American Journal of Criminal Justice

, Volume 37, Issue 2, pp 229–245

Social Disorganization and Neighborhood Fear: Examining the Intersection of Individual, Community, and County Characteristics

Authors

    • Graduate Center, Brooklyn College and Institute for Demographic ResearchCity University of New York
  • Nicole E. Rader
    • Department of SociologyMississippi State University, Mississippi State
  • Jeralynn S. Cossman
    • Department of SociologyMississippi State University, Mississippi State
    • Department of Sociology and Social Science Research CenterMississippi State University
Article

DOI: 10.1007/s12103-011-9125-3

Cite this article as:
Porter, J.R., Rader, N.E. & Cossman, J.S. Am J Crim Just (2012) 37: 229. doi:10.1007/s12103-011-9125-3

Abstract

Fear has long been studied as a consequence of crime given the consistent and ubiquitous nature of fear as a reaction and the systematic variations in its consequences. Past research has shown significant variations in fear of crime at both the individual and ecological level. Here we implement a multi-level approach to understanding potential interactions between perceived safety in one’s neighborhood in relation to social disorganization indicators at the neighborhood level and crime rates at the county level. The nationally representative sample data (n = 2,610) used in this analysis combines individual level data collected in 2006 from the Panel Study of Religion and Ethnicity (PS-ARE) with ecological level data at the tract and county level from the 2000 US Census. The findings suggest a three level interaction negating the well known protection hypothesis of marriage and crime; this essentially means that as being married or cohabitating decreases the negative effects of being in a community with a high level of familial disruption (percent of divorced) increases, but that effect is substantively negatively tempered as the violent crime rate of the county rises.

Keywords

Fear of crimeSocial disorganizationPS-ARENeighborhood effectsCounty

Fear of crime is an important social topic (Warr, 2000). As Scarborough, Like-Haislip, Novak, Lucas, and Alarid (2010) recently note “Scholars have been drawn to this topic because it is among the most overt social reactions to crime and because its consequences are so prevalent, potentially severe, and easily demonstrable” (p. 819). Since fear of crime does not necessarily match up with actual crime rates and is actually often independent of crime rates, researchers have sought to determine what factors affect fear of crime (Wyant, 2008).

Both individual factors and ecological factors at the neighborhood and community level may play a role in an individual’s fear of victimization. Historically, previous literature tended to emphasize individual factors or ecological factors (Robinson, Lawton, Taylor, & Perkins, 2003; Scarborough et al., 2010); however, with evolving statistical techniques, studies have started to jointly examine individual and ecological factors (Liu, Messner, Zhang, & Zhuo, 2009). Although these works are promising and have greatly contributed to what is known about the factors most likely to influence fear of crime, most of these studies have jointly examined only individual factors in one community (a neighborhood or street block or county) and have not used nationally representative data. We examine perceived neighborhood safety as a proxy for an individual’s fear of crime, the neighborhood level, and the county level using a nationally representative data set—Panel Study of American Race and Ethnicity (PS-ARE) supplemented with data at the neighborhood level (operationalized as Census Tracts) and macro county level data—and examine the effects of fear of crime on individual factors such as those typically associated with vulnerability (e.g., gender, race, age) and ecological factors such as those typically associated with social disorganization within communities.

Consistent with the current literature, we hypothesize that individuals are more likely to be fearful of their safety in their neighborhood based on individual level indicators and vulnerability to unsafe conditions at the neighborhood level, rather than contextual indicators of social disorganization and the occurrence of crime. We also intend to advance the current state of the literature by testing a series of two- and three-level interactions to better understand the ecological effects of social context on the respondent’s reported fear of safety in their own neighborhood. Examining the relationship between fear of crime and variables typically associated with fear at three levels of analysis using a nationally representative sample provides a more solid foundation for understanding these relationships. Furthermore, our use of a nationally representative sample allows us to better generalize these results in comparison to much of the current literature in ecological criminology, in which small geographic areas and single metropolitan areas are often the focus.

Predicting Individual Level Fear of Crime

Since the late 1980’s, fear of crime is often independent of actual crime rates (Schafer, Huebner, & Bynum, 2006; May, Rader, & Goodrum, 2009; Warr, 2000; Scarborough et al., 2010). Therefore, researchers have explored what factors within the individual would lead to an increased fear of crime. There are certain individual characteristics that make individuals feel more vulnerable to crime, regardless of their actual chances of victimization (Schafer et al., 2006). Vulnerability is typically broken down into two categories in the fear of crime literature: physical and social (Scarborough et al., 2010). Gender—with both physical and social attributes—is the strongest individual predictor of fear of crime, with women fearing crime at much higher levels than men (May et al., 2009). Although a variety of explanations have been put forth to explain this finding (see Smith & Torstensson, 1997 or C. A. Franklin & T. W. Franklin, 2009 for a full discussion), one predominant explanation is that women feel more physically vulnerable to victimization than men. It has also been proposed that women are socialized to be more fearful and to have their husbands (when married) do the “fear work” in their relationship (Rader, 2008). Age is also typically viewed as an individual trait associated with increased physical vulnerability and higher levels of fear of crime with the elderly feeling more physically vulnerable to victimization (Warr, 2000). Social vulnerability is typically described as something within the individual that makes them feel more vulnerable in society such as being a racial minority, having a lower education level, or having a lower income level (Scarborough et al., 2010).

