative efforts toward revitalizing the physical
environment are expected to signiﬁcantly improve quality-of-life throughout a city by

strengthening neighborhood cohesion and informal social control among the community

 

9 Although Sampson et al. (2002) and Taylor and Harrell (1996) attempted to review the link
between disorder and crime, the former assessed it as one portion of comprehensive
neighborhood effects and the latter focused on the relationship between physical environment and
crime. Thus, it is fair to say that no systematic attempt to review the link between neighborhood
disorder and crime based on the broken windows model has been made yet.

'0 The review in this dissertation is not a meta-analysis, which will be attempted when there is a.
sufﬁcient number of theory-driven studies. Instead, the author focuses on identifying critical
methodological issues in performing empirical research.

11

residents. Although the initial plan of randomized experiment ended prematurely, this
dissertation attempts to perform rigorous non-experimental analyses, using the two waves
of street observations and other various datasets. To accomplish the dissertation
objective, several important methodological issues are identiﬁed and applied to this
dissertation. The author expects that this research can become a valuable addition to the
scientiﬁc knowledge of how public agencies can use civil remedies to ameliorate serious
disorder and crime problems. Further, the contextual approach of this study (i.e., a cross-
level interaction model) is expected to enrich the broken windows explanation of disorder
and crime as well as other research areas about neighborhood characteristics and crime.
This dissertation includes six chapters. In chapter 11, the ecological perspective in
crime research is reviewed to identify a variety of dynamics and concepts leading to
crime within neighborhood contexts. Then the author attempts to clarify causal paths
proposed by the broken windows theory. Chapter III presents a detailed account of past
literature on the broken windows model and contextual effects. Particular attention is
paid to identifying the methodological issues that are likely to inﬂuence study results.
The author further attempts to suggest methodological standards to help work out the
issues. The suggested standards serve as the criteria for the whole research process
throughout this dissertation. Then, a brief description of the current study including such
subjects as study setting (research site), research design, limitations, and hypotheses is
presented in chapter IV. Chapter V describes the methodology. Chapter VI will describe
the study results. Findings from a hierarchical model will be highlighted. In chapter VII,
the author will discuss the weaknesses and strengths of the research and suggest ﬁrture

directions.

12

CHAPTER II: THEORY — MAIN CONCEPTS AND CAUSAL PATHS

As mentioned, a credible study requires a clear understanding of the underlying
theory(ies) and past studies. The huge controversies surrounding the broken windows
idea, research, and policy (e. g., community-oriented broken windows policing vs.
control-oriented zero tolerance policing) originate, arguably, from a lack of a ﬁrm grasp
of the theory, its developmental context, and past empirical tests (see Harcourt & Ludwig,
2006; Kelling & Sousa, 2001; Sampson & Raudenbush, 1999). Accordingly, this
dissertation presents a comprehensive overview of the theory and prior studies as an
attempt to gain a clear grasp of the key theoretical ideas and research issues. This chapter
focuses on identifying the main theoretical concepts and causal paths.

It is well known that the broken windows idea is a branch of theory within the
ecological framework that attempts to explain criminal phenomena in the milieu of
community life. Community dynamics are complex, and as such, resist any simple linear
explanation. While the broken windows model highlights the central role of disorder in
causing crime, there are dozens of important factors associated with crime suggested by
other ecological theories, such as informal social control, environmental structure, and
capable guardianship. Although each ecological theory is often discussed with no
connection with other sibling explanations, the author argues that they are interconnected
with one another; thus, understanding the whole gamut of the ecological perspective
facilitates a ﬁrm grab of each sub-theory. Therefore, this dissertation attempts to
understand the broken windows theory and literature within the ecological framework.

To this end, a brief description of the ecological perspective, sub-theories, major

works, and main causal factors is presented ﬁrst. Next, the broken windows theory is

13

reviewed in detail within the ecological framework. Causal paths of the broken windows
model are presented connecting its main components — structure, disorder, and
neighborhood (psychological) process. The next chapter presents a comprehensive

review of the past studies.

Ecological Perspective in Crime Research

Nature and Scope

The Merriam-Webster online dictionary (2007) deﬁnes ecology as “the
interrelationship of organisms and their environments.” Meanwhile, human ecology, as it
is applied to the relationship between humans and their environment, is deﬁned as “a
branch of sociology dealing especially with the spatial and temporal interrelationships
between humans and their economic, social, and political organization” (Einstadter &
Henry, 1995; see also Hawley. 1950). In a similar vein, one of the main paradigms in
social ecology, COPET, suggests that change and development of human society is an
ongoing interactive process of ﬁve multilevel components —— culture, organization,
population, environment, and technology (Hodge, 1990). Thus, the ultimate purpose of the
ecological analogy in crime explanation would be to understand the interactions of
humans with their community environment that inﬂuence and are inﬂuenced by crime
(Einstadter & Henry, 1995). Humans must be assumed to have a constrained free will, and
the interactions are expected to be complex, multilevel, ongoing, and reciprocal.
Unfortunately, however, the early ecological studies of crime (until the 19103) failed to
incorporate the assumption of soft determinism and dynamic interactions into their theory

and research. Thus, their position was sometimes labeled “environmental determinism”

l4

(Einstadter & Henry, 1995, p. 124). Fortunately, the recent successors of the ecological
explanation (since the 19205’ Chicago School) explicitly appreciate the dynamic
mechanisms involving social forces, culture, and human agency. It seems that no theory
dares to deny the existence of human will these days (Williams & McShane, 2003).
Concerning the viewpoint on society, meanwhile, the ecological framework appears to be
based on, in general, consensual viewpoint. However, most social ecologists, particularly
since the advent of the Chicago School, have recognized diverse and even paradoxical
realities at work within the seemingly whole world (Einstadter & Henry, 1995).

An important question concerns the scope in types of explanations that are regarded
as ecological in crime research. In general, it appears that the ecological framework in the
ﬁeld is simply another expression of macro—level theory. However, it must be noted that not
all macro-level theories are classiﬁed into the ecological perspective (e. g., Mertonian
structural anornie theory). Also, in light of the ecological principle of multilevel interactions,
the level of study is not likely to be an adequate criterion. According to Pratt and Cullen
(2005), spatial variation of crime rates across ecological units (e. g., nation, state, county, city,
neighborhood, etc.) and the underlying structural (or environmental) factors are the main
identiﬁers that tell whether it is an ecological theory or not. Following their idea, this
dissertation includes in the ecological perspective those theories that share common interests
in the spatial pattern of crime (or victimization) and the inﬂuence of environmental
(structural and situational) factors on uneven distribution of crime rates and criminal

decision-making.

15

Social Disorganization Theory: Chicago School

The early Chicago sociologists Robert Park and Ernest Burgess (1925) were very
concerned with the dramatic pace of heterogenic growth. Borrowing the ideas of symbiosis,
invasion, and succession from plant and animal ecology, they attempted to present a human
ecology, “interpreting people in time and space as they naturally appear” through real
observations (Williams & McShane, 2003, p. 59). As such, their basic premise was that the
spatial distribution of humans within urban areas is determined by the competition for
resources and space among them (Villarreal, 2004). The most important contribution of
their works, particularly Burgess’ (1925), was the “conception of the city as a series of
distinctive concentric circles radiating from the central business district” (Williams &
McShane, 2003, p. 59). The central business district was the area of the highest intensity of
land use and fewest residences, and growth or invasion took place when the center intruded
surrounding areas of lower intensity, resulting in intense land uses of the adjacent areas.
The next zone, referred to as the “zone of transition,” was in deteriorating condition due to
the invasion, but most immigrants settled into this area because the living expenses were
low and it was near the factories. The zone next to it was called “zone of workingmen’s
homes,” because most working-class people lived there. Living conditions were somewhat
better than in the zone of transition, and some of the immigrant workers who could afford it
moved into this area. The zone of transition was subsequently occupied by another
immigrant group. Outer areas were increasingly more expensive to live in (Williams &
McShane, 2003). As such, each area in a circle was viewed as having similar demographic,
cultural, physical and economic characteristics that tended to remain stable across time.

Burgess (1925) argued that the formation and residential distribution of the city were due

16

mainly to “the distribution of land use and values,” although cultural and economic factors
also contributed to the process (Einstadter & Henry, 1995, p. 130).

Using the basic zonal model of Park and Burgess, Shaw and McKay (1942)
investigated the spatial pattern of crime and other social problems by mapping a variety of
ofﬁcial data such as juvenile arrests, residences of truants, physical deterioration, incidence
of tuberculosis, and infant mortality. As predicted, the zone of transition had the most
serious crime problem and the problem declined as the distance from the center became
farther. The pattern remained stable for over forty years regardless of the change of racial or
ethnic composition. Subsequent studies of 1 8 other cities by Shaw and McKay (1942) and
other researchers (e. g., Longmoor & Young, 1936; Schmid, 1960) conﬁrmed the pattern
(Einstadter & Henry, 1995). The causes of this spatial pattern were then examined by
ethnographic observations and in-depth interviews with residents (life histories). The
research revealed that the conventional institutions of social control, such as family, school,
church, and voluntary community organizations, were disrupted, social life was superﬁcial
with no ties, and traditional rules or norms did not function well. The negative conditions
were caused by the dilapidated socioeconomic environment, frequent population change,
racial heterogeneity, and culture, conﬂicts among diverse ethnic groups within American
culture. Shaw and McKay named this weakened condition of primary social relationships
and its resulting lack of informal social control as “social disorganization,” which was
considered to be the main cause of crime (Williams & McShane, 2003; Pratt & Cullen,
2005). In other words, devastating structural conditions of the inner-city areas, such as poor
socioeconomic conditions, residential instability, racial heterogeneity, and cultural conﬂict

inﬂuenced the high crime rate, but only through the mediating role of social disorganization.

l7

This insight was gained from the fact that some of the similarly situated areas (e.g., the rural

South) showed low crime rates because of the strong informal social control (Einstadter &

Henry, 1995).

Contemporary Ecological Theories: Resurrection of the Chicago Tradition

After World War 11 through the 19703, the Chicago style neighborhood research lost
much of its appeal as other perspectives became more popular, speciﬁcally functionalism,
control, strain, subculture, and conﬂict theories.ll The Chicago style community-based
study did not disappear, however. In the late 19703, the ecological perspective earned
renewed interest, which has remained until today (Cullen & Agnew, 1999; Williams &
McShane, 2003). The reason for its resurrection can best be understood within the
sociohistorical contexts of the era.

A3 a reaction to the turmoil of the 19603 and the early 19703, sociopolitical
conservatism and economic nee-liberalism gained ground in the US. in the late 19703
through the 19903, which fostered the ideology of individualism. The “me generation” and
“X generation” were reﬂections of the feeling that individuals control their own lives
(Williams & McShane, 2003, p. 273). Criminology and criminal justice were not free from
the inﬂuence of individualism and sociopolitical conservatism. Since criminal behavior was
considered to be mostly dependent on individuals’ rational judgment, the positivistic search
for the cause of crime (i.e., three major positivist theories — control, strain, and learning) and

the critical agenda to unveil the inﬂuence of the power structure and relations on crime lost

 

” It would not be quite true to argue that the thoughts of the Chicago School theorists fell into
decline during the era. Note that Hirsehi’s control theory ﬁnds its origin in the works of Chicago
School theorists such as Shaw and McKay (1942), Reckless (1961), and Reiss (1951). Also,
Cohen’s subcultural theory and Cloward and Ohlin’s differential opportunity theory were
attempts to integrate the ideas of social disorganization and cultural transmission with Mertonian
anomie theory (Williams & McShane, 2003).

18

much of their appeal”. Instead, neo-classical rational perspectives highlighting the critical
role of environmental or situational factors in criminal decision-making and victimization
gained a lot of attention as a new mode of conceptualization. Accordingly, interests in
crime prevention and victim rights grew to an unprecedented degree. Also, many of the
treatment and rehabilitation programs were replaced by harsher punishment of criminals and
more rights for victims (Williams & McShane, 2003).

In this context, four ecological themes emerged. First, one of the growing interests
among criminologists and practitioners was in preventing crime through environmental
design (e. g., Jeffery, 1971; Newman, 1972) rather than in searching for the causes of crime.
It was a (physical) environmental reﬂection of the ideology of free will and general
deterrencel3 . Second, another ecological theme was developed by situational studies such as
routine activity theory (Cohen & Felson, 1979), rational choice theory (Clarke & Cornish,
1985), and crime pattern theory (Brantingham & Brantingham, 1993) that mirrored the
increased interests in victim rights and offender deterrence. It is notable that the situational
studies were fueled by the National Crime Victimization Survey initiated in 1972, which
increased criminologists’ interests in the pattern of victimization depending on various
situational and environmental factors (Williams & McShane, 2003). Third, again reﬂecting
the sociohistorical circumstances of the era, a more general concern was not with the
criminals or their traits, but with the ecology of crime, or spatial variation of crime rates by
ecological units such as states, counties, cities, and neighborhoods, which revived the value

of the Chicago-style social disorganization theory, particularly after the publication of an

 

‘2 Again, it is not true to say that theories of individual criminality fell into decline during this era.
It continued to survive through that time and still attracts considerable attention in the ﬁeld.

’3 Note that free will does not imply a complete free will. It is simply the case that the free will
aspect is more emphasized than the deterministic aspect in the continuum from hard-deterrninism
to free will.

19

article by Judith and Peter Blau in 1982 (Cullen & Agnew, 1999, p. 62). Fourth, the ﬁnal
variation in the ecological perspective came from the integration of environmental design
and social disorganization theories (e.g., Wilson & Kelling, 1982), which posited that
physical and social disorder (or incivility) increase fear of crime and weaken informal social
control mechanisms in neighborhoods, which in turn emboldens motivated offenders and
eventually leads to an increase in serious crime rates.

Although these four groups of theories have somewhat different directions and foci
from each other, they are generally classiﬁed into the ecological perspective in that they
share common interests about spatial patterns of crime (or victimization) and the central role

of environmental (structural) and situational factors.

Environmental Design

Inﬂuenced by Jane Jacob’s (I961) work on urban renewal, C. Ray Jeffery (1971)
suggested the idea of crime prevention through environmental design (CPTED),
underscoring crime prevention through changing the physical environment rather than
through changing the criminal. Further elaborating on J effery’s idea and borrowing the
notion of territoriality from animal ecology, Oscar Newman (1972), an architect, proposed
“defensible space” theory. Newman thought that “any physical area would be better
insulated against crime if those who live there recognize it as their territory and keep careful
watch over the area” (Williams & McShane, 2003, p. 67). The simple and commonsensical
idea of the defensible space theory strongly appealed to the policy makers, and thus many of
its components were adopted by the federal government in making up regulations for public
housing construction. Many modern crime prevention programs, including the

neighborhood watch program, were initiated based on the general idea that environmental

20

design is a more efﬁcient and effective crime prevention strategy than changing criminal
motivations (Williams & McShane, 2003). Newman’s idea, however, was not free from
criticism. Merry (1981), for example, argued that improving surveillance in physical
structures did not automatically result in increased surveillance of people. Further,
offenders become familiar with surveillance immediately. Merry attacked the defensible

space idea for ignoring other dimensions of social control (Eck & Weisburd, 1995).

Situational Studies

Situational studies such as routine activity theory (Cohen & F elson, 1979),
rational choice theory (Clarke & Cornish, 1985), and crime pattern theory (Brantingham
& Brantingham, 1993) are generally classiﬁed as neo-classical theories in that they
assume that individuals make their own decisions taking various situations into account.
Also, the theorists are interested in deterrence. The routine activity theory was originally
an attempt to explain the distribution and change of crime rates as a result of the
structural change of individuals’ routine activities, not as an outcome of neighborhoods’
change in their structural characteristics. The paradoxical reality that predatory crime
rates have increased in spite of the apparent improvement of the structural socioeconomic
conditions of American neighborhoods since 1960 stimulated the development of the
routine activity framework, casting doubt on the credibility of the traditional macro-level
(root-cause) explanations proposing a positive relationship between poor neighborhood
conditions and crime rates. The routine activity theory consists of two basic premises.
First, in order for a crime to occur, motivated offenders must converge in time and space
with suitabletargets in the absence of capable guardians. Second, the probability of this

occurring is inﬂuenced by our routine activities, including work, family, leisure, and

21

consumption activities. Cohen and Felson (1979, p. 589) claimed that such a
conceptualization could help to “develop an extension of the human ecological analysis
to the problem of explaining changes in crime rates overtime.” In addition to macro-
level changes in crime, the routine activity theory provides a very convincing explanation
of the differential probability of individual victimization. As such, the theory has often
been used to explain both spatial variation in crime rates and individual differences of
victimization by sociodemographic groups. Further contemporary usage of the theory
can be found in its logic about hot spots that are conceptualized as the places where
intersection of offender, target, and absence of capable guardian is most likely to occur
(Sherman & Weisburd, 1995; Williams & McShane, 2003). As the importance of places
increases, land use patterns (e. g., business, residence, entertaimnent, leisure, etc.) are
gaining more attention as a relevant proxy measure of routine activity (e. g., Kurtz, Koons,
& Taylor, 1998; Sampson & Raudenbush, 1999; Wilcox, Quisenberry, Cabrera, &
Shayne, 2004).

Do the (offenders always commit crime when they ﬁnd valuable targets that are not
protected by capable guardians? Probably not. Then why do offenders engage in certain
types of crimes in certain situations? Why is guardianship important in crime prevention?
Such conceptual questions are not answered by the routine activity theory. As a solution,
Clarke and Cornish (1985) proposed the rational choice theory (see also Cornish & Clarke,
1987), which attempts to model crime-speciﬁc decision-making processes of offenders,
underscoring the rational aspect of human activities. The rationality is bounded, however,
by the limited information available to offenders (see Simon’s discussion on bounded

rationality, 1957). Also, the rational choice theory does not aim for general deterrence

22

because both the demands of offenders and the choice structure of each crime type are
speciﬁc in time and space. Since the motivated offenders tend to seek to ﬁrlﬁll their desires
in the easiest way at the lowest cost, capable guardianship is more than likely to thwart
attempted lawbreaking, and thus guardianship is crucial in crime prevention. Another
implication is that crime can be reduced through decreasing target attractiveness or beneﬁts
of crime.

Felson (1987) extended the concept of capable guardian into three types — intimate
handlers, capable guardians, and place managers. Consistent with informal control theory,
intimate handlers are those who can exert direct inﬂuence over potential offenders, such as
parents, teachers, employers, and friends. Capable guardians are people who protect targets.
They could be either informal or formal agents. As such, they include friends, security
guards, and public police. This concept of guardians who protect targets put an emphasis on
the critical role of formal social control. Lastly, “place managers” refers to people who take
care of places such as apartments, commercial buildings, and parks. They regulate behavior
at the facilities or surrounding locations (Eek & Weisburd, 1995). The introduction of the
concept of place managers shifts the theoretical attention from large-scale place to small-
scale place and facility (Taylor, 1997). Accordingly, it leads to a connection with the
environmental design thesis, particularly with the defensible space theory.

The crime pattern theory (Brantingham & Brantingham, 1993) extends the
situational explanations already discussed to include the uneven distribution of opportunity
caused by neighborhood characteristics as an antecedent factor to routine activity and
rational choice. The main concern of this theory is to explore the interactions between

offenders and the physical and social environment that inﬂuence the choice of target.

23

Offenders are assumed to be normal actors with legitimate jobs. They become aware of
criminal opportunities while they are conducting their normal activities. Only a few, if any,
serious offenders (mostly with no jobs) would aggressively seek out unfamiliar areas for
more valuable targets. By tracking offender movements (geographic proﬁling), researchers
attempt to model where the offender will commit the next crime. This strategy can be useful
in investigations of serial crimes such as serial murder, serial rape, and serial arson (Eek &

Weisburd, 1995; Taylor, 1997).

New Social Disorganization Theory

The situational theories fueled the resurrection of scholarly interests in the location
or place of crime. Such interests were extended to apply Shaw and McKay’s ideas in
contemporary urban settings (Williams & McShane, 2003). The main idea of the new
social disorganization theory is similar to that of its traditional counterpart developed by
Chicago scholars: disadvantaged neighborhood conditions make people distrust each
other and withdraw from involvement in community activities for the common good,
which in turn increases the crime rates in the neighborhoods. Disadvantaged
neighborhood conditions refer to socioeconomic characteristics of neighborhoods that are
commonly measured by concentrated disadvantage, residential instability, and immigrant
concentration (see Sampson et al., 1997; Sampson & Raudenbush, 1999; Sampson,

Morenoff, & Earls, 1999; Morenoff, Sampson, & Raudenbush, 2001 )M. In contrast, such

 

1" Concentrated disadvantage is a combined measure of poverty, receipt of public assistance,
unemployment, female-headed households, density of children, and percentage of black residents.
Residential instability is obtained by merging the percentage of persons not living in the same
house as 5 years earlier and the percentage of renter-occupied homes. Immigrant concentration is
a factor combining the percentage of Latinos and the percentage of foreign-bom persons. These
factors are now widely accepted in many studies (e.g., Wilcox et al., 2004). However, note that
Sampson et a1. (1997) used a residential stability factor, not residential instability, combining the

24

collective neighborhood processes as mutual trust (social cohesion) and active
intervention (informal social control) are regarded as the core factors necessary to break
the link between disadvantaged neighborhood characteristics and crime rates.

Empirical evidence of the vital role of neighborhood processes has been
established by Robert Sampson and his colleagues (Kubrin & Weitzer, 2003b). For
example, in a multilevel study across 343 Chicago neighborhood clusters, Sampson et al.
(1997) found that three structural factors — concentrated disadvantage, residential stability,
and immigrant concentration — explained 70% of the variation in the neighborhood
processes, which in turn mediated the effects of residential stability and concentrated
disadvantage on violence rates. The neighborhood processes remained the most robust
predictor of lower rates of crime, even after controlling for individual characteristics and
prior violence. They termed the neighborhood psychological processes of social
cohesion and informal social control “collective efﬁcacy.” In other words, collective
efﬁcacy is an elaborated concept of social disorganization. In sum, the disadvantaged
neighborhood conditions are conceptualized to have an indirect effect on crime rates
through the mediating role of collective efﬁcacy. As a result, some neighborhoods with
poor conditions would not have high crime rates if the level of collective efﬁcacy were
strong enough to intervene in the link between social conditions and crime rates.

The causal logic of the new social disorganization theory has been expanded by
accepting the insights of the conﬂict perspective, or theory of political economy. Such

ideas as the construction of public housing in disadvantaged neighborhoods,

 

percentage of persons living in the same house as 5 years earlier and the percentage of owner-
occupied homes. The author argues that the residential instability goes better with the other two
factors and ﬁts better into the traditional interests in population mobility and socioeconomic
change.

25

deindustrialization, and racial stratiﬁcation have been explored. For example, Bursik
(1988; Bursik & Grasmick, 1993) argued that structural disadvantage in certain urban areas
is not necessarily a natural outcome of ecological competition for resources, but a political
product of public decision-making, such as the decision to construct a large-scale public
housing facility in an already dilapidated neighborhood. Wilson (1987) further argued that
the multiple forms of concentrated disadvantage were due to deindustrialization and
outmigration of middle-class residents. In particular, the exodus of middle-class people who
are the core of legitimate neighborhood processes left behind only the most disadvantaged
residents. Sampson et al. (1997) also recognized that collective efﬁcacy is imbedded in
structural contexts and a wider political economy that stratiﬁes places of residence by key

social characteristics.

Broken Windows in Brief

The basic premise of the broken windows model (Wilson & Kelling, 1982) is that
signs of disorder, either physical or social, lead to lack of neighborhood caring and
increase of fear which, in turn, further aggravate crime and disorderly conditions.
Skogan’s (1990) contemporary revision extends the theoretical outcome (i.e., crime) to
neighborhood decay and is more consistent with the ecological account. Three main
characteristics of the contemporary idea are as follows. First, disorder and crime are
parts of the cyclic development of neighborhood change, and thus they must be
understood as a dimension of the whole developmental process. Second, disorder is
theoretically conceptualized to have an indirect effect on the increase of crime. Social
psychological processes (e.g., collective efﬁcacy, social cohesion, place attachment)

serve as the mediating variable. Third, crime is not only an outcome of disorder but also

26

functions as a cause of disorder. This theme springs from the conceptualization of

reciprocal inﬂuences between crime and disorder.

Summary: Main Causal Factors

The ultimate aim of reviewing the development of the ecological perspective in
crime research is to understand the multidimensional interactions of human actors and
their neighborhood environment that inﬂuence and are inﬂuenced by crime. Humans are
assumed to have a constrained free will, and the diverse nature of neighborhood contexts
and processes are appreciated. Notwithstanding differential foci and orientations, the
various ecological theories share common interests in the spatial pattern of crime (or
victimization) and the central role of environmental (structural and situational) elements
in leading to uneven distribution of crime rates and/or criminal decision-making.

Inﬂuenced by sociopolitical conservatism, economic neo-liberalism, and cultural
individualism, the contemporary versions of the ecological analogy put an emphasis on
crime prevention through a proactive manipulation of neighborhood structures, processes,
and situational elements. Redirection of interests away from individual characteristics to
neighborhood forces has enriched the understanding of causal dynamics in criminology.
Figure 3 describes the main ecological factors and processes proposed by the ecological
theories. A general implication is that political economy inﬂuences neighborhood
structures, either social or physical, which in turn affect the level of disorder. The
capacity of informal control among residents is determined either directly by structural
factors (i.e., environmental design, new social disorganization) or indirectly through
disorder (i.e., broken windows). Informal social control is often capable of deterring

criminal decision-making of motivated offenders, but it may not be enough. The

27

potential offenders further consider various situational factors, looking for easy targets
requiring less effort (i.e., routine activity, rational choice). Disadvantaged neighborhoods
would provide more and better opportunities to the potential offenders than their
advantaged counterparts do (i.e., crime pattern). This effort to understand the
interrelationships between ecological theories and factors is expected to increase the

explanatory power of the full ecological perspective and its sub-theories.

Figure 3. Ecological Factors and Processes

 

Political Economy
- public housing in disadvantaged neighborhoods - deindustrialization - racial stratiﬁcation

 

 

 

 

 

7

 

 

 

 

 

 

 

 

 

Social Structure Physical Structure
- concentrated disadvantage - residential instability - urban design (defensible space)
- irrunigrant concentration - land use
Disorder

- vandalism - intoxication - panhandling - prostitution - abandoned building - trash - grafﬁti

 

 

 

 

 

 

Process
- social disorganization - collective efﬁcacy - territorial functioning

 

 

 

 

 

 

Situation
- guardianship (intimate handler, place manager, capable guardian) - valuable targets

 

 

 

28

Since the purpose of this dissertation is to test the broken windows model, a more
detailed description of that theory is presented in the next section. In that part of the
literature review, the ecological characteristics of the broken windows theory are
emphasized, and causal paths of the model are identiﬁed. In the next chapter, a
comprehensive review of prior research is attempted for each causal step revealed

through a consideration of the broken window’s model.

Broken Windows in Detail

Even though several theorists have highlighted the undesirable role of disorder
(e.g., Garofalo & Laub, 1978; Hunter, 1978; Wilson, 1975), their primary outcome of
interest was fear of crime (Taylor, 1999). The ﬁrst attempt to relate crime to disorder
was made by Wilson and Kelling in 1982. Wilson and Kelling argued that disorderly
street behaviors (e. g., panhandling, prostitution, loitering, vandalism, etc.) and physical
conditions (e. g., litter, grafﬁti, abandoned building, abandoned vehicle, etc.) that are in
disrepair make residents withdraw from public spaces and feel fear of crime. This
withdrawal and fear embolden local offenders and teenagers, resulting in more social and
physical disorder. Increased disorder and lack of informal social control cause residents
to become more fearful and sometimes move out of the community, which attracts
outside serious offenders into the locale and eventually the serious crime rate increases
(Taylor & Harrell, 1996). This sequential connection between disorder, fear of crime,
informal social control, and serious crime suggests that disorder is one of the key causes
of problems, and thus, it should be eradicated to restore healthy neighborhoods where
residents care for one another and volunteer to set and keep informal rules for their own

communities.

29

Extending and elaborating Wilson and Kelling’s broken windows theory, Skogan
(1990) focused on neighborhood decline as the ultimate outcome of interest (Taylor,
1999). Further, unequal distribution of disorder across communities was viewed as
rooted in structured inequality, which extended attention to social condition and urban
inequality as two causes of disorder (Taylor, 1999; Wilson, 1996). Skogan proposed
spiral processes: undesirable structural conditions give rise to disorder, which increases
fear and undermines informal social control and neighborhood satisfaction, which, in turn,
motivates residents to move out and decreases property value. Unattached outsiders as
well as offenders move into the locale primarily due to the low property value,
prevalence of disorder, and lack of informal social control, which gives rise to serious
crime, which, in turn, further worsens neighborhood cohesion and structural conditions.

After the publication of Skogan’s thesis, many scholars have further examined the
basic relationship between disorder and crime using a variety of data sources in diverse
contexts. Although research ﬁndings have not been consistent across studies and several
critical issues relating to theory and research have been raised and debated, the
contemporary theoretical model on disorder and crime is very similar to Skogan’s notion,
except for a few added neighborhood characteristics (e. g., land use) and reﬁned
psychological factors (e. g., collective efﬁcacy). It is characterized by three main themes.
First, disorder and crime are parts of the cyclic development of neighborhood change. To
Skogan, the root of the problem is in neighborhood characteristics and the ultimate
outcome is neighborhood decay. Accordingly, disorder and crime must be understood as
a dimension of the whole developmental process. Second, social psychological processes

such as collective efﬁcacy, social cohesion, place attachment, and resident-based

30

informal control mediate the effect of disorder on crime. That is, disorder is theoretically
conceptualized to have an indirect effect on the increase of crime. Finally, crime is not
only an outcome of disorder but also functions as a cause of disorder. This theme springs
from the conceptualization of reciprocal causation. That is, the spiral model must be
understood as a kind of feedback loop.

In sum, the broken windows idea was based on the assumption that street criminals,
like general individuals, have (constrained) free will and they choose to commit crimes by
considering various situational factors. Thus, criminologists and policy makers believed
that street crimes can be effectively and efﬁciently deterred by dealing with the situational
factors. In the broken windows model, physical and social disorders are representative of
the situational factors, and their distribution is assumed to be inﬂuenced by broader
structural factors, or socioeconomic conditions. Disorderly conditions and behaviors signal
to motivated offenders that their criminal behaviors would not be intervened in within the
neighborhoods (i.e., aggregate lack of capable guardianship). In contrast, systematic efforts
to reduce disorder are believed to constrain the will of motivated offenders. As such, it must
be noted that the broken windows theory emphasizes that strengthening collective efﬁcacy
via collaborative efforts among citizens, police, and govemment is the key to cut the link

between disorder and crime (Bratton & Kelling, 2006).15

Causal Paths
Figure 4 describes the contemporary broken windows model that shows the

sequential connection of neighborhood characteristics (structure), disorder, collective

 

'5 Therefore, although some researchers (e.g., Sampson & Raudenbush, 1999) argue that disorder
and crime are intrinsically the same concepts, the author would like to propose that disorder is a
concept closer to lack of collective guardianship than to crime. If this is true, researchers will be
able to use disorder as a proxy measure for the level of collective guardianship.

31

neighborhood processes, and serious crime. Although the theory highlights the role of
disorder, the other exogenous components cannot be disregarded, which would result in a
violation of the ecological assumptions. Thus, the author proposes that each component
be given equal weight in constructing and testing research models. In this vein, this
dissertation attempts to identify how variables in each component play their own roles in

leading to crime in neighborhood research.

Figure 4. Contemporary Theoretical Model Linking Disorder and Crime

Structure Disorder

Process Process

Serious
Crime

 

0 Structure (Neighborhood Characteristics): concentrated disadvantage, immigrant
concentration, residential instability, land use, etc.

0 Process (Collective Social Psychological Processes): collective efﬁcacy (social cohesion
+ informal social control), place attachment, resident-based control, etc.

