using APA format sampling research

ORIGINAL PAPER

Violence in Urban Neighborhoods: A Longitudinal Study of Collective Efficacy and Violent Crime

John R. Hipp1 • Rebecca Wickes2

Published online: 28 June 2016 � Springer Science+Business Media New York 2016

Abstract Objectives Cross-sectional studies consistently find that neighborhoods with higher levels of collective efficacy experience fewer social problems. Particularly robust is the rela-

tionship between collective efficacy and violent crime, which holds regardless of the socio-

structural conditions of neighborhoods. Yet due to the limited availability of neighborhood

panel data, the temporal relationship between neighborhood structure, collective efficacy

and crime is less well understood.

Methods In this paper, we provide an empirical test of the collective efficacy-crime association over time by bringing together multiple waves of survey and census data and

counts of violent crime incident data collected across 148 neighborhoods in Brisbane,

Australia. Utilizing three different longitudinal models that make different assumptions

about the temporal nature of these relationships, we examine the reciprocal relationships

between neighborhood features and collective efficacy with violent crime. We also con-

sider the spatial embeddedness of these neighborhood characteristics and their association

with collective efficacy and the concentration of violence longitudinally.

Results Notably, our findings reveal no direct relationship between collective efficacy and violent crime over time. However, we find a strong reciprocal relationship between col-

lective efficacy and disadvantage and between disadvantage and violence, indicating an

indirect relationship between collective efficacy and violence.

Conclusions The null direct effects for collective efficacy on crime in a longitudinal design suggest that this relationship may not be as straightforward as presumed in the

literature. More longitudinal research is needed to understand the dynamics of disadvan-

tage, collective efficacy, and violence in neighborhoods.

& John R. Hipp [email protected]

1 Department of Criminology, Law and Society, The University of California, Irvine, 3311 Social Ecology II, Irvine, CA 92697, USA

2 School of Social Science, The University of Queensland, St Lucia, Brisbane, QLD, Australia

123

J Quant Criminol (2017) 33:783–808 DOI 10.1007/s10940-016-9311-z

Keywords Collective efficacy � Violence � Disadvantage � Neighborhood

Introduction

The spatial concentration of crime is strongly associated with disadvantage and the eco-

logical structure of urban neighborhoods. For much of the last century, scholars have

explained the concentration of crime and disorder through a social disorganization lens.

From this perspective, serious crime flourishes when the neighborhood networks necessary

for maintaining informal social control have broken down (Bursik and Grasmick 1993;

Hunter 1985; Kornhauser 1978; Sampson and Groves 1989; Shaw and McKay 1942;

Skogan 1986). Contemporary reformulations of social disorganization theory suggest that a

neighborhood’s shared expectations for action are also central to the regulation of crime

and disorder (Sampson et al. 1997). While networks generate a capacity for informal social

control, collective efficacy or the collective-action orientation of a neighborhood is the

most proximate mechanism associated with lower crime (Morenoff et al. 2001; Sampson

2006, 2012; Sampson et al. 1997).

Many cross-sectional studies in both developed and developing countries find that

neighborhoods with low collective efficacy experience higher levels of social problems,

particularly crime (see for example, Browning et al. 2005; Franzini et al. 2005; Maimon

and Browning 2010; Mazerolle et al. 2010; Morenoff et al. 2001; Sampson et al. 1997;

Sampson and Wikström 2008; Zhang et al. 2007). From this literature, we know that

poverty, ethnic/racial concentration and residential mobility are negatively associated with

collective efficacy (Sampson et al. 1997) and that a neighborhood’s collective efficacy is

influenced by the socio-structural composition of surrounding neighborhoods (Sampson

2012; Sampson et al. 1999). Sampson (2012) also suggests that neighborhood structures,

collective efficacy and violence may influence each other over time. Yet there are sig-

nificant gaps in our knowledge of collective efficacy. For example, few studies examine

what generates and sustains collective efficacy over time (Wickes et al. 2013). Even fewer

consider the temporal dynamics that might be associated with the collective efficacy-crime

association.

Due to the limited availability of neighborhood panel data, 1 our understanding of the

spatial and temporal nature of collective efficacy and crime link is incomplete. In fact,

there is little theoretical guidance about the time period over which these causal rela-

tionships should occur. In this paper we draw on the pioneering work of Sampson and his

colleagues and extend the literature in three important ways. First, using measures that are

nearly identical to those from the Project for Human Development in Chicago Neigh-

borhoods (PHDCN) survey, we bring together three waves of neighborhood survey and

census data and provide the first multi-wave study of the relationship between the struc-

tural characteristics of an area and collective efficacy across 148 neighborhoods in Bris-

bane, Australia. Second, we estimate three different longitudinal models that each make

different assumptions about the temporal period in which these reciprocal relationships

take place between neighborhood characteristics and collective efficacy with violent crime.

1 At the time of writing, only the Project for Human Development in Chicago Neighborhoods (PHDCN)

and LA FANS in Los Angeles have collected more than one wave of neighborhood survey data that specifically captures collective efficacy (Sampson 2012).

784 J Quant Criminol (2017) 33:783–808

123

Third, we consider the spatial embeddedness of these neighborhood features over time and

their association with collective efficacy and the concentration of violence.

In what follows we discuss the intellectual roots of collective efficacy theory. We then

discuss the key findings emerging from the collective efficacy literature with a specific

focus on the spatial interdependency of the neighborhood composition, collective efficacy

and violence and the stability of social processes like collective efficacy over time. We

consider different possibilities regarding the temporal period over which these processes

might operate. This is followed by an overview of our key data source, the Australian

Community Capacity Study (ACCS), and our analytic approach. Our findings demonstrate

the strong reciprocal relationship between collective efficacy and disadvantage and

between disadvantage and violence. They also show that the changing context of neigh-

boring areas must also be considered in understanding violence in the focal neighborhood.

Literature Review

The study of neighborhood effects in criminology commenced in earnest with Shaw and

McKay’s (1942) citywide study of delinquency. They identified that particular zones not

only had the highest rate of delinquents, they were areas characterized by significant

residential mobility, disadvantage and ethnic and racial concentration (Shaw and McKay

1942). Shaw and McKay (1942) argued that although criminal activity was related partly to

the individual’s propensity to commit a crime, it was the constant exposure to contra-

dictory standards of behavior, combined with the breakdown of community norms and

conventional values that explained the spatial concentration of crime in the zones closest to

the city center. The main finding of Shaw and McKay (1942) research was that despite

complete population turnover, high crime neighborhoods remained crime prone decades

into the future.

In the social disorganization literature a ‘‘web of social relationships’’ was considered

necessary for regulating crime (Kornhauser 1978:45; see also Bursik 1988; Bursik and

Grasmick 1993; Hunter 1985). These social ties were deemed particularly important for

fostering informal social control that can reduce neighborhood crime, however, more

recently scholars have questioned the regulatory capacity of neighborhood social ties.

While social ties may be instrumental for informal social control (Bursik 1999; Sampson

1988; Sampson and Groves 1989), studies find that neighborhood ties have limited direct

effects on crime (Warner and Roundtree 1997) and can impede residents’ ability to engage

in the informal social control of crime and disorder (Pattillio 1998). Sampson and Groves’

(1989) research revealed that the more important mechanism for controlling crime was the

community’s capacity to enforce informal social control norms. They found that the ability

of the community to exercise control over adolescent peer groups was the more proximate

intervening process associated with lower crime (Sampson and Groves 1989). Marking

what is referred to as the ‘‘process turn’’ in the study of neighborhood effects (Sampson

2012:47), Sampson and his colleagues emphasized the importance of collective efficacy, or

the ‘‘collective capacity for social action’’ over neighborhood ties for effective crime

control (Sampson 2002, 2012; Sampson et al. 1997, 1999; Morenoff et al. 2001).

Originating from Albert Bandura’s (1995, 1997, 2001) social cognitive theory, col-

lective efficacy is formally defined as ‘‘a group’s shared belief in its conjoint capabilities to

organize and execute the courses of action required to produce given levels of attainments’’

(Bandura 1997: 477) and is separate from the sum of individual attributes. Thus collective

J Quant Criminol (2017) 33:783–808 785

123

efficacy represents the emergent property of a group that is central to group level per-

formance. Sampson views collective efficacy as a concept applicable not only to small

groups, but one that is relevant to understanding the differential capacities of neighbor-

hoods to prevent crime and disorder. Using data resulting from the Project of Human

Development in Chicago Neighborhoods (PHDCN), 2

Sampson and his colleagues

(Sampson et al. 1997) examined the perceived capacity of fellow residents to engage in

informal social control and the level of social cohesion and trust within the neighborhood,

and the relationship with neighborhood violence.

Since the publication of the 1997 Science article, the link between collective efficacy

and a range of social problems is evidenced in many studies, almost all of which have

cross-sectional designs. Residents of communities with high levels of collective efficacy

report higher levels of self-rated health (Browning and Cagney 2002; Franzini et al. 2005)

and demonstrate greater parental monitoring (Rankin and Quane 2002). Moreover, the

presence of collective efficacy appears to mediate low parental monitoring as it relates to

the timing of first intercourse for girls and boys (Browning et al. 2005) and is associated

with a higher likelihood that women will formally or informally report instances of

domestic violence (Browning 2002). Studies in Australia (Mazerolle et al. 2010), the U.K.

(Wikström et al. 2012), and Sweden (Sampson and Wikström 2008) also support the

association between collective efficacy and violence and disorder. Not surprisingly,

scholars and policy makers alike are interested in understanding the antecedents of col-

lective efficacy, the threats to collective efficacy and its relationship with social problems

over time and space.