Predicting Ecological Level Fear of Crime

Since the late 1990’s, researchers focused specifically on neighborhood and community factors that may influence fear of crime levels; work that was situated in a social disorganization tradition. Social disorganization theories argues that communities with certain characteristics—those in which internal and external social control are absent or weakened—are more likely to have crime, fear of crime, and other social problems. The specific structural characteristics discussed by social disorganization theorists include high poverty, racial heterogeneity, family disruption, and residential mobility (Markowitz, Bellair, Liska, & Liu, 2001; Paulsen & Robinson, 2004; Porter & Purser, 2010).

Most social disorganization research focuses on serious crime instead of minor infractions. The fear of crime literature has found that minor forms of crime may make a large difference in determining fear of crime within neighborhoods (Markowitz et al., 2001). Many researchers examine incivilities and disorder within neighborhoods and the effect that this may have on fear of crime levels within communities (e.g., LaGrange, Ferraro, & Supancic, 1992; Hipp, 2007; Gibson, Zhao, Lovrich, & Gaffney, 2002; Robinson et al., 2003; Skogan, 1990; Taylor, 2001; Wyant, 2008). Repeated disorder or incivilities is argued to weaken social ties, affect population turnover and racial composition and ultimately lead to socially disorganized communities (Markowitz et al., 2001; Skogan, 1990).

LaGrange et al. (1992) first defined incivilities as “low-level breaches of community standards that signal erosion of conventionally accepted norms and values” (312). Physical incivilities often includes measures of things like vandalism, litter, abandoned cars, vacant housing/lots, illegally parked cars, rundown buildings and homes, and graffiti while social incivilities includes measures of things like drinking in public, rowdy teenagers, gang presence, neighbors fighting, prostitution, loud music/parties, homelessness, begging, loitering, truancy (Kruger, Huthison, Monroe, Reischl, & Morrel-Samuels, 2007; Scarborough et al., 2010; Liu et al., 2009). Social and physical incivilities have a positive effect on fear of crime but the relationship is strong with social incivilities than with physical incivilities (LaGrange et al., 1992; Gibson et al., 2002).

Research has also identified robust community-level family structure effects on crime and delinquency (Schwartz, 2006). Wilcox et al. (2005) found murder and robbery rates to be strongly and negatively associated with the health of marriage in urban communities. In other words, low rates of marriage were usually accompanied by high rates of murder and robbery—for whites and blacks as well as for adults and juveniles. Similarly, Wooldredge and Thistlewaite (2003) suggest that neighborhoods with fewer married adults were more likely to witness higher rates of assault. These findings suggest that, along with an individual “protection” effect, there also exist an ecological “protection” from crime. As Wooldredge and Thistlewaite (2003) note, familial unions can be directly linked to the rates of assaults in a community and thus should be considered relatable to one’s fear in their own community. Therefore, if an individual lives in a community with a high degree of married and cohabitating households they should be less likely to fear being victimized given the lower rates of crime in the community. Furthermore, as we find in the preceding section that vulnerable individuals are more likely to have a fear of being unsafe in their neighborhood they should be most likely to be affected by the ecological conditions in which they live. This begs of question of a possible interaction between being a considered vulnerable (female, elderly, young, etc.) and living in a community with a familial disruption or stability. Most recently, Porter and Purser (2010) show the rate of marriage at the county level is directly related to lower rates of crime in the county even when controlling for community level indicators of racial heterogeneity, familial disruption, socioeconomic status, and urbanization (core indicators of social disorganization).

Studies that Jointly Predict Both Individual Level and Ecological Level Fear of Crime

Several studies have emphasized the multi-level nature of fear of crime, particularly in the individual and neighborhood levels. Although not meant to be exhaustive, some examples are provided that showcase the general nature of multi-level discussions of fear of crime. In a study by Kruger et al. (2007), zip codes in a Michigan county were linked to collected survey data collected at the individual to identify potential ecological linkages and through the usage HLMs, gender mattered, regardless of the level of analysis, as did age—with women and elderly reporting higher rates of fear. Furthermore, social capital at the neighborhood level decreased individual level fear of crime in Kruger et al’s models. These findings are limited in that they are drawn from data collected only in Michigan and are, therefore, may not be generalizable beyond that state.