32

CHAPTER III: LITERATURE REVIEW

The broken windows theory, as it is understood within the ecological paradigm,
suggests that the uneven distribution of crime is an outcome of the interactions among
neighborhood structure (characteristics), disorder, and process. As mentioned in the
previous chapter, each causal component must be considered equally important. Thus,
the ﬁrst section of this chapter describes how each component contributes to explaining
the distribution of crime across various communities.

However, it must be noted that, just as the interaction dynamics within various
neighborhood contexts are diverse, so are the study results. Although any meta-analytic
attempt to ﬁnd a consistent pattern of ﬁndings and its underlying cause is beyond the
scope of this dissertation, the author attempts to identify critical, mostly methodological,
issues that are likely to inﬂuence the study outcomes concerning the relationship between
disorder and crime. As such, the author further presents several examples with
unsupportive and/or mixed ﬁndings. Then, the methodological issues are discussed in
detail, which will further serve as the criteria of conducting research for this dissertation.

In the next section, the author presents several examples of contextual effects.
Context-speciﬁc understanding has great potential for theoretical elaboration and public
policy. This dissertation attempts to test a cross-level interaction model including
structure, disorder, and serious crime. Therefore, although no human subjects are
involved in this study, the review of contextual effects would provide diverse senses of

interaction dynamics among structure, disorder, process, and individuals.

33

Structure, Disorder, and Process

Neighborhood Structure and Crime

Neighborhood research begins with a consideration of the structural
characteristics of communities. Although the conﬂict dimension (i.e., political economy)
is sometimes speciﬁed as an antecedent factor to the structure, it is beyond the scope of
this dissertation. Neighborhood structure consists of social and physical dimensions,
though most literature has paid attention to only one side, to the exclusion of the other.
As mentioned above (see also Figure 3), social structure includes socioeconomic
disadvantage, residential instability (or mobility), and immigrant concentration (or ethnic
heterogeneity). On the other hand, land use and urban design (defensible space) are
representative of the physical structure. More attention is paid to land use than to urban
design, because urban design is not controlled for in this study.

Social Structure

A large body of research exists on the effects of social structure on crime.
However, a few selections would be enough to provide a good sense of the relationship
between structural factors and crime. Scholarly interests in the inﬂuence of social
conditions on crime can be traced back to the early works by moral statisticians such as
Guerry (1833) and Quetelet (1835), who ﬁrst noted the uneven geographical distribution
of crime (Cohen & Felson, 1979). In research on the effects of diverse demographic and
environmental aspects, such as season, climate, population, poverty, and geographical
distribution, on crime rate, they reached a general conclusion that societal conditions are the

causes of crime (Einstadter & Henry, 1995, p. 122, emphasis original). Meanwhile, parallel

34

to the notions of Durkheim and the Chicago School, Quetelet (1831) found the causes of
crime in rapid socioeconomic changes and relative deprivation (Einstadter & Henry, 1995).

Sampson’s early works show a similar result. In research using the National
Crime Survey for the years 1973-1975, Sampson (1985) found signiﬁcant effects of
residential mobility and female-headed households on victimization rates of personal
crime. Two years later, be (1987) reported that the high crime rates in black
neighborhoods were because of poor economic conditions and high unemployment rates
(Hayslett-McCall, 2002).

Meanwhile, holding the general ecological position that underscored place over
person, or “kinds of places” explanations, Stark (1987, p. 893) argued that crime among
black people is not universally high, but depends on where they live. As an example, the
crime rates of black people in the South were not that high because many of them lived in
suburbs and rural. areas where the physical and structural environments were not conducive
to crime. Alternatively, the high crime rates of black people in northern cities were
attributable to urban environments that induced anyone to engage in criminal activities
(Stark, 1987, p. 906, emphasis original).

Warner and Pierce (1993) and Smith et al. (2000) also reported supportive ﬁndings.
Using calls to the police during 1980 in 60 Boston neighborhoods, Warner and Pierce
discovered that poverty has a signiﬁcant effect on burglary and assault rates, after
controlling for the percentage of female-headed households and structural density. Poverty
also increased robbery rates when residential mobility was low. Interestingly, the role of
racial heterogeneity and residential mobility in their model was conditional on the level of

poverty. In an attempt to test an integrated model of social disorganization and routine

35

activities, Smith et al. (2000) discovered that the number of single-parent households was
consistently associated with more street robberies, controlling for a range of variables
relating to social disorganization, land use, and guardianship/parochial control (Hayslett-
McCall, 2002).

Finally, the devastating impact of community disadvantage has also been reported in
juvenile studies. For example, Hay and his colleagues (2006) were interested in whether
neighborhood contexts inﬂuence the link between family environment and juvenile crime.
OLS product term analysis found that community poverty had a huge impact on the effect of
family problems on juvenile delinquency (40% increase per one standard deviation increase
in poverty). Such interaction effects suggested that juveniles with family problems are
involved in considerably more delinquent activities if they live in poor neighborhoods than
in afﬂuent ones. Elliott et al. (1996) further tested for whether the effects of neighborhood
disadvantage on adolescents’ development are mediated by the organizational and cultural
features of youths’ neighborhoods. Path analysis found supportive results. Meanwhile, the
HLM analysis revealed that neighborhood factors had substantial impact on juvenile
development measured by prosocial competence. conventional ﬁiends, and problem
behavior.

Physical Structure

Regarding the link between urban design and crime, few contemporary
criminologists argue that urban design or defensible space measures are directly
associated with crime prevention. Improved territorial sense among residents through the
renovation of urban design is considered to be the key factor to deal with crime. A rare

example of the direct link between urban design and crime is found in a study by Perkins,

Wandersman, Rich, and Taylor (1993). They found that some measures of built
environment such as street visibility, private outdoor lighting, and narrow and visible
street are inversely related to crimes. However, other measures for defensible space such
as average building size and near-home barriers were not effective in reducing crimes.

The land use pattern has received much attention from ecological researchers,
especially from routine activity scholars. Routine activity theorists (e.g., Cohen & Felson,
1979; F elson, 1987) propose two ways that land use may affect crime. One is by
impairing resident-based control capacity of an. area, and the other is by increasing
particular types of activities such as alcohol consumption and illicit drug use. Both
dynamics can be explained by a rather general measure of non-residential land use, which
includes commercial (e. g., bars, convenient stores, shopping centers, etc.) and public area
land uses (e.g., parks, libraries, recreation centers, bus stops, etc.). In other words, non-
residential types of land uses tend to increase population mobility and anonymity, which
in turn inhibit informal social control. Also, the increase of particular types of routine
activities due to non-residential land use is likely to heighten the chance of intersection
between potential offenders and victims (Hayslett-McCall, 2002).

Many researchers have examined the role of non-residential land use in leading to
crime. As an example, Taylor and Harrell (1996) found a positive relationship between
the proportion of nonresidential land use and burglary rates (Hayslett-McCall, 2002).
Rountree et al. (1994) also reported that busy-place measures (e.g., schools, convenience
stores, bars, fast-food restaurants, ofﬁce buildings, parks or playgrounds, shopping malls,
hotels, bus stops) are signiﬁcantly related to violent victimization risk. They further

noted that the role of resident-based informal control is the key to connect land use and

37

victimization. Meanwhile, Wilcox et al (2004) discovered that the proportion of
business-oriented establishments increases violent crime, and the effect is partially
mediated by neighborhood disorder. ”’

Prior studies on speciﬁc types of land uses deserve attention. The link between
mass transit stops (e.g., bus stops, subway stops) and high victimization risks have been
well-established by several studies (Block & Block, 2000; Block & Davis, 1996; Roncek,
2000). Sherman, Gartin, and Buerger (1989) reported that public parks are strong
candidates as crime hot spots. Meanwhile, Smith et al. (2000) found a signiﬁcant effect
in the number of parking lots on robbery rates. Public schools also appeared to be
signiﬁcantly related to a higher number of robberies (Roncek, 2000). Convenience stores
(Gordon & Brill, 1996) and bars (Peterson et al., 2000; Roncek & Maier, 1991) have
been reported to be crime generators. However, Peterson et al. (2000) found that some
institutions such as libraries and retail establishments were not related to violent crime
rates (Hayslett-McCall, 2002). Similarly, Brantingham and Brantingham (1982)
observed a mixed pattern depending on types of facilities. While fast food restaurants,
traditional restaurants, and pubs had 2 to 2.5 times higher commercial burglaries,

supermarkets and department stores had no effects (Eck & Weisburd, 1995).

 

1" Several studies examined the effect of land use on disorder. In a study performed in two cities,
Taylor and his colleagues (1995) discovered that the level of nonresidential land use on street
blocks is signiﬁcantly related to the level of physical deterioration. A similar result was found in
Kurtz et al. (1998). They further found that the link between nonresidential land use and physical
disorder was mediated by the level of informal control.

38

Neighborhood Disorder and Crime

Positive Findings

There is a large body of research on the effect of disorder on crime, and some of it
supports the causal link between disorder and crime. The ﬁrst systematic attempt to
provide empirical evidence of this link was made by Skogan (1990). Using survey data
of 13,000 residents in 40 neighborhoods of six different cities, he found that disorder was
signiﬁcantly related to fear of crime and crime. He further reported that increased fear of
crime undermined neighborhood social control by increasing mobility, reducing
neighborhood identiﬁcation and feelings of territoriality, reducing supervision and mutual
obligation, and leading to withdrawal from neighborhood life. Skogan found that
disorder had a contagion effect and noted a spiral process where disorder led to crime and
crime fed back to higher levels of disorder and reduced capacity for neighborhood control.
Kelling and Coles (1996) supported Skogan’s ﬁndings and idea. They concluded that
Skogan’s research veriﬁed the causal link between neighborhood disorder and serious
crime.

After Skogan’s research, several studies found positive results. For example,
using Queensland Crime Victims Survey data (1991), Borooah and Carcach (1997) tested
for the relationship of perceived neighborhood incivility on increased fear of crime and
crime victimization, after controlling for neighborhood cohesion. They measured fear of
crime and victimization in two ways — personal and housing. They discovered a
consistent result that the perceived disorder had a signiﬁcant effect on both types of

crimes and fear.

39

Perkins, Meeks, and Taylor (1992) were interested in whether actual physical
disorder (measured by Block Environmental Inventory) has a signiﬁcant effect on
perceived crime problems such as drug dealing, robbery, assault, and burglary. Noting
that there are also positive, crime-reducing cues in neighborhoods reﬂecting the complex
disorder-crime dynamics, they included physical, signs of defensible space (e. g., outside
visibility, barrier on property, private outdoor lighting, public street lights, bars on
windows) and territorial functioning (e.g., private plantings, yard decorations) in their
model. The multiple regression analyses showed that the observed physical disorder was
signiﬁcantly associated with perceived crimes. Further, the defensible space and
territorial functioning measures appeared to reduce the perception of crime, controlling
for the observed physical disorder.

Signiﬁcant ﬁndings were also reported by Wilcox et al. (2004). Using a variety
of datasets such as police records of crime (Seattle PD), US. Census reports, and
telephone surveys of 5,302 Seattle residents, they attempted to examine the mediating
role of neighboring'7 and/or disorder in the link between land use and two types of crimes,
violent crime and burglary. Negative binomial regression analyses revealed that, after
controlling for neighboring, land use, and community characteristics (i.e., concentrated
disadvantage, population instability), the effect of perceived physical disorder on violent
crime and burglary remained signiﬁcant.

Negative Findings

Although Skogan’s (1990) research has been regarded as the ﬁrst systematic
effort to verify the causal links among structure, disorder, process, and crime, some

researchers cast doubt on Skogan’s conclusions. After reanalyzing Skogan’s data, for

 

'7 Neighboring may be deﬁned as “the act of being neighborly to those around one.”

40

example, Harcourt (1998) disclaimed Skogan’s conclusions. He found that the effects of
disorder on burglary, rape, and physical assault became nonsigniﬁcant after controlling
for other explanatory variables. Further, inﬂuenced by Harcourt’s ﬁndings, Eek and
Maguire (2000) concluded that Skogan’s ﬁndings are very sensitive to outliers and thus
cannot hold strong policy implications (Harcourt & Ludwig, 2006; Worrall, 2002).

Meanwhile, some studies negated the broken windows hypotheses. In a recursive
lagged model, Markowitz et al. (2001) found that the signiﬁcant inﬂuence of perceived
disorder on burglary victimization became non-signiﬁcant after controlling for prior
levels of burglary and other community variables. Another longitudinal study by Brown,
Perkins, and Brown (2004) discovered that observed physical disorder change (from time
1 to time 2) was not signiﬁcantly related to the change of police-reported crime.

Mixed Findings

Strong evidence for the complex broken windows dynamics can be found in
studies that discovered mixed results depending on model speciﬁcation, level of analysis,
crime type, and incivility type. In one of the most systematic and rigorous studies,
Sampson and Raudenbush ( 1999) reported contradictory results depending on model
speciﬁcation and crime types. In their recursive model, observed disorder had a
signiﬁcant effect on police-reported homicide and robbery. However, the effects on
victimization (burglary and violent crime) and police—reported burglary were not
signiﬁcant. In the non-recursive model, they found that police-reported robbery was
signiﬁcantly inﬂuenced by observed disorder, but police-reported homicide was not.

Perkins and his colleagues (1993) also found different results depending on

measures of disorder and crime. Controlling for informal social control, territorial

41

symbols, built environment, and neighborhood characteristics, the multiple regression
model showed that perceived disorder had no signiﬁcant effects on all types of crimes,
but that observed physical disorder showed mixed relationships with diverse crime
measures. While it had a signiﬁcantinﬂuence on perceived delinquency and self-
reported victimization, perceived index crime and police-reported serious crime were not
signiﬁcantly connected to it. Interestingly, the observed physical disorder appeared to
decrease police-reported quality-of—life crimes.

Studies by Taylor (2001) and Robinson, Lawton, Taylor, and Perkins (2003) also
produced mixed results. In a longitudinal research project in 66 neighborhoods in
Baltimore, Taylor found different outcomes depending on types of incivilities. He
suggested that different types of policies are necessary for different types of incivilities.
Meanwhile, Robinson et al.’s study reported mixed results by level of analysis. At
individual level analysis, perceived disorder had a signiﬁcant effect on fear of crime in
both lagged and change models. However, null results were found at the street-block
level.

The mediating mechanism has received mixed support. For example, in a study
on the effect of disorder on juvenile illicit drug use, Jang and Johnson (2001) found
differential results depending on types of drugs (e.g., hard drug use and marijuana use)
and types of mediators (e.g., religiosity, social bonding, and social learning). The
analyses of three waves of National Youth Survey (Elliott et al., 1989) discovered that
while juvenile religiosity mediates the effect of perceived disorder on self-reported hard
drug use, the effect of perceived disorder on self-reported marijuana use was not

mediated by the religiosity. Interestingly, and somewhat unexpectedly, social bonding

42

and social learning, which have long been regarded as crucial social psychological
processes that substantially protect juveniles from becoming deviant, did not explain the
direct effect of disorder on marijuana or hard drug use.

Another example is found in Wilcox et al.’s (2004) study. They were interested
in whether neighboring mediates the effect of disorder on crime, or disorder arbitrates the
effect of neighboring on crime. It was found that neighboring did not explain the effect
of disorder on violent crime and burglary. However, disorder intervened in the effect of
neighboring on violent crime. The results appeared to be contradictory to the general.
propositions of the collective efﬁcacy theory proposed by Sampson and his colleagues
(1997, 1999). At the same time, however, the intervening effect of disorder on the
relationship between neighboring and crime suggested an alternative method of
conceptualization with the resulting policy implication that dealing with the
neighborhood disorder ﬁrst could decrease more serious crime problems caused by the

weakened neighborhood bonding.

Methodological Issues

As mentioned, the purpose of this section. was to understand why the study
outcomes are inconsistent. The review of theory and prior studies revealed several
important and sensitive issues with regard to measure, data source, control factors, and
study design. In a concise format, Table 1 presents the diverse characteristics of past
studies linking disorder and crime that are discussed above. The studies took inconsistent
methodological approaches. which seems to have inﬂuenced their ﬁndings. This sub-
section illustrates those issues, and the author’s suggestions are presented. The issues

form the basis of the research plan of this dissertation.

43

 

moumtoeoﬁmso

 

 

 

 

 

 

 

3:239: <3 .233m
a. beige gala
access 52 E =§oaoz
we U> 3% m T 3.: . 058m <2 Mmo seam 8:88:05 3: on: $5 a. 3:322
mum—wen: Horn O<E cam—mam
.m: ”moi E .CﬂmSm I wwa E Oman . ”mozmceoﬂago mAmnv AmOmV A: :83 :a
.mc ”30— E mum—wezm .I Ewa— E Omnn ._ bingou :23on 0: 25m 5:322:85 Gav Omnn ~8st 3 $35me
6mm Q
Beau m9: 28.825
.wﬁ ”3:0; 8 823555 l 363 052 OZ Q C VOE $3:
883. 3:8ch . <2 385:: Mm0 “cabana mEND :a Are; OnO ”OMB :a 6 520:
.m: ”wade Em: co wEEmo: Q mEvcom .
.m: “meanness. :o wEEmo: Q wEecom .
.mﬁ ”mac Em: :o EROOwL—om . wEEmB .m.O
.mc manganese no 56293: . bwmewzon Eoom Cool
scum wanes: .N $868220 @588 22 6?: 22V 2.: angel.
.wﬁ uoobm :82 .: 3:239: Eoom ”O: 25m toaomﬁom Omnn :OCmZ Q mama
aim... 9 :poEEOlAman—COaO A: .
.2me 9 .CoECOTAmHozbOnO 4m . mocmcgoﬁmzu 352865 .35 8:5 :mm
.m: UAm... 9 TC oEtO .l E. a OnO 1:: . 5:58:50 Legions/x :32 058% 05 Gav Omnn mm; Avoomv
.wi Ame 9 TC 22:0 l ; 8 OnO :m . Q .2629: 2:03 MO: EEontO tone: oozon x<v OnO Mem .2 no 2265
258 Bioouon $8322: 35m «2833:.
E35 .m> :83: oEoonm Q 6:28830
.wﬁ ”uEto weasel .1 Oman .. ”mocwtoaoﬂmco 0:2 :2 Q C 05 mMDV Anaanzomocau
.wi SEE Ecoﬂon .l Omnn . .9636: ”:23on ”m0 oEmm 20:35:65 Omnn VOE Q smoocom
3.59: n 3:05.80 3805 cwnmoO mousom 2:302 2:282 chom com—n
a o c o a a
532 m2 ”Oar—0&5: Eonwo—osuxmm amt—5m SNQ DECO .535an um Toezaﬁ
$50 Eoom

 

oECO so 5385 no mpoonnm 2: :0 363m 33.56:: no meEsm ._ 2an

44

 

88 ”>8 so 9: .

 

 

 

.8 H> a. m 8 888882 . 6mm 8
8mm 86882 .N 8858828 mbnv <3 6.88m
.8 U> a. m T 88 . 32.5888 o<2 6 52 Goals
8&8 =82 ._ 88 83 888882 Mmu 388:5 :88 8:8 E on: m$5 a 8 883
.8 ”EV tone oozoa l OmnO ..
.wﬁ. A”: roam: oo:on l OmaO .
Zebu 288903: 02853352 .N 562% mm:

.8 ”8V :88 8:8 T ammo . 8:288 :2 a c a a .8820
as ”E a E :88 82.8 T 080 . 8: .83 :88 8:8 2V own: man: gram
.8 ”9 a. E 888885 T 080 . ”8582888 38:8 52 E mg on: ﬂag

Hoots”: c.2853: ._ 3:58:50 9:88:00 ”m0 EBotO mousing; OmnO MOMS 3%
89:28:30
8385

Q 5:55.800 <3 .233m
82238 52 E aolodd
as H> a m T own: . 888 <2 ”mo 88 8:85.85 3 own: mm: 8 8828

82838883 :2 Q C

.m: HoEto no 3582: :< l Oman . bEzEEoU 38:8 :onoa uB—on
.m: a tan oozon Q _ BZoBoa .1 OnO . ”ccoEcoFEo 368 A: >2
3 .wﬁ :2 tone» oozon .1 OaO .. :SO 7258:: soumﬁEcog 25> >62
8.8 £88... 88288 o<2 :2 a. 2 AS 88 m8 Gala
8:88:85 a. .2 828.8TEO . 88:8. 88 no; 8885 8:888 :3 80 M$5 gala
oomnm O2
o_£m:onoO 68952.5
8888 032 :2 a. : a
.ma 888 8288 T 80 . 888a <2 M8 .8885 8888 23 80 mm: 43%
8:65.: 3:08:80 mmuooun :wBoO 850m 2332 83.82 mcﬁom uoﬂn

: o e u n a
. :82 WE 8“:an— 882888 beam 85 28:0 8885 33%
.550 Enoom

 

883 ._ 28H

45

88:0 8303 “> $83300 um 40.87850: J: 69080: ”I 8870—003 ”Am .bﬂwbﬁ “m a
3:8 3:8 ”mm 8.8888 ”ADE 88780.8 ”0:2 8.67888 UU<E .3868880— ”04 880808-880 ”m0 .0
8.85m 88> 3:082 Hm>2 £38m 88280 885 ”mum .20
0880 808-80: ”22 .0880 868 a .0
8886 :88 08 28589 830088 ”Own—m .8886
30538 8.28088 ”GAE .8886 308 98 188.38 88890 ”ammo .8886 .8853 088830 ”n50 .3888 88888 ”<2 .2888 8888 ”< b
.m

835 "ED .8888 “mDm 488 ”MUM .88 338288 ”mm—d .83 0:88 “mDm .0828 032

88:0 08a 8886 88389 0?: 2: .20 8328 2: 8 80.“ 8:08:00 00: 83 8 8:83 8 8088 00: EU 8.5 Ammo: .E 8 82M

5 83.88 888808 88088_ 8 ﬂ 8: 08: cwsonzm 63898 Sm .8328 9808088800 2: 8 8:8: 8? 8m 88800880 >088 8:08: 2: 85
80: 8:8 828.2 .0880 was 8886 88380 3880822 88:08 98 88:0 80% 3:0 Ba 8508.“ 88082 on. 00.8 .233882 .8828 83 2:8
mm 8883 of :23 8008 8588 «89 88 88> 5:88 a E .252 is 8 8305 rmov .8808: 08>» 3008 2868882 2: 80¢ .880 880080 .3008
1868880— 98 380.508-8080 503 88: 3:8 .80 (z .8 35. 88395 0288.888 808 2: 80¢ 8:850 805 Ba 228 85 E 88088 8.868.“ 852

8.53 ._ 28...

46

Deﬁning and Measuring Disorder
Although Wilson and Kelling (1982) and Skogan (1990) eloquently explained the
crucial role of disorder giving rise to crime and facilitating neighborhood decline, they
never systematically deﬁned neighborhood disorder (Ross & Mirowsky, 1999). This lack
of deﬁnition and theoretical operationalization resulted in many variations of measures,
some of which did not appear to be appropriate. For example, J ang and Johnson (2001)
included burglaries and assaults to measure perceived disorder of residents, but they
seemed to be closer to crime than to disorder. Borooah and Carcach (1997) simply asked
residents whether they perceived fairly common signs of incivilities without specifying
theoretically relevant indicators.
Responding to this issue, Kelling and Coles (1996) provided their own deﬁnition:
In its broadest social sense, disorder is incivility, boorish and threatening
behavior that disturbs life, especially urban life . . . Most citizens have little
difﬁculty balancing civility, which implies self-imposed restraint and obligation,
with freedom. Yet, a few are either unable or unwilling to accept any limitations
upon their own behavior. At the extreme are predatory criminals who murder,
assault, rape, rob, and steal . . . Less extreme is disorderly behavior that, while

not as serious as the crime noted above, nonetheless can threaten social order by
creating fear and criminogenic conditions.

(pp. 14-15)
They speciﬁcally indicated that disorder includes aggressive panhandling, street
prostitution, drunkenness and public drinking, menacing behavior, harassment,
obstruction of streets and public spaces, vandalism and grafﬁti, public urination and
defecation, unlicensed vending and peddling, unsolicited window washing of cars, and
other such acts. Most state laws and city ordinances classify these behaviors as petty
offenses or misdemeanors that are most often punishable only by ﬁnes or community

service.

47

Adding to Skogan’s (1990) research, Ross and Mirowsky (1999) deﬁned disorder
as “visible cues indicating a lack of order and social control in the community” (p. 413).
The visible cues included both social and physical signs. Social disorder was deﬁned as
“signs indicating a lack of social control that involve people,” which included ﬁghts and
trouble among neighbors and the presence of people hanging out on the streets, drinking,
taking drugs, panhandling, and creating a sense of danger (p. 413). Physical disorder was
referred to as “overall physical appearance of a neighborhood,” which contained noisy,
dirty, and rundown places, unrepaired or abandoned buildings, vandalism, grafﬁti, and
litter, indicating that social control has broken down (p. 413).

Sampson and Raudenbush (1999) argued that disorder and crime have the same
origins (exogenous factors) and, thus, are not separate concepts at least in terms of their
roots. They insisted that disorder and crime are different only in their seriousness, saying
“Although ordinance violations like drinking in public and many ‘soft crimes’ like
grafﬁti may not be judged as particularly serious, this is an evaluation or classiﬁcation
issue and not a statement on etiology” (p. 608). Although Sampson and Raudenbush’s
argument makes sense to the degree that disorder and crime have the same roots, they
seem to overlook the point that we should distinguish offensive behaviors or conditions
on the basis of their seriousness and prevalence. Further, considering the differential
level of tolerance toward different types of offenses across neighborhoods, disorder and
crime do not necessarily appear to spring from the same source. Thus, combining the
ﬁrst two deﬁnitions by Kelling and Coles (1996) and Ross and Mirowsky (1999),
neighborhood disorder may well be deﬁned as “visible social or physical cues that disturb

life and threaten infomial social control and that are classiﬁed as petty offenses or

48

misdemeanors punishable only by ﬁnes or community service.” This effort to clarify the
meaning of disorder is expected to reduce measurement error committed from
inappropriate conceptualization of disorder.

Presence or Perception ?: A sensitive issue with regard to measuring disorder is
whether we have to observe the actual presence of disorder or if residential perception is
a better measure. Taking the ﬁrst stance, Sampson and Raudenbush (1999) proposed
Systematic Social Observation ($80) as a key measurement strategy for natural social
phenomena.18 Brown and her colleagues (2004) further argued that the objective
presence of disorderly conditions should be observed instead of using perceived disorder
from residential surveys, which is consistent with the theoretical assumption that
troublesome disorder starts with its actual presence in the neighborhood environment.
However, some researchers (e. g., Bratton & Kelling, 2006) raised a question about the
validity of observed disorder, particularly of social disorder, arguing that ﬁeld
observations tend to miss real world disorderly activities in the late evening hours from
dusk to dawn. It appears that the debate will continue for a while (J ang & Johnson, 2001).

An additional issue is whether observed disorder and perceived disorder measure
the same underlying concept. Several studies examined the empirical association
between them, and the results appeared to be non-supportive. For example, Perkins et al.
(1992) found that observed physical incivilities explained only 28% of the variance in
perceived disorder and the explained variance decreased further to 16% after accounting
for four control variables: racial composition, education, home ownership, and block size.

Based on the results, they argued that perceived disorder is determined by other factors

 

'8 The necessity of systematic social observation was originated with Reiss (1971), and Sampson
and Raudenbush supported for his idea.

49

than objective disorder. Sampson and Raudenbush (2004) discovered an analogous
ﬁnding. Although actual disorder signiﬁcantly increased the perception, a stronger
inﬂuence was found for social structure, or racial and economic composition.

In sum, the convergent validity of perceived and observed disorder remains
unveriﬁed due to the weak empirical association between them, which implies that the
two measures of disorder might not assess the same concept (Taylor, 1999). Therefore,
study ﬁndings may be different depending on how disorder is measured, although it is
premature to conclude which measure is better (J ang & Johnson, 2001). Leaving the
issue to future research, I suggest an altemative approach: using ofﬁcial data (e. g., 911
calls), especially for social disorder, could be another reliable and valid approach because
it signiﬁes both presence and perception.

Deﬁning and Measuring Crime

In the broken windows literature, crime must be understood as a separate concept
from disorder. Otherwise, we cannot expect much theoretical and practical utility
because the factors for cause and effect could be redundant. Judging from the above
discussion on disorder, therefore, it is obvious that crime as the endogenous variable in
the disorder related literature must consist of serious offenses that “society almost
uniformly condemns” and which are punishable by incarceration ordered by criminal law
(Kelling & Coles, 1996, p. 15; Worrall, 2002). According to Durkheim (1933), the
criminal law is a particular expression of the collective conscience to protect the
commonly shared values of people. Thus, criminal offenders are punished by repressive

sanctions, which go beyond restoring the status quo and seek to impose pain, because

50

what is being violated by the offenders is so sacred and large, as opposed to the interest
of each individual.

Measuring crime as it is conceptualized by theory is as important as measuring
disorder, since they should be separate concepts. However, some studies have not paid
much attention to the crucial dimension of crime measurement, with the results that there
are serious methodological ﬂaws and results are questionable. For example, J ang and
Johnson (2001) regressed juvenile drug abuse on perceived disorder that includes
vandalism, burglary, assault, and the like. Although their main concern is the
constraining mechanism of religion, the dependent variable, or drug abuse, clearly does
not match with the broken windows notion. Rather, it is closer to disorder, which makes
their model tautological. Some measures in other studies (Borooah & Carcach, 1997;
Kurtz et al., 1998; Perkins et al., 1992; Perkins et al., 1993) also appeared to be
inappropriate, for example, disturbance at house, drug dealing, and car tampering.
Following a conventional approach, I suggest that serious crime be operationalized to
include the FBI index crimes, such as murder, assault, rape, robbery, burglary, larceny,
motor vehicle theft, and arson.

Theory-driven measures of disorder and crime are very important in hypothesis
testing. One further step the author proposes is for researchers to check statistically
whether they truly indicate separate concepts. No matter how well-measured they are, it
would cause a serious problem if they measure the same concepts. Taylor (2002) argued
that establishing discriminant validity between two concepts (e. g., social integration and
collective efﬁcacy) is the ﬁrst step in statistical modeling. An exploratory factor analysis

technique (EF A) would be a nice tool to explicitly check the issue, which is useful in

51

identifying latent variables out of measured indicators. Further, it can successfully solve
the multicollinearity problem prevalent in sociological research that has multiple
indicators for a certain concept, although the factors are still allowed to covary unlike the
principal component approach (Tabachnick & F idell, 2006).“)

Data Sources for Indicators of Disorder and Crime

Another controversial issue in this area is that the disorder-crime link could be
spurious if perceived crime or victimization experience (or fear) is measured in the same
survey used to assess perceived disorder (J ang & Johnson, 2001; Sampson &
Raudenbush, 1999). If the two constructs are indeed conceptually inseparable as
discussed in the above section, the observed relationship between disorder and crime
might be tautological (Harcourt, 1998). Another reason, more technically, may stem
from the same data source problem (Perkins et al., 1992; Perkins et al., 1993). In other
words, self-reported perception of disorder is very likely to reﬂect the level of fear of
crime itself if they are included in the same survey (Jang & Johnson, 2001; Sampson &
Raudenbush, 1999; Taylor, 1999). Therefore, researchers need to adopt different data

sources for disorder and crime measure.

 

'9 Some social scientists seem to have confusion over the differential usage of principal
component analysis (PCA) and exploratory factor analysis (EF A) mainly because most of the
current software packages use the same name of factor analysis for both techniques. Although
they are both frequently used for the purpose of data reduction, they are distinct from each other
in several key aspects. For example, while PCA accounts for the variance of observed variables,
EFA deals with the interrelationships (correlations or covariances) of them. Thus, the main
purpose of PCA is to reduce the multiplicity of observed variables to a smaller number of
components, which are not related to each other. Whereas, EFA attempts to identify latent factors
that explain the interrelationships between observed measures, which could be assumed to be
either correlated or independent. It is notable that researchers can meaningfully interpret factors,
but not components (Fabrigar et al., 1999; Tabachnick & F idell, 2006).