Poverty, Collective Efficacy and Violence: Their Spatial and Temporal Dynamics

Collective efficacy is strongly influenced by the socio-demographic composition of the

focal neighborhood. Concentrated disadvantage, residential instability and immigrant

concentration correspond with lower levels of collective efficacy at the neighborhood level

in cross-sectional studies. Taken together, these variables account for 70 % of the vari-

ability in collective efficacy across the 343 neighborhoods in Chicago (Sampson et al.

1997). Of these structural characteristics, concentrated disadvantage appears the most

deleterious for collective efficacy. Scholars have argued that poverty significantly reduces

the informal social control capacity of the neighborhood (Bursik and Grasmick 1993;

Morenoff et al. 2001) and as a consequence increases one’s exposure to violence and

victimization (Bingenheimer et al. 2005; Morenoff et al. 2001; Sampson et al. 1997). In

their study of Chicago neighborhoods, Morenoff et al. (2001) find that a one standard

deviation increase in disadvantage was associated with a 40 % higher homicide rate.

Sampson (2006) argues that poverty sets in motion a cycle that undermines collective

efficacy, which in turn sets in play a range of problems that deepens and reinforces poverty.

The association between poverty, collective efficacy and violence holds across cities in

the U.S. and internationally. In a cross-sectional study of Australian neighborhoods,

Mazerolle et al. (2010) find that concentrated disadvantage and residential mobility

2 The PHDCN is a longitudinal project sponsored by the MacArthur Foundation in partnership with the

National Institutes of Justice and Mental Health, Harvard School of Public Health, the Administration on Children, Youth and Families of the U.S. Department of Health and Human Services and the U.S. Department of Education. It is a multi-million dollar project examining the social, criminological, economic, organizational, political and cultural structures of Chicago’s communities.

786 J Quant Criminol (2017) 33:783–808

123

influence collective efficacy in much the same way as they do in Chicago and that col-

lective efficacy appeared to mediate the direct effect of these structural characteristics on

violence. Later studies in Australia suggest that disadvantage, the percentage of residents

speaking a language other than English and population density also negatively influence

collective efficacy for task specific problems like the control of children, violence and

more civic related problems (Wickes et al. 2013).

Although neighborhood socio-demographic characteristics, such as poverty, demon-

strate significant durability, neighborhood change does occur (Sampson 2012). Yet for the

most part, this change is predominantly in the direction of more deeply entrenched poverty

and violence (Sampson 2012; Taylor and Covington 1988; Wilson 1987). As Sampson and

Raudenbush (2006: 177) suggest ‘‘despite the vulnerabilities or assets associated with a

neighborhood’s internal characteristics, its rate of poverty change is directly linked to

changes in the surrounding network of neighborhood poverty’’. Theoretically then the

relationship between neighborhood characteristics, collective efficacy and crime is likely to

be temporal and reciprocal.

While we know that increasing poverty leads to an increase in a range of social

problems, including violence, we do not fully understand the reciprocal relationship

between disadvantage, collective efficacy and violence over time. Due to the limited

availability of panel data, there are no studies that examine the reciprocal relationship

between neighborhood structure, collective efficacy and violence across more than two

time points (Sampson 2012). Thus it is unclear whether collective efficacy is a protective

factor that prevents future violence or if violence undermines the development of collective

efficacy, which may then lead to greater levels of violence. Social disorganization per-

spectives would argue for the former: structural characteristics of the neighborhood break

down the regulatory mechanisms, like collective efficacy, which in turn leads to higher

crime (Sampson et al. 1999). There is some support for this position. Several studies find

that economic disadvantage, racial composition and residential instability significantly

predict violence, in particular homicide (Hipp 2010; Kubrin and Herting 2003; Land et al.

1990; Sampson et al. 1997). In their longitudinal examination of homicide in St. Louis,

Kubrin and Herting (2003) found that disadvantage and residential mobility was associated

with both the initial levels and trends of different types of homicide.

Yet others argue that crime is actually the catalyst for neighborhood change. Skogan

(1990) disorder and decline model posits that crime has a direct influence on residents’

behavior (residents leave the area) and perceptions (residents feel less safe and retreat from

social life). As Boggess and Hipp (2010) argue, crime encourages dissatisfaction with the

neighborhood which in turn promotes an exodus of those residents with the means to

relocate. That crime itself may play a role in how neighborhoods change (Bursik 1988;

Felson 2002; Miethe and Meier 1994; Skogan 1990) suggests the possibility of a reciprocal

relationship between crime, the socio-structural characteristics of the neighborhood and the

neighborhood processes necessary for regulating unwanted behavior. Thus violence might

lead to greater instability and increase disadvantage, which will in turn diminish collective

efficacy, leading to even more crime. Indeed, a longitudinal study of neighborhoods across

13 cities found that neighborhoods with higher levels of violent crime at the initial time

point experienced larger increases in concentrated disadvantage over the subsequent

decade, and that this effect was particularly pronounced for neighborhoods with very high

levels of violence (Hipp 2010).

The degree to which neighborhoods can successfully prevent violence is not only

determined by the ecological structure of the focal neighborhood, but is potentially

dependent upon nearby neighbors (Morenoff et al. 2001). Spatial dynamics are arguably

J Quant Criminol (2017) 33:783–808 787

123

important in explaining the dynamic relationship between concentrated disadvantage,

collective efficacy and violence and evidence suggests that disadvantaged neighborhoods

with high levels of violence tend to co-exist in space (Morenoff et al. 2001; Mears and

Bhati 2006; Sampson 2012; Tita and Greenbaum 2009).

Poverty and crime are strongly related to each other and exhibit significant spatial

clustering (Peterson and Krivo 2010). This is perhaps unsurprising given the strong rela-

tionship between poverty and violence more generally, yet even after controlling for other

community conditions, including previous violence, disadvantage in the focal and neigh-

boring areas can lead to higher violence. In their study of the spatial relationship between

resource deprivation and homicide, Mears and Bhati (2006) find that the spatial proximity

of disadvantage is more consequential for homicide in the focal neighborhood than the

level of homicide in neighboring communities. This leads them to conclude that the

‘‘spatial diffusion mechanism often found in the homicide literature could be an artefact of

omitting the spatially lagged resource deprivation measure’’ (Mears and Bhati 2006: 528).

Similarly, Kubrin and Hipp (2014) find that concentrated disadvantage in the area sur-

rounding a block group has a positive relationship with crime in blocks.

It is possible that neighborhood collective efficacy may also be influenced by ecological

conditions of neighboring areas. In the U.S., collectively efficacious neighborhoods are

spatially proximate to other neighborhoods with low levels of violence, reasonable levels

of affluence and low levels of mobility and ethnic/racial concentration (Morenoff et al.

2001; Sampson 2012; Sampson et al. 1999). Conversely, neighborhoods with low levels of

collective efficacy are closer to disadvantaged neighborhoods with high levels of violence,

residential mobility and racial/ethnic concentration. Thus ‘‘concentrated disadvantage,

crime and collective efficacy are spatially interrelated in ways that go beyond chance

expectations’’ (Sampson 2012: 240).

Temporal Periods and Causal Order

For each of the relationships of import to collective efficacy theory, there is little theo-

retical guidance regarding the time period in which they should play out. Taylor (2015)

discusses at length the importance of considering the proper causal time length when

specifying statistical models—failing to do so can result in improper specifications. Given

this theoretical uncertainty, we exploit the longitudinal nature of our data (in which each

wave of the survey data were collected 2–3 years apart) to empirically test three possible

time periods in which these processes might operate (rather than assuming any one in

particular).

One perspective argues that the relationship between collective efficacy and violence in

neighborhoods plays out in a much shorter time period than 2 years: thus, an increase in

collective efficacy would be expected to result in reduced violence in a manner of weeks or

months, not 2 years. And likewise, increases in violence would have a relatively quick and

negative effect on perceived collective efficacy (again, in a matter of weeks or months, not

2 years). Such a model effectively requires a simultaneous equations model given that our

data are coarser-grained temporally. Our longitudinal data is nonetheless useful in that it

provides plausible instrumental variables for such a model (MacDonald et al. 2013; Paxton

et al. 2011). This model is shown in Fig. 1a, in which time 1 violence serves as an

instrumental variable for the effect of time 2 violence on time 2 collective efficacy. This is

because we would expect that violence at one time point will be strongly dependent on the

level of violence at the prior time point, but collective efficacy at the present time point

would be dependent on current levels of violence but not levels of violence at prior time

788 J Quant Criminol (2017) 33:783–808

123

points. And likewise, time 1 collective efficacy serves as an instrumental variable for the

effect of time 2 collective efficacy on time 2 violence. In this case, a neighborhood with

high levels of collective efficacy at one time point is likely to remain a high collective

efficacy neighborhood, suggesting a strong positive relationship, but whereas the level of

violence will be dependent on current levels of collective efficacy in this model, there is no

reason to expect it to be impacted by prior levels of collective efficacy.

A second perspective argues that the causal process actually takes longer than weeks or

months. Thus, it takes a period of time after which collective efficacy has increased to

result in action that makes offenders aware that this location is not a suitable target given

Crime (t1) Crime (t2) Crime (t3)

Collective efficacy (t1)

Collective efficacy (t2)

Collective efficacy (t3)

Crime (t1) Crime (t3)

Collective efficacy (t1)

Collective efficacy (t3)

Crime (t1) Crime (t2) Crime (t3)

Collective efficacy (t1)

Collective efficacy (t2)

Collective efficacy (t3)

(a)

(b)

(c)

Fig. 1 Three hypothesized and tested temporal models

J Quant Criminol (2017) 33:783–808 789

123

the likely response through informal social control. Likewise, residents are not necessarily

immediately aware that crime events are increasing, but instead it takes a period of time to

become aware of this change (Hipp et al. 2009). As residents become aware of this change,

levels of collective efficacy would go down. This perspective suggests that this time period

may be in the range of 2 years, which implies that a two-year cross-lagged model may be

appropriate. This model is shown in Fig. 1b (Berry 1984).