Another recent study by Scarborough et al. (2010), conducted OLS regression using neighborhood and individual level data from surveys collected in Kansas City, Missouri. The authors argued that they were examining both levels of analysis because “most studies have not examined individual and neighborhood level fear of crime at the same time” and therefore “exploring the relevance of demographic characteristics on fear when neighborhood context is considered and thus expand the understanding of the proximate causes of fear or crime” (819–820). Social cohesion had a negative effect on fear of crime controlling for demographic variables and disorder was positively related to fear of crime (Scarborough et al. 2010). Again, these results are limited by the fact that they are rooted in data collected in one city—Kansas City—and therefore would not be generalizable to any other aggregate populations that vary significantly from the area under examination. Also, their analyses did not use more advanced statistics to adequately account for the nested nature of individuals within neighborhoods.

Robinson et al. (2003) examined the effect of incivilities on fear of crime at both the individual level and the street block level in Baltimore, Maryland. This study had the added advantage of a longitudinal focus since the researchers gave the survey on two occasions within a year and also used HLM as an analysis strategy. Robinson et al. (2003) found that incivilities predicted individual level fear of crime but did not show lagged effects on fear of crime. This study can only account for individual and neighborhood variation in Baltimore but does not have any level three variables to further test these relationships.

Within a similar line of research, Schafer et al. (2006) examined gender differences in fear of crime among individuals in a large mid-western city. The study examined variations in fear of crime at the individual level and nested these data within patrol beats of the targeted city. Their HLM analyses suggested that fear facilitators and inhibitors varied by gender so that demographic predictors such as race and class mattered more for men than women. Neighborhood integration mattered for men while perception of neighborhood conditions mattered more for women. With its focus on one mid-western city and neighborhood level control variables sans level three analyses, these results are similarly limited.

To address the issue of nested ecological effects, Wyant (2008) examined both individual level and neighborhood level fear of crime and perceived incivilities in four New Jersey counties and five Pennsylvania counties so that all 45 Philadelphia neighborhoods were equally represented. This research aggregated individual fear of crime responses to the neighborhood level and supplemented those data with 2000 Census county-level data. Individual level incivility was important while at the ecological level, perceived incivilities did not directly affect fear of crime but instead was mediated by crime risk (Wyant, 2008). The findings are based on data from just nine Northeastern counties and, therefore, are—like many other studies—limited in their generalizability.

Two studies which examine multiple levels of analyses and fear of crime were conducted in Britain and China. Markowitz et al. (2001) were interested in the social disorganization and how fear of crime related to cohesion and disorder. Using data from 151 neighborhoods across three waves of the British Crime Survey and conducting panel analysis in non-recursive models, Markowitz and colleagues found that cohesion affected disorder which affected fear which in turn affected cohesion. Their longitudinal data permitted them to examine the reciprocal nature of crime and social conditions, such as fear of crime, and found that the effect of disorder on cohesion was partially mediated by fear of crime. This national and longitudinal study is enlightening. The data were collected in Britain; so, it is unclear whether these results are applicable to residents of neighborhoods or counties within the United States.

Liu et al. (2009) also examined multiple levels of fear of crime in a large city in China. At the individual level, young, educated, women, and previous victims were more fearful than their counterparts. At the ecological level, living near black or Latino neighbors was positively correlated with perceived disorder which lead to fear of crime. Since these findings are only applicable to one city in China, it is difficult to determine whether similar findings would be seen for United States residents.

As the above examples indicate, most studies that examine multiple levels of analysis cannot simultaneously report three levels of analysis nor are they able to report on a nationally representative sample of American residents. In this paper, we add to the existing literature on fear of crime by testing individual, neighborhood, and county level predictors of fear of crime using a nationally representative sample.

Data and Methods

Sources of Data

Individual level data were obtained from the Panel Study on American Religion and Ethnicity (PS-ARE), an in-home survey collected in 2006, with oversamples of people of color. The survey collection used a multi-stage approach. Three-digit zip code areas were selected with probabilities proportional to a composite size measure weighting areas with high minority concentrations to result in over-samples of non-whites. From each three-digit area, two five-digit zip codes were randomly selected, from which about 90 addresses were randomly selected, and households were screened for eligibility. The final sample size was 2,610, and represents a 58% response rate (83% of eligible respondents successfully contacted x 86% of contacted persons successfully screened x 82% of persons screened and selected for an interview who completed the survey). Once weighted to account for oversampling, the PS-ARE closely mirrors the Census Bureau’s American Community Survey 3-year average (2005–2007) estimates. For more details about the survey, see the Researchers section at www.ps-are.org.