52

Defining and Measuring Neighborhood

A very important but largely ignored issue is how to deﬁne and measure
neighborhood as conceptualized in theory. It has become increasingly important as the role
of neighborhood processes, such as collective efﬁcacy and social capital, gains more
attention. That is, if community is more than just a simple aggregation of individual
characteristics, it is problematic to use artiﬁcially deﬁned boundaries, such as cities, census
tracts, and block groups, as neighborhood units. In this line of thought, Sampson et al.
(2002) argued that community researchers must appreciate the street-level dynamics of
neighborhood interaction that would have the strongest inﬂuence on residents’ lives and
behaviors. Similarly, Brown and her colleagues (2004) insisted that street blocks best
capture the social, environmental, and psychological neighborhood processes relevant to
crime prevention and community development. This new approach can be seen in several
contemporary studies (e. g., Brown et al., 2004; Grannis, 1998; Miethe & McDowall, 1993;
Rice & Smith, 2002; Rountree et al., 1994; Sampson et al., 1999).

A contradictory argument appears to be gaining support recently, however. For
instance, Sampson (2002) argued that the romantic deﬁnition of neighborhood as a primary
group based on intimate relations and social bonds is of little relevance, particularly in
modern urban contexts. This conceptualization is expected to provide a more ﬂexible way
of measuring neighborhood.

Control of Neighborhood Process

As the theoretical model suggests, collective neighborhood processes are expected
to mediate the effect of disorder on crime. Accordingly, empirical research needs to

specify them as intervening factors, or at least control for them as confounding variables.

53

One cannot overemphasize the importance of model speciﬁcation based on theory and
careful review of prior research, because otherwise the parameter estimates are more than
likely to be biased and unreliable.

Following the above discussion on neighborhood process, the author would like to
propose that researchers use collective efﬁcacy as the measure for informal control,
particularly when they use artiﬁcially deﬁned areas as their neighborhood measure, for two
reasons. First, collective efﬁcacy goes well with the concept of formal social control (e.g.,
broken windows or community policing) because it does not require social ties or kinship.
Formal control may have indirect effects on crime by inﬂuencing or interacting with
residents’ informal control practices. It seems to be obvious that proactive police control
strengthens the willingness and capacity of residents to ﬁght crime. Broken windows or
community policing is based on such a theoretical premise, namely, that informal norms and
rules set by residents can be effectively realized and maintained by the cooperative efforts
between residents and police. Although there is little empirical evidence of the relationship
between formal and informal social control, the policing literature illustrates a possible
linkage (Greene, 1999; Skogan, 1990; Skogan & Hartnett, 1997). In contrast, inadequate
policing in highly disadvantaged neighborhoods would undermine the informal control
capacity of residents and could further allow offenders to set their own norms and rules in
the communities. Kubrin and Weitzer’s (2003a) study on retaliatory homicide demonstrates
that residents in poor neighborhoods have little expectation of proactive police intervention,
which fosters retributive justice by some residents.

Second, collective efﬁcacy seems to be able to provide far more ﬂexibility in

measuring neighborhood than social ties or social capital do. In other words, research with

54

social ties or social capital emphasizing the sense of community are likely to have a serious
limitation in measuring neighborhood because they appear to ﬁt only into small-scale areas
such as street blocks. However, researchers measuring collective efﬁcacy would be able to
try a variety of units for their neighborhood measures such as block groups, census tracts,
and cities. Thus, although Sampson himself supports using street blocks as the best unit (see
Sampson et al., 2002), his construct of collective efﬁcacy seems to be able to allow him and
other scholars to expand the unit of analysis to larger areas such as census tracts and cities.
However, such a strategy would not make sense to other researchers who follow the
traditional deﬁnition of neighborhood, which emphasizes intimate social bonds and a shared
sense of community.

Control of Neighborhood Structure (as Confounding Factors)

The contemporary model of broken windows theory (e.g., Skogan, 1990) put an
emphasis on the need for control of community characteristics to identify the unique or
mediating effect of disorder on serious crime. Especially, studies with non-experimental
design must control for the confounding factors. Major factors that are widely accepted
in recent neighborhood studies include concentrated disadvantage, residential instability,
and immigrant concentration, as proposed by Sampson and his colleagues in 1997
(Wilcox et al., 2004). Noticing that they represent only social dimensions of
neighborhood structure, some scholars (e. g., Sampson & Raudenbush, 1999; Wilcox et
al., 2004) have proposed including additional measures for physical structure, speciﬁcally

the percentage of commercial buildings out of total properties.20 It is suggested that

 

20 In fact, as distinct from the routine activity propositions, the interests in the role of land use as a
leading factor in crime date back to the Chicago sociologists (e.g., Burgess, 1925; Shaw &
McKay, 1942).

55

measuring both social and physical structure would better reﬂect the theoretical concept
of community characteristics than measuring only social structure.

Longitudinal Design

The contemporary model of the broken windows theory indicates that a
longitudinal design is desirable in looking at the developmental process relating to
disorder and crime. Without clarifying time order, it is hard to argue that disorder causes
crime. Deﬁning causal order is particularly crucial, since crime is also assumed to have a
signiﬁcant effect on disorder in a reciprocal way. However, only a few studies (e. g.,
Brown et al., 2004; Jang & Johnson, 2001; Markowitz et al., 2001; Perkins et al., 1993)
used longitudinal analysis. The most common reason is the unavailability of community-
level panel data (Markowitz et al., 2001).

Another plausible reason for the lack of longitudinal analysis is a lack of
appropriate analysis technique to explicitly model neighborhood change including
disorder and crime (Kubrin & Weitzer, 2003b). The common techniques used in
longitudinal studies until recently, such as residual change scores and cross-lagged
correlation approach, have certain limitations. First of all, they provide information mainly
on the between-variable relations rather than on neighborhood change over time (Bursik &
Grasmick, 1992; Kubrin & Weitzer, 2003b; Rogosa, 1995). Second, they are too
cumbersome to be used to analyze multiple waves of data, and thus their usefulness is
limited to a description of short-term trends. As an attractive alternative, Kubrin and
Weitzer (2003b) suggested a (hierarchical) growth-curve model that has often been
employed in the life-course literature. The model has some advantages. First, it models ﬁrst

level regression coefﬁcients as random variables at the second level so that it can exactly

56

sort out the random effect and contextual effect of any level of exogenous factors. Second,
it can specify the nonlinear trends, which are often the case in neighborhood change. And
third, full information for a certain period of time rather than truncated information for the
two time points is utilized in the model. Therefore, it is suggested that ﬁrture research
collect longitudinal panel data and utilize the growth-curve modeling technique so that the
theoretical model can be fully tested.

Spatial Dynamics

Macro-level criminological inquiries need to consider the spatial dynamics of
crime and its structural covariates or else analyses will not meet critical assumptions of
traditional, econometric regression models. In particular, the assumptions regarding the
independence of cases and the homoscedasticity of errors are questioned because, in
practice, crime is often non-randomly distributed across areas, and because causal
processes work differently within these areas. Violation of such assumptions results in
larger and biased standard error of estimates and the property of BLUE (Best Linear
Unbiased Estimate) of the OLS estimates does not hold any more (Bailey and Gatrell,
1995; Baller et al., 2001; Chainey and Ratcliffe, 2005). Notwithstanding the importance
of spatial analysis in the ﬁeld, however, no reviewed study carried it out. Hence, it is
suggested that future research in this review area make a serious attempt to examine

possible spatial processes.

Neighborhood Process and Crime: Social Disorganization, Systemic Control,
Social Capital, and Collective Efﬁcacy
Neighborhood process has been a central component in community research since

the advent of the social disorganization theory (i.e., Shaw & McKay, 1942). As

57

articulated in chapter II, the broken windows theory appreciates the role of collective
process as a mediator of the relationship between disorder and crime. Since there have
been many variations of the concept with no clear understanding of their deﬁnitions and
interrelationships (i.e., similarity and difference), this dissertation allocates some space
for a discussion of it, particularly focusing on collective efﬁcacy.”

Social Disorganization

Shaw and McKay (1942) named “the weakened condition of primary social
relationships and its resulting lack of informal social control” as social disorganization.
However, the authors simply assumed that the breakdown of conventional institutions
(i.e., family, school, church, and voluntary community organizations) and rules
automatically cause the lack of informal social control, which in turn increases crime
rates (Kubrin & Weitzer, 2003b; Williams & McShane, 2003). As such, they did not pay
much attention to the mechanism through which social networks and rules are transmitted
to social control in neighborhoods (Bursik, 1999). Further, they did not even attempt to
test their theoretical model in a quantitative manner.

In criminology, the ﬁrst rigorous empirical examination of their assumption was
performed after several decades by Sampson and Groves in 1989. Their ﬁndings were
largely supportive of the mediating role of social disorganization in the link between
structural characteristics (residential mobility, ethnic heterogeneity, low SES, family
disruption) and crime rates. Many subsequent works (e. g., Bellair, 1997, 2000; Elliott et

al., 1996; Markowitz, Bellair, Liska, & Liu, 2001; Morenoff et al., 2001; Sampson et al.,

 

2' Taylor (2002) argues that many of the various concepts for neighborhood process are
overlapped in terms of their measuring practices. For example, social integration and social ties
(networks) and territorial functioning and collective efﬁcacy actually represent the same
dynamics. Thus, this dissertation focuses on identifying differences between the four concepts —
social disorganization, systemic control, social capital, and collective efﬁcacy.

58

1997; Sampson & Raudenbush, 1999) provided further support for the social
disorganization hypotheses (Kubrin & Weitzer, 2003b). Unfortunately, however, these
studies failed to pay attention to the dynamics of social control that are exercised through
neighborhood network structures.22
Systemic Control
Contemporary systemic theorists redeﬁned the deﬁnition of social disorganization
to be “the regulatory capacity of a neighborhood that is imbedded in the structure of that
community’s afﬁliational, interactional, and communication ties among the residents”
(Bursik, 1999, p. 86). In other words, they succeeded in separating informal social
control capacity from neighborhood networks (or ties). Further, they identiﬁed three
types of networks and examined how the effect of structural conditions on informal
control varies across different types of networks: (1) private networks refer to intimate
friendship and kinship relationships; (2) parochial networks include less intimate and
secondary group relationships; (3) public networks are linkages to groups and institutions
located outside of the neighborhood (Bursik & Grasmick, 1993; Bursik & Grasmick,
1995; Hunter, 1985 in Bursik, 1999, p. 86).23 Their main proposition was that “rapid
residential turnover and population heterogeneity make it difﬁcult to establish relational

network structures that can serve as the sources of effective social control” (Brusik, 1999,

p. 87).

 

22 Bursik and Grasmick (1993) noted that there have been many conceptual variations of social
disorganization and integration. A few examples include community integration, responsiveness,
and social networks (McGarrell, Giacomazzi, & Thurman, 1997). In addition, Taylor (2002)
argued that there have been similar conceptualizations and research in other disciplines such as
community psychology, environmental psychology, and sociology.

23 Since the general systemic model of urban structure has had a great inﬂuence on the
development of this new orientation (see Berry & Kasarda, 1977), the reformulated version of
social disorganization often is referred to as the systemic theory of neighborhoods and crime
(Bursik, 1999, p. 86).

59

Like Shaw and McKay, however, the systemic theorists have paid little attention
to the dynamics of social control that are exercised through network structures. This
issue has gradually gained attention because of several studies that have found that some
forms of social networks (ties) do not serve as informal controllers of crime. For
example, Pattillo-McCoy (1998) discovered that, in a black neighborhood in Chicago,
some networks undermine residential efforts to ﬁght crime because they include both
law-abiding residents and active criminals such as drug dealers and gang members.
Bellair (1997) found that frequent interaction is not related to the capacity of informal
control. Wilson (1996) also noted that many poor urban neighborhoods do not show
strong informal control even though they are tightly interconnected. These studies raised
questions about the assumption of social disorganization and systemic theory that
neighborhood networks and interaction automatically foster informal control among
residents.

Social Capital

Responding to this issue, Bursik (1999) attempted to explain how neighborhood
networks serve as informal regulators against non-conforming behaviors. To this end, he
adopted Coleman’s (1988, 1994) social capital theory. Coleman (1988: 8100) deﬁned
social capital as intangible resources produced in ‘relations among persons that facilitate
action’ for mutual beneﬁt (Kubrin & Weitzer, 2003b). Bursik argued that the social
capital theory is useful in explaining the macro (networks) to micro (self-control)
transition dynamics because of its “dual focus on the overall structure of relationships
among actors and the nature of the exchanges that occur with that structure” (p. 87). In

other words, the social capital obtained through the various relationships can either be

60

related to social control or not, depending on the nature of networks and interactions. For
example, “J anowitz (1976: 9-10) noted that some forms of nonconforming behavior may
be tolerated by the members of a group as long as they do not interfere with the
attainment of some other common goal” (p. 87).
Drawing on a simple random survey of 386 adult residents in Oklahoma City,

Bursik tested a model in which the ecological positions of an individual inﬂuence the
systemic variables (i.e., various forms of networks), which in turn affect the social
control measured by loss of respect. He found that neighborhood-based private and
parochial networks increased informal social control. However, the friendship/family
networks outside the neighborhood were not associated with social control.

Collective Efficacy

A criticism of the social capital explanation is that, although important, the
resources obtained through social interactions are not sufﬁcient conditions for social
control to ﬁght neighborhood problems including crime (Kubrin & Weitzer, 2003b).
Although Bursik (1999) articulated that some forms of social capital are effective and
others are not, his study appears to focus on individuals’ self-regulation mechanism to the
exclusion of collective intervention dynamics. In other words, the key implication of his
idea and research is that people would not commit crime because they are afraid of losing
valuable resources.

The other side of informal social control is the willingness of residents to intervene
with nonconforming conditions and behaviors. The author would like to argue that while
social capital could be a nice self-controller, it has only a limited utility in explaining why

people do not tolerate others’ anti-social behaviors. That is, proactive intervention efforts

61

of individuals might entail the risk of losing certain types of social capital. A concept
highlighting the dimension of collective intervention, called “collective efﬁcacy,” was
developed by Sampson and his colleagues (1997, 1999). Collective efﬁcacy is a
combination of informal social control and social cohesion. They combined the measures
of informal social control and social cohesion and trust because “willingness and
intentions to intervene on behalf of the neighborhood would be enhanced under
conditions of mutual trust and cohesion” (p. 921 in Gibson, Zhao, Lovrich, & Gaffney,
2002, p. 544). It was proposed that the differential level of collective efﬁcacy is the
determining factor of neighborhood variation in interpersonal violence, without regard to
neighborhood— and individual-level characteristics. They found that three neighborhood
characteristics — concentrated disadvantage, residential stability, immigrant concentration —
explained 70% of the neighborhood variation in collective efﬁcacy, which in turn mediated
the effects of residential stability and concentrated disadvantage on violence rates.
Collective efﬁcacy remained the most robust predictor of lower rates of crime, even after
controlling for the individual characteristics and prior violence, which was consistent with
the causal logic of social disorganization theory.

The constraining effects of collective efﬁcacy on fear of crime and crime have
generally enjoyed strong supports. For example, using survey data from three medium-
sized cities, Gibson et al. (2002) found that collective efﬁcacy signiﬁcantly reduced fear of
crime. The results were consistent across the cities. Morenoff et al. (2001) also discovered
positive results. The measure of collective efﬁcacy was related to signiﬁcantly reduced
homicide rates in Chicago neighborhoods. Interestingly, the causal mechanism was similar

across racial groups.

62

Contextual Effects

As mentioned in the introduction chapter, examining contextual effects is important
because neighborhood is more than the sum of individuals. Besides individuals, collective
process, physical structure, and some elements of the social structure make up the entity of
neighborhood, which is why neighborhood dynamics are so diverse and complex. The
contextual model has a further advantage for theory elaboration and policy implication.
Meanwhile, the contextual effect can be tested via two approaches. One is to control for
individual characteristics as confounding factors, which allows researchers to sort out the
direct and unique effect of neighborhoods. The other is to test an interaction model of
individual and neighborhood factors, which shows how the relationship between individual
variables and outcome measures (e. g., crime) differs depending on neighborhood
characteristics (Rountree et al., 1994). Since this dissertation attempts to test a cross-level
interaction model of disorder and structure on crime, several examples on interaction effects
are reviewed.

In a grth curve model involving perceived disorder, juvenile religiosity, and illicit
drug use, J ang and Johnson (2001) found a signiﬁcant interaction effect of disorder and
religiosity both on marijuana and hard drug use. This result implied that adolescents’
religiosity weakens the effect of disorder on illicit drug use, or alternatively, religiosity
has a stronger constraining effect on drug use for adolescents living in a disorderly
neighborhood than for their counterparts living in an ordered neighborhood.

On the other hand, Miethe and McDowall (1993) used telephone survey data
(1989, part of a larger project on changes in crime over the last three decades in Seattle,

WA) to examine whether macro-level variables (e. g., population density, ethnic

63

heterogeneity, disorder) interact with the individual-level risk factors (e.g., home
unoccupied, family income and expensive goods, safety precautions and living alone) to
impact individuals’ risks of crime victimization (burglary & violent crime). Whereas
there was no such interaction effect on victimization of violent crime, perceived disorder
interacted with safety precautions to affect victimization of burglary. This interaction
effect indicated that while the safety precautions appear to have little effect in reducing
the victimization risks of burglary for residents in disordered neighbors, the protective
actions may be substantially effective for their counterparts living in ordered
neighborhoods. Put another way, safety precautions weaken the effect of disorderly
conditions on burglary victimization.

Using the same data and measure as Miethe and McDowall (1993), Rountree et al.
(1994) ﬁrrther examined whether disorder interacts with ethnic heterogeneity and safety
precautions to affect victimization risks. One distinct feature of their study, different
from Miethe and McDowall, was that they adopted a hierarchical logistic linear model to
capture the multilevel characteristics of the study variables. Similar to Miethe and
McDowall, they found that perceived disorder interacts with safety precautions to have a
signiﬁcant effect on victimization of burglary. Further, they discovered a negative
interaction effect of disorder and ethnic heterogeneity on burglary victimization, which
suggested that the effect of heterogeneity on burglary decreases in more disordered
neighborhoods, or alternatively, the positive effect of disorder on burglary victimization

is signiﬁcantly tempered in heterogeneous neighborhoods.

64

Implications

Based on the above results, it seemed to be valid to argue that individual
characteristics such as religious belief and safety precautions can effectively buffer the
harmful effect of disorder on crime. This result underscored the crucial role of human
agency in overcoming the harmful effect of dilapidated neighborhood conditions. It also
suggested that adolescents can be protected by local institutions such as churches from
the effect of disorder (J ang & Johnson, 2001). Meanwhile, individual lifestyle or
preventive strategy (i.e., safety precautions) appeared to be effective only in an ordered
neighborhood. That is to say, individual efforts to reduce victimization risk of burglary
by increasing safety precautions (and other forms of guardianship) seemed to be more
useful in ordered neighborhoods than in already disordered neighborhoods (Miethe &
McDowall, 1993). However, this did not necessarily mean that the individual preventive
strategy is completely ineffective in disordered neighborhoods. Rather, residents in more
disordered neighborhoods must try harder to avoid victimization than their counterparts
do in less disordered areas (Rountree et al., 1994). Lastly, the negative interaction effect
of ethnic heterogeneity and disorder on burglary risk suggested that in more
heterogeneous neighborhoods, disorder does not necessarily appear to contribute to more
serious crime. Thus, a control-oriented policing strategy in those areas would not be as
effective as was suggested by the theory. Because the root cause of serious crime in
disordered neighborhoods appears to be in their disorganized characteristics themselves,
policies would be better-oriented toward improving overall neighborhood conditions.

As implied by these interaction mechanisms, the differential effect of disorder on

crime depending on different contexts is of particular importance, because it suggests that

65

a context-speciﬁc approach is likely to better reﬂect the reality. As Miethe and
McDowall argued, failing to consider such contextual effects would result in distorted
conclusions. Further, these context-speciﬁc effects are important because they illustrate
the necessity of bridging the macro-micro gap in developing integrative theories of
criminal victimization (Miethe & McDowall, 1993). This line of thought is crucial in that
it stimulates theoretical elaboration and enables practitioners to take a more realistic,

sophisticated approach.

Summary

As proposed by the ecological perspective, particularly by the broken windows
theory, much of the literature shows that structure, disorder, and process play important
roles in determining the uneven spatial distribution of crime rates. A closer look at the
empirical evidence, however, indicates that the contemporary broken windows model has
received only moderate, though not i gnorab‘le, supports. But, the author suggests that it
does not necessarily cast doubts on the veracity of the theory itself. Rather, the
systematic review of theory and research implied that the mixed and inconsistent ﬁndings
appeared to be related to several, mostly methodological, issues such as measurement,
data source, research design, model speciﬁcation, and so on. Identifying those issues was
of great value in that they could guide more theory-driven and rigorous approaches.

Meanwhile, the review of research on contextual effects (or interaction
mechanisms) suggested that the link between disorder and crime varies across diverse
neighborhood contexts. For instance, the effect of disorder on burglary appeared stronger
in homogeneous neighborhoods than in heterogeneous ones (Rountree et al., 1994).

Despite such an important observation, actual testing for a cross-level interaction effect

66

has seldom been performed. Thus, the attempt of this dissertation to examine cross-level
interaction effects in Lansing neighborhoods deserves a lot of credit. If the hypotheses
for contextual effects prove to be true, policy makers will have to pay more attention to
the differential dynamics in order to distribute limited resources in a more efﬁcient and

effective manner.

67

CHAPTER IV: THE CURRENT STUDY

As mentioned in the ﬁrst chapter, this dissertation has two main purposes. First, it
attempts to test the broken windows theory within the Lansing, Michigan, context, using
the two waves of ﬁeld observation data and other various ofﬁcial datasets such as CAD,
US Census reports, and Select Phone USA records. Second, cross-level interaction
dynamics are examined via a hierarchical model. In order to make the multivariate
contextual analyses more credible, this study further attempted to clarify sensitive
methodological issues that are likely to inﬂuence study outcomes. The previous chapter
identiﬁed several critical issues regarding measurement, data sources, model
speciﬁcation, and spatial and temporal analysis, which serve as the methodological
guideline for this research.

Prior to a detailed description of the methodology, this chapter provides a general
outline of this dissertation, in order to make a smooth transition from the review of theory
and literature to methods and analyses. First, the history and characteristics of the study
place, Lansing, are described. The unique value of studying a mid-sized city is further
discussed. Next, the Lansing Dilapidated Housing Project (LDHP) is revisited to explain
the study subjects (sample), unit of analysis, and research design of this dissertation.
Then, several limitations of the current research are discussed. Finally, a study model is

proposed and the main research hypotheses are identiﬁed.

Study Setting: Lansing
Lansing is the capital city of the State of Michigan. It was originally named after

the village of Lansing in New York State, from which the original settlers of Lansing

68

arrived in the early 18005. There is an interesting history behind its status as state capital.
The state capital was temporarily located in Detroit in the mid-18005. Afraid of foreign
invasion (e. g., the War in 1812), however, the Michigan legislators tried to ﬁnd another
place. James Seymour, a rich immigrant to Detroit from New York, suggested Lansing
as a candidate. Although ignored at ﬁrst, his idea ﬁnally prevailed, and Lansing became
the new hub of Michigan’s government in 1847 (City-Data.com, 2006).

Today, Lansing is a mid-sized city with a population of 118,379 as of 2003.
Although the population has decreased gradually in recent years, it is still a great
community where government, business, education, and culture ﬂourish. As an
illustration, the active business climate was recognized by Entrepreneur magazine in
2003 as number seven on its list of "Best Cities for Entrepreneurs: Top Midsize Cities in

the Midwest" (City-Data.com, 2006).

Table 2. Racial and Economic Conditions of Lansing, Michigan, and US.

 

Race (%) Income

 

White Black Hispanic Others Median ($) Poverty (%)

 

 

Lansing (2000) 65.3 21.9 10 2.8 34,833 16.9
Michigan (2000) 80.2 14.2 3.3 2.3 44,667 10.5
US. (2004) 67.4 12.8 14.1 5.7 43,318 12.5

 

Compared to other Michigan areas, however, Lansing has a large variation in
racial composition and income distribution. The percentage of persons living below the
poverty line is also much larger than the state average (US. Census, 2000). Along with
the economic downturn (mainly due to the staggering automobile industry), it seems to

threaten the quality-of-life in the Lansing community. At the same time, however, the

69

large variation of racial and economic conditions makes Lansing a suitable site for study
testing the ecological hypothesis. Table 2 describes the racial and economic conditions
of Lansing, along with those of Michigan and the US.

Meanwhile, it would be valuable to examine how different settings have different
dynamics in the formation of the relationship between disorder and crime, since the well—
known studies have focused only on metropolitan urban areas (e. g., Sampson &
Raudenbush, 1999; Skogan, 1990; Wilson & Kelling, 1982; see also Table 1). For
example, Brown et al. (2004) argued that disorder such as grafﬁti and vandalism may
exist both in urban areas and suburbs, but unkempt lawns and ill-maintained homes may
signify more salient disorderly conditions in the suburbs than in urban areas. In a similar
vein, Markowitz et al. (2001) highlighted the importance of cultural understanding,
insisting that neighborhoods have differential levels of tolerance toward various criminal
activities. Some activities are not considered to be all bad (Patillo-McCoy, 1998) and
some are simply regarded as adaptive behaviors to devastated socioeconomic conditions
(Anderson, 1999). Such interests in differential dynamics across various levels of
communities are closely related to the contextual understanding that is presented in the
previous chapters. As such, this study appreciates a strong caveat against simple

generalization.

Lansing Dilapidated Housing Project Revisited

As explained already, the initial plan of randomized experiment of the LDHP
prematurely ended after treating 50 tagged houses in 40 street segments. To make up for
that, the researchers decided to take a non—experimental longitudinal approach, where

they attempted to observe and model the natural change of disorder and crime across the

70

study segments. To this end, the author performed another ﬁeld survey in 2005 to
measure physical conditions of the street segments.

The purpose of this dissertation is to test a longitudinal multilevel model linking
disorder and crime, using the two waves of ﬁeld surveys and other datasets (e. g.,
computer aided dispatch, the US. Census). Although neither explicit nor direct, this
study is likely to provide a good sense of whether the cooperative efforts involving city
government, council, businesses, and residents have been effective in reducing physical
disorder as well as social disorder and crime. A detailed description of the analytic
technique is presented in the next chapter. This section focuses on describing the sample,
unit of analysis, and general characteristics (i.e., strengths and weaknesses) of the

research design of this study.

Study Subjects (Sample) and Unit of Analysis

The street surveys of physical conditions of the segments were performed twice,
in the summer of 2003 and in the fall of 2005. In 2003, there were 386 segments
observed and 378 segments were surveyed in 2005. A total of 377 segments were
commonly surveyed in both years, and the multilevel analysis is based on them. The
sample segments represent 6.48% of all street segments of the city. Figure 5 shows the

study segments.

71

 

 

rFigure 5. Study Segments (n=377) ]

 

 

 

 

 

 

Meanwhile, the city of Lansing consists of 109 block groups. But only 84 block
groups in 2003 and 83 block groups in 2005 contained at least one street segment. Thus,
83 block groups are used in this study for the hierarchical model. Figure 6 shows the

block groups of Mid-Michigan and Lansing for the 2000 Census.24

 

2" The block group 1 (ofﬁcial ID - 260650055019) does not exist in the Census, and each of the
four block groups - 78 (260650010002), 97 (260650034001), 101 (260650009001), 103
(260650031021) » is a part of a larger block group overlapped with another block group within
the boundary of Lansing. Thus, the exact total number is 109, although it seems there are 114
block groups. Meanwhile, the block groups marked with x (n=26) are those that do not contain

72

 

1
l x,

l \

[ Figure 6. Mid-Michigan & Lansing Block Groups I 1

 

’ l ‘ M1d»MIchIgan Blockgroup
Larrsrng Blockgroup(rr=‘114)

 

 

 

The study model adopts street segments as the level-1 unit of analysis, and block
groups serve as the level-2 unit. As explained in the previous section about

methodological issues, using street segments as the measure of neighborhood could be a

 

any study segment: 12 (260450214003), 82 (260650009002), 71 (260650014001), 64
(260650016001), 52 (260650017011), 48 (260650022001), 34 (260650023002), 45
(260650025003), 22 (260650027003), 16 (260650028002), 53 (260650029011), 113
(260650033011), 108 (260650033021), 106 (260650033022), 100 (260650034001), 88
(260650034002), 54 (260650035002), 19 (260650036012), 33 (260650037005), 84
(260650038012), 70 (260650040003), 36 (260650044041), 11 (260650051002), 9
(260650052011), 8 (260650053033), 4 (260650053042).

73

more valid approach. It would be especially the case in a mid-sized urban context such

as Lansing.

Non-Experimental Longitudinal Design

Menard (1991, p. 4) deﬁnes longitudinal research as “a research in which (a) data
are collected for each item or variable for two or more distinct time periods; (b) the
subjects or cases analyzed are the same or at least comparable from one period to the
next; and (c) the analysis involves some comparison of data between or among periods.”
He further argues that “at a bare minimum, any truly longitudinal design would permit
the measurement of differences or change in a variable from one period to another” (p.4).
There are three types of longitudinal studies —— trend, cohort, and panel studies. First,
trend studies examine a trend or pattern of an event(s) among general populations over a
certain period of time. Examining burglary rates over the last four decades in Lansing
would be an example. Meanwhile, cohort studies look at a speciﬁc group of population
(e.g., birth cohort) as they change in their certain characteristics or behaviors over time.
Although the study subjects at each time point are not the same, they are usually
comparable in many of the core study characteristics. Finally, the panel studies follow
the same subjects across time, and thus, they provide the most comprehensive picture of
temporal changes (Maxﬁeld & Babbie, 2001).

Compared to a cross-sectional design, a longitudinal study is superior in terms of
the internal validity because it clearly has the element of temporal order between
independent variable and dependent variable. However, the longitudinal design,

particularly the panel study, could suffer from participant attrition and other problems

74

such as maturation and instrumentation.25 Regarding the external validity, meanwhile,
longitudinal studies, particularly the panel ones, have more disadvantages than cross-
sectional studies in dealing with the threats to external validity because they have limited
population, in general, with which to make an inference. However, it is notable that a
random (probability) sampling could solve many of the threats to external validity just as
the random assignment does most of the threats to internal validity (Maxﬁeld & Babbie,
2001; Menard, 1991; Shadish et al., 2002).

In short, the design of this research ﬁts into the category of a panel study, which
has the greatest potential to verify the causal order among non-experimental studies. The
loss of nine segments, however, might raise a substantial concern. Also, it must be noted
that this study is not based on a random sample. Accordingly, the author does not
attempt to generalize the study results to other contexts. Such consideration is closely

related to appreciating the importance of context-speciﬁc understanding.

Limitations

As shown in the previous chapter, study results can be contaminated by
neglecting such-methodological issues as measurement, model speciﬁcation, and design.
Accordingly, this study tried not to be trapped in such controversies. Detailed research
processes are described in the next chapter. Unfortunately, however, this research project
could not be free from limitations. First, the neighborhood process could not be
controlled for because no residential survey was conducted due to budget constraints.

Accordingly, study ﬁndings must be interpreted with some caution. Another limitation

 

2’ Maturation: Developmental changes in the subjects (e.g., growing in a long-term study or
becoming bored and being tired in short-term research) could be confused with a treatment effect.

Instrumentation: Change of measurement process due to, for example, different questionnaires
or changed standards of experimenters could be confused with a treatment effect.