A third perspective argues that the process takes even longer than 2 years. When we

consider the effect of crime on neighborhood characteristics such as residential instability,

racial composition, or income composition, it may be that levels of crime need to rise for a

period of time even longer than a year or two to induce mobility. Likewise, it may take

longer for residents to change their assessment of collective efficacy. And it may take

higher levels of collective efficacy even longer to actually reduce neighborhood violence.

This perspective suggests that a 2-year lag is too short, and that a longer period is required

to model these processes. We model this by excluding our middle wave of data, and

modeling a 5-year cross lagged model between waves 1 and 3 of our data as shown in

Fig. 1c.

The Present Research

In this paper we bring together three waves of panel data survey data (collected in 2007,

2010 and 2012) and two waves of census data (collected in 2006 and 2011) to examine

spatial and temporal relationship between the socio-structural composition of the neigh-

borhood, collective efficacy and rates of violence across 148 neighborhoods in Brisbane,

Australia. The time frame of this longitudinal study provides a unique context in which to

examine the reciprocal relationships of interest to this study. In the last decade Brisbane

has experienced significant population growth due to increases in both immigration and

internal migration (ABS 2012; Hugo et al. 2013), resulting in increased diversity. In 2001,

9.2 % of Brisbane’s residents spoke a language other than English at home, in 2011, this

figure reached 13.2 %. While historically the bulk of Brisbane’s immigrant population has

come from the United Kingdom and New Zealand, recent trends have seen substantial

growth in Asian immigration, particularly in the Indian and Chinese population (ABS

2012). Affiliations with non-Christian religions like Hinduism and Islam have also been on

the rise (ABS 2012). According to social disorganization theory and its more contemporary

reformulation, collective efficacy theory, changes in these socio-demographic character-

istics can negatively impact social cohesion and generate confusion around norms of

informal social control (Sampson et al. 1997).

Methods

The Research Site

Located in South East Queensland, the longitudinal component of the ACCS focuses on the

Greater Brisbane area. Brisbane, the state capital of Queensland, is Australia’s third largest

city with an estimated population of over 2 million people covering 15,825.9 square

kilometers. The five biggest industries in Brisbane are health care, retail, manufacturing,

professional scientific and technical services and education. The Brisbane region provides

an optimal research site to test the core tenets of collective efficacy theory. Brisbane has

790 J Quant Criminol (2017) 33:783–808

123

experienced considerable growth in recent years with the population increasing by 23 %

between 2001 and 2011. Brisbane also has one of the largest metropolitan Indigenous

populations in Australia. In recent years, the immigrant population of Queensland broadly,

and Brisbane specifically, has steadily increased and diversified boasting residents from

over 300 different countries speaking nearly 30 languages. Brisbane has also become home

to many immigrants from war torn countries (ABS 2012).

Longitudinal Sample Design

We focus on waves 2, 3 and 4 of the Brisbane ACCS sample given that wave 1 used

neighborhood units of analysis that are not comparable to those in the later waves. The

survey comprises 148 randomly drawn neighborhoods 3 from a possible 429 neighborhoods

in the Brisbane Division. The average population of the ACCS neighborhoods is about

6000 (Appendix 1) (for further information on the ACCS study design please see http://

www.uq.edu.au/accs). The ACCS neighborhoods are somewhat larger than census tracts in

the U.S., where the average size of the census tract is approximately 4000 inhabitants with

a minimum of around 1200 and a maximum of 8000 residents. Yet we note that research

examining the effects of neighborhood collective efficacy on perceptions of violence and

rates of violence have relied on much larger units of analysis. Sampson and his colleagues

employed neighborhood clusters with an average size of 8000 residents (Sampson 2012;

Sampson et al. 1997). In later analyses of the PHDCN data, these neighborhood clusters

were aggregated up to territorial communities with an average of 11,000 respondents. The

ecometric properties for these larger territorial communities were ‘‘virtually equivalent’’ to

the neighborhood clusters (Sampson 2012: 443).

Survey Process and Participant Sample

The ACCS surveys were conducted by the Institute for Social Science Research at the

University of Queensland. Trained interviewers utilized computer-assisted telephone

interviewing to administer the survey. The in-scope survey population comprised all

people aged 18 years or over who were usually resident in private dwellings with tele-

phones in the selected neighborhoods in Brisbane. 4 Particular focus was placed on con-

tacting those who had participated in previous waves. All participants were randomly

selected. Response rates over the three waves ranged from 36 to 43 %, and cooperation

rates ranged from 46 to 62 %. 5 Appendix 2 shows that the sample is similar to the Census

measures of these neighborhoods for several measures; however, the sample does over-

3 In Australia, the term ‘‘suburb’’ is used to refer to a feature that in the U.S. would be referred to as a

‘‘neighborhood’’. Throughout, we use the more familiar term ‘‘neighborhood’’ to refer to these. The suburbs in the ACCS sample include those that are adjacent to the main city center and those located in peri-urban areas which have experienced large increases in population growth. 4 In Australia, the number of mobile phone only users has only increased recently. 90 % of the population

was covered by landline phones in 2008, and in 2011 (the last wave of our sample) the number of mobile phone-only users was estimated to still be just 19 % (Australian Communications and Media Authority 2012). By comparison, in the US there were over 45 % mobile only users in 2014 (Blumberg and Luke, 2015). 5 In contrast to face to face surveys like those used in the PHDCN or the Los Angeles Family and

Neighborhood Study, phone response rates tend to be lower. This is true for the ACCS survey’s response rate. Yet the response rates for ACCS are on par with or indeed higher than other studies in Australia and the United States using phone contact (Duncan and Mummery 2005; Pickett et al. 2012; Lai, Zhao and Longmire, 2012; Larsen et al., 2004; Wood et al. 2012.

J Quant Criminol (2017) 33:783–808 791

123

represent females, home owners, English only speakers, older, married, highly educated,

and those who have not moved recently. We therefore account for these compositional

effects in our collective efficacy measure as described shortly, an approach that is

preferable to using survey weights (Winship and Radbill 1994). Information on the final

sample composition is provided in Appendix 1.

Additional Data Sources

Australian Bureau of Statistics (ABS) census data from 2006 to 2011 were merged with the

ACCS longitudinal study. Census data include a range of variables empirically derived

from the neighborhood effects literature (Bursik 1988; McMillan and Chavis 1986;

Sampson et al. 1997; Shaw and McKay 1942). As the Australian Geographical Classifi-

cation System changed substantially at the 2011 census, the ACCS team contracted the

ABS to provide all census data concorded to the 2006 census boundaries, allowing for

geographical consistency across our time periods.

The Queensland Police Service (QPS) provided crime incident information aggregated

to the suburb level for the Brisbane area. The Queensland Police Service (QPS) crime

incident data represents monthly counts of reported offences in all suburbs in South-East

Queensland from 2005 to 2013. In this paper we used crime incident data for violent crime

which includes homicide, other homicide (e.g. manslaughter), assaults and robbery.

Measures

Dependent Variable

Our key outcome measure is violent crime in the neighborhood. This measure is based on

official crime reports to the police, and is the violent crime rate per 1000 persons. To

smooth out yearly fluctuations in crime, the violent crime rate is averaged over the 2 years

nearest each survey wave. The violent crime count includes homicide, attempted murder,

other homicide, conspiracy to murder, manslaughter (excluding by driving), driving

causing death, grievous assault, assaults (excluding sexual), serious assault, serious assault

(other), common assault, armed robbery and robbery. Although official crime reports suffer

from under-reporting, a concern would be systematic under-reporting in certain neigh-

borhoods; however, Baumer (2002) tested and found no relationship between the rate of

reporting such serious violent events and key structural measures of neighborhoods,

including economic disadvantage.

Independent Variables

Our primary variable of interest is collective efficacy. We constructed a measure of col-

lective efficacy based on survey responses. This measure contains items that are identical to

those used in the PHDCN in the original study of collective efficacy (Sampson et al. 1997).

These items are reliable at both the individual (a = 0.75) and neighborhood level (a = 0.93). Approximately 18 % of the variation in this measure is between neighbor- hoods. Our measure of collective efficacy was adjusted for compositional effects. We

correct for individual-level biases by accounting for compositional effects in which

neighborhood assessments may be systematically affected by the characteristics of

respondents in the neighborhood. This measure was constructed based on factor scores

792 J Quant Criminol (2017) 33:783–808

123

from a maximum likelihood factor analysis, and then standardized factor scores were

constructed with mean of 0 and standard deviations of 1. 6 We estimated fixed effects

models in which the outcome measures were the previously computed factor score, and

included indicator variables for all neighborhoods, as well as several individual charac-

teristics that might systematically impact our measure of collective efficacy. 7 We then used

the estimated coefficients for each of the neighborhoods from this analysis as estimates of

the amount of collective efficacy in the neighborhood in the models once accounting for

these household characteristics.