With access to respondents’ street addresses, all respondents were geocoded to their exact latitude/longitude location and linked to relevant Census geographies using ArcGIS spatial “join” techniques. We linked each survey respondent to contextual level data from the US Census Bureau at both the tract and county level. Data from the 2000 decennial Census were added to the dataset, resulting in a nationally representative sample of 2,610 respondents linked to relevant neighborhood and county-level indicators of social disorganization and crime rates.

Measurement

The dependent variable is a proxy for fear of crime indicating the perceived safety one feels in their neighborhood. This question asks “On average, how often have you felt unsafe, if ever, in your current neighborhood within the last year?” About 64% of respondents reported “never feeling unsafe.” Given the results of ancillary analyses, the variable that is used in this examination has been collapsed into a dichotomous variable indicating feeling unsafe in the last year (yes = 1, no = 0).1

Relevant individual determinants of fear were identified via the above literature review. These variables represent a standard set of demographic controls and include, race/ethnicity (two dummies for Hispanic and Black in reference to White), age (in years), gender (male = 1), education (three dummies for high school graduate, some college/associates degree, and college graduate in reference to non-high school graduate), existence of children in the home (yes = 1, no = 0), and marital status (married or cohabitating = 1 else = 0).2

At the neighborhood level a series of census tract-level indicators were identified as measures of the four components of social disorganization. The total census tract population was included as a measure of the level of urbanicity in the immediate community. In measuring the ethnic/racial heterogeneity, the tract percent non-white was included. The socioeconomic status component of social disorganization was measured here via the tract’s percent of the population in poverty and the percent of the tract’s housing units that are occupied, as a proxy for the existence of dilapidated and abandoned housing. The familial disruption component was measured via the percent of tract households that are female-headed and the percent of the tract population that is divorced. The inclusion of these neighborhood (tract) level indicators allows us to test for the existence of cross-level individual and neighborhood interactions as well as the degree to which individual level variations in fear can be accounted for given the inclusion of neighborhood level indicators in the modeling strategy.

At the county level, a series of controls are used to control for the level of crime (property and violent rate per 1,000 residents in the county) and the racial/ethnic relations in the county (as measured by the dissimilarity index, indicating residential segregation). The three county-level determinants of fear will permit testing of cross-level interactions between individual-level and county-level indicators, as well as tract-level and county-level indicators. Finally, and perhaps the most significant portion of this analysis is the ability to test potential three-level interactions at a nationally representative scope.

Analytic Techniques

The analytic approach for this research involves multiple phases. Variables are examined descriptively to test for compliance with modeling assumptions (see Table 1). Bivariate relationships for all variables are examined to understand baseline relationships and to test for the potential of multicolinearity. These baseline relationships help to guide, and validate, the testing of potential two- and three-way cross-level interactions. Correlations are only presented for exploratory purposes and ancillary analyses, using crosstabulations and chi-square tests, are used to confirm any preliminary relationships identified among nominal and binary variables.
Table 1

Descriptive statistics (n = 2,610)

 

Mean (%)

St. dev.

Dependent variable (individual)

 Ever felt unsafe in your neighborhood, %

36.6

Level 1 (individual)

 White, %

55.0

 Black, %

23.0

 Hispanic, %

22.7

 Age

44.1

16.5

 Male, %

40.2

 High-school dropout, %

14.6

 High school graduate, %

40.9

 Some college/associates degree, %

18.9

 4 years college graduate or +, %

15.1

 Children in the home, %

42.1

 Married/cohabitating, %

52.7

Level 2 (census tract)

 Total population

3,654.1

753.5

 Percent minority

15.0

 Percent in poverty

15.2

 Percent of housing occupied

76.2

 Percent of total households female-headed

13.1

 Percent divorced

6.6

Level 3 (county)

 Property crime rate per 10,000

41.2

27.3

 Violent crime rate per 10,000

17.1

13.6

 Residential segregation (dissimilarity)

0.5

0.2

The final stage of analysis involves the multi-level statistical modeling of fear for one’s safety in their neighborhood. Given the natural hierarchical relationship of the survey respondents from the PS-ARE to specific neighborhood- and county-level contexts, we employ a hierarchical linear modeling (HLM) approach, which allows us to examine individual- and contextual- level coefficient effects as well as cross-level interactions. Also, given the fact that our dependent variable is binary in nature, we employ a logistic approach and, in turn, results will be reported in terms of odds ratios of the respondent’s perceptions of safety in their respective community and county given their demographic profile.

Traditionally, there are three steps to this analysis. First, we use HLM Random Coefficients Model to estimate the respondent-level effects on the measures of perceived safety. This procedure is different from a traditional single-level model because, within HLM, the level-one effects are modeled for each respondent and then the average intercept and slope are reported. This allows for the fact that the survey respondents are selected in a stratified, multi-level context and is, therefore, included in the modeling process (Raudenbush & Bryk, 2002).