75

was that this study could not perform a systematic longitudinal analysis such as the
growth curve modeling, because only two observations have been made for the same
subjects. To make up for this weakness, the author followed the approach by Markowitz
et al. (2001) and Sampson and Raudenbush (1999), where they controlled for prior levels
of crime. This strategy was expected to be able to control for the time order in the

hierarchical model.

A Study Model and Hypotheses

After reviewing the theory and previous research, several general hypotheses
were identiﬁed for cross-sectional and longitudinal models. Figure 7 shows the
conceptual hierarchical model to be tested in this dissertation. For the cross-sectional
model, the author expects:

H1: Disorder, physical and social, has a significant effect on serious crime at
the first level.

H2: Structural characteristics have signiﬁcant effects on block group means of
crime rates at the second level.

H3: The effect of disorder on serious crime remains significant after
controlling for neighborhood characteristics (multivariate significance).

H4: The relationship between disorder and crime at the first level is
significantly inﬂuenced by structural covariates at the second level (cross-
level interaction).

H5: The mean crime rate of a block group is further inﬂuenced by the mean
crime rates of neighboring block groups, after controlling for the
structural covariates (spatial lag).

On the other hand, for the longitudinal model, the author expects:

H1: Change of disorder, physical and social, has a significant effect on change
of serious crime at the first level.

76

H2:

H3:

H4:

H5."

Structural characteristics have significant effects on block group means of
crime rates change at the second level.

The effect of disorder change on serious crime change remains significant
after controlling for neighborhood characteristics (multivariate
significance).

The relationship between disorder change and crime change at the first

level is significantly inﬂuenced by structural covariates at the second level
(cross- level interaction).

The change of mean crime rate ofa block group is further inﬂuenced by
the change of mean crime rates of neighboring block groups, after
controlling for the structural covariates (spatial lag).

Figure 7. A Hierarchical Model Linking Disorder and Crime”

 

 

 

 

 

 

 

 

 

 

 

 

Concentrated Residential Immigrant Land Spatial
Disadvantage Instability Concentration Use Lag
V
Disorder, 4 Serious
Land Use 7 Crime
Summary

 

 

 

 

 

 

 

 

 

 

The city of Lansing appears to be a nice candidate for ecological research in that

there is a large variation in racial and economic composition. Further, in light of the fact

that most of the well-known studies have been carried out in the contexts of large

 

2° Spatial lag represents spatially lagged crime rates of each block group, which amounts to a
simple average of the crime rates for the neighboring block groups. In other words, the model
controls for the spatial autocorrelation process (Anselin, 2005). Analytic strategies in the next
chapter provide a more detailed discussion on the spatial model.

77

metropolitan urban areas, this study of the Lansing area provides a good opportunity to
examine how the ecological dynamics in a mid-sized Mid-Westem city are different from
those in large urban areas.

This dissertation constructs a hierarchical model for the non-experimental two—
waves of datasets. Main effects and cross-level interaction dynamics are tested via the
multilevel model. Although the panel design can successfully construct internal validity,
it has a limited capacity to verify external validity or generalization.

Meanwhile, two main limitations are likely to constrain the conclusion validity
and reliability of the study results. First, one of the most critical mediators, collective
efﬁcacy, is not controlled for in the non-experimental model. And, a more systematic
longitudinal analysis (e. g., growth curve model) is not performed due to the absence of
neighborhood panel data. Notwithstanding these limitations, however, strict applications
of the sensitive issues and cross-level interaction tests are expected to enrich the theory,

research, and policy implications.

78

CHAPTER V: METHODOLOGY

This study attempts to test the above hypotheses, using the two waves of street

surveys and various ofﬁcial datasets. This chapter describes methods for rigorous

hypotheses testing. The ﬁrst section explains the characteristics of the datasets. Then,

the author illustrates how the study variables were measured. Particular attention is paid

to examining whether the measures of social disorder and crime represent the same

underlying concept or not. Finally, the analytic strategies are presented, including a

detailed description of spatial analysis and a hierarchical mixed Poisson model.

Data

This research project utilized four types of datasets to measure the variables in the

study model. Table 3 provides a general description of the datasets. Each of them is

explained below in more detail.

Table 3. Description of Data

 

 

 

 

Type Source Year Basic Unit Measure
2003, Crime,
CAD (91 1) LPD 2005 Address Social Disorder
Christopher D. Maxwell, Ph. D., 2003, Address, . .
Street Survey MSU Land Policy Institute 2005 Segment Physrcal Disorder
Census US. Census Bureau 2000 Block Group Social Structure
Phone Select Phone CD-ROM 2000 Address Land Use
CAD (911)

Citywide computer-aided dispatch (CAD), or 911, data were collected for 2003

and 2005. There were 85,550 calls for service in 2003. The number increased slightly to

79

85,646 in 2005. Various types of crimes and social disorder were available to be utilized.
The LPD provided geocoded data with the last two digits of address deleted for privacy
purposes. The author planned to aggregate them into the street segment level using a GIS
program.27 For some unknown reasons, however, much of the data was contaminated,
and thus the researcher could not match some of the cases with corresponding segments.
Left with no choice, the author decided to do it manually using the SPSS program.

This study utilized the CAD data instead of other police records such as incident
report and arrest data for two reasons. First, since the basic level of analysis was street
segment, it was very likely that most of the segments did not have incident or arrest
records. In other words, the author attempted to increase the statistical validity by
expanding the variation of crime and social disorder, which was possible by taking
advantage of the CAD data (Kurtz et al., 1998). The other reason was that the CAD is
generally regarded as a more reliable measure of crime (Green, 1995; Sherman et al.,
1989; Sherman & Weisburd, 1995, in Katz et al., 2001). Incident or arrest records are a
product of police discretion, and thus highly vulnerable to distortions of reality. In
particular, an intervention effort (i.e., initiation of a new program or policy) would
aggravate the distortion by inﬂuencing the police discretion process.

However, several potential weaknesses of the CAD data must be noted, too. First,
the CAD data may not reﬂect the location of the criminal incident but the caller’s address.
This will raise a serious concern when one carries out research at the address level. But,
the problem will be less severe if the CAD data is aggregated into larger areas such as
block groups and census tracts (Warner & Pierce, 1993). Since the basic level of analysis

of this study involving crime and social disorder is street segments, the author would like

 

2’ ArcGIS 9.0 (a product of ESRI)

80

to say that the issue might not be that serious. Second, there could be multiple calls for
service recorded for a single incident. Lastly, the wrong type of crime may be recorded
because many citizens are not able to tell the exact nature of a crime. For example, a call
could be recorded as a robbery when it actually meets the burglary criteria (Hayslett-
McCall, 2002). Thus, researchers have to be cautious in using the CAD data, although it
is generally considered to be a better measure of crime than the incident or arrest record,
a product of police discretion. If possible, it would be desirable to assess the validity of
the CAD data by, for example, comparing them to incident or arrest data. Unfortunately,
this study could not cross-check the CAD data due to limited resources and data

unavailability.

Street Survey28

Researchers including the author conducted street surveys in 2003 and 2005 with
an aim to collect data on the physical disorder of tagged houses and street segments.29 In
particular, the survey in 2005 was made possible through a grant from the Land Policy
Program. The initial plan was to use the ﬁrst survey in 2001 as a proxy of pretest for the
randomized experiment. Despite the premature ending of the initial scheme, however,

the researchers decided to carry out the same survey again after two years in order to

 

28 As will be explained later, the number of houses recorded in the surveys (item 8) was used for
two purposes. First, the author combined it with the number of business properties from the
phone data to create the denominator (i.e., total number of buildings) for calculations of crime,
social disorder, and physical disorder rates. It was based on the fact that the street surveys
contained more exact information of the number of houses than the phone data due to the
underreporting problem. Second, land use at the segment level was measured by the percentage
of (crime attracting) business properties out of the sum of houses and (crime attracting)
businesses. In the summation, the number of houses from the street surveys was utilized. Refer
to the measurement section for more detailed discussion.

29 Besides physical disorder, the street surveys attempted to measure other dimensions of physical
structure, such as crime watch sign, street lights, and volume of motorized trafﬁc. However, they
were not used in this dissertation in light of the study purpose and model.

81

make comparisons of any changes in physical disorder and crime in a non-experimental
fashion. For visual evidence, they also took pictures of all houses.30

The researchers created two separate survey questionnaires (coding sheets) for
houses and segments, respectively. The questionnaires contained items measuring as
many dimensions of physical conditions as possible. For a complete list of the items,
refer to Appendix A. To increase the inter-rater reliability, a detailed guideline for
coding (or evaluation) was developed for the ﬁeld observers, who were then educated on
how to use it. The evaluation scheme is attached as Appendix B.3 ’

The ﬁeld survey was designed following the theoretical notion that troublesome
disorder begins with its actual presence in the neighborhood environment (Brown et al.,
2004). Consistent with this assumption and following the idea of Reiss (1971), Sampson
and Raudenbush (1999) supported systematic social observation (880) as a key measure
for natural social phenomena. Although it is premature to conclude which one is a better
measure between perceived and objective disorder, study results would be inﬂuenced by

the way of measuring disorder (J ang & Johnson, 2001).

Census

The US. Census 2000 was utilized to measure the social characteristics of
neighborhoods. The author downloaded the dataset from the ICPSR database.32 Since no
information was available on the level of street segment, the data were aggregated into
the block group level. Although the block group did not appear to be a good measure of

neighborhood as was discussed above, the author believed that such practice would not

 

30 Figure 1 shows an example of a renovated housing unit.

3 ' Since this study does not use the survey data of houses, the evaluation guideline for housing
survey is not included.

32 Inter-University Consortium for Political and Social Research. http://www.icpsr.umich.edu/

82

cause a serious problem because the block group represented a higher level of

neighborhood. Otherwise, it might well have raised a severe concern.

Phone Data

To measure land use, phone data from Select Phone CD-ROM 2000 were
collected. The data contained a variety of information on the listed phone number, such
as name (resident and business), address, types of business, and so on. In other words,
both residential and business information were available, which made the source very
useﬁil in measuring land use. A total of 34,980 households and 5,642 businesses were
listed as of 2000.

However, the dataset appeared to suffer from an underreporting problem. That is,
it was obvious that some of the houses and businesses did not register their phone
numbers}3 The author ﬁgured out this problem by comparing the dataset with the US.
Census. The Census indicated there were 49,505 households in 2000 and 8,047
businesses in 1997. Assuming the number of businesses remained the same, around 29%
of the households and 30% of the businesses seem to have been missed.

The phone data had aspects of both strength and weakness. Since the address was
geocoded, the researcher planned to aggregate houses and businesses into segments and
block groups using the ArcGIS program. As was the case with the CAD data, however,
some of the cases could not be matched with corresponding segments for unknown
reasons. Thus, the SPSS program was utilized to match each address with the study

segments manually. Unlike the CAD data, exact addresses were readily available.

 

3’ It was also possible that some of the poor households did not own a phone. A rather trendy
reason was that more and more people rely on mobile phones, disconnecting the traditional phone
service.

83

Considering the underreporting problem, however, it did not appear to be
appropriate to use only the phone data in order to measure land use at the segment level.34
Since the street surveys contained more exact information on the number of houses for
each study segment than the phone data, it seemed to be more valid to use the street
surveys for counting houses and the phone data for counting businesses. The sum of
houses and businesses obtained from the two sources was further utilized as a
denominator for calculations of crime, social disorder, and physical disorder rates.

In sum, the author took advantage of the ﬁeld surveys and phone data to measure
land use at the segment level. On the other hand, the ArcGIS was still useful for the
author to aggregate houses and businesses in the phone data into the block group level.

No problem of mismatching was detected. Also, the underreporting issue did not appear

to cause a serious threat in light of the large scale of measurement unit, block groups.

Measurement

It is well known to social scientists that measurement error is one of the most
salient and inevitable sources of the weakness or ﬂaw of social science research. For this
reason, it is recommended that researchers “describe the measurement process explicitly”
(Maxﬁeld & Babbie, 2001, p. 101). Basically, the author followed the above discussions
on methodological issues to measure each concept, with an aim to increase their

reliability and validity.

 

3" In fact, the phone data contained no information on house or business for twelve study
segments, which did not make sense at all because the street surveys indicated that all segments
had at least one building, house or business.

84

Dependent Variable: Serious Crime

As suggested above, the researcher decided to use FBI index crime as the measure
of serious crime, using the CAD data. Serious crime was measured in two ways — violent
and property. Violent crime was a sum of homicide, rape, robbery, and assault. Burglary,
larceny, vehicle theft, and arson were combined to measure property crime. Since the
information on population was not available at the segment level, the author used total
number of buildings in each segment as the denominator in calculating crime rates.35

In general, the dynamics involving violent and property crimes are regarded to be
different from each other, which might well be explained by the examples of mixed
ﬁndings depending on crime types. Accordingly, one related research interest was
whether the results are different depending on the outcome measure. In fact, noting the
unique trend and mechanism of each type of crime, Clarke (1997) suggested that
criminologists and practitioners need to focus on more narrowly deﬁned types of crime
(Maxﬁeld, 1999). Unfortunately, however, it did not appear to be appropriate to further
narrow down the crime measure in this study because too many segments did not have a
single record of crime. Thus, the author used the combined measure of violent and
property crime.

As mentioned before, the SPSS program was utilized to calculate the number of
crimes for each segment. The basic scheme was to compare the addresses of crimes to
the range of addresses of each segment. By assigning the unique number of a segment
(i.e., segment id) to the address of a crime incident, the researcher attempted to total the

number of crimes for each segment. Unfortunately, however, the strategy faced two

 

3’ As explained above, the number of houses was obtained from the street surveys and the number
of businesses was acquired from the phone data.

85

challenging issues. The ﬁrst issue was that, since the last two digits of each crime
address were deleted, the number of assigned crimes became somewhat inﬂated in those
segments that do not have address ranges from xx00 to xx99 (i.e., segments in which the
last two digits of the starting address are not 00 and/or the last two digits of the ending
address are not 99).” For those segments, the inﬂated sum had to be adjusted by a
reliable standard. The author decided to use “the percentage of buildings (sum of houses
and crime-attracting businesses”) within the exact segment ranges out of total buildings
(sum of houses and crime-attracting businesses) within the modiﬁed segment ranges with
003 at the starting point and 99s at the ending point” as the weighting variable. The
phone data was used to construct the weight because it contained information on the
exact address of each house and business for the study segments as well as the other
segments. It must be noted, however, that the weighting scheme might have been
incorrect, because the number of properties was somewhat underreported in the phone
data.

The other issue was that some of the incidents were recorded at intersections. As
Table 4 shows, 8.6% of violent crimes and 6.4% of property crimes were recorded at
intersections in 2003. In 2005, 10.8% of violent crime and 6.0% of property crime were

located at intersections.38 Instead of dropping those cases, the author decided to count

 

3" Sixty ﬁve segments out of 377 (17.24%) were those with inﬂated crime rates.

37 In this case, both the number of houses and crime attracting businesses were extracted from the
phone data, because the street survey data had housing information only for the 377 study
segments, which made it impossible to compare the number of buildings of a study segment to its
neighboring ones within the adjusted address ranges. Refer to the measure of land use for more
information on crime attracting businesses.

38 The percentage was calculated by dividing the crime number on intersections by the total crime
number on study segments and intersections (e.g., percentage of 2003 violent crime = 42/
(449+42) = .086).

86

one incident as half in light of two rationales. One reason was that, compared to the
incidents on segments, those at intersections took place relatively far away ﬁ'om the study
segments and were likely to have less impact on residents than their counterparts on
segments do. The other rationale was related to avoiding the possibility of double- or
triple-counting of the incidents that occurred at intersections with two or three study

39
segments.

Table 4. Percentage of Crimes and Social Disorders Recorded on Intersections

 

Study Segments : Intersections (percentage)

 

 

2003 2005
Violent Crime 449 : 42 (8.6%) 472 : 57 (10.8%)
Property Crime 960 : 66 (6.4%) 1015 :65 (6.0%)
Social Disorder 930 : 321 (25.7%) 829 : 329 (28.6%)

 

Primary Independent Variables of Interest

Physical Disorder

Through the street surveys, the researchers collected a lot of information on the
physical conditions of the street segments in 2003 and 2005. Surveyed items for physical
disorder measure included cigarettes (item 9), litter (item 10), garbage (item 1]), empty
bottles (item 12), abandoned or disabled cars (item 14 or 15), condoms (item 17), needles

and syringes (item 18), grafﬁti (items 19 and 20), and dilapidated houses (items 22, 23,

 

Appendix C contains a complete list of the number of crime and social disorder for the total
segments, 377 study segments, and intersections.
’9 One case of social disorder (noise complaint) in 2003 and two cases of social disorder (juvenile
complaint) in 2005 were recorded at intersections with four study segments.

87

and 24).40 Unfortunately but not unexpectedly, many of the indicators, such as empty
bottles, abandoned or disabled cars, condoms, needles and syringes, and grafﬁti, were not
found for most segments."l

One research interest was whether the multiple indicators represented the same
concept. To examine the issue, an exploratory factor analysis technique (EFA) was
attempted.42 Before that, the author took two steps to make the EF A valid. First, the net
counts of indicators were transformed into rates. For litter and garbage, the number of
houses was used as the denominator (i.e., number of houses with litter or garbage per 10

houses). The total number of houses and commercial buildings was used as the

 

’0 Cars parked on the street (item 13) and cars with parking violations (item 16) were excluded
from the analyses since they did not seem to be adequate measures of physical disorder. FYI,
item 16, cars with parking violations, was found only at one segment in 2003 and two segments
in 2005.

4' For example, empty bottles, the most prevalent item among those rarely found indicators, were
located only at 36 segments in 2003. Although it was found at 211 segments in 2005, the author
decided not to use the item as an independent variable for the EF A to make the independent
variables for the EF A consistent across the years.

42 The author presents several methodological points concerning EF A that were taken advantage
of in this study. (1) The standard method for factor extraction in EFA is maximum likelihood.
Compared to the principal component analysis (PCA), it has an additional advantage that a
goodness of ﬁt test is presented. In other words, one can decide how many factors he/she needs
to extract out of the observed variables. It saves researchers from arbitrarily setting the minimum
of the eigenvalue. It must be noted, however, that the decision on the adequate number of factors
to retain needs to be based on substantive issues (i.g., theory and prior research) as well as
statistical issues (Fabrigar, Wegener, MacCallum, and Strahan, 1999). (2) Assuming the
identiﬁed factors are correlated with each other, the author chooses oblique (i.e., promax) rotation.
In general, this approach is regarded to be better representative of reality in social science than
the orthogonal rotation technique with the assumption of independent factors. (3) Concerning the
minimum factor loading of an indicator for its inclusion in factor interpretation, the researcher
followed the rule suggested by Norman and Streiner (1994): Minimum factor loading when the
sample size is larger than 100 = 5.152 / sqrt(n-2). Applying this rule, in case of disorder for the
377 segments, an indicator has to have at least .266 factor loading to be considered as a pertinent
variable in factor interpretation: 5.152 / sqrt(375) = .266. For social structure of 83 block groups,
an indicator would need at least .572 factor loading, although the sample size is smaller than 100:
5.152 / sqrt(81) = .5 72. The bolded numbers in Table 5 and Table 6 indicate non-meaningful
factor loading in interpretation. FYI, the factor loadings in EF A represent how much each
indicator is correlated with the factor. (4) For factor score estimation, the author selected a
weighted least squares method (Bartlett scores) instead of regression method following the
general trend. However, it must be noted that none is uniformly better than the other.

88

denominator for the other indicators (i.e., number of the indicators per 10 buildings).43
Meanwhile, the score for cigarettes was standardized to z-scores so as to make the scores
comparable across the two collection years, which was necessary because the raters were
not the same people for each survey. Second, the rarely found indicators were combined
into one variable — “others.” In a similar vein, litter and garbage were aggregated into
one variable because they appeared to measure similar phenomenon. Model 1 in Table 5
shows the EF A results. It suggested that the various indicators represented one
underlying concept, namely, physical disorder. Created factor scores were further

utilized as the primary independent variable for the hierarchical model.

Table 5. Factor Loadings of Physical Disorder and Serious Crime

 

 

 

 

Model 1 Mode12
2003 2005 2003 2005
(1) (2) (1) (2) (1) (2) (1) (2)

Physical Disorder

Cigarettes .54 .50 .54 .59 .33 .54 .52 .69

Litter or Garbage .62 .84 .62 .58 .79 .80 .44

Dilapidated Houses .44 .48 .41 .32 .46 .49 .53

Others .12 .38 .29 .13 .40 .37 .37
Violent Crime .38 .18
Property Crime .78 .36

 

Another question to deal with, probably a more important issue than the ﬁrst one
in the broken windows ﬁeld, was to conﬁrm whether the measures of serious crime and
disorder (physical and social) represent distinctly separate concepts. Modeling with

different concepts was crucial because otherwise the conclusion validity of study

 

4’ The denominator used in this case was the same as that of crime and social disorder. Refer to
footnote 33.

89

outcomes was highly likely to be questioned. To analyze the issue, the EFA was
performed again. As Model 2 in Table 5 shows, both violent and property crimes did not
appear to represent the same underlying concept as the indicators of physical disorder do.

Social Disorder

The CAD data was used to measure social disorder. Following the above
discussion on measurement, six indicators were extracted — drunken subject, juvenile
complaint, noise complaint, panhandling, loud party, and prostitution — that appeared to
ﬁt well into the suggested deﬁnition: “visible social cues that disturb life and threaten
informal social control and that are classiﬁed as petty offenses or misdemeanors
punishable only by ﬁnes or community service.” Through the identical processes of
measuring crime (i.e., denominator for crime rates, weight for inﬂated crime rates, half-
counting for incidents at intersections), those multiple indicators were aggregated into
street segments. One notable issue was that social disorder seemed to have been recorded
at intersections much more frequently than serious crime (see Table 4), and dropping
those cases would have caused a serious threat to the validity of the measure.

As with physical disorder, two issues had to be addressed. One was if the
multiple indicators of social disorder represent one underlying concept. The other was to
examine if the indicators of social disorder and serious crimes represent an identical
concept. It was a particularly important question in this study because the same data
source, CAD, was used to measure both crime and social disorder.

To examine the issues, the EF A was performed again. Since drunk subjects,

panhandling, and prostitution were rarely found events, they were combined into a

90

variable — “others.”44 Model 1 and Model 2 in Table 5 describe the results for the two
questions, respectively. Model 1 conﬁrmed that the indicators of social disorder measure
one latent factor for both years. Model 2 appeared to prove that the measures of serious
crime and social disorder did not represent the same underlying concept. Thus, the factor

of social disorder created in Model 1. was further included in the study model.

Table 6. Factor Loadings of Social Disorder and Serious Crime

 

Model 1 Model 2

 

2003 2005 2003 2005

 

(1) (2) (1) (2) (1) (2) (1) (2)

 

Social Disorder

Juvenile .29 .62 .39 .29 .43 .34
Noise .99 .30 .72 .99 .31 .32
Loud Party .31 .28 .41 .32 .78 .75
Others .14 .37 .99 .53 .58 .62
Violent Crime .43 .44 .53
Property Crime .99 .88

 

Control Variables: Neighborhood Structure (Characteristics)

Social Structure

The author attempted to measure neighborhood structure with two dimensions —
social and physical. For social structure, this dissertation followed the approach of
Sampson and his colleagues (1997). Using the Census data, the author created very
similar indicators to those used by Sampson et al. (1997). The indicators were as
follows: percent of families below the poverty level, percent of households receiving

public assistance, percent female-headed households, percent youth, percent black,

 

4" Refer to Appendix C.

91

percent unemployment, percent Hispanic, percent foreign-bom, percent of residents who

lived in a different house 5 years ago, and percent of renter-occupied residential units.45

Table 7. Component Loadings of Community Characteristics

 

 

 

Model (2)
l 2 3

Concentrated Disadvantage

Households below poverty level .81

Households on public assistance .85

Female headed households .80

Youth (less than age 18) .62

African Americans .67

Unemployment .65
Immigrant Concentration

Hispanic .79

Foreign born .79
Residential Instability

Different house from 1995 .85

Renter occupied house .85

 

Since the three neighborhood factors — concentrated disadvantage, immigrant
concentration (ethnic heterogeneity), residential instability (mobility) — had received
enough support, theoretical and empirical, within the neighborhood research, this study
attempted to reduce the indicators to those corresponding factors that share the common
variance of the indicators. In other words, since no hypothesis (or model) testing was

attempted to ﬁnd latent factors here, the Principal Component Analysis (PCA), instead of

 

45 As of the year of 2000, about 10 percent of Lansing’s population had Hispanic origins, and
around ﬁve percent were born in foreign countries. The author could not get any information on
racial makeup of the people born in foreign countries. As such, the author did not attempt to give
further meanings to the ﬁndings related to immigrant concentration.

92

the EFA, was performed three times to extract the three factors.46 Table 7 shows the
component loadings of each indicator for its corresponding factor.

Physical Structure: Land Use

Next, the author used the phone data mainly and the ﬁeld survey data also to
measure land use for street segments and block groups. As explained above, the basic
scheme was to use the survey data for a housing count and the phone data for a business
count at the segment level. Only the phone data was utilized for the block group level,
however.

The basic approach to operationalize the land use was to calculate the percentage
of commercial properties out of all buildings, following Wilcox et al. (2004). However, a
more conservative approach was taken in this study. In other words, the researcher
restricted the business types to those that are generally considered to be crime-attracting.
Put more speciﬁcally, the author referred to the SIC (Standard Industrial Classiﬁcation)
Code and counted businesses with numbers of 50 through 60, 70, 72, 75, 76, 78, 79, 80
and 82.47 Thus, the land use measure of this dissertation is better conceptualized as
criminogenic land use. Considering the study results that certain types of commercial

and public land uses (e. g., library) are not related to high crime rates (Brantingham &

 

4" Refer to footnote 42 for more discussions on the PCA vs. EFA. As mentioned before, the ﬁrst
six indicators were reduced to concentrated disadvantage, the next two to immigrant
concentration (heterogeneity), and the last two to residential instability.

47 50: Wholesale Trade — durable goods, 51: Wholesale Trade -— nondurable goods, 52: Building
Materials, Hardware, Garden Supply,and Mobile Home Dealers, 53: General Merchandise Stores,
54: Food Stores, 55: Automotive Dealers and Gasoline Service Stations, 56: Apparel and
Accessory Stores, 57: Home Furniture, Furnishings, and Equipment Stores, 58: Eating and
Drinking Places, 59: Miscellaneous Retail, 60: Depository Institutions, 70: Hotels, Rooming
Houses, Camps, and Other Lodging Places, 72: Personal Services, 75: Automotive Repair, 76:
Miscellaneous Repair Services, 78: Motion Pictures, 79: Amusement and recreation Services, 80:
Health Services, 82: Educational Services

93

Brantingham, 1982; Peterson et al., 2000), this approach would not cause a serious

measurement problem.

Analytic Strategies

As mentioned already, this dissertation attempts to test a hierarchical model
linking disorder and crime, controlling for the spatial autocorrelation. Adjusting for
autocorrelated crime rates is crucial in hypothesis testing because otherwise the estimated
parameters become unreliable. Meanwhile, since the outcome variable of this
dissertation is count data, the author plans to construct a hierarchical mixed Poisson
model. Both cross-sectional and longitudinal analyses are conducted. Following
Markowitz et al. (2001) and Sampson and Raudenbush (1999), the key strategy for
longitudinal analysis is to control for the prior level of crime. The strategy appears to
further ﬁt well into the hierarchical mixed Poisson model because it requires that
dependent variable(s) have non-negative values. Detailed descriptions of the spatial

analysis and non-linear hierarchical model (HGLM) are presented below.

Spatial Analysis

Every single social phenomenon has a spatial (and temporal) dimension. It seems
that the spatial patterns of most of the social phenomena tend to follow the pattern of
population distribution, which is not random or even. Crime is not an exception and thus
its distribution tends to be spatially dependent. This spatial dependence is well-illustrated
by Tobler’s ﬁrst law of Geography — “everything is related to everything else but nearby

objects are more related than distant objects” (Tobler, 1969).

94

Spatial analysis of crime is important both statistically and theoretically.
Regarding statistical inference, the clustered nature of crime distribution is highly likely
to produce correlated prediction mom in statistical analysis, which is called spatial
autocorrelation of errors. The spatial autocorrelation produces inﬂated and biased
standard errors of regression coefﬁcients, and so, if the spatial process is not accounted
for, statistical inference will be inaccurate. Concerning its theoretical (substantial)
importance, on the other hand, the causal processes involving crime do not necessarily
operate identically in all places (spatial heterogeneity) and spatial analysis can reveal
subareas of geography in which the effects of predictor variables differ (Baller et al.,
2001; Messner and Anselin, 2004).

Put more technically, the behavior of a spatial phenomenon consists of ﬁrst-order
process or large scale variation in mean value (i.e., overall trend) and second-order
process or stochastic deviations from the mean (i.e., local dependence of error) (Bailey
and Gatrell, 1995). The key element in the spatial analysis is to appropriately model the
second-order process. For this, three sources of the second-order process have been
suggested by Anselin and his colleagues — spatial heterogeneity, spatial lag (or spatial
effect), and spatial error (or spatial disturbance) (Anselin, 2002, 2005; Baller et al., 2001).
As explained above, the spatial heterogeneity process represents the unexplained nature
of heterogeneity across study areas. Meanwhile, the spatial lag process signiﬁes that the
crime rate of an area is inﬂuenced by that of neighboring areas. Finally, the spatial error
process indicates the inﬂuence of omitted covariates that are spatially correlated.

Anselin and other researchers suggested a convincing spatial modeling process.

Accordingto them, the ﬁrst step is an exploratory analysis to examine whether crime

95

rates are distributed randomly. Signiﬁcant Global Moran’s 1 indicates non-randomness
of crime distribution, which further suggests that spatial modeling is necessary. In
contrast, a notable spatial process would not exist if the measure of global autocorrelation
(i.e., Global Moran’s I) were not signiﬁcant. Next, a pure structural model or the linear
regression with OLS estimation is run, and it is examined if the residuals are
homogeneous and independent across study areas. Finally, following the results of
diagnostics for autocorrelation, a spatial lag or a spatial error model is tested.

Under the assumptions of independence and homogeneity (i.e., no second-order

. 2 . - -
process), the error matrrx takes a form of G I, which represents the 1dent1cally and

independently distributed (iid) errors. The spatial heterogeneity, however, makes the
diagonal of the error matrix have diverse values.

Meanwhile, the spatial lag process implying a possible diffusion process can be
expressed (in matrix form) as

y = pWy + X13 + u
Where, p is the spatial lag parameter, W is a weight matrix for neighboring units, Wy is a
lagged outcome variable weighted by W, and u is identically and independently
distributed (iid) errors (0'21).48

Lastly, since the spatial error process indicates the inﬂuence of omitted covariates

that are spatially correlated, the error temr can be expressed (in matrix form) as

 

’8 A simple matrix operation transforms the formula into y = (I — pW)'IXB + (I — pW)'lu.
Where, I is identity matrix and (I — pW)l is spatial multiplier. The formula indicates that crime

at one location is inﬂuenced not only by covariates at that location but also by covariates at
neighboring areas through the spatial multiplier. Also, the whole portion for residuals does not

take the iid form (0'21) anymore. The author presents the simpler formula because it appears to
imply the diffusion process better.

96

8 = kWe + u
Where, it is the spatial error parameter, W is a weight matrix, and u is iid. By substituting
the correlated error term into the traditional linear model (y = X13 + s), a new model is

constructed for the spatial error process as below.
y = x8 + (I — )tW)"u
Where, (I — kW)"I is spatial multiplier and u is iid. This model indicates that crime at a

location is inﬂuenced by errors at neighboring areas through the spatial multiplier.

Construction and Control of Spatial Lag

To examine the plausible spatial processes, this dissertation uses a user-friendly
software package for spatial analysis, GeoDa, which was developed by Anselin with the
support of the Center for Spatially Integrated Social Science (CSISS). Unfortunately,
however, it is not feasible to follow the whole modeling process suggested above because
it does not ﬁt with the hierarchical modeling process. Instead, Anselin suggests that a
researcher can create a spatially lagged variable within GeoDa and include the lag in
another software as a control for the neighboring effects. This way one can test spatial
processes of interest.