Drawing on the social disorganization and collective action literatures, we included

several neighborhood level control variables based on census data from the Australian

Bureau of Statistics (ABS) from 2006 to 2011. We constructed a measure of residential

instability as the percentage new households in the last 5 years. Concentrated disadvan-

tage was constructed as a factor score combining three measures: median household

income; unemployment rate; percent one parent households. 8 A measure of language

heterogeneity was constructed as a Herfindahl index based on nine language groups. 9 To

capture those in the prime offender age group, we constructed a measure of the percent

aged 15–24. Table 1 presents the summary statistics for the variables used in the analyses

at each wave. To demonstrate the amount of variability in these measures over the study

period, the last two columns show: 1) the standard deviation of a measure of the difference

between wave 1 and wave 3 for each neighborhood for the variable of interest; 2) the ratio

of this standard deviation of change over the time period to the standard deviation of the

measure at a single time point (wave 3). There is a fair amount of change even over this

limited period of time, as the amount of variability is about � that across neighborhoods at a single point in time.

Analytic Approach

As mentioned earlier, we estimated three different models that each make different

assumptions about the temporal period in which these processes operate. All of the models

account for possible feedback effects. Two of our models were estimated as cross-lagged

6 Factor analysis provides specific weights to each of the variables that compose the measure, which are

analogous to an item response theory (IRT) approach; see Kamata and Bauer (2008) for the analytical proof that these approaches are identical. 7 This equation is: yij = a ? NjCN ? XijCX ? eij; where yij is the factor score of collective efficacy as

reported by the i-th respondent of I respondents in the j-th neighborhood, a is an intercept, Nj is an indicator of the neighborhood in which the respondent lives, CN is a vector of the effects of these neighborhoods on collective efficacy, X is a matrix of the exogenous household-level predictors, CX is a vector of the effects of these predictors on the subjective assessment, and eij is a disturbance term. The following individual level characteristics are included in the model: household income, education level, length of residence in the neighborhood, female, age, homeowner, marital status (single, widowed, divorced, and married as the reference category), presence of children, and speaking only English in the home. Previous research found very high correlations between measures using a frequentist approach, as we do here, and those using a Bayesian approach (see Steenbeek and Hipp, 2011, footnote 12 on page 846). 8 Note that using factor scores for some of our measures that are standardized to a mean value of 0 at each

time point is not problematic given that we are not estimating latent trajectory models attempting to capture change over time. Instead, our models are interested in marginal change in the outcome variable at a point in time given a marginal change in the covariate. Thus, the centering of the variables does not impact the substantive interpretation of our results, and is captured in the intercept terms estimated at each time point. 9 This measure was based on the following language groups: indigenous; East Asian; South-central Asian;

Southeast Asian; Southern Asian; Eastern European; Northern European; Southern European; other languages.

J Quant Criminol (2017) 33:783–808 793

123

simultaneous equation models (Berry 1984). This approach allows us to take into account

the possibility of autocorrelated error structures over time (by allowing correlations

between the error terms for each outcome variable in adjacent time periods), correlated

errors among measures at the same time point (by allowing correlations between the error

terms for each outcome variable at the same time point) and changing levels of the

outcome variables over time (by estimating a unique intercept value at each time point).

For each outcome variable, we can write the equation as:

y1t ¼ at þ B1Yt�1 þ B2WYt�1 þ B3Xt�1 þ e1t ð1Þ

where y1t is, for example, the violent crime rate which is measured at time t, at is an intercept at each time point, Yt-1 is a matrix of the endogenous variables in the model

(including violent crime, in this case) measured at the previous time point, B1 is a vector

that captures the effect of these other measures on the violent crime rate, WYt-1 is a matrix

of spatially lagged variables at the previous time point and B2 is a vector of parameters that

capture their effects on the violent victimization rate, X is a vector of other control

variables in the model at the previous time point whose effects are contained in the B3 matrix, and e1 is an error term with an assumed normal distribution. Given that we have three waves (t = 1–3) for the collective efficacy and crime data, the equations for col-

lective efficacy and violence appear two times (as they cannot be estimated for the first

wave as there are no t - 1 observations at that point) in our 2-year lag model (Fig. 1b). 10

In the other cross-lagged models—using just waves 1 and 3 as shown in Fig. 1c—the

equations for collective efficacy and violence appear just once. The three census outcome

variables (concentrated disadvantage, residential instability, and language heterogeneity)

10 The 2006 census measures are included as covariates in each of these equations, as they are clearly

temporally prior to the 2010 and 2012 survey waves.

Table 1 Summary statistics of variables used in analyses

Wave 1 Wave 2 Wave 3 Change W1–W3

Mean SD Mean SD Mean SD SD RLCV (%)

Dependent variable

Logged violent crime rate 0.549 0.373 0.572 0.379 0.532 0.367 0.139 37.9

Independent variables

Collective efficacy 0.008 0.140 0.019 0.149 0.017 0.157 0.087 55.1

Concentrated disadvantage 0.000 1.000 0.000 1.000 0.456 45.6

Residential instability 42.297 8.910 38.173 10.292 7.711 74.9

Language heterogeneity 0.181 0.098 0.124 0.108 0.067 62.2

Percent aged 15–24 13.972 2.829 14.028 2.878 1.596 55.4

Spatial lag variables

Concentrated disadvantage 0.000 1.000 0.000 1.000 0.396 39.6

Residential instability 42.571 3.781 38.137 5.298 3.237 61.1

Language heterogeneity 0.186 0.064 0.125 0.080 0.041 51.0

Percent aged 15–24 14.484 2.433 14.264 1.974 1.019 51.6

N = 148 neighborhoods. RLCV ratio of longitudinal to cross-sectional variance in measure as a percentage

794 J Quant Criminol (2017) 33:783–808

123

have one equation each in both models which regresses the 2011 measures on the wave 1

(2007) survey measures and 2006 Census measures.

Our other approach estimates these as simultaneous equations models (Berry 1984).

Although we estimate these with a maximum likelihood estimator in Stata 13.1 in which all

equations are estimated simultaneously in a single model, the intuition is identical to a two

stage least squares estimator using instrumental variables (for a complete discussion of

this, see Paxton et al. 2011). For example, we can estimate collective efficacy (y2) as a

function of violence (y1) at the same time period as:

y1t ¼ at þ C1y1t�1 þ C2WYt�1 þ C3Xt�1 þ e1t ð2Þ

y2t ¼ at þ b1y2t�1 þ b2ŷ1t þ B3WYt�1 þ B4Xt�1 þ e1t ð3Þ

The first equation regresses violence at time t on violence at time t - 1 and the control

variables (X) and the spatially lagged measures (WY). This provides an estimate of y1 (ŷ1)

at time t which is included in Eq. 3 (the structural equation of interest). Here, b1 captures the effect of collective efficacy at one time point on collective efficacy at the next time

point, and b2 estimates the simultaneous effect of violence on collective efficacy at time t. These equations assume that the only effect of violence on collective efficacy is within

the same time period; to be a suitable instrument, violence at time t - 1 must have no

effect on collective efficacy at time t (other than its effect on the level of violence at time

t). We are therefore assuming in this model that the effect of violence on collective efficacy

is relatively short-term and there is no additional effect of violence from 2 years ago on

collective efficacy once accounting for the current level of violence. A similar set of

equations capture the simultaneous effect of collective efficacy on violence, and analo-

gously assume that the level of collective efficacy 2 years ago does not impact violence

once accounting for the current level of collective efficacy. This model is identified; for a

discussion of this, see chapter 5 of (Finkel 1995).

To account for spatial lag effects, we first created a spatial weights matrix in which each

neighborhood was linked to all neighbors within 5 miles (weighted by an inverse distance

decay), and then computed spatially lagged measures by multiplying the values of key

measures in these neighborhoods by this weight matrix (row standardized). We then

temporally lagged these spatially lagged measures, which mirrors the approach adopted by

other studies (Hipp et al. 2009; Bernasco and Block 2011; Steenbeek and Hipp 2011).

Results

We first describe the results of the two wave cross-lagged model assuming a 2–3 year

period for the causal effect (Fig. 1b). Given the importance of collective efficacy in the

recent neighborhoods and crime literature, we begin by focusing on the equation with

collective efficacy as the outcome measure. In Table 2, we see strong stasis effects, as

neighborhoods with higher levels of collective efficacy at one time point have higher levels

of collective efficacy, on average, at the next time point (b = 0.72). There is no evidence

in this model of a negative feedback effect of crime on the level of collective efficacy, as

the coefficient for the violent crime rate is actually slightly positive, although not signif-

icant. Of the neighborhood characteristics, the one significant effect detected is for con-

centrated disadvantage: neighborhoods with one standard deviation higher concentrated

disadvantage at one time point have about .21 standard deviations lower collective efficacy

at the next time point (b = -0.032), controlling for the other measures in the model.

J Quant Criminol (2017) 33:783–808 795

123

Furthermore, higher levels of concentrated disadvantage in the surrounding area also lead

to lower collective efficacy at the next time point.

We next turn to the equation in which the violent crime rate is the outcome measure.

There is no evidence that neighborhoods with higher levels of collective efficacy at one

time point have lower levels of violent crime as reported by the police, as the coefficient is

actually positive. Note that these models are accounting for the violent crime rate at the

previous time point, and there is strong evidence that more violence at one time point is

associated with higher levels of violence at the next time point. There is, however, strong

evidence of a relationship between concentrated disadvantage and violence. Neighbor-

hoods with one standard deviation higher concentrated disadvantage at one time point have

0.24 standard deviations more violent crime (b = 0.088; b = 0.24) at the next time point. There is also a spatial effect in which more young adults in the surrounding area lead to

more violence at the next time point.