The Random Coefficients Model is specified as:
$$ {Y_{{ij}}} = {\gamma_{{00}}} + {\gamma_{{i0}}} + {u_{{ij}}} + {u_{{0j}}} + {r_{{ij}}} $$
Where the perceived safety (Yij) is equal to the grand mean level of the variable (γ00) plus the average regression slope for each individual-level independent variable on the response variable (γi0) plus the unique effect of the variations in ecological context on the associated individual-level regression slope (uij). Also, the random error at both the contextual level (u0j) and individual level (rij) are taken into account.

In the second step, we use the HLM Regression with Means-as-Outcomes Model, to test the effects of the tract- and county-level determinants of perceived safety isolated from each other and the individual-level determinants.

HLM Regression with Means-as-Outcomes Model is specified as:
$$ {Y_{{ij}}} = {\gamma_{{00}}} + {\gamma_{{0i}}} + {u_{{0j}}} + {r_{{ij}}} $$
Where the respondent’s level of perceived safety (Yij) is equal to the grand mean level of perceived safety (γ00) plus random error at both the contextual (u0j) and individual level (rij). In addition the fixed effect of the contextual-level independent variables are taken into account (γ0i) in order to identify potential effects between perceived safety and context in an isolated level-two (tract) and level-three (county) model.

Finally, we use the HLM Intercepts- and Slopes-as-Outcomes Model to examine cross-level interactions of the respondent- and contextual-level variables.

The full Intercepts- and Slopes-as-Outcomes model is specified as:
$$ {Y_{{ij}}} = {\gamma_{{00}}} + {\gamma_{{0i}}} + {\gamma_{{i0}}} + {u_{{ij}}} + {\gamma_{{ij}}} + {u_{{0j}}} + {r_{{ij}}} $$
where the respondent’s perceived level of safety (Yij) is equal to the grand mean level of the respective safety measure (γ00) plus the main effects of all contextual-level independent variables (γ0i), all respondent-level independent variables (γi0), and all cross-level interactions (γij) (two- and three-way). Lastly, the unique error associated with the level-one slopes (uij) and the random error at both the contextual-(u0j) and individual-level (rij) are again modeled.

Ultimately, the analysis includes six nested models in which Model 1 introduces the respondent-level random coefficients, including all demographic controls. Model 2 presents the isolated, fixed main effects of neighborhood-level (tract) measures with Model 3 incorporating the isolated and fixed main effects of county-level (level-three) determinants of perceived safety. Model 4 presents the full model without interactions including both random coefficient and fixed effects controlling for each. Model 5 introduces cross-level interactions between all tested two-way interactions and Model 6 introduces all three-way cross-level interactions.

Results

Following the initial statistical description, all variables are examined for baseline relationships using bivariate correlations (Table 2). Respondent’s perceived safety is related to a number of independent variables. The baseline bivariate likelihood of feeling unsafe in one’s neighborhood is directly related to being Black, younger, women, being a college graduate, not having children at home, living in a highly population area, having a high percent minority, a high percent in poverty, a low percentage of housing occupied, a high percentage of households female-headed, a high percentage of the population divorced, higher property crime rates, higher violent crime rates, and a higher rate of residential segregation. These relationships are in line with existing literature. Furthermore, none of the variables exhibits multicolinearity, but a few do indicate potential issues that will be further examined in the modeling stages of this analysis.
Table 2

Bivariate correlations of all variables in analysis (n = 2,610)

 

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

[14]

[15]

[16]

[17]

[18]

[19]

Dependent variable (individual)

[1]

Ever felt unsafe in neighborhood

0.05*

−0.04*

−0.10*

−0.12*

−0.01

0.02

0.04*

−0.07*

−0.07

0.11*

0.12*

0.16*

−0.14*

0.17*

0.07*

0.06*

0.07*

0.08*

Level 1 (individual)

[2]

Black

1

−0.29*

−0.06*

−0.02

0.05*

0.04*

−0.08*

0.05*

−0.19*

0.07

0.61*

0.38*

−0.17*

0.51*

−0.01

0.08*

0.12*

0.11*

[3]

Hispanic

1

−0.18*

−0.01

−0.01

−0.05*

−0.09*

−0.13*

0.04

0.31*

−0.11*

0.03

−0.08*

0.08*

−0.03

0.11*

0.11*

−0.01

[4]

Age

1

−0.04*

−0.11*

0.01

−0.01

0.57*

0.03

−0.04

−0.04*

−0.07*

0.05*

−0.09*

−0.01

−0.08*

−0.08*

0.06*

[5]