This study has interests in whether the crime rates of an area (i.e., block group
mean of crime rates) are inﬂuenced by those of neighboring areas above and beyond the
impact of structural covariates. If this is the case, it suggests a spatial difﬁrsion process,
although a more formal test is required to verify that. Meanwhile, the study has less
interest in the spatial error process, or the impact of omitted covariates, because the study
model of this dissertation appears to be well-speciﬁed as the theory suggests, with the

possible exception of neighborhood process. In fact, few models in social science

97

research can be perfectly speciﬁed. Further, it is difﬁcult to model the spatial error
process in HLM due to the author’s lack of knowledge and the limitation of GeoDa’s
ﬁrnctionality. Therefore, this research project attempts to test the spatial lag process in
HLM (HGLM, put more speciﬁcally) after constructing the lag variable of crime rates in

GeoDa.

A Hierarchical Mixed Poisson Model

When an event of interest is discrete, truncated at zero, and distinctly skewed (i.e.,
count data), the traditional linear model with OLS estimation produces biased and
unreliable estimates due to violations of several assumptions. First, the event is highly
likely to have a non-linear relationship with independent variables. Thus, a linear model
might predict a negative value, which does not make sense at all. Second, the prediction
residuals tend to have a differential variability depending on the level of independent
variables (heteroscedasticity). Finally, the residuals are likely to be correlated with
independent variables (Raudenbush & Bryk, 2002; Sturman, 1999). A related issue is
that any transformation of the event cannot solve the problems. For example, a square
root transformation keeps the value greater than zero (truncated) and still suffers from
heteroscedasticity (Sturman, 1999). The 10 git transformation that is useful in a binary
outcome model does not solve the heteroscedasticity and non-linearity, although it makes
the value of an event unlimited. Therefore, an effective solution is to use a nonlinear
model based on a relevant distribution function for the count data, such as the Poisson

model (Raudenbush & Bryk, 2002). 49

 

4” To examine whether the non-linear model of this study better ﬁts the data distribution than its
linear counterpart does, the author tested a linear (unconditional) model and compared the results
(reliability estimates) of both models. For the linear model, the author tried to normalize the

98

Various Poisson Models

The Poisson distribution is a discrete probability distribution discovered by the
French statistician Simeon-Denis Poisson (1781-1840). The number of events in an
interval of given length or in a spatial area is Poisson distributed with the below
probability density function. In other words, the function speciﬁes the probability that

there are exactly k occurrences of Y in a given time or area (Land et al., 1996).50
_ _ -7. k ,
P(Y—k)—e is /k.
Where:
k is a natural number including 0, k = 0, l, 2, > 0,
e is the base of the natural logarithm (e = 2.71828. . .),
k! is the factorial of k,
7t. (intensity or rate parameter) is a positive real number, equal to the
expected number (mean) of occurrences that occur during the given
interval (or in the given area).
The formula speciﬁes an unconditional model that the probability does not
depend on factors other than it. A standard application of the Poisson model speciﬁes a
conditional Poisson process where each unit is permitted to have its own unique intensity

parameter, and differences across units in these parameters are a function of a set of

known explanatory variables. Under this conditional Poisson speciﬁcation, the

probability that the ith area will experience It events is given by:
P(Yi=kIX)=e"‘)tk/k!

And the intensity parameter is specified by:

 

distribution of the dependent variables (i.e., weighted crime rates) by adding 1 and taking the
natural log. The results are presented in Appendix F. The relatively low reliabilities (particularly
for the violent crime model) indicated that the block group mean estimates in the linear model are
much less reliable than those in the non—linear model, which further suggested that the latter
model better reﬂects the data (i.e., real world phenomenon).

50 In light of the purpose of this project, my discussion focuses on the area units rather than
temporal periods.

99

Eon 1 X) = ), = e<z (W).
In this speciﬁcation, ﬂjs are effect parameters to be estimated. Thus, the analytical

framework is based on the premise that the likelihood any area will experience a given
number of discrete events is a function ofa set of exogenous factors that affect the
intensity parameter governing the area. Variation in the number of events across areas,
therefore, must be due to corresponding spatial variation in the explanatory factors
(Cameron & Trivedi, 1998; Beck & Tolnay, 1995).

The Poisson distribution possesses an interesting feature in that the mean
(expected value) of the event is the same as its variance, which is called “equidispersion”
(Cameron & Trivedi, 1998). In many instances, however, count data displays
“overdispersion,” with its variance being larger than the mean. Thus, applying the
Poisson model to overdispersed dat'a results in a violation of the crucial assumption,
leading to smaller standard errors. It eventually increases the probability of
misidentiﬁcation of non-existent relationships (Type I error). Further, the Poisson model
assumes that an occurrence of the event is independent of the number of previous
occurrences. This assumption is also inconsistent with many real cases, which provides
another source of controversy.

There is a modiﬁed Poisson approach that deals with the overdispersion problem

by adding one unknown parameter ‘1’ to the model.
E(Yi l X) = in = 9(2 ﬂny)
Var (Y, 1 X) = rm,
In this model, the estimated effect coefficients are not affected, but only their standard

errors become larger by the factor \tl. Hence, the asymptotic t ratios will be

100

proportionally smaller. However, the model assumes that there are no excluded
explanatory variables or other sources of randomness.
To relax this strong assumption, one can respecify the regression component to

include a term 8 for speciﬁcation errors or other origins of randomness.

E(Yi l X, 8) = in = 3Q: [Br/Yd + 8")
This model is known as a mixed Poisson regression because it accounts for the two
stochastic processes: (a) the number of events experienced by each unit is assumed to
follow a Poisson process, but (b) the Poisson parameter itself has a gamma distribution.5 1
The model is very attractive because it not only corrects for the overdispersion problem
but also allows for random errors and random inﬂuences of excluded variables (Land et
aL,l996)

Construction of H GLM

A handy software for the hierarchical modeling is HLM 6.0 developed by
Raudenbush, Bryk, and Congdon (2005). It has a function to handle various forms of
hierarchical generalized linear models (HGLM), including the hierarchical mixed Poisson
model. For a valid test for the hierarchical mixed Poisson model, it is suggested that
researchers select the “LaPlace” approximation and “full PQL (Penalized Quasi-
Likelihood)” estimation method. Furthermore, in general, only one level 2 random effect
is estimated for HGLM. Following that, this dissertation attempts to ﬁt the various
datasets to the hierarchical mixed Poisson model.

It is customary to construct an unconditional model (i.e., one-way ANOVA) ﬁrst

to examine the variability of the outcome measure across level 2 units. Then, a random

 

5' The gamma family of distributions is a very general class of continuous, unimodal distributions
for random variables that take on only nonnegative values (Land et al., 1996, p. 393).

101

coefﬁcient model is constructed to test the ﬁrst hypothesis. The other hypotheses are

0

tested via a conditional model that controls for all variables at both levels. Below is an

example of cross-sectional conditional model for violent crime 2003.

Where:

Level 1: Tlij = log (VC03) = [301- + tin-*PD03 + [32J-*SD03 + B3j*LUO3

LCVC] 23 [30) = 700 + 701*CD + 702’"1C + 703*RI + Y()4*LU +
Y05*W(VCO3) + Uoj
BU = Yto + 711*CD + 102’le + Y13’kRI + 714*LU
[32) = 720 + Y21*CD + Y22*IC + 1(23’"RI + Y24*LU
[33) = 730

,qu = the average log of VC 2003 of a block group j when there is no PD, SD, or
LU

1611' = the average slope of PD 2003 in a block group j

,ng = the average slope of SD 2003 in a block group j

,6’3j = the average slope of LU in a block group j

1’00 = the average log of VC 2003 across 83 block groups with no PD, SD, or LU
ylg = the average slope of PD 2003 across 83 block groups

720 = the average slope of SD 2003 across 83 block groups

3130 = the average slope of LU across 83 block groups

ya 1, 702, y03’ yo.) = the effects of structural covariates on the average log of VC 2003
in a block group j

y I 1.3) )4. y; ,--y34 = the effects of structural covariates on the average slope of SD
2003 and PD 2003 in a block group j
V05 = the effect of neighboring VC 2003 on the average log of VC 2003 in a block
groupj
u(),- = level-2 random effect (deviation of block group j’s mean from the grand
mean)

Therefore, the ﬁrst hypothesis (H 1 .' Disorder, physical and social, has a significant effect

on serious crime at the ﬁrst level.) is tested via BU and [394.52 The second hypothesis (H2:

 

52 It must be noted that the ﬁrst hypothesis is tested by a random coefﬁcient model that has
control variables (i.e., physical disorder, social disorder, and land use) only at the segment level.

Refer to

the chapter on results.

102

Structural characteristics have signiﬁcant effects on block group means of crime rates at
the second level.) is examined via YOI , 1’02, 703, 'Y04. The third hypothesis (H3: The effect
of disorder on crime remains significant after controlling for neighborhood
characteristics — multivariate significance.) is analyzed via Yin and Y20- The fourth

hypothesis (H4: The relationship between disorder and crime at the first level is

significantly inﬂuenced by structural covariates at the second level — cross—level
interaction.) is tested via ”Yr 1, 712, 713, 714, Y2], 'Yzz, Y23, ’Yz4. The ﬁﬁh hypothesis (H5:

The mean crime rates of a block group is further influenced by the mean crime rates of

neighboring block groups, after controlling for the structural covariates — spatial lag.) is
examined via 705.

Meanwhile, an example of longitudinal conditional model for violent crime 2005
is presented below. It must be noted that land use 2005 is controlled for in the model,
instead of the land use change, because only 11 segments experienced a decrease or
increase of commercial land use. Also, as suggested above, the author does not use the
change of crime as the dependent variable because the Poisson model requires the
outcome variable to be non-negative numbers. Instead, logged crime rates of the

previous year (i.e., year 2003) will be controlled for in the model.

Level 1: n,- =1og(VC05)=80,- + p.,*(PD05-PD03) + 132,-*(SD05-
SD03) + p,,-*L1105 + p..,-*vc03

LCV61 23 Biij = Yno + 'Y01*CD + Y()2*IC + “M3":RI + Y04*LU +
Y05*W(VC05-VC03) + 1.10)"
Blj = Yto + Y11*CD + 712*IC + ’Y13*RI + 714*LU
52) = 720 + Y21*CD + ‘Yzz’kIC + 1’23*RI + Y24*LU
B3) = 730
B4) 2 Y40

103

Where:

ﬁg,- = the average log of VC 2005 of a block groupj when there is no PD change
(from 2003 to 2005), SD change (from 2003 to 2005), LU 2005, or VC 2003.

,8 U = the average slope of PD change (from 2003 to 2005) in a block group j
,ng = the average slope of SD change (from 2003 to 2005) in a block group j
,83j = the average slope of LU 2005 in a block group j

yog = the average log of VC 2005 across 83 block groups with no PD change, SD
change, LU 2005, or VC 2003

y”) = the average slope of PD change across 83 block groups
720 = the average slope of SD change across 83 block groups
y3() = the average slope of LU 2005 across 83 block groups

yo), yo; ”)3, My = the effects of structural covariates on the average log of VC
change in a block group j

y) /,y)4_ y31~y34 = the effects of structural covariates on the average slope of SD
change and PD change in a block group j

3105 = the effect of neighboring VC change on the average log of VC change in a
block group j

The ﬁrst hypothesis (H1: Change of disorder, physical and social, has a significant effect
on change of serious crime at the first level.) is tested via [31 J- and [321-.53 The second
hypothesis (H2: Structural characteristics have significant effects on block group means
of crime rates change at the second level.) is examined via Ynt , ’Yoz, 'Y03, 'Y()4. The third

hypothesis (H3: The effect of disorder change on crime change remains significant after

controlling for neighborhood characteristics — multivariate significance.) is analyzed via
y“) and 730. The fourth hypothesis (H4: The relationship between disorder change and
crime change at the first level is signiﬁcantly inﬂuenced by structural covariates at the
second level — cross-level interaction.) is tested via 71 t, 'le, 713, 714, 72), 722, 723, 724.

The ﬁfth hypothesis (H5: The change of mean crime rates of a block group is further

 

5" As with the cross-sectional model, the ﬁrst hypothesis is tested by a random coefﬁcient model.

104

influenced by the change of mean crime rates of neighboring block groups, after

controlling for the structural covariates — spatial lag.) is analyzed via 705.

105

CHAPTER VI: RESULTS
This chapter presents the results of various analyses. First, general characteristics
of the study subjects are described for the study variables. A simple examination was
further attempted to check whether there had been a notable pattern of change of disorder
and crime. Next, bivariate relationships among the variables are presented. Third,
detailed procedures for constructing spatial lag and examining autocorrelation of crime
are described. Finally, the results of multivariate hierarchical models, both cross—

sectional and longitudinal, are discussed.

General Characteristics of Study Subjects (Street Segments and Block Groups)

Table 8 shows the descriptive statistics of the variables for both street segments
and block groups. The average violent crime rate for 2003 was .93 (per 10 buildings)
within a range of 0 and 11.1 1. That for 2005 was .90 within a range of 0 and 13.33.
Meanwhile, the average property crime rate was 2.05 (per 10 buildings) for both years.
The minimum was 0 and maximum was 33.33. Both violent and property crimes were
highly and positively skewed, which was consistent with the distribution of the count
data.54

The mean factor scores of social disorder were 1.00 and 1.43 for 2003 and 2005.
Interestingly, social disorder appeared to take place as rarely as property crime in light of

the average rates of 2. l 8 and 2.24 for 2003 and 2005.55 It also was highly and positively

 

5" Although not shown in Table, 166 segments out of 377 study segments (44%) had no
occurrence of violent crime in 2003. Similarly, 160 (42.4%), 78 (20.7%), and 85 (22.5%)
segments showed no case of violent crime 2005, property crime 2003, and property crime 2005,
respectively. It also appeared to be somewhat consistent with the general frequency distribution
of the count data.

55 Similar to property crinre, 84 (22.3%) segments in 2003 and 93 (24.7%) segments in 2005
showed no occurrence of social disorder.

106

skewed. Meanwhile, physical disorder showed mean factor scores of 1.33 and 1.14. It
was not notably skewed, which was understandable because all segments contained a

certain amount of physical disorder.

Table 8. Descriptive Statistics

 

N Mean SD. Min Max Skewness

 

Segment-Level Variables

Violent Crime 2003 377 .93 1.51 .00 11.11 3.22
Violent Crime 2005 377 .90 1.47 .00 13.33 4.01
Property Crime 2003 377 2.05 3.27 .00 33.33 4.94
Property Crime 2005 377 2.05 2.79 .00 33.33 4.92
Social Disorder 2003 (rates) 377 2.18 3.48 .00 25.75 3.54
Social Disorder 2005 (rates) 377 2.24 3.95 .00 41.67 5.26
Social Disorder 2003 (factor) 377 .00 1.00 -.46 8.84 4.72
Social Disorder 2005 (factor) 377 .00 1.43 -.90 13.00 3.79
Physical Disorder 2003 (factor) 377 .00 1.33 -3.11 3.98 .43
Physical Disorder 2005 (factor) 377 .00 1.14 -2.09 4.01 .67
Land Use 2003 377 5.54 13.95 .00 100.00 4.03
Land Use 2005 377 5.62 14.56 .00 100.00 4.18
Block Group-Level Variables

Concentrated Disadvantage 83 .00 1.00 -l .65 2.56

Immigrant Concentration 83 .00 1.00 -1.67 3.16

Residential Instability 83 .00 1.00 -1.39 2.93

Land Use 83 5.74 5.62 .00 33.62

 

The average land use was 5.54 and 5.62 for 2003 and 2005 at the segment level,
which indicated that 5.54% and 5.62% of all buildings at the study segments were crime-

attracting commercial properties for 2003 and 2005.56 The study block groups had an

 

5° For both years, 264 (70%) segments did not have crime-attracting businesses at the segment
leveL

107

average of 5.74% of crime-attracting commercial properties for 2003. Similar to crime
and disorder variables, the land use was highly skewed.

Finally, the three social structural covariates — concentrated disadvantage,
immigrant concentration, and residential instability — were converted to mean component

scores and thus were all zeros with one standard deviation.

Change of Disorder and Crime

Since the main focus of this study was whether the change in crime is attributable
to the change in disorder, the author examined several descriptive statistics for evidence
of change in crime and disorder. First, the overall rates of crime and social disorder were
examined. As Table 9 shows, violent crime rates decreased from .93 in 2003 to .90 in
2005, which amounted to a reduction of 3.33%. However, property crime rates remained
the same. In contrast, social disorder rates increased from 2.18 in 2003 to 2.24 in 2005,
which was equal to an increase of 2.75%. Overall, crime and social disorder appeared to
have remained stable.

Second, the total sum of physical disorder indicators appeared to have decreased
from 8.22 in 2003 to 7.00 in 2005, which represented a reduction of 14.84%.57 Although
the change seems to be impressive, the results for the two surveys should not be
compared because, though every effort was made to maintain inter-rater reliability and
follow coding guidelines (Appendix B), the two ﬁeld surveys for physical disorder were

conducted. by different researchers. Thus, 10% of street segments with the lowest sum of

 

57 Total sum of 2003 (or 2005) = z-score of cigarette (item 9) + rates of litter (item 10) + rates of
garbage (item 11) + rates of empty bottles (item 12) + rates of abandoned or disabled cars (item
14 or 15) + rates of condoms (item 17) + rates of needles and syringes (item 18) + rates of grafﬁti
(items 19 and 20) + rates of dilapidated houses (items 22, 23, and 24). These items are those that
were used to measure physical disorder. Refer to pages 85 and 86.

108

physical disorder (n=3 8) and 10% of street segments with the highest sum of physical
disorder (n=38) were examined to check whether there had been notable changes in

physical disorder, social disorder, and crime on those segments during the study period.

Table 9. Change of Disorderand Crime for Total and Selected Segments58

 

 

2003 2005 Change
Physical Disorder 8.22 7.00 -l .22 14.84%
Social Disorder 2.1.8 2.24 .06 2.75%
Total Segments
(n=377) Violent Crime .93 .90 -.03 3.33%
Property Crime 2.05 2.05 .00 0.00%
Physical Disorder 16.30 10.03 -6.27 38.47%
High PD Social Disorder 2.17 3.00 .83 38.25%
Segments Violent Crime 1.15 1.42 .27 23.48%
(“‘38) Property Crime 2.18 2.39 .21 9.63%
Physical Disorder 1.84 3.50 1.66 90.22%
150‘” mt Social Disorder .57 1.55 .98 171.93%
($323“ 5 Violent Crime .45 .55 .10 22.22%
Property Crime 2.37 1.69 -.68 28.69%

 

As Table 9 reveals, the segments with high physical disorder experienced a
stronger reduction than the total did (38.47% vs. 14.84%), which seemed to be a good
sign. However, those with low physical disorder went through an opposite trend:
physical disorder increased 90.22% in those segments. Overall, it appeared that the level
of physical disorder tended to converge toward the mean.

Interestingly, social disorder and the two types of crimes increased on the high

physical disorder segments, which foreshadowed that physical disorder might not be

 

’8 A series of paired samples t-tests were conducted to statistically check the changes of violent
crime and social disorder for the total segments. No signiﬁcant change was discovered (not
shown in table).

109

highly correlated with social disorder and crimes. On the low physical disorder segments,
meanwhile, social disorder and violent crime went up 171.93% and 22.22%, respectively.
But, property crime decreased 28.69%. Overall, no notable pattern of change was

detected.

Bivariate Relationships

As Table 10-1 shows, the zero-order correlations among independent variables at
the segment level were not that high, even though some of them were statistically
signiﬁcant (e. g., PD 2003 & SD 2003, PD 2003 & LU 2003, SD 2005 & LU 2005).
Except land use, the structural covariates at the block group level were all highly
correlated at the .01 level, which indicated that there could be a multicollinearity problem

(Table 10-2).

Table 10-1. Correlation Matrix (Level 1)

 

. PD 2003 1.00
. SD 2003 .13‘ 1.00
. LU 2003 .10‘ .05 1.00

‘Q It i.

. VC 2003 .23 .29 .16 1.00

O ti *0 $0

.PC2003 .11 .24 .57 .47 1.00
PD2005 .44“ .14" -04 .10‘ .04 1.00

it 3 it Q.

.SD2005 .19” .18 .16‘ .19 .22 .03 1.00

t t. 0. .t O.

. LU 2005 .11 .04 .98 .14 .53 -.05 .24 1.00

.vc2005 .23" .11‘ .22“ .38" .32 .13 .31 .26 1.00

it it

.40 1 .00

\omqgmhuws—

it .‘

10.PC2005 .21“ .21“ .40" .25" .39 .07 .45 .49

 

*p < .05 u p < .01 (2-tailed)

110

Table 10-2. Correlation Matrix (Level 2)

 

 

 

 

1 2 3 4
LCD 1.00
2. IC .48” 1.00
3. R1 .38" .42“ 1.00
4. LU -.04 .15 .14 1.00
‘p<.05 ”p<.01(2-tailed)

Diagnostics of Multicollinearity

One of the important assumptions of multivariate analysis is that there should be
no perfect collinearity between independent variables, for a perfect collinearity makes it
impossible to calculate parameter estimates. In practical situations, however, a perfect
collinearity problem rarely happens, and if it takes place, computer programs warn
researchers, thus it is easy to ﬁgure out the problem. The real issue is that sometimes
independent variables are either highly correlated or linearly dependent with each other,
which is called multicollinearity. It is problematic because it tends to result in very large
standard errors and unreliable or unusual regression coefﬁcients (Allen, 1997; Berry &
Feldman, 1985; Fox, 1991).

Besides high bivariate correlations, as mentioned above, the multicollinearity
problem can spring from a linear combination of more than two other independent
variables. Notwithstanding the relatively low correlations among the level 1 and level 2
independent variables,” therefore, the author further conducted more rigorous

diagnostics by examining tolerance statistic, VIF (Variance Inﬂation Factor), and

 

5” It would be fair to say that the bivariate correlations were relatively low in that the usual cutoff
value for multicollinearity is .80 (Berry & Feldman, 1985: 43).

111

condition index. For this, OLS regressions were performed via the SPSS program, and
the results for the level 2 variables are discussed below."0 (’1
First, tolerance and VIF (Variance Inﬂation Factor) statistics were examined,
which were more likely than the correlation matrix to detect a multicollinearity problem,
especially when there were multiple correlations between independent variables. Table

11 describes the results. All of the level 2 covariates did not appear to be problematic,

since both tolerance and variance inﬂation factors were close to 1.0. (’2

Table 11. Tolerance and Variance Inﬂation Factors (VIF)

 

 

Independent Variables Tolerance VIF

Concentrated Disadvantage .72 1.34
Immigrant Concentration .68 1.46
Residential Instability .77 1.30
Land Use .95 1.01

 

Table 12. Collinearity Diagnostics

 

Variance Proportions

 

comet Eigen- Condition
P value Index Intercept CD IC RI LU

 

l 1.89 1.00 .01 .10 .11 .ll .02
2 1.69 1.06 .12 .03 .02 .02 .ll
3 .64 1.72 .02 .38 .03 .74 .00
4 .52 1.90 .03 .40 .77 .11 .01
5 .26 2.69 .82 .09 .07 .03 .85

 

 

(’0 Since the study model included (i.e., controlled for) the average crime rates of neighboring
block groups (= lagged crime rates), the author also examined the various collinearity statistics
for the level 2 variables including the lagged crime rates. The results were little different than
what is presented in the text. Refer to Appendix D.

6' The results for the level 1 variables are presented in Appendix E. As was the case with the
level 2 variables, no substantial sign of multicollinearity was detected.

62 The general rule of thumb is that VIF over 10 is problematic, though there is not a cutting-point
(DeJong, 2004).

112

To further check the magnitude of linear dependency and the amount of
variation that each variable contributes to each concept, the author examined condition
index and variance proportions. Table 12 contains the results. No variable appeared to
have abnormal values in the condition index, a measure of linear dependency."3 Also, the
variance proportions indicated that none of the variables appeared to measure one same
concept

In sum, the various diagnostic statistics indicated that there seems to be no
substantial multicollinearity problem in the multivariate model. Also, the four level 2
covariates were those that have been theoretically important and empirically supported.

Therefore, the author decided to keep all the covariates in the model.

Correlations between Dependent Variables and Independent Variables

The dependent variables, violent crime rates and property crime rates, were
signiﬁcantly and positively associated with most of the segment level independent
variables (Table 10-1). It was remarkable that both types of crimes were correlated more
highly with social disorder than with physical disorder. However, it was not unexpected
because social disorder is generally regarded as more closely related to serious crime than
is physical disorder. Also, the issue of “same data source” could have been inﬂuential
although the EFA conﬁrmed that social disorder and crimes represent different concepts.

The relationships were ﬁrrther tested via multivariate models in the next sections.

 

(’3 In general, condition index over 30 is problematic, although it is not a cutting-point (DeJong,
2004).

113

Spatial Lag and Autocorrelation of Crime

Before examining the spatial autocorrelation of crime rates, the author attempted
to create lagged variables of them, using GeoDa. For this, crime rates were ﬁrst
aggregated into the block groups, since they were measured at the segment level. Figure
8 shows the overall distribution of violent crime rates and property crime rates. Although
several areas, including the “Chestnut area” (northwestern area of the State Capitol),
appeared to suffer from the concentrated crime problem, it was hard to tell ﬁom the maps

whether the crime rates were autocorrelated or not.

Figure 8-1. Quantile Maps of Crime Rates (Violent Crime 2003)

 

Quantile: VC03
E] 1St range (20)
2nd range (21)
E”; 3rd range (21)
- 4th range (21)

 

 

 

 

 

114

Figure 8-2. Quantile Maps of Crime Rates (Property Crime 2003)

 

Quantile: PC03

[:1 15trange (20)
ndrange (21)

N

 
   

rd range (21)
- 4th range (21)

Lo.)

 

 

 

 

 

 

115

Figure 8-3. Quantile Maps of Crime Rates (Violent Crime 2005)

 

Quantile: VC05

[:1 1S" range (20)
"I I ”i 2nd range (20)
rd

3 range (22)
- 4th range (21)

 

 

r ..

1;.

 
   

 

 

 

 

 

 

116

Figure 8-4. Quantile Maps of Crime Rates (Property Crime 2005)

 

 

Quantile: PC05
C] 151 range (20)
2nd range (21)
3: range (21)

- 4h range (21)

 

 

 

 

 

 

Next, spatial weights were assigned to each block group, considering two
technical points. (1) By deﬁnition, while the rook criterion counts neighboring units with
common boundaries as the neighboring units, the queen criterion considers neighboring
units with either common boundaries or common comers as the neighboring units
(Anselin, 2005). In this study, the queen criterion rather than the rook criterion was used
to determine neighboring block groups, because the suggested difﬁlsion process was not
necessarily supposed to take place in surrounding block groups with common boundaries.
(2) By deﬁnition, the ﬁrst-order neighboring units based on the queen criterion include

the neighbors with common boundaries or comers. Whereas, the hi gher—order

117

neighboring units based on the queen criterion (e. g., the second-order neighboring units)
include not only the ﬁrst-order neighbors but also the neighbors having common
boundaries or comers with the ﬁrst-order neighbors (Anselin, 2005). In this study, only
the ﬁrst-order neighboring units were weighted considering the small number of block
groups.

Through the procedures, the spatial lags were created for both types of crime rates
for both years. Further, the spatial lags for the change in crime rates from 2003 to 2005
were created through the same procedures.I It is notable that spatial lag corresponds to “a

simple average of the values for the neighboring units” (Anselin, 2005, p. 125).2

Figure 9-1. Moran’s I and Permutation Tests for Crime Rates

 

 

Moran’s I=-0.0470 Moran’s l=0.0054

 

 

 

 

 

w_VC03
W PCO3
+90
1-
+
I

 

 

 

 

 

 

 

 

 

 

VC03 PC03

P = .25 (999 permutations) p = .34 (999 permutations)

 

' The spatial lags for the change in crime rates are used in the longitudinal model, and the other
lags are controlled for in the cross-sectional model.

2 One block group (260650065004) had no neighboring units (i.e., island) and the spatial lags of
crime rates of the block group were zeros. It caused a problem when the author attempted to
transform the spatial lags to logged lags in order to control for the logged lags in the conditional
models (see p. 121). Thus, the aggregated average rates of crimes were assigned to the block
group as its spatial lags.

118

Figure 9-1. (cont’d)

 

 

w_vc05

 

P = .1 1 (999 permutations)

Moran’s I=-0.0942

 

 

 

 

 

 

VC05

 

Moran’s l=-0.0026
8
8 at».
3
PC05
p = .59 (999 permutations)

 

 

 

 

 

 

 

Figure 9-2. Moran’s I and Permutation Tests for Crime Rates Change

 

w_vcs3

 

p = .53 (999 permutations)

rates were spatially dependent or not. To this end, the Global Moran’s 1 statistics were

calculated and then tested through a random permutation (or simulation) procedure. The

Moran’s 1=-0.0049

 

 

 

 

 

 

VC5 3

 

Moran’s l=0.001 l
in“
B . ‘
l * ‘.
3 we.
PCS 3
p = .39 (999 permutations)

119

 

 

 

 

 

 

Using the original and lagged crime rates, the author tested whether the crime

results are presented in Figure 9. No signiﬁcant results were detected either for cross-
sectional lags (Figure 8-1) or longitudinal lags (Figure 82),“ which suggested that the
crime rates were not spatially correlated to each other. Further, results suggested that
there might be little utility to control for the lagged variables in the HGLM model.
However, the author decided to include them in the model to obtain parameter estimates

that were as valid as possible.

The Hierarchical Mixed Poisson Model
HGLM uses the log link function when the level-1 sampling model is Poisson.
’7ij = 108(k))

The purpose is to handle the truncated nature of count data. In other words, while tel-J- is

always a nomregative number, log(t\,-J-) can take on any value. For an adequate interpretation

of the coefﬁcient, however, researchers need to transform the logged value back to the
original form by using the exponential function. Fortunately, the HLM program produces
results with both the logged and original units (Raudenbush and Bryk, 2002).

To examine the variability of violent crime and property crime across the block
groups, an unconditional model (one-way ANOVA) was tested ﬁrst. Next, cross-sectional
analyses were performed, including a random coefﬁcient model and a conditional model, to

test the study hypotheses. And then, the author conducted longitudinal analyses.

 

6" A plausible reason was that the crime rates were unstable for two reasons. First, since crimes
were rare events at the segment level, the crime rates had much larger variances compared to the
means (refer to Table 8). Second, the number of study segments in each block group varied a lot.

120

Unconditional Model (One-Way ANOVA)

The author tested an unconditional model to check whether the variances of crimes
are statistically signiﬁcant across the level 2 units, block groups. It was the ﬁrst step
required to verify the utility of a hierarchical approach to ﬁt the data. Table 13 presents the

results.

Level 1: 770- = Log(crime rates) = ,80,
Level 2: ﬂu,- = 1’00 + uaj, in), ~ N (0, too)

Where:
,60} = the average log of crime of a block group j
1’00 = the average log of crime across 83 block groups
up,- = level-2 random effect (deviation of block group j ’s mean from the grand mean)
T0() = the variance between block groups in block group-average log of crime

Table 13-1. Unconditional Model (Violent Crime 2003)(’7

 

Coefﬁcient Standard

 

Fixed Effect

 

 

Log Rate Error 1?
1’00 -.14 .87 .10 .17
Variance
Random Effect Component P
Block group mean, ugj .53 .00

Reliability of C oefficient Estimate
,Boy' .56

As mentioned before, the reported statistics are based on the LaPlace approximation. Also, the
author reported the results of a population average model rather than of a unit speciﬁc model for
two reasons: ﬁrst, the research interest of this study was in how a unit change in the independent
variables is expected to affect the overall mean of the outcome variables, not in how a unit change
in the independent variables is expected to affect a particular level 1 unit’s mean. Second,
population average inferences do not require as many assumptions about the distribution of
random effects as unit speciﬁc inferences do, and tend to be more robust to erroneous
assumptions. However, it must be noted that the HLM program does not produce deviance
statistics for the population average model, which makes it difﬁcult to compare the superiority of
various models through a multivariate likelihood-ratio test (Raudenbush & Bryk, 2002).