We next turn to the equations in which the neighborhood structural characteristics are

the outcome measures. These equations test whether there are feedback effects from

Table 2 Three-wave longitudinal models for Brisbane, using combined collective efficacy measure, including spatial lag measures

(1) (2) (3) (4) (5) Collective efficacy

Violent crime

Concentrated disadvantage

Residential instability

Language heterogeneity

Collective efficacy (t - 1) 0.720** (14.13)

0.254* (2.15)

-0.521� -(1.65)

-7.660 -(1.14)

-0.061 -(1.30)

Violent crime (t - 1) 0.011 (1.55)

0.822** (37.35)

0.567** (9.15)

-1.055 -(0.79)

-0.003 -(0.33)

Concentrated disadvantage (t - 1)

-0.032** -(4.79)

0.088** (5.11)

0.704** (15.75)

0.638 (0.67)

0.003 (0.50)

Residential instability (t - 1)

0.000 -(0.29)

-0.003� -(1.93)

0.003 (0.83)

0.762** (10.46)

0.001* (2.47)

Language heterogeneity (t - 1)

-0.054 -(0.91)

-0.160 -(0.99)

-0.079 -(0.23)

-7.196 -(0.99)

0.544** (10.72)

Percent aged 15–24 (t - 1)

-0.003 -(1.40)

0.007 (1.54)

0.010 (0.71)

0.322 (1.14)

0.006** (3.26)

Population density (t - 1) 0.001 (1.59)

-0.002 -(1.06)

0.006 (1.07)

-0.111 -(1.00)

0.000 -(0.51)

Nearby concentrated disadvantage (t - 1)

-0.012* -(2.16)

-0.006 -(0.42)

0.059 (1.58)

-0.205 -(0.26)

0.016** (2.88)

Nearby residential instability (t - 1)

-0.001 -(0.96)

-0.005� -(1.71)

-0.007 -(0.79)

0.315 (1.61)

-0.003* -(2.38)

Nearby language heterogeneity (t - 1)

-0.112 -(1.26)

0.031 (0.13)

-0.732 -(1.30)

5.190 (0.44)

0.573** (6.93)

Nearby aged 15–24 (t - 1) -0.001 -(0.24)

0.015* (2.08)

-0.042* -(2.43)

0.302 (0.82)

0.003 (1.22)

Intercept 0.097� 0.096 0.565 -14.048 0.125�

(1.86) (0.71) (1.23) -(1.43) -(1.84)

R-square 0.74 0.90 0.89 0.56 0.82

Testing cross-lagged effects of collective efficacy and violence

T-values in parentheses. N = 148 neighborhoods

** p \ 0.01(two-tail test), * p \ 0.05 (two-tail test), � p \ 0.05 (one-tail test)

796 J Quant Criminol (2017) 33:783–808

123

violence or collective efficacy on these measures. In the equation with concentrated dis-

advantage as the outcome measure, we find evidence that neighborhoods with one standard

deviation higher violent crime at one time point have 0.32 standard deviations more

concentrated disadvantage at the next time point. Thus, there appears to be a reciprocal

relationship between concentrated disadvantage and violence in that both are impacting

each other over time in this longitudinal analysis. We also see evidence in these models

that neighborhoods with higher levels of collective efficacy at one time point have

somewhat lower levels of concentrated disadvantage at the next time point (b = -0.521, p \ 0.10). Whereas higher levels of collective efficacy do not appear to impact the level of crime at the next time point directly, they do impact it indirectly by reducing the level of

concentrated disadvantage (presumably through a selective mobility process). And

neighborhoods surrounded by fewer young adults have higher levels of disadvantage at the

next time point.

In the equation with residential instability as the outcome measure, we see no evidence

that the level of violence, collective efficacy, or concentrated disadvantage impact the level

of residential instability at the next time point. Finally, in the equation with language

heterogeneity as the outcome measure, there is also no evidence that the level of violence,

collective efficacy, or concentrated disadvantage impact this characteristic of neighbor-

hoods. However, neighborhoods with more residential instability and more young adults

experience larger increases in language heterogeneity by the next time point. Neighbor-

hoods surrounded by higher levels of language heterogeneity are more likely to experience

larger increases in language heterogeneity themselves. This indicates that there are spa-

tially clustered locations of immigrants with higher levels of language heterogeneity, with

increasing language heterogeneity over the period of this study. The fact that neighbor-

hoods surrounded by more concentrated disadvantage and residential stability also expe-

rience an increase in language heterogeneity suggests that this broader spatial pattern

impacts the location of where these immigrant communities develop.

Additional Temporal Patterns

We next assess the extent to which the results are different when assuming either a shorter

temporal causal relationship (Table 3) or a longer temporal causal relationship (Table 4).

In general, we find that our results are generally quite robust regardless of the assumption

about the temporal causal period. For example, in none of the three models is there

evidence of a negative feedback effect of violence on the level of collective efficacy, as the

coefficient for violent crime is not significant in Tables 2, 3, and 4. Instead, concentrated

disadvantage has an important impact on collective efficacy, as higher levels of concen-

trated disadvantage in the neighborhood or the surrounding area lead to lower collective

efficacy regardless of the model specification.

In the equation in which violence is the outcome measure, there is no evidence in any of

these three models that neighborhoods with higher levels of collective efficacy result in

lower levels of violence as reported by the police. In fact, the coefficient is significantly

positive in the 2-year cross-lagged model (Table 2) and the simultaneous effect model

(Table 3) and positive, but not significant, in the 5-year cross-lagged model (Table 4).

Despite the body of evidence finding a negative relationship between collective efficacy

and violence in cross-sectional analyses in Australia (Mazerolle et al. 2010) and elsewhere

(Sampson et al. 1997; Sampson and Wikström 2008), we find no evidence for this rela-

tionship when assessing it longitudinally regardless of the temporal assumption of the

J Quant Criminol (2017) 33:783–808 797

123

causal relationship. There is, however, strong evidence in all three models of a positive

relationship between concentrated disadvantage and violence.

In all three models, higher levels of violence lead to higher levels of concentrated

disadvantage. Thus, we see a robust reciprocal relationship between concentrated disad-

vantage and violence in that both are impacting each other over time—regardless of the

specified temporal period—in this longitudinal analysis. There is some evidence that

higher levels of collective efficacy lead to lower levels of concentrated disadvantage,

although this process may occur over a longer period as the effect is only significant at

p \ 0.05 in the five year lag model. And there is a positive relationship between more 15–24 year olds in the surrounding area and concentrated disadvantage in the neighbor-

hood when assuming a simultaneous or two-year causal relationship; the effect is weaker

when assuming a 5-year causal relationship.

Table 3 Three-wave longitudinal models for Brisbane, using combined collective efficacy measure, including spatial lag measures

(1) (2) (3) (4) (5) Collective efficacy

Violent crime

Concentrated disadvantage

Residential instability

Language heterogeneity

Collective efficacy (t) 0.367* (2.03)

Collective efficacy (t - 1) 0.711** (13.63)

-0.517 -(1.64)

-7.819 -(1.16)

-0.061 -(1.30)

Violent crime (t) 0.014 (1.44)

Violent crime (t - 1) 0.816** (36.03)

0.565** (9.11)

-1.077 -(0.81)

-0.003 -(0.35)

Concentrated disadvantage (t - 1)

-0.033** -(4.68)

0.100** (4.65)

0.705** (15.82)

0.640 (0.68)

0.003 (0.52)

Residential instability (t - 1)

0.000 -(0.20)

-0.003� -(1.94)

0.003 (0.81)

0.762** (10.43)

0.001* (2.46)

Language heterogeneity (t - 1)

-0.054 -(0.91)

-0.142 -(0.86)

-0.081 -(0.24)

-7.261 -(1.00)

0.543** (10.71)

Percent aged 15–24 (t - 1)

-0.003 -(1.45)

0.008� (1.66)

0.010 (0.72)

0.320 (1.14)

0.006** (3.26)

Population density (t - 1) 0.001 (1.60)

-0.002 -(1.18)

0.005 (1.05)

-0.113 -(1.02)

0.000 -(0.53)

Nearby concentrated disadvantage (t – 1)

-0.012* -(2.15)

0.000 -(0.02)

0.059 (1.59)

-0.211 -(0.27)

0.016** (2.88)

Nearby residential instability (t - 1)

-0.001 -(0.92)

-0.005 -(1.56)

-0.007 -(0.78)

0.315 (1.61)

-0.003* -(2.38)

Nearby language heterogeneity (t - 1)

-0.108 -(1.20)

0.062 (0.25)

-0.727 -(1.30)

5.304 (0.45)

0.574** (6.94)

Nearby aged 15–24 (t - 1)

-0.001 -(0.33)

0.015* (2.06)

-0.042* -(2.41)

0.309 (0.84)

0.003 (1.23)

Intercept 0.098� (1.86)

0.068 (0.46)

0.559 (1.21)

-14.101 -(1.43)

-0.126� -(1.84)

R-square 0.74 0.90 0.89 0.56 0.82

Testing simultaneous effects of collective efficacy and violence

T values in parentheses. N = 148 neighborhoods

** p \ 0.01 (two-tail test), * p \ 0.05 (two-tail test), � p \ 0.05 (one-tail test)

798 J Quant Criminol (2017) 33:783–808

123

Whereas there is no evidence that that the level of violence, collective efficacy, or

concentrated disadvantage impact language heterogeneity in neighborhoods, neighbor-

hoods with more residential instability and more young adults experience larger increases

in language heterogeneity in all three models. Likewise, neighborhoods surrounded by

higher levels of language heterogeneity, concentrated disadvantage, and residential sta-

bility tend to experience larger increases in language heterogeneity. There is no evidence in

any of these models that the level of violence, collective efficacy, or concentrated disad-

vantage impacts the level of residential instability in the neighborhood. 11

Table 4 Two-wave longitudinal models for Brisbane, using combined collective efficacy measure, including spatial lag measures

(1) (2) (3) (4) (5) Collective efficacy

Violent crime

Concentrated disadvantage

Residential instability

Language heterogeneity

Collective efficacy (t - 2) 0.626** (8.74)