Male

1

0.02

−0.01

−0.01

0.01

0.09*

−0.01

−0.03

−0.07*

0.04*

−0.06*

0.06*

−0.01

−0.01

−0.03*

[6]

High school graduate

1

−0.41*

−0.35*

0.01

−0.06*

−0.04

0.05*

0.05*

0.01

0.04*

−0.02

−0.03

0.01

0.03

[7]

Some college/associates degree

1

−0.21*

0.03

0.01

−0.02

0.01

−0.01

0.01

−0.01

0.04*

−0.01

−0.01

−0.01

[8]

4 years college graduate or +

1

−0.12*

0.06*

−0.01

−0.12*

−0.18

0.09*

−0.17

−0.02

0.01

−0.01

−0.05*

[9]

Children in the home

1

0.09*

−0.07*

0.06*

0.02

0.03

0.01

−0.04*

−0.02

−0.02

0.05*

[10]

Married/cohabitating

1

−0.03

−0.15*

−0.18*

0.14*

−0.19*

−0.12*

−0.04*

−0.05*

−0.07*

Level 2 (census tract)

[11]

Total population

1

0.11*

0.11*

−0.31*

0.25*

−0.06*

−0.16*

−0.15*

0.37*

[12]

% Minority

1

0.65*

−0.36*

0.54*

−0.02

0.01

0.06*

0.24*

[13]

% in poverty

1

−0.49*

0.56*

0.12*

0.13*

0.16*

0.13*

[14]

% of housing occupied

1

−0.43*

−0.31*

0.11*

0.11*

−0.32*

[15]

% of households female-headed

1

0.04*

0.07*

0.12*

0.28*

[16]

% Divorced

1

0.13*

0.12*

−0.12*

Level 3 (county)

[17]

Property crime per 10,000

1

0.67*

−0.18*

[18]

Violent crime per 10,000

1

−0.21*

[19]

Residential Seg(dissimilarity)

1

*= p < 0.05

The next phase of the analysis makes use of an HLM logistic modeling approach (Table 3). The isolated effects of the individual level determinants of the respondent feeling unsafe in their neighborhood are seen in Model 1. Hispanics are less likely to report feeling unsafe when compared to Whites. Older respondents, men, and those that are married or cohabitating are all less likely to report feeling unsafe. In relation to college dropouts those who completed a 4 year college degree are more likely to report feeling unsafe. In some cases these relationships are expected to be linked to contextual factors that are not measured in the current model.
Table 3

Odds ratios of self-reported feelings on being unsafe in one’s neighborhood (n = 2,610)

 

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6

Level 1 (Individual)

 Black

1.118

0.609***

0.605***

0.605***

 Hispanic

0.769*

0.553***

0.583***

0.563***

 Age

0.985***

0.986***

0.986***

0.986***

 Male

0.593***

0.611***

0.614***

0.613***

 High school graduate

1.007

1.081

1.077

1.084

 Some college/associates degree

1.123

1.246

1.251

1.255

 4 years college graduate or +

1.264*

1.510**

1.507**

1.517**

 Children in the home

1.016

1.012

1.011

1.013

 Married/cohabitating

0.793*

0.871*

0.827*

0.873*

Level 2 (census tract)

 Total population

1.007**

1.006**

1.006**

1.006**

 Percent black

1.001

1.003

1.002

1.003

 Percent in poverty

1.008

1.009

1.009

1.008

 Percent of housing occupied

0.993*

0.993*

0.994*

0.993*

 Percent of total households female-headed

1.016

1.019

1.022

1.025*

 Percent divorced

1.046*

1.047*

1.079*

1.075*

Level 3 (county)

 Property crime rate per 1,000

1.001

1.002

1.002

1.002

 Violent crime rate per 1,000

1.013*

1.007

1.007

1.006

 Residential segregation (dissimilarity)

2.922***

1.894*

1.799*

1.857*

Cross-level interactions

 Female HH * married

0.988

 %t Divorced * married

0.911*

 % Divorced * married * violent crime rate

1.004*

 -2 log likelihood

2946

2795

2998

2705

2702

2702

 Chi-square

80.97***

89.52***

28.99***

178.76***

182.067***

181.63***

*= p < 0.05, **= p < 0.10, ***= p < 0.001

Model 2 introduces these contextual factors at the neighborhood, or census tract, level. Being in a census tract with higher population and a higher percent divorced, both significantly increase the likelihood of the respondent feeling unsafe. In contrast, being in a community with a higher percent of the housing units being occupied lowers the likelihood of feeling unsafe. All of these results support a social disorganization theoretical framework. Most directly, the more urban, the lower the socioeconomic status, and the greater the familial disruption the more likely an individual is to feel unsafe. There is no support of the racial/ethnic heterogeneity hypothesis, which states that the higher the degree of racial and ethnic heterogeneity the higher the feelings of unsafety should be, in this model.