 

 

()7

121

Table 13—2. Unconditional Model (Property Crime 2003)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Coefﬁcient Standard
Fixed Effect Log Rate EITOI' P
)100 .61 1.84 .10 .00
R d m Eff r variance P
on 0 cc Component
Block group mean, up] .59 .00
Reliability of Coefficient Estimate
’60]. .70
Table 13-3. Unconditional Model (Violent Crime 2005)
Coefﬁcient Standard
Fixed E [feet Log Rate Error P
V00 -.20 .82 .09 .03
Ra dom Eﬂe t Variance P
n C Component
Block group mean, up] .34 .00
Reliability of C oeﬂicient Estimate
ﬁaj .46
Table 13-4. Unconditional Model (Property Crime 2005)
Coefﬁcient Standard
Fixed Effect Log Rate EI‘I‘OI‘ P
3200 .74 2.09 .09 .00
Random Eﬂect Variance P
Component
Block group mean, ugj .49 .00
Reliability of Coefficient Estimate
.70

.301

 

122

The average logs of violent crime 2003, violent crime 2005, property crime 2003,
and property crime 2005 across 83 block groups were -. 14, .61, -.20, and .74, respectively.
Each of them corresponded to the rates of .87, 1.84, .82, and 2.09.

The signiﬁcant variance components for both types of crimes of both years indicated
that the block groups have signiﬁcantly different averages of crimes. This was ﬁlrther
supported by the relatively higher reliability of the level 1 coefﬁcient estimates. These

results justiﬁed the hierarchical approach in this study.

Cross-Sectional Analysis - Random Coefficient Model

Next, a random coefﬁcient model was ﬁtted to examine whether the covariates at
street segments have a signiﬁcant effect on crime without controlling for the block group
level structural covariates (Hypothesis 1). First, a cross-sectional analysis was conducted.
In other words, physical disorder, social disorder, and land use of the same year as the crime

were controlled for. The results are displayed in Table 14.

Level 1. I?” :ﬂ0j+ ﬂ/I’I‘PD + ﬂ2j*SD + ﬂ3/*LU68

Level 2: ﬁg,- 2 1’00 + 1’0)

ﬁt] = 1’10
,621 = 1’20
’53)“: 1’30

Where:

180/ = the average log of crime rates of a block group j when there is no PD, SD, or
LU

,B/j = the average slope of PD 2003 in a block groupj
,ng = the average slope of SD 2003 in a block groupj
,33, = the average slope of LU in a block groupj

 

b8 PD = Physical Disorder, SD = Social Disorder, LU = Land Use

123

The average block group mean was .74, .69, 1.34, and 1.65 for violent crime 2003,
violent crime 2005, property crime 2003, and property crime 2005, respectively, when
physical disorder, social disorder, and land use were zero. After controlling for social
disorder and land use, one unit (i.e., one standard deviation) increase of physical disorder
was associated with .22 log increase of violent crime 2003, .21 log increase of violent crime
2005, .03 log increase of property crime 2003, and .17 log increase of property crime

2005.69 Except property crime 2003, the effects were highly signiﬁcant.

Table 14-1. Random Coefﬁcient Model (Cross-Sectional - Violent Crime)

 

 

 

 

2003 2005
Fixed Effect Log Count Se P Log Count Se P
Average BIOCK Gl‘OUp Mean, 1’00 -.31 .74 .08 .001 -.37 .69 .09 .000
PD-Crime Slope, )1”) .22 1.24 .04 .000 .21 1.23 .05 .000
SD-Crime S10pe, ygo .19 1.21 .04 .000 .14 1.15 .03 .000
LU-Crime Slope, no .01 1.01 .00 .000 .01 1.02 .00 .000

 

Table 14-2. Random Coefﬁcient Model (Cross-Sectional - Property Crime)

 

 

 

 

2003 2005
Fixed Effect Log Count Se P Log Count Se P
Average Block Group Mean, 1’00 .29 1.34 .08 .000 .50 1.65 .07 .000
PD-Crime Slope, ylg .03 1.03 .03 .376 .17 1.18 .04 .000
SD-Crime Slope, )130 .19 1.20 .03 .000 .11 1.11 .02 .000
LU-Crime Slope, y30 .03 1.03 .00 .000 .02 1.02 .00 .000

 

 

(’9 Although log .22, log .21, log .03, and log .17 correspond to net count of 1.24, 1.23, 1.03, and
1.18, researchers should not use the net count because the Poisson model is a non-linear model.

124

One unit (i.e., one standard deviation) increase of social disorder increased the rates
of violent crime 2003, violent crime 2005, property crime 2003, and property crime 2005
by .19 log, .14 log, .19 log, and .11 log, respectively. The effects were highly signiﬁcant for
all outcome variables. Finally, land use also had a signiﬁcant effect on all dependent
variables, controlling for physical disorder and social disorder. In sum, except for the
inﬂuence of physical disorder 2003 on property crime 2003, the results appeared to be

supportive of the ﬁrst research hypothesis of the cross-sectional model.

Cross-Sectional Analysis - Conditional Model

Finally, the author controlled for the exogenous variables at both levels. A cross-
sectional model was ﬁrst constructed. Four main interests were examined: the ﬁrst interest
was in what community factors affected the block group means of crime, controlling for the
other factors (Hypothesis 2). Another interest was whether the signiﬁcant relationships
between disorder and crime, conﬁrmed by the above random coefﬁcient model, remained
signiﬁcant even after controlling for the level 2 structural covariates (Hypothesis 3 —
multivariate signiﬁcance). Third, and one of the most critical interests, was whether the
relationships between disorder and crime at the ﬁrst level were signiﬁcantly moderated by
the structural factors (Hypothesis 4 ._ cross-level interaction). The last interest was whether
the average crime rates of a block group were inﬂuenced by the mean crime rates of
neighboring block groups, after adjusting for the structural covariates (Hypothesis 5 — spatial

lag or diffusion). Table 15 displays the results.

LCVCI 1: TM 2 [30) + 0”,ka + sz*SD + B3j*LU

LeVel 23 1301' = 100 + 101*CD + Y02*IC + Y03*RI + Yo4*LU + 1’05*WY + 110170
131) = 1'10 + Y11*CD + 112*IC + 1’13*RI + Y14*LU
132) = 1’20 + 1’21*CD + Y22*IC + 1’23*RI + 124*LU

133)“ = Y30

Table 15-1. Conditional Model (Cross-Sectional) — Violent Crime

 

 

 

Violent Crime 2003 Violent Crime 2005
Fixed Effect Log Count Se P Log Count Se P
Model for Block Group Means
INTERCEPT, mo -.47 .63 .12 .000 -.62 .54 .12 .000
CD, yo, .19 1.20 .09 .049 .13 1.14 .09 .161
IC, 1’02 .08 1.09 .08 .285 .06 1.06 .08 .439
R1, )203 -.00 1.00 .09 .981 .12 1.12 .09 .183
LU, 1’04 .00 1.00 .01 .917 .00 1.00 .01 .881
Wy, y05 -.16 .85 .08 .064 -.28 .76 .09 .004
Modelfor I’D-Crime Slope
INTERCEPT, 1’10 .12 1.12 .07 .098 .03 1.03 .08 .710
CD, )1], .01 1.01 .05 .879 -.06 .95 .06 .374
1C, )1); -.05 .95 .05 .280 .05 1.05 .05 .347
R1, )113 -.04 .96 .06 .439 .06 1.07 .06 .316
LU, 1’l4 .02 1.02 .01 .021 .01 1.01 .01 .019
Model for SD- Crime Slope
INTERCEPT, yy) .23 1.26 .06 .001 .14 1.15 .07 .039
CD, jig) -.09 .91 .05 .089 -.02 .98 .05 .665
1C, ygz -.09 .91 .05 .046 -.05 .95 .04 .279
R1, W3 .06 1.06 .05 .219 .08 1.08 .04 .072
LU, y34 .00 1.00 .01 .925 -.00 1.00 .01 .740
Model for LU—Crime Slope
INTERCEPT, )130 .01 1.01 .00 .000 .02 1.02 .00 .000

 

 

70 CD = Concentrated Disadvantage, IC = Immigrant Concentration, R1 = Residential Instability,
LU = Land Use, Wy = Logged Spatial Lag of Crime

cf., The slope for land use was ﬁxed assuming the slope does not vary across block groups. It was
mainly to simplify the model in light of the principle of parsimony. Another reason was that cross-
level interaction involving land use was not in the main interests (hypotheses).

126

Table 15-2. Conditional Model (Cross-Sectional) — Property Crime

 

 

 

Property Crime 2003 Property Crime 2005
Fixed Effect Log Count se p ’ Log Count Se P
Model for Block Group Means
INTERCEPT, )100 .34 1.40 .12 .007 .45 1.56 .12 .001
CD, 1’0] .12 1.13 .08 .116 .08 1.08 .07 .267
[C y”: -.04 .96 .07 .577 -.05 .95 .07 .426
R1, 1’03 .30 1.35 .08 .000 .25 1.28 .07 .001
LU, MM -.01 .99 .01 .572 .00 1.00 .01 .810
Wy, 1’05 -.19 .83 .l l .091 -.ll .90 .14 .433
Model for PD-Crime Slope
INTERCEPT, 1’10 -.01 .99 .05 .805 .11 1.12 .06 .048
CD, )1), .03 1.03 .04 .398 .02 1.02 .04 .554
1C, 1’12 -.03 .97 .03 .387 -.01 .99 .04 .726
R1, y,3 -.05 .95 .04 .204 .00 1.00 .05 .978
LU, 31/4 .01 1.01 .01 .183 .00 1.00 .00 .885
Model for SD- Crime Slope
INTERCEPT, 3’30 .21 1.23 .05 .000 .15 1.16 .05 .002
CD, )13/ .01 1.01 .04 .841 -.08 .92 .03 .015
IC, )9; -.08 .92 .03 .011 -.04 .96 .03 .195
R1, )133 .04 1.04 .03 .273 .04 1.04 .03 .151
LU, 1’34 -.00 1.00 .00 .958 -.01 .99 .01 .426
Model for L U— Crime Slope
INTERCEPT, )130 .03 1.03 .00 .000 .02 1.02 .00 .000

 

Regarding the second hypothesis, concentrated disadvantage had a positively

signiﬁcant effect on violent crime 2003, and residential instability signiﬁcantly increased

property crime in both years.

With regard to the third hypothesis, while the effect of social disorder remained

signiﬁcant for all outcome measures, the inﬂuence of physical disorder remained

127

signiﬁcant only for property crime 2005. In other words, the signiﬁcant relationships
between physical disorder and violent crime 2003 and 2005 at the segment level became
null after controlling for the community factors. The stronger multivariate relationship
between social disorder and crime than that between physical disorder and crime was
anticipated, in a way, by the results of zero-order correlations.

Concerning the fourth hypothesis, several interesting results were discovered. First,
the effect of physical disorder on violent crime varied across block groups depending on
land use. Put more speciﬁcally, one unit increase of land use increased the inﬂuence of
physical disorder on violent crime 2003 and 2005 by .02 log and .01 log. This suggested
that physical disorder has a stronger impact on violent crime in the neighborhoods with
more crime-attracting commercial properties. Second, the relationships between social
disorder 2003 and violent crime 2003 and social disorder 2003 and property crime 2003
were negatively inﬂuenced by immigrant concentration. Put in more detail, one unit (one
standard deviation) increase of immigrant concentration reduced the impact of social
disorder 2003 on violent crime 2003 and property crime 2003 by .09 log and .08 log,
respectively. In other words, social disorder appeared to have a weaker effect on violent
crime 2003 and property crime 2003 in the neighborhoods with more heterogeneous racial
distribution. Or, social disorder seemed to be more problematic in homogenous
neighborhoods. Third, the slope for social disorder 2005 and property crime 2005 varied
signiﬁcantly depending on the level of concentrated disadvantage. Put more speciﬁcally,
one unit increase of concentrated disadvantage lowered the inﬂuence of social disorder 2005

on property crime 2005 by .08 log, which suggested that social disorder has a weaker impact

128

on property crime 2005 in more disadvantaged neighborhoods. Or, social disorder appeared
to be more problematic in advantaged communities.

As to the ﬁfth hypothesis, the level of crime in neighboring block groups (i.e.,
logged spatial lag of crime rates) did not have a positively signiﬁcant effect on any outcome
measure. That is, the ﬁfth hypothesis was not supported at all. This was anticipated by the
extremely weak result of the autocorrelation measure, or Global Moran’s I.

In sum, the cross-sectional analysis produced several notable results. (1)
Concentrated disadvantage appeared to have the most salient effect on violent crime,“ but
property crime seemed to be most significantly inﬂuenced by residential instability. (2)
While the signiﬁcant relationship between physical disorder and crime became mostly non-
signiﬁcant (except for property crime 2005) after controlling for the structural covariates,
the impact of social disorder remained signiﬁcant for all types of outcome measures. (3)
Regarding the cross-level interaction effect, the impact of physical disorder on violent crime
appeared to become stronger as the percentage of crime—attracting businesses increased.
Meanwhile, the effect of social disorder on violent crime 2003 and property crime 2003 got
weaker as the level of immigrant concentration rose. Lastly, the inﬂuence of social disorder
on property crime 2005 decreased as the degree of concentrated disadvantage became

stronger. (4) As expected, no signiﬁcant spatial diffusion was detected.

Longitudinal Analysis - Random Coefficient Model
As the theory suggests, the focal point of community research must be in the
longitudinal relationship among factors. Thus, this study attempted to examine the various

types of hypotheses concerning the link between disorder change and crime change during

 

7' Although the effect of concentrated disadvantage on violent crime 2005 was not statistically
signiﬁcant, it had the strongest impact on the outcome variable relatively (see Table 13-1).

129

the study periods, 2003-2005. First, a longitudinal random coefﬁcient model was ﬁtted to
test whether the change in disorder had a signiﬁcant effect on change in serious crime at the
segment level (Hypothesis 1). Several notes deserve attention: (1) the changes in disorders
were measured by subtracting the levels of them in 2003 from those in 2005; (2) since this
study was utilizing the mixed Poisson model, which requires a non-negative outcome
variable, the author controlled for the 2003 crime instead of creating the change variable by
subtracting 2003 crime from 2005 crime; (3) as noted already (see p. 108), land use 2005
was controlled for in the model, instead of the land use change, because only 11 segments
experienced a decrease or increase in commercial land use. The results are presented in

Table 16.

Level 1: 17,,- = log(VC05) = 401+ ,8),*(PD05-PD03) + ,82j*(SD05-SD03) +
p3,*LU05 + 11,,*VC03 7’

Level 22 1801' = 1’00 + “of

181) : 1’10
1621 = 1’20
1831: 1’30
1841: 1’40

Where:
,ng = the average log of crime rates of a block group j when there is no change of
PD, no change of SD, no LU, or no previous crime

,Blj = the average slope of PD change in a block group j
,ng- = the average slope of SD change in a block group j
,831- = the average slope of LU 2005 in a block groupj

The average lags of block group means of violent crime 2005 and property crime
2005 were .59 and 1.66, respectively, when there was no change in disorder, no land use,

and no previous crime. Unlike the cross-sectional counterpart, physical disorder change was

 

‘72 The same type of model was applied to property crime 2005-

130

not signiﬁcantly related to the two-year lags of violent and property crimes.73 Social
disorder change did not have a signiﬁcant effect on the lag of property crime either. The
only signiﬁcant relationship was found between social disorder change and violent crime lag.
In sum, it appeared that, except for the impact of social disorder change on the lag of violent

crime, the ﬁrst longitudinal hypothesis was not supported.

Table 16. Random Coefﬁcient Model (longitudinal)

 

 

 

 

Violent Crime 2005 Property Crime 2005

Fixed Effect Log Count Se p Log Count Se P
Average Lag of Block Group ,

Mean, 1’00 -.53 .59 .09 .000 .51 1.66 .08 .000
PD Change Slope, y”) -.02 .98 .04 .621 -.03 .97 .03 .341
SD Change Slope, ygo .08 1.08 .03 .008 .03 1.03 .02 .180
LU-Crime Slope, )230 .01 1.01 .00 .001 .01 1.01 .00 .000
Previous Crime (2003) .19 1.21 .02 .000 .03 1.03 .01 .010

 

Longitudinal Analysis — Conditional Model

The ﬁnal model controlled for all exogenous factors at both levels. It must be noted
that the slopes for land use and previous crime were ﬁxed, assuming the slopes do not vary
across block groups for the same reasons as those in the cross-sectional model. As was done
in the cross-sectional analysis, four main interests were tested: the ﬁrst interest was in what
community factors affected the block group means of crime lags, controlling for the other
exogenous variables and previous crime (Hypothesis 2). Another interest was whether the

signiﬁcant relationship between social disorder change and crime change, proved by the

 

73 The two-year lags of crimes are sometimes expressed as crime changes in the text just to ﬁt the
interpretation of results into the study hypotheses (cf., hereinafter, lag means two-year lag).

131

above random coefﬁcient model, remained signiﬁcant after controlling for the level 2
structural. covariates (Hypothesis 3 — multivariate signiﬁcance). This particular question
was quite limited because the other relationships between disorder change and crime change
at level 1 were not signiﬁcant. Meanwhile, the third, and one of the most critical interests,
was in whether the relationships between disorder change and crime change at the ﬁrst level
were signiﬁcantly moderated by structural factors (Hypothesis 4 — cross-level interaction).
The last interest was whether the average crime rates change in a block group was
inﬂuenced by the mean crime rates change in neighboring block groups, after adjusting for

the structural covariates (Hypothesis 5 — spatial lag or diffusion). Table 17 shows the results.

Level 1: 17,, = log(VC05) = poj+ p,,*(PD05-PD03) + .52/*(SD05-SD03) +
,B3j-*LU05 + p,,*vc03

LCVC] 2: [301' 2 YOU + Y0]*CD + 702*IC + Y03*RI + 'Y()4*LU +
Y05*W(VCOS-VC03) + uO, ’4
Bu = 710 + 711*CD + 112*1C + Y13*RI + Y14*L1U
132) = Y20 + ’er’kCD + Y22*IC + 723*RI + 724*LU

1331 = 1’30
134j = 1’40
With regard to the second hypothesis (i.e., ﬁrst interest in the conditional model),
immigrant concentration, residential instability, and land use had a signiﬁcant effect on the
lag of violent crime at the block group level. The result was quite different from that of the
cross-sectional counterpart. On the other hand, the lag of property crime was signiﬁcantly
related to residential instability, which was consistent with the result of the cross-sectional

model.

 

7” Unlike the cross-sectional conditional model, the author did not transform the spatial lag of
crime change, or W(VC05-VC03), by taking the natural log, for the lag had a number of negative
values.

132

Concerning the third hypothesis, no positively signiﬁcant effect was found after
controlling for the structural covariates.75 In other words, the signiﬁcant effect of social

disorder change on violent crime change failed to remain signiﬁcant. In sum, the third

hypothesis did not receive empirical support.

Table 17. Conditional Model (Longitudinal)

 

 

 

Violent Crime 2005 Property Crime 2005

Fixed Effect Log Count Se p Log Count Se P
Model for Block Group Means

INT ERCEPT , 1’00 -.78 .46 .11 .000 .33 1.39 .09 .001

CD, ya, .03 1.03 .08 .688 .10 1.10 .07 .161

IC, )103 .13 1.14 .07 .048 -.04 .96 .06 .520

R1, )103 .17 1.18 .08 .030 .24 1.27 .07 .001

LU, ya.) .02 1.02 .01 .029 .01 1.01 .01 .406

W(Crime Change), )105 .17 1.19 .12 .168 .09 1.10 .06 .142
Model for PD Change Slope

INTERCEPT, mg --.11 .90 .07 .126 -.13 .88 .05 .012

CD, y, I -.12 .89 .06 .043 .03 1.03 .04 .466

IC, )1 I 2 .06 1.07 .05 .153 .06 1.06 .03 .065

R1, )113 .08 1.09 .06 .154 .02 1.02 .04 .646

LU, )114 .01 1.01 .00 .075 .00 1.00 .00 .232
Model for SD Change Slope

INTERCEPT, )720 .01 1.01 .05 .850 .02 1.02 .04 .534

CD, )12, -.04 .96 .04 .351 -.01 .99 .03 .675

IC, ygg -.07 .93 .03 .045 -.02 .98 .03 .350

R1, y23 .06 1.06 .04 .116 -.06 .94 .03 .031

LU, y24 .01 1.01 .01 .051 .00 1.00 .00 .435
Modelfor LU Slope

INTERCEPT, 3230 .01 l .01 .00 .001 .01 1.01 .00 .005
Model for Previous Crime Slope

INTERCEPT, y40 .23 1.26 .03 .000 .04 1.04 .01 .001

 

 

7’ Although the effect of physical disorder change on property crime change was signiﬁcant, it

was negatively so. And thus, the author lest it out of the discussion.

133

As to the fourth hypothesis, three cross-level interaction effects were discovered.
First, the effect of physical disorder change on violent crime change was moderated by
concentrated disadvantage. Put in detail, one unit increase of concentrated disadvantage
reduced the inﬂuence of physical disorder change on violent crime change by .12 log. The
result implied that physical disorder has a stronger impact on violent crime in less
disadvantaged neighborhoods. Second, the relationship between social disorder change and
violent crime change was negatively inﬂuenced by immigrant concentration. Put more
speciﬁcally, one unit increase of immigrant concentration lessened the impact of social
disorder change on violent crime change by .07 log. It suggested that social disorder has a
stronger effect on violent crime in less heterogeneous neighborhoods. Third, the slope for
social disorder change and property crime change varied signiﬁcantly across block groups,
depending on the level of residential instability. Put in more detail, one unit increase of
residential instability lowered the inﬂuence of social disorder change on property crime
change by .06 log, which implied that social disorder has a stronger impact on property
crime in more stable neighborhoods.

Regarding the ﬁfth hypothesis, the level of crime change in neighboring block
groups (i.e., spatial lag of crime rates) did not have a positively signiﬁcant effect on any type
of crime change. Like the result of the cross-sectional model, the hypothesis on spatial lag
was not supported.

In sum, the longitudinal analysis generated a few noteworthy results. (1) While
violent crime change appeared to be signiﬁcantly associated with immigrant concentration,
residential instability, and land use, property crime change seemed to be signiﬁcantly related

only to residential instability. (2) No relationship between disorder change and crime

134

change remained signiﬁcant after controlling for the community characteristics. (3)
Concerning the cross-level interaction effect, the impact of physical disorder change on
violent crime change appeared to become weaker as the level of concentrated disadvantage
rose. Meanwhile, the effect of social disorder change on violent crime change got weaker in
more heterogeneous neighborhoods. Lastly, the inﬂuence of social disorder change on
property crime change became weaker as the degree of residential mobility increased. (4)

Like the cross-sectional model, no signiﬁcant spatial lag effect was found.

Summary

A variety of analyses were conducted to examine what the data and models
demonstrate about the relationship between disorder and crime in Lansing communities.
Some results were as expected but others were not. Several notable ﬁndings are
summarized. (1) No consistent pattern of change in disorder and crime was found at either
the high physical disorder segments (n=3 8) or the low physical disorder segments (n=38).
(2) Non-signiﬁcant Moran’s 1 indicated that crime was not spatially autocorrelated. Further
multivariate tests for the spatial diffusion process (hypothesis 5) conﬁrmed the result. (3)
While the ﬁrst hypothesis (i.e., signiﬁcant relationship between disorder and crime at
segment level) was generally supported by the cross-sectional random coefﬁcient model, the
longitudinal random coefﬁcient model produced opposing results. (4) Regarding the
second hypothesis (i.e., signiﬁcant relationship between community characteristics and
crime at block group level), the cross-sectional model and the longitudinal model turned out
different ﬁndings for violent crime: while concentrated disadvantage appeared to exert the
strongest, effect on violent crime in the cross-sectional model, all of the structural covariates

except concentrated disadvantage (i.e., immigrant concentration, residential instability, and

135

land use) signiﬁcantly increased violent crime in the longitudinal model. On the other hand,
both models produced the same results for property crime, namely, that it was signiﬁcantly
inﬂuenced only by residential instability. (5) Concerning the third hypothesis (i.e.,
multivariate signiﬁcance of the impact of disorder on crime), the impact of social disorder
on all outcome measures remained signiﬁcant in the cross-sectional model. However, no
relationship between disorder and crime stayed signiﬁcant in the longitudinal model. (6)
Differential types of cross-level interaction mechanisms were found for the cross-sectional
and longitudinal models. For example, while the impact of physical disorder on violent
crime was positively moderated by land use in the cross-section model, it was negatively
moderated by concentrated disadvantage in the longitudinal model. Another example was
that while the relationship between social disorder and property crime varied signiﬁcantly
across block groups depending on the level of concentrated disadvantage in the cross-
sectional model, this varied signiﬁcantly, relying on the degree of residential instability in
the longitudinal model. However, the relationship between social disorder and violent crime
was similarly moderated by immigrant concentration in both the cross-sectional and

longitudinal models.

136

CHAPTER VII: CONCLUSION

Researchers have had high interest in many key theoretical disputes related to crime
and disorder. However, it is surprising that, despite such high interest, only slightly more
than 10 studies have been conducted to examine the theoretical link between disorder and
crime since the publication of Wilson and Kelling’s work in 1982. Thus, before taking any
side, either pro or con, the author believes more studies need to be performed, which was the
main motivation of this study. Going one step further, this dissertation attempted to clarify
sensitive, mostly methodological, issues that would help researchers to perform more
systematic research. This was necessary because the past studies had adopted inconsistent
procedures for measurement, design, and model speciﬁcation, which, in turn, resulted in
various, inconsistent results above and beyond such substantive factors as sample
characteristics. For these reasons, the author wanted this dissertation to serve as a valuable
addition to the broken windows literature.

As a conclusion, this chapter deals with four topics. First, the results of diverse
analyses are summarized and their implications for theory, research, and policy are
discussed. Next, the limitations of the models and analyses are presented to ensure that the
readers correctly understand the results. Third, the author explains what contributions this
study made to the broken windows ﬁeld regarding theory, research, and policy. Finally, the

author presents a humble opinion for the future directions of the broken windows research.

Summary and Implications for Theory, Research, and Policy
As a branch of the ecological perspective, the broken windows thesis shares the
assumption that crime is inﬂuenced by the structural covariates of neighborhoods, such as

concentrated disadvantage, residential instability, immigrant concentration, and land use. In

137

the broken windows model, however, the relationship between community factors and crime
is mediated by social and physical disorder. This conceptual model highlights the
importance of dealing with disorderly conditions as an effective alternative to confronting
serious crime.76

A hierarchical multivariate approach was taken to test the effectiveness of the
alternative, with special attention to the contextual effect. The hierarchical mixed Poisson
model discovered several notable ﬁndings. The author describes each ﬁnding and discusses
its implications for theory, research, and policy. First, the results for violent crime were
different from those for property crime. This ﬁnding appeared to be supportive of Clarke’s
(1997) notion that more crime-speciﬁc approaches need to be taken in research and crime
prevention policy. For example, while concentrated disadvantage had the strongest effect
on violent crime, property crime was most signiﬁcantly inﬂuenced by residential instability
(in the cross-sectional model). The ﬁndings suggested that violent crime could be
effectively controlled by improving socioeconomic conditions, but property crime would be
better dealt with by reducing residential instability. With more evidence accumulated,77 this
would further stimulate theoretical elaborations of the broken windows theory just as the

situational theories postulate crime-speciﬁc propositions.

 

7" It must be noted, however, that the study model in this dissertation does not explicitly test the
mediating role of disorder in the link between neighborhood characteristics and crime rates. Instead,
this study attempted to test the multivariate direct effect of disorder on crime, after controlling for the
structural covariates, and cross-level interaction effect of disorder and structural covariates on crime.
Nevertheless, the author would say the former test was able to serve as an alternative to the
mediating model in that the multivariate signiﬁcance implied that serious crime can be effectively
handled by dealing with disorder.

77 To give another example, Weisburd and his colleagues (1993) found that calls for service for
such minor offenses as public morals and drunks are more strongly correlated with certain serious
crimes such as robberies than with other types of serious offenses in crime hot sopts (Sousa &
Kelling, 2006).

Second, and related to the above discussion, property crime appeared to be most
signiﬁcantly related to residential instability in both the cross-sectional and longitudinal
models. The consistently signiﬁcant inﬂuence of residential instability on property crime
implied that the lack of social bonds from residential anonymity may be the most critical
factor leading to crime. It further seemed to be related to the issue of contextuality. That is,
while a measure of economic devastation such as concentrated disadvantage is generally
regarded as the ﬁrst candidate in metropolitan urban areas (e. g., Chicago neighborhoods), a
neighborhood bond measure such as residential instability might well deserve more attention
in mid-sized cities where the residential bond and cohesion is still widely believed to be an
indispensable element in living a high-quality life. The author believes that something
unfamiliar to the residents in a neighborhood or inconsistent across near areas (i.e., highly
variable across nearby neighborhoods) has more psychological and behavioral impact on
them than do things well-known to them or consistent throughout neighboring areas.
Accordingly, residential mobility, which metropolitan urban residents have already become
used to, would not have much inﬂuence on their mindset and behavior. Neither does
socioeconomic devastation in mid-sized city neighborhoods have a big impact on residents’
attitude and conduct, assuming the economic conditions are relatively similar across near
areas. Although not explicitly modeled, one could further speculate that, while the
neighborhoods’ capacity of informal social control in metropolitan urban neighborhoods is
mostly dependent on the degree of concentrated disadvantage, in mid-sized cities with urban
and suburban areas, it is mostly inﬂuenced by residential instability, which in turn inﬂuences

the crime rates.

139

Third, the results of the test of the cross-sectional model and the longitudinal
counterpart were not consistent. It was particularly notable that, while the effect of social
disorder on all types of outcome measures remained signiﬁcant in the cross-sectional model,
no signiﬁcant multivariate effect of disorder change on crime change was discovered in the
longitudinal model. Since a longitudinal model is generally regarded as a more explanatory
model, better reﬂecting the real world than its cross-sectional counterpart does, a plausible
conclusion would be that disorder does not have a causal impact on crime.78 Such a
conclusion would be somewhat hasty, however, since the longitudinal model itself has
weaknesses as was discussed in Chapter IV”. A more convincing conclusion would have
been possible from a more systematic longitudinal model such as the growth curve model.
Further, considering the reciprocal relationship suggested by the contemporary broken
windows theory, it would also be possible to argue that crime occurred ﬁrst and then
inﬂuenced social disorder. It must be noted that the cross-sectional model could suggest that
some other things that the longitudinal model does not show might be working in the real
world. Therefore, the author decided to discuss the results of both the cross-sectional and
longitudinal analyses with an equal weight. It is simply recommended that more effort be
exerted to systematically collect and analyze data in order to reach a more persuasive
conclusion relying on the longitudinal model.

Fourth, social disorder had a stronger impact on both types of crimes than physical
disorder did, regardless of temporal types of models. In the cross-sectional model, put in
more detail, the impact of social disorder on crime remained signiﬁcant after controlling for

the structural covariates, but that of physical disorder stayed signiﬁcant only for property

 

78 In other words, the lack ofcollective guardianship is not causing serious crime.
7” See pages 73-74.

140

crime 2005 (conditional model). Meanwhile, the longitudinal model found that only social
disorder change was signiﬁcantly related to violent crime change, with the other three
models having null results (random coefﬁcient model). Such results were somewhat
anticipated by the same data source issue and by a commonly accepted sense in the ﬁeld as
well. It had a policy implication that focusing solely on physical disorder might have a
limited utility in reducing crime, and thus, it might be more effective to deal with social
disorder as well as physical disorder.