0.130 (0.65)

-0.719* -(2.45)

-0.274 -(0.04)

-0.050 -(1.13)

Violent crime (t - 2) 0.006 (0.55)

0.768** (25.29)

0.437** (9.80)

0.352 (0.36)

0.003 (0.38)

Concentrated disadvantage (t - 1)

-0.037** -(3.90)

0.094** (3.56)

0.728** (18.68)

1.052 (1.24)

0.005 (0.83)

Residential instability (t - 1)

-0.001 -(0.66)

-0.003 -(1.41)

-0.001 -(0.28)

0.783** (11.09)

0.001** (2.67)

Language heterogeneity (t - 1)

-0.124 -(1.51)

-0.187 -(0.82)

-0.131 -(0.39)

-6.488 -(0.88)

0.547** (10.78)

Percent aged 15–24 (t - 1)

-0.006� -(1.85)

0.002 (0.25)

0.008 (0.60)

0.351 (1.23)

0.006** (3.09)

Population density (t - 1) 0.001 (1.06)

-0.001 -(0.30)

0.003 (0.59)

-0.126 -(1.16)

-0.001 -(0.89)

Nearby concentrated disadvantage (t - 1)

-0.020* -(2.24)

-0.012 -(0.48)

0.064� (1.75)

-0.136 -(0.17)

0.015** (2.78)

Nearby residential instability (t - 1)

-0.001 -(0.64)

-0.012� -(1.93)

-0.011 -(1.26)

0.369� (1.90)

-0.003* -(2.19)

Nearby language heterogeneity (t - 1)

-0.133 -(0.98)

-0.147 -(0.39)

-0.829 -(1.50)

5.777 (0.48)

0.584** (6.98)

Nearby aged 15–24 (t - 1) 0.004 (1.08)

0.014 (1.22)

-0.027 -(1.63)

0.406 (1.13)

0.004� (1.71)

Intercept 0.141 (1.37)

0.515� (1.79)

0.754� (1.79)

-19.931* -(2.17)

-0.155* -(2.44)

R-square 0.78 0.87 0.91 0.55 0.83

Testing cross-lagged effects of collective efficacy and violence between waves 1 and 3

T values in parentheses. N = 148 neighborhoods

** p \ 0.01 (two-tail test), * p \ 0.05 (two-tail test), � p \ 0.05 (one-tail test)

11 Regarding model fit, for the two-year lag model (Table 2), the v2 of 173.2 on 73 df (p \ 0.01) implies a

nonperfect fit, although the RMSEA of 0.098 and the CFI of 0.945 suggest a reasonable approximate fit for this model. Simulation studies have shown that model fit can be impacted by various characteristics of the model, and therefore strictly employing cutoff values is not wise; nonetheless, rough guidelines includes RMSEA values below 0.08 and CFI values above 0.95 (Hu and Bentler 1999). Inspection of modification indices suggested only that estimating the effect of lagged violence on current violence should be freed over the two waves; when freeing this path, or constraining the error covariances at the same time point to be zero

J Quant Criminol (2017) 33:783–808 799

123

Sensitivity Analyses

In our main analytic models we focused on violent crime as a neighborhood outcome. A

criticism is that this measure combines not only homicides, robberies, and aggravated

assaults, but also simple assaults. Given the evidence of Baumer (2002) that simple

assaults may be more susceptible to reporting bias, we estimated additional models for

three crime types separately: aggravated assaults, robberies, and simple assaults (there

were too few homicides to provide stable estimates). The general pattern of results was

very similar to those already presented, and similar across these three separate crime types.

There were only two differences worth noting. First, in both the cross-lagged and simul-

taneous models collective efficacy had a weaker positive relationship with the subsequent

level of these three individual crime types compared to the relationship with violent crime

in general in the main models. Second, in these ancillary models collective efficacy had a

stronger negative relationship with subsequent concentrated disadvantage for all three of

these separate crime type models. Thus, we see no evidence that our results are impacted

by combining these different crime types into a single measure of violence.

Discussion

This study has contributed to the neighborhoods and crime literature by exploring the

relationship between concentrated disadvantage, collective efficacy, and violence both

longitudinally and spatially. Whereas there have been numerous studies exploring these

characteristics of neighborhoods in cross-sectional designs, a longitudinal strategy is

crucial for assessing whether collective efficacy indeed operates to reduce crime. As it is

also plausible that levels of violence may reduce collective efficacy over time, teasing apart

these possible reciprocal relationships is necessary. Given the lack of existing theoretical

specificity about the time scale of these processes, an important contribution of this study

was exploring three different model specifications of the temporal pattern of this rela-

tionship, rather than simply assuming that one is correct. The study was also able to

account for the possible spatial patterning of these processes, and assess whether the

composition of these measures nearby was related to changes in these measures in the

neighborhood itself over time. This study failed to find any evidence that higher levels of

collective efficacy in a neighborhood at one point in time are directly associated with

greater decreases in violence over time. There was some evidence, however, that con-

centrated disadvantage plays an important role in linking the level of collective efficacy in

a neighborhood with the level of violence.

Footnote 11 continued (given their nonsignificance), the model fit only improved somewhat (RMSEA = 0.081, CFI = 0.958), and, most importantly, the substantive results remained unchanged. It is important to recall the insights of Browne and colleagues (Browne, MacCallum, Kim, Andersen, and Glaser 2002) that model fit will be negatively impacted due to high statistical power when the R-squares of the equations are quite high, as they are here, ranging from 0.56 to 0.90. In the simultaneous effects model (Table 3) the fit was similar: v2 of 174.1 on 72 df (p \ 0.01) implies a nonperfect fit, although the RMSEA of 0.10 and the CFI of 0.944 suggest a reasonable approximate fit for this model. Freeing the lagged violence effect and constraining the error covariances at the same timepoint to zero again resulted in modest improvement to model fit, but with the substantive results remaining unchanged. The 5-year lag model (Table 4) was exactly identified, and therefore model fit could not be assessed.

800 J Quant Criminol (2017) 33:783–808

123

A contribution of this study was exploring three possible time periods for these causal

processes to operate. Given that existing theories are typically quite unclear on the tem-

poral period over which these various processes ought to operate, we adopted an approach

that tested three different possible lags rather than simply assuming that one of them was

correct. It was reassuring to see that our results were quite robust over these three different

temporal specifications. Thus, the relationship between violence and concentrated disad-

vantage was relatively similar regardless whether we used a simultaneous model, a 2-year

lag, or a 5-year lag. Nonetheless, certain relationships appeared to operate only over a

longer time scale: for example, the relationship between higher levels of collective efficacy

at one time point at subsequently lower concentrated disadvantage was only statistically

significant in the five-year lag model. Although our results were generally robust over these

different temporal model specifications, the ecological crime literature would be well

served to carefully consider the time period of these proposed causal relationships in future

work.

One notable finding was the lack of a direct relationship between collective efficacy and

violence over time. Whereas prior research has often found a cross-sectional relationship

between collective efficacy and violence—including research using ACCS data (Mazerolle

et al. 2010)—no such effect was found here in these longitudinal analyses. This is an

important finding, given that cross-sectional models are a particularly weak assessment of

potential causal relationships. We similarly found no relationship when estimating models

assuming a longer or a shorter time period for this causal relationship. Although only a

single study of one particular urban area, it does raise questions about the causal impact of

this relationship and imply the need for additional longitudinal analyses. We attempted to

assess different causal lags of this possible relationship, given the discussion of (Sampson

2012) regarding the distinction between enduring effects and situational effects of col-

lective efficacy. Our simultaneous equations model was attempting to capture situational

effects in which residents are able to intervene when observing possible instances of crime

or disorder. Our 2- and 5-year lag models were attempting to capture enduring effects of

collective efficacy.

Although a robust finding in the existing literature is the importance of concentrated

disadvantage for levels of violence in neighborhoods, this study reinforces the importance

of this reciprocal relationship in a longitudinal framework. In our study, higher levels of

concentrated disadvantage had one of the strongest relationships with violence at later

time points. This is consistent with longitudinal studies of this relationship in the U.S.

(e.g., Hipp 2010; Kubrin and Herting 2003). But at the same time, we found that higher

levels of violence were the strongest predictor of increased concentrated disadvantage at

the next time point. This reciprocal relationship is consistent with research from the U.S.

(Hipp 2010) and implies a downward spiral for these neighborhoods. This reciprocal

relationship was robust over our three different specifications of the temporal period of

these processes.

It is interesting to note that we also detected a strong reciprocal relationship between

collective efficacy and concentrated disadvantage over time, and there was a spatial pat-

terning to this relationship. Concentrated disadvantage and collective efficacy in neigh-

borhoods tended to move in opposite directions over time. Neighborhoods with more

concentrated disadvantage at one point in time experienced a subsequent decrease in

collective efficacy, and this relationship was robust to our three different model specifi-

cations of the temporality of this process. Furthermore, concentrated disadvantage in the

surrounding area had an additional negative effect on residents’ sense of collective effi-

cacy, implying that spatially disadvantaged neighborhoods are particularly hard hit

J Quant Criminol (2017) 33:783–808 801

123

regarding their sense of an ability to work together. The fact that neighborhoods with

higher levels of collective efficacy experienced lower levels of disadvantage over time has

not been detected previously in the literature but speaks to the enduring effects of col-

lective efficacy’s presence, or absence (Sampson 2012); we emphasize that this relation-

ship was not statistically significant in the models specifying a shorter time lag, but was

only significant in the five-year lag model. This suggests that this may be a process that

plays out more slowly, implying that it may not be detected in studies focusing on shorter

time periods. Conversely, it appears that neighborhoods with lower collective efficacy

experience larger increases in concentrated disadvantage over time, which then leads to

greater increases in violence. This implies a possible different mechanism through which

collective efficacy may impact crime in neighborhoods: a differential mobility pattern in

the neighborhood that increases concentrated disadvantage, and then subsequently more

violence. This indicates a dynamic relationship that requires further empirical exploration.