Model 3 introduces the level-three contextual factors at the county level. The results show that higher rates of violent crime and higher levels of residential segregation are linked to an increased likelihood of reporting feeling unsafe. Interestingly, the property crime rate of the county has no significant effect. Even more interesting is the indication that the racial segregation plays a significant effect in heightening the likelihood of feeling unsafe. While this does not directly measure heterogeneity, it indicates a high level of division among those in the community by race. Thus, demonstrating support for the racial/ethnic heterogeneity component of the social disorganization framework.

Model 4 includes all variables from all three levels of data. When controlling for contextual factors, both Blacks and Hispanics are less likely to report feeling unsafe in their neighborhoods compared to Whites. The effects for age, gender, and being a college graduate remain consistent from Model 1. All of the effects at the tract level remain consistent with those in Model 2, but the effect of the violent crime rate becomes insignificant when introducing the individual level controls. This is not uncommon in research concerning fear as it is commonly shown to not be effected by the crime rate, but instead by individual and neighborhood traits. The residential segregation measure remains a positive indicator of the likelihood of reporting feelings of unsafety in one’s neighborhood.

Analyses (not shown here) were undertaken to understand which variables were responsible for the slight changes identified in the levels of significance across some of the independent variables. These analyses identified the familial disruption tract variables as accounting for these differences, namely the percent of households that were female-headed and the percent divorced. Based on this evidence two interactions were created between the married/cohabitating dummy and these two variables in centered form. The results are presented in Model 5. While the interaction between marital status and the percent of female-headed households is not significant, a significant interaction between the percent of the tract divorced and marital status is significant in relation to its effect on the respondent reporting feelings of unsafety. Ultimately, for those that are married or cohabitating, the likelihood of reporting feelings of unsafety are further decreased as the percent of the neighborhood that is divorced decreases. This finding provides further support for the protection effects of marriage in a multi-level context, which is further discussed later.

Finally, this two-way interaction is further examined as part of a three-way interaction to test the effect of the violent crime rate in the county on the protection effect of being married in relation to the effect of the neighborhood percent divorced and feelings of unsafety. The three-way interaction is presented in Model 6, showing a protection effect of marriage and cohabitation depending on the violent crime rate of the county. This positive effect on feelings of safety of being married or cohabitating in neighborhoods with a high divorce rate are actually reversed as the violent crime rate climbs in the county. A complicated three-way interaction, this essentially means that as being married or cohabitating decreases the negative effects of being in a community with a high level of familial disruption (percent of divorced) increases, however that effect is substantively negatively tempered as the violent crime rate of the county rises.

It is important at this point to address the effect that the dependent variable may play in this finding. As noted, we employ a measure of ‘safety’ as our proxy for ‘fear of crime’. Ultimately, this indicator may be more appropriately noted a ‘fear of victimization’, given the relationship of victimization and safety to violent crime. That being said, that the violent crime rate is significant in this interaction may not be all that surprising given that we employ this indicator which is directly related to the fear of violent victimization, or more directly, personal safety. On that point, we have been able to identify a three-way relationship in which the protection effect of marriage/cohabitation is undermined as the community becomes more violent in terms of the types of crimes committed, reported and documented.

Discussion

Individual, neighborhood, and county level factors all influence the likelihood of feeling unsafe in one’s neighborhood. We used a nationally representative sample of Americans to examine different levels of analysis to better understand the mechanisms of fear of crime to test whether, at the contextual level, social disorganization variables may play a central role in predicting fear of crime. Social disorganization theory tends to highlight contextual level variables such as high poverty, racial heterogeneity, family disruption, and residential mobility. We find that in a few cases, the effects of the individual demographics are eliminated with the introduction of neighborhood characteristics, such as is the case for the relationship between marital status and its variation being directly related to the degree of familial disruption in the census tract. This finding supports social disorganization theory and its effect on fear of crime. Few studies examine family disruption and its effect on fear of crime at both the contextual and the individual level. Therefore, examining the dual effect of contextual and individual level predictors can provide important insight into this relationship.

Another finding of interest at the contextual level is race, which doesn’t matter at the individual level until level two and three variables are added, suggesting that racial heterogeneity plays more of a role than race itself. That is, in racially-segregated communities, collective fear of crime is more likely, despite individual levels of fear.

On the other hand, the contextual factors themselves are eliminated with the introduction of individual characteristics. For instance, violent crime rates were associated with increased likelihood of reporting feelings of unsafety in isolated form, but were insignificant once the demographic profile of the individual was controlled. This was especially true for women, older, and more educated respondents. As other studies have suggested, certain individual characteristics such as gender or age are significant, regardless of other contextual factors (Schafer et al., 2006). Therefore, fear of crime cannot be studied only ecologically and must include individual level analyses.