Fifth, a variety of cross-level interaction effects were discovered, each of which
contained an important policy implication. ( 1) Physical disorder and land use had a positive
interaction effect on violent crime in both years (cross-sectional model), which indicated
that the link between physical disorder and violent crime may be less valid in neighborhoods
with fewer crime-attracting business properties, or more valid in neighborhoods with more
crime—attracting businesses. It further implied that dealing with physical disorder would be
more effective in handling violent crime in neighborhoods with many commercial properties,
and less useful in neighborhoods with a few businesses.

(2) The effects of social disorder 2003 on violent crime 2003 and property crime
2003, and those of social disorder change on violent crime change, were negatively
moderated by the level of immigrant concentration (cross-sectional and longitudinal model).
In other words, the effects of social disorder 2003 on violent crime 2003 and property crime
2003 became stronger as the level of racial heterogeneity decreased, which implied that a
policy to reduce crime by dealing with social disorder might be less effective in

heterogeneous neighborhoods than in homogenous ones.

141

(3) The relationship between social disorder 2005 and property crime 2005 became
less valid as the level of socioeconomic disadvantage increased (cross-sectional model).
That is, the inﬂuence of social disorder on property crime was weaker in disadvantaged
neighborhoods than in prosperous counterparts. And thus, it was suggested that an order-
maintenance or control-oriented policy in poor neighborhoods would not be as successﬁil as
is suggested by the theory.

(4) The inﬂuence of physical disorder change on violent crime change was
negatively moderated by concentrated disadvantage (longitudinal model), which showed
that the link between physical disorder and violent crime is less valid in prosperous
neighborhoods than in disadvantaged counterparts. Thus, strategies focusing only on
reducing physical disorder might have little effect on dropping the level of violent crime in
poor neighborhoods.

(5) Social disorder change and residential instability had a negative interaction effect
on property crime change (longitudinal model). It indicated that the impact of social
disorder change on property crime change became weaker as the level of residential
instability increased, which further implied that dealing with social disorder might be less
effective in reducing property crime in non-stable communities than in stable communities.

Overall, the cross—level interaction effects made a common suggestion that the link
between disorder and crime could be different depending on various community contexts,
and thus more context-speciﬁc programs need to be developed. More speciﬁcally, the
negative interaction mechanisms seemed to suggest that improving structural conditions
may be a more valid approach to reducing crime in poor, heterogeneous, or non-stable

neighborhoods, but restoring order would be more effective in afﬂuent, homogeneous, or

stable communities. On the other hand, the positive interaction dynamics involving physical
disorder, land use, and violent crime appeared to imply that, in high-density neighborhoods
with lots of businesses, physical disorder tends to develop more easily into serious violent
crime than in residential areas with few, if any, commercial properties. Thus, a proactive
approach to improve the physical environment would be a good strategy for solving serious
crime problems. However, the results do not necessarily support for one approach to the
exclusion of the other. Approaches for both structure and disorder would have to be taken
in order to obtain better results.

The broken windows theory suggests that the link between disorder and serious
crime is mediated by neighborhood processes such as fear of crime and/or self-regulation
capacity. In that case, one plausible question related to the above negative cross-level
interaction dynamics is: why is the relationship between disorder and crime weaker in
structurally disadvantaged neighborhoods than in advantaged counterparts? There could be
two answers. One answer could be found in the theoretical model including the self-
regulation capacity: it may be because the self-regulation capacity is stronger in
disadvantaged neighborhoods than in advantaged ones. hr other words, a disorder problem
does not easily develop into a crime problem in disadvantaged neighborhoods due to the
strong informal social control. In contrast, disorder is strongly connected to crime in
prosperous neighborhoods because of the weak self-regulation capacity. However, this is
less likely to be the case, although the author cannot say with certainty.

The other and better answer would be that the weak link between disorder and crime
in disadvantaged neighborhoods is because the crime problem is already prevalent in the

disadvantaged neighborhoods, and thus the crime rates of such areas are not much

143

inﬂuenced by the level of disorder. In other words, it could be a statistical issue that the
slight variation of the crime rates in disadvantaged neighborhoods tends to produce little
association between disorder and crime regardless of the level of disorder.80 In contrast,
assuming that the crime problem is generally rare in afﬂuent neighborhoods, even one
incident could have a big impact on the crime rates of those neighborhoods, and thus the
crime rates could be highly variable probably depending on the level of disorder, which is
likely to produce a high correlation between disorder and crime. Also, as suggested above,“
it is possible that disorder causes little fear in those disadvantaged neighborhoods simply
because disorder (and possibly crime) is so prevalent that people have already become
accustomed to such situations. But, even a low level of disorder could make people in
afﬂuent neighborhoods feel a lot of fear, which in turn lowers the self-control capacity and
increases crime rates. At any rate, this explanation suggests the same policy implication as
above: improving structural conditions would be a better approach to deal with serious
crime problems in disadvantaged neighborhoods. In contrast, solving disorder is important
in reducing crime in advantaged counterparts. As a matter of course, taking both
approaches together would have a synergy effect. In addition, as the theory suggests,
collective efforts among the residents, local governments, businesses, and police to improve
the structural conditions and deal with disorder and crime problems would eventually

strengthen the neighborhood informal social control, which could serve as a long-lasting

strong factor in curbing the serious crime problem.

 

8” In fact, it appears that the initial status of crime is one of the strongest and most consistent
predictor of later crime (e.g., Taylor, 2001). The longitudinal (random coefﬁcient) model in this
study showed that the previous crime in 2003 is the best predictor of the crime in 2005 (see Table
16).

3' See the ﬁrst paragraph, p. 133.

144

Sixth, no spatial effect was detected, which showed that the level of crime in a
neighborhood (i.e., block group) in Lansing is not dependent on that of its neighboring areas.
Although not explicitly modeled, no spatial diffusion (or displacement) effect was expected
either. However, the results were not that credible, since the crime rates of block groups
were measured by the averaged sum of crime rates of street segments within each block
group. As shown in the previous chapter, for example, 18 block groups had only one study
segment and 11 block groups contained two segments,82 which made the reliability and
validity of crime rates in many neighborhoods doubtful.

In sum, although one of the main hypotheses, the multivariate signiﬁcant effect of
disorder on crime was only half-supported in this study, various models of this research
produced several noteworthy ﬁndings, which further suggested viable directions for future
research and policy and theoretical elaboration. Put brieﬂy, crime-speciﬁc and context-
speciﬁc understanding is required for theory, research, and policy. As Thacher (2004) noted,
scholars of the broken windows research appear to have been “preoccupied with the search
for strong causal relationships between disorder and criminal activity” (Sousa & Kelling,
2006, p. 88). Although it is important, the author would say understanding the complex and
curvilinear real world dynamics under the seemingly simple and linear causal processes
suggested by the broken windows logic is more valuable.

Meanwhile, the suggestions made above should be interpreted with a caveat, for the
results were not consistent across the models, especially depending on temporal types.
Additional discussions on some other limitations need to be made to help readers interpret
the ﬁndings of this dissertation. The next section explains some of the important constraints

of this study.

 

82 See pages 78-79.

145

Limitations

Two representative limitations of this study were brieﬂy introduced already - no
control for psychological process and non—systematic longitudinal analysis. This section
presents a more detailed discussion of them. An additional discussion on spatial analysis is
made to highlight the importance of understanding spatial dynamics. Meanwhile, it must be
noted that there could be other dimensions that undermine the validity of our ﬁndings and

policy implications.

Psychological Process

The models of this dissertation did not control for the psychological process that is
considered to be a mediator linking disorder and crime. One contemporary measure that is
widely accepted to be representative of the psychological process is collective efﬁcacy as
developed by Sampson and his colleagues (1997, 1999). Collective efﬁcacy refers to
underlying mechanisms of social control, including mutual trust and shared expectations and
willingness to intervene in antisocial activities for the common good. Although social
networks (ties) and resources (social capital) are important, they are not sufﬁcient conditions
for social control to confront neighborhood problems (Kubrin & Weitzer, 2003b).
Interestingly, some forms of social networks (e. g., gangs and drug dealers) even impede the
efforts to ﬁght crime (Pattillo-McCoy, 1998). Hence, Sampson (2002) argued that
collective efﬁcacy, highlighting informal norms and expectations about the exercise of
social control without regard to the level of social bonds, must be the focal point of
neighborhood processes, particularly in modern urban areas in which the romantic deﬁnition
of neighborhood as a primary group based on intimate relations and social bonds is not

relevant at all. Accordingly, the control for collective efﬁcacy could have provided a useful

146

opportunity to test whether the effect of disorder on crime is still signiﬁcant or not. If it had

remained signiﬁcant, the policy implication would have been given more credit.

Temporal Dynamics

As mentioned in Chapter IV, although this study tried to control for the temporal
order by following the approach of previous studies (i.e., controlling for the previous year’s
crime rates as was done in Markowitz et al., 2001 and Sampson & Raudenbush, 1999), the
approach had certain limitations that did not fully incorporate the theoretical notion. The
ecological perspective has a clear implication for the temporal change of neighborhood
characteristics and mechanisms. This theme dates back to the thesis of Shaw and McKay
(1942), who argued that the changing nature of crime distribution in urban areas is the
product of larger socioeconomic processes that inﬂuence the formation and change of
neighborhood characteristics and processes (Sampson et al., 2002; Kubrin & Weitzer,
2003b). Applying the idea of criminal careers (Blumstein & Cohen, 1987; Blumstein,
Barnett, & F anington, 1987) to ecological studies, Albert Reiss, Jr., and others worked
intensively on writing about the longitudinal change in crime rates responding to the long-
terrn pattern (trajectory) and short-term change (transition) in community characteristics
(Bottoms & Wiles, 1986; Reiss, 1986; Schuerrnan & Kobrin, 1986). Their efforts directed
more attention to the theoretical viability of community crime careers (Williams &
McShane, 2003). To the author’s knowledge, the best statistical model for describing and
testing the idea of community crime careers is the growth curve model, which requires

sequential data with at least three time points.83 Unfortunately, however, only two waves of

 

83 It applies to the dependent variable. In other words, the independent variables need not to be
sequential data. As a matter of course, the more time points the data have, the better results the
model will produce.

147

data were available to this study and thus the technique could not be applied. Assuming it
takes time, say more than two years, for the change in disorder to affect the change in crime,

applying the growth curve model might have discovered different results.

Spatial Dynamics

As mentioned above, the spatial analysis of this study was quite limited because the
measure of the dependent variable, or crime rates of block groups, was simply the averaged
sum of crime rates of street segments within each block group, which might well cast doubt
on the reliability and validity of outcome measures. More reliable and valid measures of
crime and systematic models are required for future research to get a better understanding of
the spatial dynamics of crime, disorder, neighborhood process, and structural characteristics.
As explained in the previous chapter, the theoretical factors do not exist independent of
those of other neighborhoods, however deﬁned. Thus, the assumption of spatial
independence must be corrected in theory and research (Sampson, 2002). For instance,
residents who live right across the street and thus feel like the most intimate neighbors to
each other could be divided into different census tracts. Moreover, many interpersonal
crimes are rooted in social interactions that tend to cross neighborhood boundaries, which
causes spatial diffusion (e.g., Rosenfeld et al., 1999). Therefore, spatial models that reﬂect
and correct the inﬂuence of surrounding neighborhoods (i.e., spatial dependence) must be
counted on (Kubrin & Weitzer, 2003b).84 Such a mechanism of spatial interdependence has
further implications for understanding residential stratiﬁcation. That is to say, two

neighborhoods that have similar characteristics and processes could have different levels of

 

8" Baller et al. (2001) provide the most complete analysis of spatial dependence to date by
comparing ﬁndings from a spatial lag and spatial error model in their analysis of structural
correlates of county homicide rates from 1960 to 1990 (Kubrin & Weitzer, 2003b: 395)

148

crime rates depending on their surrounding communities, which fosters the need for
contextual analysis using a method such as HLM (Sampson et al., 2002). On the other hand,
it is notable that the recent development of methods for the spatial econometrics of panel
data is very promising (Messner& Anselin, 2004) in that it accounts for both spatial and

temporal dynamics.

Contributions to the Broken Windows Field

Despite these limitations, the study made several contributions to the broken
windows research. First, the author attempted to reﬁne and distinguish the two key concepts
of disorder and crime. Since the concepts were very close to each other, it was necessary
that researchers use clearly distinguished measures for each of them. Otherwise, there was a
high chance of tautological ﬁndings. Further, as a methodological buffer, the author
suggested that an exploratory factor analysis should be conducted to statistically tell whether
the measures of disorder and crime represent their corresponding concepts. Finally, another
suggestion was made that different data sources be used to measure disorder and crime in a
clearly distinguished manner.

Second, besides the measures for disorder and crime, an enthusiastic attempt was
made to suggest a methodological standard for future quantitative research. As mentioned
before, the lack of methodological standard appeared to have inﬂuenced the study results,
which, in turn, heated the controversies on the authenticity of the broken windows theory
and policy. Several examples of the standard included how to deﬁne and measure
neighborhood, the necessity of controlling neighborhood process as a mediating factor, the
necessity of controlling neighborhood structure as a confounding factor, the necessity of a

systematic longitudinal design, and the necessity of examining spatial dynamics. It must be

149

noted that they were mostly identiﬁed via deep theoretical considerations. Therefore, even
though meeting all the criteria seems to be extremely difﬁcult, it is expected that even a
minor consensus would facilitate more theory-driven research and policy.

Third, the author tried to understand the broken windows theory within the
ecological framework. By doing so, the interconnectivity of the broken windows theory
with the other ecological theories, such as new social disorganization, defensible space,
routine activity, and rational choice, were clariﬁed, as was its uniqueness among the diverse
explanations. Figure 2 in Chapter 2 represents the outcome of the author’s effort, which is
expected to help readers clearly understand the theoretical model of this study as well as the
other studies on neighborhood structure and process.

Fourth, the efforts to understand contextual dynamics through multilevel models
were rare, if ever, in the broken windows ﬁeld. To the best knowledge of the author, only
one study (i.e., Rountree et al., 1994) modeled cross—level interaction relationships among
disorder, crime, and community characteristics. Further, the study setting of Lansing, a mid-
sized city with urban and suburban nelghborhoods, provided a unique chance to examine
how differential settings produce differential outcomes. Considering the importance of
contextual understanding in theory development and policy implementation, the hierarchical
models with various cross-level interaction results and the study setting of this study were
highly valuable. Context-speciﬁc understanding would be much more effective and

efﬁcient in crime prevention than a rather general approach.

150

Future Directions of the Broken Windows Research and Policy

More Emphasis on the Role of Formal Social Control

It is true that Shaw and McKay’s social disorganization theory and contemporary
variations of it highlight the central role of informal social control in mediating the effects of
cormnunity characteristics on crime, largely ignoring the possible effectiveness of formal
control such as police law enforcement. Such a theoretical formation is understandable in
that informal control is an expression of the community-based self-regulation and
prevention mechanism, whereas formal control is mostly exercised after the event by
institutions placed outside the community. However, formal control may add valuable
insights to neighborhood research including the broken windows ﬁeld by inﬂuencing crime
rates both directly and indirectly (Kubrin & Weitzer, 2003b, p. 381-382).

Direct effects of formal control on crime and victimization rates are well-
documented. For example, Smith (1986) found that police activity varies signiﬁcantly
across neighborhoods. Police tend to underpolice disadvantaged hi gh-crime neighborhoods
because they think the residents in those communities deserve victimization in light of their
lifestyles (Klinger, 1997). High levels of police misconduct and abuse in those areas (F agan
& Davies, 2000; Mastrofski, Reisig, & McCluskey, 2002) further lowers residents’
satisfaction with the police, which in turn may hamper residential efforts to reduce crime.
On the other hand, vigorous law enforcement seems to be effective in reducing crime by
increasing the perceptive probability of arrest for illegal activities (Sampson, 1986, p. 281;
Sampson & Cohen, 1988; Stark, 1987). Velez (2001) further discovered that the effects are
more salient in poor neighborhoods than in afﬂuent communities (Kubrin & Weitzer, 2003b,

p. 382-383).

151

Formal control may have indirect effects on crime by inﬂuencing or interacting with
residents’ informal control practices. It seems to be obvious that proactive police control
strengthens the willingness and capacity of residents to ﬁght crime. Community policing is
based on the theoretical premise that informal norms and rules set by residents can be
effectively realized and maintained by cooperative efforts between residents and police.
Although there is little empirical evidence on the relationship between formal and informal
social control, community policing literature illustrates the possible linkage (Greene, 1999;
Skogan, 1990; Skogan & Hartnett, 1997): “Community policing thus illustrates one way in
which formal and informal social control can reinforce each other, helping to reduce crime”
(Kubrin & Weitzer, 2003b, p. 383). In contrast, inadequate policing in highly disadvantaged
neighborhoods would undermine the informal control capacity of residents and could further
allow offenders to set their own norms and rules in the communities. Kubrin and Weitzer’s
(2003 a) study 011 retaliatory homicide demonstrates that residents in poor neighborhoods
have little expectation of proactive police intervention, which fosters retributive justice by
some residents.

Another mechanism of indirect effects of formal control is suggested by
incarceration research. That is, high incarceration rates could undermine neighborhood
cohesion and social control by causing a variety of unintended consequences such as “(1)
disrupting family cohesion and ﬁnancial resources, (2) reducing the supply of marriageable
partners, (3) depleting labor markets, (4) decreasing the number of adults available to
supervise neighborhood youth and intervene in neighborhood problems, and (5)

compounding neighborhood problems after these incarcerated offenders are released into the

152

community” (see Clear et al., 2003; Rose & Clear, 1998, in Kubrin & Weitzer, 2003b, p.
384).

The author is particularly interested in the interaction dynamics of formal and
informal controls, which have been little studied to date. Although several studies on
fomral control in the ﬁeld, or broken windows policing, have been conducted, none of
them consider the possible link between formal and informal controls. Thus, the author
suggests that more interest be paid to the role of formal control in strengthening the self-
regulation mechanism among residents. This way, the zero tolerance approach to
disorder would fade away and the critics of the broken windows idea and policing would

begin to appreciate the true underlying idea of the theory and policy.

Comparative Study with another Country

The second purpose of this dissertation was to examine whether the relationship
between disorder and crime was dependent on neighborhood characteristics, which was
expected to facilitate contextual understanding. The results suggested that a context-
speciﬁc policy would be more effective in reducing crime than a general one. It further
encouraged theoretical elaborations to the direction that the relationships among structure,
disorder, process, and crime might depend on various neighborhood contexts. For
instance, while concentrated disadvantage appeared to be a leading factor in crime in
metropolitan urban areas due to its rareness and inconsistent distribution across
neighborhoods, residential instability seemed to be the best indicator of crime (especially
property crime) for similar reasons. Also, the theoretical link between disorder and crime
appeared to be weaker in structurally disadvantaged neighborhoods than in advantaged

counterparts, the reasons for which, the author speculated, was because crime and

possibly disorder was so prevalent in disadvantaged areas that the level of disorder did
not have a statistical impact on the link between disorder and crime, and relatedly,
disorder did not increase fear of crime among residents.

In this vein, it would be valuable to apply such contextual understanding to
another setting, or another country. Compared to the US, for example, South Korea
tends to be more tolerant toward disorderly behaviors and conditions. One plausible
reason is that, notwithstanding the rapid economic growth of recent years, Koreans are
still oriented more toward economic prosperity than toward quality of life, which makes
disorder not as problematic to Koreans as to Americans. Therefore, the author speculates
that the link between disorder and crime would be weaker in Korean communities than in
American counterparts, an idea which well-deserves empirical research.

Within Korean society, however, there must be a large variation of tolerance
toward disorder across various types of communities, just as within the US. Also, it is
obvious that the relationship between disorder and crime is open to temporal change.

The cross-sectional, variation and longitudinal change in disorder and crime seems to be
getting more apparent within Korean society 111 light of the fact that residential
segregation or stratiﬁcation has been becoming more and more serious in recent years.
That is, people with similar socioeconomic backgrounds cluster together around a certain
school district. However close a block is to a school, if residents in that block have a
different level of socioeconomic status than. others in neighboring blocks, then they
cannot send their children to the school. These mechanisms in another country deserve

attention in order for the broken windows model to be more thoroughly tested within and.

154

across various contexts. This way, the idea and policy would be able to contribute to the

gradual improvement of human society.

155

APPENDICES

156

APPENDIX A

Coding Sheet for Street and Housing Survey

Table 18. Coding Sheet for Street Survey

 

City of Lansing=s Tagged Housing Revitalization Program (MSU, CJ)

 

Observer: Date: Time:
Block Number: Street Name Street Segment #

 

 

Street Segment Measurements

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1 Presence of a Neighborhood Crime Watch Sign Yes No
2 Vehicular trafﬁc direction l-way 2-way
3 Number of trafﬁc lanes
4 Number of overhead street lights
5 Number of trafﬁc control lights
6 Volume of motorized trafﬁc (number of cars per minute)
7 Condition of the street ( l=all very poor 5 =Ave lO=All very good)
Block Face Measure (3)51: [83:18:
8 Number of houses
Extent of cigarettes and/or cigars on sidewalks, street or in gutters (e.g.
9 public space)

0=None 1=Very light 2=Light 3=Moderate 4=Heavy 5=Very Heavy
10 Number of houses with litter in front (smaller than newspaper size)
1 1 Number of houses with garbage in front (larger than newspaper size)
12 # of empty liquor and beer bottles or cans on sidewalks, street or gutters
l3 # of cars parked on the street
14 # of abandoned cars visible from the street
15 # of cars appearing from the street to be disabled
l6 # of cars with parking violations
l7 # of condoms on sidewalks, street or in gutters (e.g. pttblic space)
18 # of needles & syringes on sidewalks, street or gutters (e.g. public space)

 

 

157

 

Table 18. (cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

19 # of incidents of grafﬁti visible from the street

20 # of incidents of painted over grafﬁti visible from the street

21 Overall condition of most sidewalks

22 # of houses damaged by ﬁre or vandalism

23 # of houses with boarded/covered windows

24 # of houses with broken windows NOT boarded up/NOT covered/N OT
25 i of houses/property with window bars or other visible security beyond
26 # of residential units being renovated/rehabilitated/current maintenance
27 Overall condition of most residential units

28 # of vacant lots or open/unused lots

29 # of commercial units being renovated/rehabilitated/current maintenance
30 # of commercial units/property with window bars or visible security

31 # of boarded/covered/vacant commercial units

32 # of commercial units

 

158

 

 

 

Table 19. Coding Sheet for Housing Survey

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

City of Lansing=s Tagged Housing Revitalization Program
(MSU, School of Criminal Justice Evaluation)
Observer: Date: Time: Street Segment #
Building Number: Street Name Parcel #
Type of residence: Single Family, Multiple Dwelling; (other)
Condition of the principal exterior structure/resident
l % of roof or drainage that is missing /damaged (e.g., allows in weather or _%
creates nuisance)
2 % of facade/siding that is physically damaged and needing repair _%
3 % of facade/siding that is defaced with markings, carvings, or grafﬁti (property _%
defaced)
4 % of facade/siding with absent or peeling paint such that material underneath is %
visible
5 % of lSl ﬂoor window trim with peeling or absent paint _%
6 % of 15‘ ﬂoor window glass broken (out) or substantially (more than 1") cracked _%
7 % of 1” ﬂoor windows partially or completely boarded or covered by building _%
materials
8 % pathways/walkways leading from sidewalk to entryway needing repair (V
because of hazardous condition 0
9 % entryway/path/porch/stoops to front door needing repair because of _%
hazardous condition
10 % of front door (main entry way) missing or in poor condition and needing _%
repair
11 Is residence address (premises identiﬁcation) number in plain view from the Les
street 0
Red
12 Is City Notice/Card that the structure is tagged visible from street (circle) Gray
Neither
. . . Yes
12a Is that a building permit on the structure No
13 General condition of the exterior of the house
1=Very poor 5: Ave 10=Very Good
14 . % of visible (from public place) grass and/or weeds grown to 8+ _%

 

159

 

Table 19. (cont’d)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Condition of the grounds/property area:
15 % of visible property with trash, junk, debris, or garbage on the grounds
(newspapers, bags, cans, etc.) _%
l6 # of appliances (e. g. refrigerators) and indoor-type furniture on the grounds _#
l7 # of buildings or structures that are open at a door or window leaving the
interior exposed _#
18 # of abandon or disabled cars (no license, 2+ missing tires, 2+ tickets) _#
l9 # of vehicle, motorized recreation vehicle, watercraft or trailer parked in the
front yard _#
20 # of watercraft or trailers mechanically and structurally unable to operate _%
21 # of accessory structures (such as a detached garage or shed) in need of repair _%
Any damaged (structurally unsound) fences or structures; or lumber/building Yes
22 . . .
materials, hazardous trees, broken and/or hanging tree limbs or hazardous No
vegetation (text)
23 Any object obstructing the vision of motorist including fences, hedges, bushes, Yes
trees No
24 General condition of the grounds (1=Very poor 5: Ave 10=Very Good)
25 Does residence appear occupied (e.g. someone currently living at the address, Yes
looks lived-in/in-use etc). No

 

160

 

APPENDIX B

Evaluation Guidelines for Street Survey

Street Segment Evaluation Criteria
Street Segment Measurements

1. Presence of a Neighborhood Crime Watch sign : Yes if there is a sign present at either

end of the street segment
Code YES or NO

2. Vehicular Trafﬁc direction: l-way for a one-way road, 2-way for a two direction street
Code 1 or 2

3. Number of trafﬁc lanes: Count of the number of lanes on the street to include a

universal turn lane counting as a lane.
Code Number

4. Number of overhead street lights
Code Number

5. Number of trafﬁc control lights: the number of trafﬁc lights including either end of the

street segment.
Code Number

6. Volume of Motorized trafﬁc: the average numbers of cars to pass per minute
Code x cars in y minutes (need place for x and y to calc average)

7. Condition of the Street
Code 1 — 10
-- Three or more potholes (larger than a basketball)
-- one or two potholes
-— three or more noticeable patched spaces of asphalt or concrete
-- one or two patched areas of roadway
-- Three or more noticeable cracks (at least halfway across the width of the
road) in the roadway
-- One or two cracks in the roadway
-- Dirt
— Under construction
0 —Excellent Road condition — none of the above listed problems

G-hDJN—e

H000“

161

Block Face Measuremen_ts

Odd side refers to the dies of the block face with odd numbers, while even side
refers to that which contains the even numbered houses.

8. Number of houses on the street segment (to include comer houses that may fall on

either street)
Code Number

9. Extent of cigarettes and /or cigars on sidewalks, streets or in gutters (e. g Public space)
0 — None -- No butts seen when inspecting street
1 — Very Light -- Average one or two butts per 10 feet segment
2 — Moderate -- More than three butts per ten foot segment
3 - High -- More than three butts per ten foot segment with several concentrations
of butt disposal along the route

CodeOorl or20r3

10. Number of houses with litter in front (smaller than newspaper size)
Code Number

11. Number of Houses with garbage in front (Larger than newspaper size)
Code Number

12. Number of liquor and beer bottles or cans on sidewalks, streets, or gutters
Code Number

13. Number of cars parked on the street
Code Number

14. Number of abandoned cars visible from the street (missing registrations, missing
more than two tires, broken windows, grass growing around the wheels etc)
Code Number

15. Number of cars appearing from the street to be disabled (with license plates but
having clear mechanical problems such as ﬂat tires, on blocks, power train items

removed)
Code Number

16. Number of cars with visible parking violations posted
Code Number

17. Number of condoms on sidewalks, street or in gutters
Code Number

18. Number of needles and syringes on sidewalks, street, or gutters
Code Number

162

19. Number of incidents of grafﬁti visible from the street
Code Number

20. Number of incidents of painted over grafﬁti visible from the street
Code Number

21. Overall condition of sidewalks
Percentage of Sidewalk with grass growing in cracks
Percentage of Sidewalk with additional cracks

Percentage Code
0 0
0-10% 10
10-25% 25
25-50% 50
50-75% 75
75-100% 100

22. Number of Houses damaged by ﬁre or vandalism
Code Number (Calculation: using number of houses on street this should become a
percent)

23. Number of houses with boarded/covered windows
Code Number (Calculation: using number of houses on street this should become a
percent)

24. Number of houses with broken windows that are not boarded, covered up or cared for
Code Number (Calculation: using number of houses on street this should become a
percent)

25. Number of Houses/properties with window bars or other visible security beyond
locked doors
Code Number (Calculation: using number of houses on street this should become a

percent)

26. Number of residential units being renovated / rehabilitated / current maintenance
activity

Code Number (Calculation: using number of houses on street this should become a
percent)

27. Overall condition of most residential units
Code Number (number of poorly conditioned houses.....will turn into percent)
(Calculation: using number of houses on street this should become a percent)

Number of houses in visible sign of disrepair, needing maintenance, needing
paint, needing lawnwork (greater than 8”)

163

29. Number of vacant lots or open / unused lots
Code Number

30. (Q30) Number of commercial unities renovated/rehabilitated etc.
Code Number

31. (Q31) Number of commercial units/property with window bars or visible security
Code Number

32. (Q32) Number of boarded / covered / vacant commercial units
Code Number

33. (Q33) Number of businesses
Code Number

164

APPENDIX C
Number of Serious Crime and Social Disorder for Study Segments and Intersections85

Table 20. Number of Serious Crime and Social Disorder for Study Segments and
Intersections

 

 

 

2003 2005
total : study study : inter total : study study : inter
(percentage 1) (percentage 2) (percentage 1) (percentage 2)
Total 16957 : 2356 2356 : 430 16554 : 2310 2310 : 451
(13.89) (15.43) (13.95) (16.33)
Violent Crime 3292 : 449 449 : 42 3668 : 472 472 : 57
(13.64) (8.55) (12.87) (10.78)
Homicide 8:1 1:0 11:0 0:0
(12.50) (0.00) (0.00) (0.00)
Rape 251 :29 29:0 262: 12 12:1
" (11.55) (0.00) (4.58) (7.69)
Robbery 196228 28:6 224: 16 16:11
(14.28) (17.65) (7.14) (40.74)
Assault 2837 : 391 391 :36 3171 : 408 403 :45
(13.78) (8.43) (12.87) (10.04)
Pro ert Crime 7737 : 960 960 : 66 7631:1015 1015 :65
- p y ' (12.41) (6.43) (13.30) (6.02)
Bur lar 1591 :288 288: 12 1694 : 321 321:6
‘5 V (18.10) (4.00) (18.95) (1.83)
Larceny 4868 : 535 535 : 25 4899 : 550 550 : 34
' (10.99) (4.46) (11.23) (5.82)
VT 1260: 134 134:29 1023: 125 125:25
(10.63) (17.79) (12.22) (16.67)
Arson 18:3 3:0 15:0 0:0
(16.67) (0.00) (0.00) (0.00)
Social Disorder 5928 : 930 930 : 321 5255 : 823 823 : 329
(15.69) (25.66) (15.66) (28.56)
Drunk subject 393 : 32 32 : 33 347 :25 25 : 39
(8.14) (50.77) (7.20) (60.94)
Juvenile 1641 :222 222 : 119 1467: 189 189: 106
(13.53) (34.90) (12.88) (35.93)
Noise 2828 : 520 520 : 93 2250 : 439 439 : 71
(18.39) (15.17) (19.51) (13.92)
Panhandling 151 : 29 29 : 2 163 : 22 22 : 4
(19.20) (6.45) (13.50) (15.38)
Loudparty 608: 101 101 :23 609: 107 107236
(16.61) (18.55) (17.57) (25.17)
Prostitution 307 : 26 26 :44 419 : 41 41 : 69
(8.47) (62.86) (9.78) (62.73)

 

 

85 * total: number of serious crime or social disorder for total segments (n=5,953) including intersections
* study: number of serious crime or social disorder for study segments (n=3 77) excluding intersections

(Each case was counted as l in this study.)