Another notable finding was the substantial clustering of ethnic diversity detected over

time. Although ethnic clustering has been detected in other settings such as the U.S., it

has only been detected in Australia within Sydney and Melbourne. Indeed, prior to the

2011 census Brisbane had the lowest concentration of immigrants when compared to

Sydney, Melbourne and Perth (Markus et al. 2009). But in our study, neighborhoods

surrounded by higher concentrations of immigrants were more likely to see an increase in

immigrants themselves. But another interesting non-pattern was detected: this increased

clustering of immigrants is not associated with negative consequences over time of any

sort. Furthermore, there is no evidence that neighborhoods with more language hetero-

geneity experience greater increases in violent crime or greater decreases in collective

efficacy over time.

We acknowledge four limitations of the current research design. First, although lon-

gitudinal data allows for exploring how these patterns change over time, it is nonetheless

the case that we were constrained to discrete observation points that may or may not match

the true causal process (Taylor 2015). We therefore used three different temporal speci-

fications to assess the robustness of the results. Second, we were constrained to measuring

these processes at the larger unit of analysis of neighborhoods. Although a growing body of

research focuses on street segments (Groff et al. 2010; Weisburd et al. 2004), we were

limited to larger units given the challenge of collecting information on levels of collective

efficacy in such small units (Weisburd et al. 2012). Third, our obtained sample over-

represented certain types of households in the neighborhood and therefore may have

affected our collective efficacy measure. We attempted to account for this with an eco-

metrics approach that corrects for the biases of certain types of persons (Sampson and

Raudenbush 1999), but this possible limitation should be kept in mind. Fourth, although we

had a measure of collective efficacy, we had no measures of actual informal social control

behavior in these neighborhoods to assess whether it might impact violence over time

(Steenbeek and Hipp 2011). This is a limitation of many studies in the collective efficacy

literature, as they are reliant upon the assumption that these perceptions of collective

efficacy indeed lead to collective action on the part of residents. It is possible that the more

proximate mechanism associated with crime rates is what residents do when presented with

a problem in their neighborhood.

Although we did not find a direct relationship between collective efficacy and violence

over time, existing theory presumes that shared perceptions of collective efficacy lead to

informal actions to reduce crime. For example, Reynald (2009) argues that although res-

idents may be available and willing to supervise their environment, they must also able to

take direct action (such as contacting the police or directly intervening to stop the incident)

802 J Quant Criminol (2017) 33:783–808

123

to prevent crime. We believe unpacking the relationship between collective efficacy, cit-

izen oriented actions and crime, and how this might unfold over time, is the next frontier

for ecology of crime scholarship.

Acknowledgments This work was supported by the Australian Research Council (LP0453763; fDP0771785; RO700002; DP1093960; DP1094589 and DE130100958).

Appendix 1

See Table 5 and Fig. 2.

Table 5 Resident population size for ACCS suburbs and sample size by wave and cohort

Average no. of residents

Min Max ACCS longitudinal sample N

Top-up sample N

Total sample N

ACCS wave 2

5690 245 20,999 1077 3247 4324

ACCS wave 3

6046 241 21,001 2286 1935 4221

ACCS wave 4

6633 258 22,807 2473 1659 4132

c1 c2 ε1 c3 ε2

d1 d2 ε3 d3 ε4

ce1 ce2 ε5 ce3 ε6

Fig. 2 Full path model showing crime (c), disadvantage (d), and collective efficacy (ce)

J Quant Criminol (2017) 33:783–808 803

123

Appendix 2

See Table 6.

Table 6 ABS census and ACCS sample demographics

Demographics Census 2006

Census 2011

ACCS wave 2

ACCS wave 3

ACCS wave 4

Age 46.0 years 46.3 years 49.9 years 51.2 years 53.4 years

Gender (male) 49.1 % 49.3 % 39.8 % 40.8 % 40.1 %

Home ownership (own) 66.5 % 65.3 % 85.8 % 86.8 % 87.9 %

Aboriginal and Torres Strait Islander

1.9 % 2.3 % 1.5 % 0.9 % 1.1 %

Language other than english

English only 85.3 % 83.7 % 93.3 % 89.0 % 93.8 %

Non English 8.9 % 11.0 % 6.7 % 11.0 % 6.2 %

Country of birth

Born in Australia 73.4 % 72.4 % 75.5 % 75.9 % 75.6 %

Employment

Employed (full time) 39.7 % 39.0 % 42.0 % 38.5 % 37.1 %

Employed (part time) 17.5 % 18.1 % 20.5 % 19.2 % 18.3 %

Unemployed 2.7 % 3.7 % 2.1 % 3.1 % 3.5 %

University education 11.7 % 13.7 % 31.2 % 35.5 % 33.9 %

Different address 5 years ago 42.1 % 37.6 % 25.5 % 23.3 % 17.1 %

Married 48.1 % 47.3 %* 63.8 % 66.9 % 66.6 %

Income

Median household income (yearly)

$58,953 $75,414 $60,000– $79,999

$60,000– $79,999

$80,000– $99,999

Religion

Buddhism 1.7 % 1.9 % 1.7 % 1.7 % 0.6 %

Christianity 64.7 % 62.6 % 63.2 % 67.6 % 70.5 %

Hinduism 0.5 % 0.9 % 1.3 % 1.4 % 0.5 %

Islam 0.6 % 1.0 % 1.6 % 1.8 % 0.6 %

Judaism 0.1 % 0.1 % 0.8 % 0.2 % 0.1 %

Other religion 0.5 % 0.8 % 1.0 % 0.9 % 1.1 %

No religion 19.3 % 22.9 % 30.4 % 26.3 % 26.4 %

804 J Quant Criminol (2017) 33:783–808

123

Appendix 3

See Table 7.

Appendix 4

See Table 8.

Table 7 Comparison of neighborhood characteristics for ACCS suburbs compared to state and national averages

2001 2006 2011

ACCS QLD AUST ACCS QLD AUST ACCS QLD AUST

% Unemployed 4.8 % 8.2 % 7.4 % 2.8 % 4.7 % 5.2 % 3.8 % 6.1 % 5.6 %

% Renting 25.2 % 31.6 % 27.6 % 23.0 % 30.0 % 27.2 % 23.1 % 32.0 % 28.7 %

% ATSI 1.5 % 3.1 % 2.2 % 1.6 % 3.2 % 2.3 % 1.9 % 3.6 % 2.5 %

% Born overseas 20.8 % 16.9 % 21.6 % 26.9 % 17.58 % 22.0 % 24.4 % 20.2 % 24.4 %

% LOTE 8.7 % 6.9 % 15.0 % 10.0 % 7.6 % 15.7 % 12.4 % 9.3 % 18.0 %

Median household income

$929 $739 $784 1225 $1036 $1029 $1559 $1227 $1230

Table 8 Items from the collective efficacy scale from waves 2, 3 and 4 of the ACCS

The collective efficacy adapted from the project of human development in Chicago neighborhoods

Willingness to intervene If a group of community children were skipping school… If some children were spray painting graffiti… If there was a fight in front of your house If a child was showing disrespect… Suppose that because of budget cuts the fire station… Response categories Very unlikely, unlikely, neither likely/unlikely, likely, very likely

Social cohesion and trust People in this community are willing to help their neighbors This is a close-knit community People in this community can be trusted People in this community do not share the same values Response categories Strongly disagree, disagree, neither agree nor disagree, agree, strongly agree

J Quant Criminol (2017) 33:783–808 805

123

References

Australian Bureau of Statistics (2012). Greater Brisbane (QLD) (Greater Capital City Statistical Division), retrieved from http://www.censusdata.abs.gov.au/census_services/getproduct/census/2011/community profile/3GBRI?opendocument&navpos=220

Australian Communications and Media Authority (2012) Communications report 2011–12. ACMA, Canberra

Bandura Albert (1995) Self efficacy in changing communities. Cambridge University Press, New York Bandura Albert (1997) The exercise of control. W.H. Freeman & Company, New York Bandura A (2001) Social cognitive theory: an agentic perspective. Annu Rev Psychol 52:1–26 Baumer EP (2002) Neighborhood disadvantage and police notification by victims of violence. Criminology

40:579–616 Bernasco W, Block R (2011) Robberies in Chicago: a block-level analysis of the influence of crime

generators, crime attractors, and offender anchor points. J Res Crime Delinq 48:33–57 Berry WD (1984). In: Niemi RG (ed) Nonrecursive causal models, vol 37. Sage, Beverly Hills Bingenheimer JB, Robert TB, Earls FJ (2005) Firearm violence exposure and serious violent behavior.