* in_teg: number of serious crime or social disorder for intersections that include the study segments
(Intersections are generally made up of four segments, and thus technically ‘inter’ can contain up to
four study segments. Actually, one case of social disorder in 2003 and two cases of social disorder in
2005 were recorded on intersections with four study segments. Each case was counted as half.)

* percentage 1: study / total

* percentage 2: inter / (inter + study)

165

APPENDIX D

Multicollinearity Diagnostics of Level 2 Variables Including Lagged Crime Rates86

Table 21. Tolerance and Variance Inﬂation Factors

 

Independent Variables Tolerance VIF

 

Concentrated Disadvantage .71 1.40
Immigrant Concentration .68 1.48
Residential Instability .76 1.31
Land Use .94 1.06
Lagged Violent Crime Rates 2003 .96 1.04

 

Table 22. Collinearity Diagnostics

 

Variance Proportions

 

Eigen- Condition

 

Concept value Index Intercept CD IC RI LU Wy
1 2.46 1.00 .04 .00 .00 .00 .06 .04
2 1.84 1.16 .00 .13 .12 .12 .00 .00
3 .64 1.96 .01 .38 .03 .72 .01 .00
4 .52 2.17 .01 .37 .75 .13 .03 .00
5 .37 2.56 .01 .08 .04 .01 .70 .31
6 .16 3.87 .94 .04 .06 .02 .20 .64

 

 

8" The results for the other types of lagged crime rates were almost the same as those in the table.

'1 66

APPENDIX E

Multicollinearity Diagnostics of Level 1 Variables (2003 and 2005)

Table 23. Tolerance and Variance Inﬂation Factors (2003)

 

 

Independent Variables Tolerance VIF

Physical Disorder .97 1.03
Social Disorder .98 1.02
Land Use .99 1.01

 

Table 24. Collinearity Diagnostics (2003)

 

Variance Proportions

 

Eigen- Condition

 

Concept value Index Intercept PD SD LU
l 1.39 1.00 .27 .03 .02 .30
2 1.12 1.12 .06 .38 .41 .01.
3 .87 1.26 .01 .54 .57 .00
4 .62 1.50 .65 .04 .00 .69

 

Table 25. Tolerance and Variance Inﬂation Factors (2005)

 

 

Independent Variables Tolerance VIF

Physical Disorder .99 1.00
Social Disorder .94 1.06
Land Use .94 1.06

 

Table 26. Collinearity Diagnostics (2005)

 

Variance Proportions

 

Eigen- Condition

 

Concept value Index Intercept PD SD LU
l 1.42 1.00 .22 .00 .09 .29
2 1.03 1.18 .07 .51 .36 .00
3 .98 1.21 .17 .47 .33 .00
4 .57 1.58 .54 .02 .22 .71

 

167

APPENDIX F

Results for Hierarchical Linear Models87

Table 27. Unconditional Model (Violent Crime 2003)

 

 

 

 

 

 

 

Standard
Fixed Effect Coefﬁcient Error p
1’00 .46 .03 .00
Random Effect Variance
Component p
Block group mean, 110]“ .01 .11
Reliability of Coefficient Estimate
.301 .09
Table 28. Unconditional Model (Property Crime 2003)
Standard
Fixed Effect Coefﬁcient Error p
3’00 .79 .05 .00
Random Effect Variance
Component p
Block group mean, uoj .08 .00
Reliability of Coefficient Estimate
[301. .42

 

 

87 Instead of the LaPlace approximation method used in the non-linear model, the reported statistics in the
linear model are based on the restricted maximum likelihood estimation.

l68

Table 29. Unconditional Model (Violent Crime 2005)

 

 

 

 

 

 

 

Standard
Fixed Effect Coefﬁcient Error P
37,0 .46 .03 .00
Random Effect Variance
Component p
Block group mean, l-l(),' .02 .07
Reliability of Coefficient Estimate
[30)- .19
Table 30. Unconditional Model (Property Crime 2005)
Standard
F ixed Effect Coefﬁcient Error p
}’()() .85 .05 .00
Random Eﬂect Variance p
Component
Block group mean, no,- .07 .00
Reliability of C oeﬂicient Estimate
301 .39

 

169

REFERENCES

170

REFERENCES

Allen, M. P. (1997). Understanding regression analysis. New York: Plenum Press.

Anderson, E. (1999). Code of street: Decency, violence, and the moral life of the inner
city. New York: W. W. Norton.

Anselin, L. (2002). Under the hood: Issues in the speciﬁcation and interpretation of
spatial regression models. Agricultural Economics, 27, 247-67.

Anselin, L. (2005). Exploring spatial data with GeoDaTM : A workbook. Spatial Analysis
Laboratory (SAL). Department of Agricultural and Consumer Economics,
University of Illinois, Urbana-Champaign, IL.

Bailey, T. C., & Gatrell, A. C. (1995). Interactive spatial analysis. Essex: Prentice
Hall.

Beck, E. M., & Tolnay, S. E. (1995). Analyzing historical count data: Poisson and
negative binomial regression models. Historical Methods, 28(3), 125-131.

Baller, R. D., Anselin, L., Messner, S. F ., Deane, G., & Hawkins, D. F. (2001).
Structural covariates of US. county homicide rates: Incorporating spatial effects.
Criminology, 39, 561-590.

Bellair, P. E. (1997). Social interaction and community crime. Criminology, 35(4), 677-
703.

Bellair, P. E. (2000). Informal surveillance and street crime: A complex relationship.
Criminology, 38, 137-167.

Berry, W. D., & Feldman, S. (1985). Multiple regression in practice. In M. S. Lewis-
Beck (Ed.), Sage university paper series on quantitative applications in the social
sciences, Vol 07-050. Newbury Park, CA: Sage.

Berry, B. J ., & Kasarda, J. D. (1977). C ontemporary urban ecology. New York:
Macmillan.

Blau, J. R., & Blau, P. M. (1982). The cost of inequality: Metropolitan structure and
violent crime. American Sociological Review, 47, 114-129.

Block, R., & Block, C. R. (2000). The Bronx and Chicago: Street robbery in the environs
of rapid transit stations. In V. Goldsmith, P. G. McGuire, J. H. Mollenkopf, & T.
A. Ross (Eds.), Analyzing crime patterns: Frontiers of practice. Thousand Oaks,
CA: Sage.

171

Block, R., & Davis, S. (1996). The environs of rapid transit stations: A focus for street
crime or just another risky place. In R. V. Clarke (Ed), Crime and place, Vol 6
(pp. 236-257). Monsey, NY: Criminal Justice Press.

Blumstein, A., Barnett, A., & F arrington, D. (1987). Probabilistic models of youthful
criminal careers. Criminology, 25, 83-107.

Blumstein, A., & Cohen, J. (1987). Characterizing criminal careers. Science, 238 (No.
4818), 985-991.

Borooah, V. K., & Carcach, C. A. (1997). Crime and fear: Evidence from Australia.
British Journal of Criminology, 37, 635-657.

Bottoms, A. E., & Wiles, P. (1986). Housing tenure and residential community crime
careers in Britain. In A. J. Reiss, Jr. & M. Tonry (Eds), Crime and Justice, V8
(pp. 101-162). Chicago: University of Chicago Press.

Brantingham, P. L., & Brantingham, P. J. (1982). Mobility, notoriety, and crime: A study
of crime patterns in urban nodal points. Journal of Environmental System, 11, 89-
99.

Brantingham, P. L., & Brantingham, P. J. (1993). Environmental, routine, and situation:
Toward a pattern theory of crime. In R. V. Clarke & M. F elson (Eds.), Routine
activity and rational choice: Advances in criminological theory. New Brunswick,
NJ: Transaction Publishers.

Bratton, W., & Kelling, G. (2006). There are no cracks in the broken windows. National
Review Online, February 28. http://artic1e.nationalreview.com/ (accessed March
24,2006)

Brown, B. B., Perkins, D. D., & Brown, G. (2004). Crime, new housing, and housing
incivilities in a ﬁrst-ring suburb: Multilevel relationships across time. Housing
Policy Debate, 15, 301 -346.

Burgess, E. W. (1925). The growth of the city. In R. E. Park, E. W. Burgess, & R. D.
McKenzie (Eds), The city. Chicago: University of Chicago Press.

Bursik, R. J ., Jr. (1988). Social disorganization and theories of crime and delinquency:
Problems and prospects. Criminology, 26, 519-551.

Bursik, R. J ., Jr. (1999). The informal control of crime through neighborhood networks.
Sociological Focus, 32(1), 85-97.

Bursik, R. J ., J r., & Grasmick, H. G. (1992). Longitudinal neighborhood proﬁles of
delinquency: The decomposition of change. Journal of Quantitative Criminology,
8, 247-264.

Bursik, R. J ., Jr., & Grasmick, H. G. (1993). Neighborhood and crime: The dimensions of
effective community control. New York: Lexington Books.

172

Bursik, R. J ., Jr., & Grasmick, H. G. (1995). Neighborhood-based networks and the
control of crime and delinquency. In H. D. Barlow (Ed.), Crime and public policy.
Boulder: Westview Press.

Cameron, A. C., & Trivedi, P. K. (1998). Regression analysis of count data. Cambridge
University Press.

Chainey, S. P., & Ratcliffe, J .H. (2005). GIS and Crime Mapping. London: Wiley and
Sons

City-Data.com (2006). Lansing: History. http://www.city-data.com/us-cities/The-
Midwest/Lansing-History.html (accessed September 7, 2006).

Clarke, R. V. (1997). Introduction. In R. Clarke (Ed.), Situational crime prevention:
Successful case studies, 2nd ed. New York: Harrow and Heston.

Clarke, R. V., & Cornish, D. B. (1985). Modeling offenders’ decisions: A framework for
research and policy. In M. Tonry & N. Morris (Eds), Crime and justice: An
annual review of research, Vol 7. Chicago: University of Chicago Press.

Clear, T., Rose, D., Waring, E., & Scully, K. (2003). Coercive mobility and crime.
Justice Quarterly, 20, 33-63.

Cohen, L. E., & F elson, M. (1979). Social change and crime rate trends: A routine
activities approach. American Sociological Review, 44, 588-608.

Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal
of Sociology, 94, 895-120.

Coleman, J. S. (1994). A vision for sociology. Society, November, 29-34.

Cornish, D. B., & Clarke, R. V. (1987). Understanding crime displacement: An
application of rational choice theory. Criminology, 25(4), 933-947.

Cullen, F. T., & Agnew, R. (1999). C riminological theory: Post to present. Los Angeles,
CA: Roxbury.

Durkheim, E. (1933, originally published in 1893). The division of labor in society. New
York: Free Press.

Eek, J. E., & Maguire, E. (2000). Have changes in policing reduced violent crime? An
assessment of the evidence. In A. Blumstein & J. Wallman (Eds), The crime drop
in America. Cambridge, UK: Cambridge University Press.

Eck, J. E., & Weisburd, D. (1995). Crime places in crime theory. In J. Eck & D.
Weisburd (Eds), Crime and place. Monsey, NY: Criminal Justice Press and
Police Executive Research Forum.

173

Einstadter, W., & Henry, S. (1995). C riminological theory: An analysis of its underlying
assumptions. Orlando, FL: Harcourt Brace.

Elliott, D. S., Huizinga, D., & Menard, S. (1989). Multiple problem youth: Delinquency,
substance use and mental health problems. New York: Springer-Verlag.

Elliott, D. S., Wilson, W. J ., Huizinga, D., Sampson, R. J ., Elliott, A., & Rankin, B.
(1996). The effects of neighborhood disadvantage on adolescent development.
Journal of Research in Crime and Delinquency, 33(4), 389-426.

F abrigar, L. R., Wegener, D. T., MacCallum, R. C., & Strahan, E. J. (1999). Evaluating
the use of Exploratory Factor Analysis in psychological research. Psychological
Methods, 4(3), 272-299.

F agan, J ., & Davies, G. (2000). Street stops and broken windows: Terry, race, and
disorder in New York City. F ordham Urban Law Journal, 28, 457-504.

F elson, M. (1987). Routine activities and crime prevention in the developing metropolis.
Criminology, 25(4), 911-931.

Fox, J. (1991). Regression diagnostics. In M. S. Lewis-Beck (Ed.), Sage university paper
series on quantitative applications in the social sciences, Vol 07-079. Newbury
Park, CA: Sage.

Garofalo, J ., & Laub, J. (1978). The fear of crime: Broadening our perspective.
Victimology, 3, 242-253.

Gibson, C. L., Zhao, J ., Lovrich, N. P., & Gafﬁrey, M. J. (2002). Social integration,
individual perceptions of collective efﬁcacy, and fear of crime in three cities.”
Justice Quarterly, 19, 537-564.

Gordon, C. L., & Brill, W. (1996). The expanding role of crime prevention through
environmental design in premises liability. Washington, D. C.: US. Department

of Justice, Ofﬁce of Justice Programs, National Institute of Justice.

Grannis, R. (1998). The importance of trivial streets: Residential streets and residential
segregation. American Journal of Sociology, 103(6), 1530-1564.

Green, L. (1995). Cleaning up drug hot spots in Oakland, California: The displacement
and diffusion effects. Justice Quarterly, 12(4), 73 7-754.

Greene, J. A. (1999). Zero tolerance: A case study of police policies and practices in New
York City. Crime and Delinquency, 45(2), 171-187.

Guerry, A. M. (1833). Essai sur la statistique morale de la France. Paris: Crochard.

174

Harcourt, B. E. (1998). Reﬂecting on the subject: A critique of the social inﬂuence
conception of deterrence, the broken windows theory, and order-maintenance
policing New York style. Michigan Law Review, 97, 291-389.

Harcourt, B. E., & Ludwig J. (2006). Broken windows: New evidence from New York
City and a ﬁve-city social experiment. The University of Chicago Law Review, 73,
271-320.

Hawley, A. H. (1950). Human ecologv: A theory of community structure. New York:
Ronald Press.

Hay, C., Fortson, E. N., Hollist, D. R., Altheimer, I., & Schaible, L. M. (2006). The
impact of community disadvantage on the relationship between the family and
juvenile crime. Journal of Research in Crime and Delinquency, 43(4), 326-356.

Hayslett-McCall, K. L. (2002). Neighborhoods, land-use, and robbery rates: A test of
routine activity theory. A Dissertation for Ph.D., Pennsylvania State University.

Hodge, S. S. (1990). The C-O-P-E-T framework, a methodology for analyzing complex
human-environmental interactions in the process of adaptation: Its application in
social forestry. A Thesis for MS, Michigan State University.

Hunter, A. (1978). Symbols of incivility. Paper presented at the 1978 meeting of the
American Society of Criminology. Dallas, Texas.

Jacobs, J. (1961). The death and life of great American cities. New York: Random House.

J ang, S. J ., & Johnson, B. R. (2001). Neighborhood disorder, individual religiosity, and
adolescent use of illicit drugs: A test of multilevel hypotheses. Criminology, 39,
109-143.

J anowitz, M. (1976). Social control of the welfare state. New York: Elsevier.

Jeffery, C. R. (1971). Crime prevention through environmental design. Beverly Hills,
CA: Sage.

Katz, C. M., Webb, V. J ., & Schaefer, D. R. (2001). An assessment of the impact of
quality-of-life policing on crime and disorder. Justice Quarterly, 18(4), 825-876.

Kelling, G. L., & Coles, C. M. (1996). Fixing broken windows: Restoring order and
reducing crime in our communities. New York: Simon and Schuster.

Kelling, G. L., & Sousa, W. H., Jr. (2001). Do police matter? An analysis of the impact
of New York City’s police reforms. Civic Report, No. 22. New York: Center for
Civic Innovation.

Klinger, D. (1997). Negotiating order in police work: An ecological theory of police
response to deviance. Criminology, 35, 277-306.

175

Kubrin, C. E., & Weitzer, R. (2003a). Retaliatory homicide: Concentrated disadvantage
and neighborhood culture. Social Problems, 50, 157-180.

Kubrin, C. E., & Weitzer, R. (2003b). New directions in social disorganization theory.
Journal of Research in Crime and Delinquency, 40(4), 374-402.

Kurtz, E. M., Koons, B. A., & Taylor, R. B. (1998). Land use, physical deterioration,
resident-based control and calls for service on urban streetblocks. Justice
Quarterly, 15, 121-149.

Lab, S. (1988). Crime prevention: Approaches, practices, and evaluations. Cincinnati,
OH: Anderson Publishing Co.

Land, K. C., McCall, P. L., & Nagin, D. S. (1996). A comparison of Poisson, negative
binomial, and semiparametic mixed Poisson regression models with empirical
applications to criminal careers data. Sociological Methods and Research, 24(4),
387-442.

Longmoor, E. S., & Young, E. F. (1936). Ecological interpretations of juvenile
delinquency, dependency and population mobility: A cartographic analysis of data
from Long Beach, California. American Journal of Sociology, 41, 598-610.

MacDonald, C. (2003). Red tape hindering ﬁght against blight. Lansing State Journal,
May 19, 2003.

Markowitz, F. E., Bellair, P. E., Liska, A. E., & Liu, J. (2001). Extending social
disorganization theory: Modeling the relationships between cohesion, disorder,
and fear. Criminology, 39, 293-320.

Mastrofski, S., Reisig, M., & McClusky, J. D. (2002). Police disrespect toward the public:
An encounter-based analysis. Criminology, 40, 519-551.

Maxﬁeld, M. G. (1999). The national incident-based reporting system: Research and policy
implications. Journal of Quantitative Criminology, 15, 119-149.

Maxﬁeld, M. G., & Babbie, E. (2001). Research methods for criminal justice and
criminology. Belmont, CA: Wadsworth.

McGarrell, E., Giacomazzi, A, & Thurman, Q. (1997). Neighborhood disorder,
integration, and fear of crime. Justice Quarterly, 14, 479-500.

Menard, S. (1991). Longitudinal research. In M. S. Lewis-Beck (Ed), Quantitative
Applications in the Social Science, Vol. 07- 76. Beverly Hills, CA: Sage.

Merry, S. F. (1981). Defensible space undefended: Social factors in crime prevention
through environmental design. Urban Affairs Quarterly, 16, 397-422.

176

Messner, S. F., & Anselin, L. (2004). Spatial analyses of homicide with areal data. In M.
Goodchild & D. Janelle (Eds), Spatially Integrated Social Science (pp. 127-141).
New York: Oxfords University Press.

Miethe, T. D., & McDowall, D. (1993). Contextual effects in models of criminal
victimization. Social Forces, 7], 741-759.

Morenoff, J. D., Sampson, R. J ., & Raudenbush, S. W. (2001). Neighborhood inequality,
collective efﬁcacy, and the spatial dynamics of urban violence. Criminology,
39(3), 517-559.

Newman, 0. (1972). Defensible space: Crime prevention through urban design. New
York: Macmillan.

Norman, G. R., & Streiner, D. L. (1994). Biostatistics: The bare essentials. St. Louis,
MO: Mosby.

Park, R. E., & Burgess, E. (1925). The city. Chicago: University of Chicago Press.

Pattillo-McCoy, M. E. (1998). Sweet mothers and gangbangers: Managing crime in a
black middle-class neighborhood. Social Forces, 76, 747-774.

Perkins, D. D., Meeks, J. W., & Taylor, R. B. (1992). The physical environment of street
blocks and resident perceptions of crime and disorder: Implications for theory and
measurement. Journal of Environmental Psychology, 12, 21-34.

Perkins, D. D., Wandersman, A., Rich, R. C., & Taylor, R. B. (1993). The physical
environment of street crime: Defensible space, territoriality, and incivilities.
Journal of Environmental Psychology, I 3, 29-49.

Peterson, R. D., Krivo, L. J ., & Harris, M. A. (2000). Disadvantage and neighborhood
violent crime: Do local institutions matter? Journal of Research in Crime and
Delinquency, 37, 31-63.

Pratt, T. C., & Cullen, F. T. (2005). Assessing macro-level predictors and theories of
crime: A meta-analysis. In M. Tonry (Ed), Crime and justice: A review of
research, Vol. 32 (pp. 373-450). Chicago: University of Chicago Press.

Quetelet, L. A. (1831). Research on the propensity for crime at different ages.
Cincinnati: Anderson, 1984.

Quetelet, L. A. (1835). Physique sociale: Ou, essai sur le developpement desfacultes
de l ’Homme, Vol 2. Brussels: Muquardt.

Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and
data analysis methods. Thousand Oaks, CA: Sage.

177

Reckless, W. C. (1961). A new theory of delinquency and crime. Federal Probation, 25,
42-46.

Reiss, A. J. (1951). Delinquency as the failure of personal and social control. American
Sociological Review, 16, 196-207.

Reiss, A. J ., Jr. (1971). Systematic observations of natural social phenomena. In H. L.
Costner (Ed.), Sociological methodology, vol. 3. San Francisco: Jossey-Bass.

Reiss, A. J. Jr. (1986). Why are communities important in understanding crime? In A. J.
Reiss, Jr. & M. Tonry (Eds), Crime and Justice, V8 (pp. 1-33). Chicago:
University of Chicago Press.

Rice, K. J., & Smith, W. R. (2002). Socioecological models of automotive theft:
Integrating routine activity and social disorganization approaches. Journal of
Research in Crime and Delinquency 39, 304-336.

Rider, S. W. (2003). Abandoned housing and crime: A policy approach. Policy Paper at
Michigan State University.

Robinson, J. B., Lawton, B. A., Taylor, R. B., & Perkins, D. D. (2003). Multilevel
longitudinal impacts of incivilities: Fear of crime, expected safety, and block
satisfaction. Journal of Quantitative Criminology 19, 237-274.

Rogosa, D. R. (1995). Myths and methods: Myths about longitudinal research plus
supplemental questions. In J. M. Gottrnan (Ed.), The analysis of change. Mahwah,
NJ: Lawrence Erlbaum.

Roncek, D. W. (2000). Schools and crime. In V. Goldsmith, P. G. McGuire, J. H.
Mollenkopf, & T. A. Ross (Eds), Analyzing crime patterns: Frontiers of practice
(pp. 153-166). Thousand Oaks, CA: Sage.

Roncek, D. W., & Maier, P. ( 1991). Bars, blocks, and crimes revisited: Linking the
theory of routine activities to the empiricism of “hot spots”. Criminology, 29,
725-753.

Rose, D., & Clear, T. (1998). Incarceration, social capital, and crime: Implications for
social disorganization theory. Criminology, 36, 441-479.

Rosenfeld, R., Bray, T. M., & Egley, A. (1999). Facilitating violence: A comparison of
gang-motivated, gang-afﬁliated, and non-gang youth homicides. Journal of
Quantitative C riminology, 15, 495-516.

Ross, C. E., & Mirowsky, J. (1999). Disorder and decay: The concept and measurement
of perceived neighborhood disorder. Urban Affairs Review, 34(3), 412-432.

178

Rountree, P. W., Land, K. C., & Miethe, T. D. (1994). Macro-micro integration in the
study of victimization: A hierarchical logistic model analysis across Seattle
neighborhoods. Criminology, 32, 387- 404.

Sampson, R. J. (1985). Neighborhood and crime: The structural determinants of personal
victimization. Journal of Research in Crime and Delinquency, 22, 7-40.

Sampson, R. J. (1986). Crime in cities: The effects of formal and informal social control.
In A. J. Reiss, Jr. & M. Tonry (Eds), Crime and Justice, V8 (pp. 271-311).
Chicago: University of Chicago Press.

Sampson, R. J. (1987). Does an intact family reduce burglary risk for its neighbors?
Sociology and Social Research, 71, 204-207

Sampson, R. J. (2002). Transcending tradition: New directions in community research,
Chicago style. Criminology, 40(2), 213-230.

Sampson, R. J ., & Cohen, J. (1988). Different effects of the police on crime. Law and
Society Review, 22, 163-189.

Sampson, R. J ., & Groves, W. B. (1989). Community structure and crime: Testing social
disorganization theory. American Journal of Sociology, 94, 774-802.

Sampson, R. J ., Morenoff, J. D., & Earls, F. (1999). Beyond social capital: Spatial
dynamics of collective efﬁcacy for children. American Sociological Review, 64,
633-660.

Sampson, R. J ., Morenoff, J. D., & Gannon-Rowley, T. (2002). Assessing “neighborhood
effects”: Social processes and new directions in research. Annual Review of
Sociology, 28, 443-478.

Sampson, R. J ., Morenoff, J. D., & Raudenbush, S. (2005). Social anatomy of racial and
ethnic disparities in violence. American Journal of Public Health, 95(2), 224-232.

Sampson, R. J ., & Raudenbush, S. W. (1999). Systematic social observation of public
spaces: A new look at disorder in urban neighborhoods. The American Journal of
Sociology, 105, 603-651.

Sampson, R. J. & Raudenbush, S. W. (2004). Social disorder: Neighborhood stigma and
the social construction of “broken windows.” Social Psychology Quarterly, 67(4),
319-342.

Sampson, R. J ., Raudenbush, S. W., & Earls, F. (1997) Neighborhoods and violent crime:
A multilevel study of collective efﬁcacy. Science, 27 7(August 15), 918-924.

Schmid, C. (1960). Urban crime areas: Part 1. American Sociological Review, 25, 527-42.

179

Schuerman, L. & Kobrin, S. (1986). Community careers in crime. In A. J. Reiss, Jr. & M.
Tonry (Eds), Crime and Justice, V8 (pp. 67-100). Chicago: University of
Chicago Press.

Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and quasi-
experimental designs for generalized causal inference. Boston, MA: Houghton
Mifﬂin Company.

Shaw, C. R., & McKay, H. D. (1942). Juvenile delinquency and urban areas: A study of
delinquents in relation to differential characteristics of local communities in
American cities. Chicago: University of Chicago Press.

Sherman, L. W., Gartin, P. R., & Buerger, M. E. (1989). Hot spots of predatory crime:
Routine activities and the criminology of place. Criminology, 27(1), 27-55.

Sherman, L. W., & Weisburd, D. (1995). General deterrent effects of police patrol in
crime hot spots: A randomized, controlled trial. Justice Quarterly, 12, 625-648.

Silver, E. (2000). Extending social disorganization theory: A multilevel approach to the
study of violence among persons with mental illnesses. Criminology, 38, 1043-
1074.

Simcha-Fagan, O., & Schwartz, J. E. (1986). Neighborhood and delinquency: An
assessment of contextual effects. Criminology, 24, 667-699.

Simon, H. A. (195 7). Models of man: Social and rational; Mathematical essays on
rational human behavior in society setting. New York: Wiley.

Skogan, W. G. (1990). Disorder and decline: Crime and the spiral of decay in American
neighborhoods. New York: Free Press.

Skogan, W. G., & Hartnett, S. M. (1997). Community policing, Chicago style. Chicago:
University of Chicago Press.

Smith, D. (1986). The neighborhood context of police behavior. In A. J. Reiss, Jr. & M.
Tonry (Eds), Crime and Justice, V8 (pp. 313-341). Chicago: University of
Chicago Press.

Smith, W. R., Frazee, S. G., & Davison, E. L. (2000). Furthering the integration of
routine activity and social disorganization theories: Small units of analysis and the
study of street robbery as a diffusion process. Criminology, 38(2), 489-523.

Sousa, W. H., & Kelling, G. L. (2006). Of “broken windows,” criminology, and criminal

justice. In D. Weisburd & A. A. Braga (Eds), Police Innovation: Contrasting
Perspectives (pp. 77-97). Cambridge: Cambridge University Press.

180

Spelman, W. (1993). Abandoned buildings: Magnets for crime? Journal of Criminal
Justice, 21(5), 481-495.

Stark, R. (1987). Deviant places: A theory of the ecology of crime. Criminology, 25(4), 893-
909.

Sturman, M. C. (1999). Multiple approaches to analyzing count data in studies of
individual differences: The propensity for type I errors, illustrated with the case of
absenteeism prediction. Educational and Psychological Measurement, 59(3), 414-
430.

Tabachnick, B. G., & Fidell, L. S. (2006). Using multivariate statistics. Boston, MA:
Allyn & Bacon.

Taylor, R. B. (1997). Crime and small-scale places. Crime and place: Plenary papers of
the 1997 Conference on Criminal Justice Research and Evaluation. Washington,
DC: National Institute of Justice.

Taylor, R. B. (1999). The incivilities thesis: Theory, measurement, and policy. In R. H.
Langworthy (Ed.), Measuring what matters: Proceedings from the policing
research institute meetings (pp. 65-88). Research Report. Washington, DC: US.
Department of Justice, National Institute of Justice.

Taylor, R. B. (2001). Breaking away from broken windows. Boulder, CO: Westview.
Taylor, R. B. (2002). Fear of Crime, social ties, and collective efﬁcacy: Maybe
masquerading measurement, maybe déja vu all over again. Justice Quarterly,

19(4), 773-792.

Taylor, R. B., & Harrell, A. V. (1996). Physical environment and crime. Final Summary
Report. Washington, DC: US. Department of Justice, National Institute of Justice.

Taylor, R. B., Koons, B. A., Kurtz, E. M., Greene, J. R., & Perkins, D. D. (1995). Street
blocks with more nonresidential land use have more physical deterioration:

Evidence from Baltimore and Philadelphia. Urban Affairs Review, 31, 120-136.

Thacher, D. (2004). Order maintenance reconsidered: Moving beyond strong causal
reasoning. The Journal of Criminal Law and Criminology, 94(2), 101-133.

Tiechert, N. (1999). Safer streets: Block by block, the tide of crime receded in city.
Sacramento Bee, Sept. 19, 1999. Retrieved on 9/23/2003 from www.Sacbee.com.

Tobler, W. R. (1969). Geographical ﬁlters and their inverses. Geographical Analysis, 1,
234-253.

181

US. Census (2000). http://www.census.gov/main/www/cen2000.htm1. Retrieved on
September 7, 2006.

Velez, M. (2001). The role of public social control in urban neighborhoods. Criminology,
39, 837-863.

Villarreal, A. (2004). The social ecology of rural violence: Land scarcity, the
organization of agricultural production and the presence of the state. American
Journal of Sociology, 110, 313-348.

Warner, B. D., & Pierce, G. L. (1993). Reexamining social disorganization theory using
calls to the police as a measure of crime. Criminology, 31(4), 493-517.

Weisburd, D., Maher, L., & Sherman, L. (1993). Contrasting crime general and crime
speciﬁc theory: the case of hot spots of crime. Advances in Criminological T heory.
4, 45-70.

White, M. D., Fyfe, J. J ., Campbell, S. P., & Goldkamp, J. S. (2003). The police role in
preventing homicide: Considering the impact of problem-oriented policing on the
prevalence of murder. The Journal of Research in Crime and Delinquency, 40(2),
194-225.

Wikstrom, P. H., & Loeber, R. (2000). Do disadvantaged neighborhoods cause well-
adjusted children to become adolescent delinquents? Criminology, 38, 1109-1142.

Wilcox, P., Quisenberry, N., Cabrera, D. T., & Shayne, J. (2004). Busy places and broken
windows? Toward deﬁning the role of physical structure and process in
community crime models. Sociological Quarterly, 45, 185-207.

Williams, F. P., & McShane, M. D. (2003). Criminological theory. Upper Saddle River,
NJ: Pearson Prentice Hall.

Wilson, J. Q. (1975). Thinking about crime. New York: Basic Books.

Wilson, J. Q., & Kelling, G. (1982). The police and neighborhood safety: Broken
windows. Atlantic Monthly, 127, 29-38.

Wilson, W. J. (1987). The Truly Disadvantaged: The Inner City, the Underclass, and Public
Policy. Chicago: University of Chicago Press.

Wilson, W. J. (1996). When work disappears: The world of the new urban poor. New
York: Knopf.

Worrall, J. L. (2002). Does “broken windows” law enforcement reduce serious crime?
Research Reports, California Institute for County Government.

182

Worrall, J. L., & Zhao, J. H. (2003). The role of the COPS ofﬁce in community policing.
Policing, 26(1), 64-87.

183