Science 308:1323–1326 Blumberg SJ, Luke JV (2015) Wireless substitution: early release of estimates from the national health

interview survey, National Center for Health Statistics. July–December 2014. Retrieved from: http:// www.cdc.gov/nchs/data/nhis/earlyrelease/wireless201506.pdf

Boggess LN, Hipp JR (2010) Violent crime, residential instability and mobility: does the relationship differ in minority neighborhoods? J Quant Criminol 26:351–370

Browne MW, MacCallum RC, Kim CT, Andersen BL, Glaser R (2002) When fit indices and residuals are incompatible. Psychol Methods 7:403–421

Browning CR (2002) The span of collective efficacy: extending social disorganization theory to partner violence. J Marriage Family 64:833–850

Browning CR, Cagney KA (2002) Neighborhood structural disadvantage, collective efficacy, and self-rated physical health in an urban setting. J Health Soc Behav 43:383–399

Browning CR, Leventhal T, Brooks-Gunn J (2005) Sexual initiation in early adolescence: the nexus of parental and community control. American Soc Rev 70:758–778

Bursik RJ (1988) Social disorganization and theories of crime and delinquency. Criminology 26:519–551 Bursik RJ (1999) The informal control of crime through neighborhood networks. Sociol Focus 32:85–97 Bursik RJ, Grasmick HG (1993) Neighborhoods and crime. Lexington Books, New York Duncan M, Mummery K (2005) Psychosocial and environmental factors associated with physical activity

among city dwellers in regional Queensland. Prev Med 40(4):363–372 Felson M (2002) Crime and everyday life. Sage, Thousand Oaks Finkel SE (1995). In: Lewis-Beck MS (ed) Causal analysis with panel data, Sage, Thousand Oaks Franzini L, Caughy M, Spears W, Esquer MEF (2005) Neighborhood economic conditions, social processes

and self-rated health in low-income neighborhoods in Texas: a multilevel latent variables model. Soc Sci Med 61:1135–1150

Groff ER, Weisburd D, Yang SM (2010) Is it important to examine crime trends at a local ‘‘Micro’’ level?: a longitudinal analysis of street to street variability in crime trajectories. J Quant Criminol 26:7–32

Hipp JR (2010) A dynamic view of neighborhoods: the reciprocal relationship between crime and neigh- borhood structural characteristics. Soc Probl 57:205–230

Hipp JR, Tita GE, Greenbaum RT (2009) Drive-bys and trade-ups: examining the directionality of the crime and residential instability relationship. Soc Forces 87:1777–1812

Hu L, Bentler PM (1999) Cutoff criteria for fit indexes in covariance structure analysis: conventional criteria versus new alternatives. Struct Equ Model 6:1–55

Hugo G, Helen F, George T (2013) Internal migration and regional Australia, 2006, 2011. Australian population and migration research centre policy brief. Retrieved from http://www.adelaide.edu.au/ apmrc/pubs/policy-briefs/APMRC_Policy_Brief_Vol_1_6_2013.pdf

Hunter A (1985) Private, parochial and public social orders: the problem of crime and incivility in urban communities. In: Gerald D, Suttles G, Mayer N, Zald N (eds) The challenge of social control: citi- zenship and institution building in modern society. Ablex Publishing, NJ

Kamata A, Bauer DJ (2008) A note on the relationship between factor analytic and item response theory models. Struct Equ Model 15:136–153

Kornhauser R (1978) Social sources of delinquency. University of Chicago Press, Chicago Kubrin CE, Herting JR (2003) Neighborhood correlates of homicide trends. Sociol Q 44:329–355 Kubrin CE, Hipp JR (2014) Do fringe banks create fringe neighborhoods? Examining the spatial relationship

between fringe banking and neighborhood crime rates. Justice Q

806 J Quant Criminol (2017) 33:783–808

123

Lai Y-L, Zhao S, Longmire D (2012) Specific crime-fear linkage: the effect of actual burglary incidents reported to the police on residents’ fear of burglary. J Crime J 35(1):13. doi:10.1080/0735648X.2011. 631408

Land KC, McCall PL, Cohen LE (1990) Structural covariates of homicide rates: are there any invariances across time and social space? Am J Sociol 95:922–963

Larsen L, Harlan SL, Bolin B, Hackett EJ, Hope D, Kirby A, Nelson A, R TR, Wolf S (2004) Bonding and bridging: understanding the relationship between social capital and civic action. J Plan Educ Res 24(1):64–77

MacDonald JM, Hipp JR, Gill C (2013) Neighborhood effects of immigrant concentration on changes in neighborhood crime rates. J Quant Criminol 29:191–215

Maimon D, Browning CR (2010) Unstructured socializing, collective efficacy, and violent behavior among urban youth. Criminology 48:443–474

Markus A, Jupp J, McDonald Pr (2009) Australia’s immigration revolution. Allen & Unwin, Australia Mazerolle L, Wickes R, McBroom J (2010) Community variations in violence: the role of social ties and

collective efficacy in comparative context. J Res Crime Delinq 47:3–30 McMillan DW, Chavis DM (1986) Sense of community: a definition and theory. J Commun Psychol

14:6–23 Mears DP, Bhati AS (2006) No community is an island: the effects of resource deprivation on urban

violence in spatially and socially proximate communities. Criminology 44:509–548 Miethe TD, Meier RF (1994) Crime and its social context: toward an integrated theory of offenders, victims,

and situations. State University of New York Press, New York Morenoff JD, Sampson RJ, Raudenbush SW (2001) Neighborhood inequality, collective efficacy and the

spatial dynamics of urban violence. Criminology 39:517–560 Pattillio ME (1998) Sweet mothers and gangbangers: managing crime in a black middle-class neighborhood.

Soc Forces 76:747–774 Paxton P, Hipp JR, Marquart-Pyatt S (2011) Nonrecursive models: endogeneity, reciprocal relationships,

and feedback loops. Sage, Los Angeles Peterson RD, Krivo LJ (2010) Divergent social worlds: neighborhood crime and the racial-spatial divide.

Russell Sage Foundation, New York Pickett JT, Ted C, Kristin MG, Marc G (2012) Reconsidering the relationship between perceived neigh-

borhood racial composition and whites’ perceptions of victimization risk: do racial stereotypes matter? Criminology 50(1):145–186

Rankin BH, Quane JM (2002) Social contexts and urban adolescent outcomes: the interrelated effects of neighborhoods, families, and peers on African-American youth. Soc Probl 49:79–100

Reynald DM (2009) Guardianship in action: developing a new tool for measurement. Crime Prev Commun Saf 11:1–20

Sampson RJ (1988) Local friendship ties and community attachment in mass society: a multilevel systemic model. Am Sociol Rev 53:766–779

Sampson RJ (2002) Transcending tradition: new directions in community research, Chicago style. Crimi- nology 40:213–230

Sampson RJ (2006) Collective efficacy theory: lessons learned and directions for future inquiry. In: Cullen FT, Wright J, Blevins K (eds) Taking stock: the status of criminological theory advances in crimi- nological theory. Transaction Publishers, New Brunswick

Sampson RJ (2012) Great American city: Chicago and the enduring neighborhood effects. The University of Chicago Press, Chicago

Sampson RJ, Groves WB (1989) Community structure and crime: testing social-disorganization theory. Am J Sociol 94:774–802

Sampson RJ, Wikström POH (2008) The social order of violence in Chicago and Stockholm neighborhoods: a comparative inquiry. In: Stathis N (ed) Order, conflict and violence. Cambridge University Press, Kalyvas

Sampson RJ, Raudenbush SW, Earls F (1997) Neighborhoods and violent crime: a multilevel study of collective efficacy. Science 277:918–924

Sampson RJ, Morenoff JD, Earls F (1999) Beyond social capital: spatial dynamics of collective efficacy for children. Am Sociol Rev 64:633–660

Shaw C, McKay H (1942) Juvenile delinquency and urban areas. University Press, Chicago Skogan WG (1986) Fear of crime and neighborhood change. In: Albert J (ed) Communities and crime.

University of Chicago Press, Reiss Skogan WG (1990) Disorder and decline: Crime and the spiral of decay in American neighborhoods. The

Free Press, New York

J Quant Criminol (2017) 33:783–808 807

123

Steenbeek W, Hipp John R (2011) A longitudinal test of social disorganization theory: feedback effects among cohesion, social control and disorder. Criminology 49:833–871

Taylor RB (2015) Community criminology: Fundamentals of spatial and temporal scaling, ecological indicators, and selectivity bias. New York University Press, New York

Taylor RB, Covington Jeanette (1988) Neighborhood changes in ecology and violence. Criminology 26:553–590

Tita GE, Greenbaum R (2009) Crime, neighborhoods, and units of analysis: putting space in its place. In: Weisburd D, Bernasco W, Bruinsma G (eds) Putting crime in its place. Springer, New York

Warner BD, Roundtree PW (1997) Local social ties in a community and crime model: questioning the systemic nature of informal social control. Soc Probl 44:520–536

Weisburd D, Bushway S, Lum C, Yang SM (2004) Trajectories of crime at places: a longitudinal study of street segments in the city of Seattle. Criminology 42:283–321

Weisburd D, Elizabeth G, Sue-Ming Y (2012) The Criminology of Place, Oxford, New York Wickes R, Hipp JR, Sargeant E, Homel R (2013) Collective efficacy as a task specific process: examining

the relationship between social ties, neighborhood cohesion and the capacity to respond to violence, delinquency and civic problems. Am J Commun Psychol 52:115–127

Wikström POH, Oberwittler D, Treiber K, Hardie B (2012) Breaking rules: the social and situational dynamics of young people’s urban crime. Oxford University Press, Oxford

Wilson WJ (1987) The truly disadvantaged: the inner city, the underclass and public policy. The University of Chicago Press, Chicago

Winship C, Radbill L (1994) Sampling weights and regression analysis. Sociol Methods Res 23:230–257 Wood L, Billie G-C, Max B (2012) Streets apart: does social capital vary with neighbourhood design? Urban

Studies Research Zhang L, Messner SF, Liu J (2007) A multilevel analysis of the risk of household burglary in the city of

Tianjin, China. Br J Criminol 47:918–937

808 J Quant Criminol (2017) 33:783–808

123

Reproduced with permission of copyright owner. Further reproduction prohibited without permission.