Finally, we found significant interactions across individuals, neighborhoods and counties in regards to the respondent’s feeling of safety in their own neighborhood, indicating the importance of multi-level modeling in examining fear of crime. Studies that use multi-level representative samples that can merge both contextual and individual level data will be better able to understand and explain fear of crime.

At the county level, violent crime was a significant predictor but property crime was not; this is not unexpected since people are often fearful of being a victim of violent crime. Although people may be fearful of property crime occurring to their property, they are not necessarily fearful in regards to their perceived personal safety. This finding is in line with fear of crime and socialization literature which suggests individuals (particularly women) are socialized to be fearful of violent crime more than property crime (Chiricos, Hogan, & Gertz, 1997; Madriz, 1997; Reid & Konrad, 2004). Fear seems to be learned from primary sources such as parents and friends as well as from the media, which teaches individuals to fear things unnecessarily (Chiricos et al., 1997; De Groof, 2008). Therefore, the reality of crime may suggest that individuals are not very likely to experience violent crime, but discussions of the possibility of violent crime increase individuals’ fears of violent crime more than property crime. Violent crime only effects individual fear until individual demographics are controlled—this interaction provides support for the idea this idea as it highlights the tempered nature of the protection that certain sociodemographic indicators provide at both the individual and ecological level as violent crime increases at the third-level (county-level).

Support for the intensification of the marriage or cohabitation’s protection effect on perceived feelings of safety occurred at both the individual and neighborhood level. At the individual level, marriage serves as a protective effect against fear of crime, especially for women (Rader, 2008). Given what is known about the contextual effects of social disorganization in the form of family disruption, the logical next step was to examine the role of both contextual and individual indicators of family makeup on fear of crime. We did so and found that in socially disorganized communities, via the familial disruption component, the positive effect of being married remains significant and becomes more protective as the familial disruption of the neighborhood increases. When adding level three variables, this relationship is further linked to the violent crime rate of the county which tempers the relationship. This is a complex relationship that needs further investigation in the future; however, this finding is likely linked to the operationalization of our dependent variable. Future analyses should employ more direct measures of ‘fear of crime’ in order to test for similar findings.

This work adds to the existing literature on fear of crime by providing a national sample of Americans, but we recognize there are limitations as well. Fear of crime researchers have spent the last 30 years debating the definition of fear of crime. Our measure of fear of crime assesses safety more than worry; however, this weakness is overshadowed by our use of a national sample and data at three levels of analysis. Future studies that examine multi-level data of fear of crime should use a more comprehensive measure that can take into account worry versus safety, as well as fear for multiple types of crimes.

Our findings significantly contribute to the previous literature on fear of crime. We suggest here that fear of crime can best be explained by examining individual level and contextual level variables simultaneously. When doing so with a nationally representative sample, we found that family makeup, race, and type of crime all matter, along with traditionally known predictors such as gender and age. Future fear of crime research will need to be able to assess all of these variables at various levels. In doing so, a more complete view of fear of crime will emerge. Furthermore, the fact that females receive a more beneficial effect from being married is very interesting and should be further explored with more direct measures of “fear of crime” and among other vulnerable populations. For instance, does the protection of being in a stable family have similar effects on elderly, children, newly arrived immigrants, or other potentially vulnerable groups?

From a policy perspective, these questions highlight a number of relevant issues, including family formation/dissolution, protection of vulnerable population, and prevention of crime among others. Perhaps most interesting is our finding that even in the face of high rates of violent crime, being in a union of some sort decreases the likelihood that one is likely to feel unsafe in their neighborhood. In relation, being married even in a socially disorganized community with high rates of divorce and single-parent families provides a sense of safety. This is tempered somewhat by the overall level of violent crime in the community, but still remains a consistent interaction. In general, our findings emphasize an important relationship between family structure at both the individual and ecological level. At the ecological level, our research highlights variations in perceived feelings of safety that those in socially disorganized communities tend not to benefit from in the same way as those in more “socially-organized” communities. At the individual level, we find that those in some form of union with a partner do are also much more likely to be perceptively less fearful of their neighborhood. That being said, policy makers should continue to support the formation of families, whether through marriage or less formal measures, as a way to provide a sense of safety to their constituents and, by relation, healthier communities.

Footnotes
1

We recognize that our measure of fear of crime is limited by focusing on safety instead of worry as well as not focusing on specific crime types. This critique of fear of crime measures is well document in the literature (see Warr, 2000 for this discussion). Although clearly a limitation of our study, this is the best indicator in our data set and is also the same question used in a variety of large data sets including the General Social Survey and the National Crime Victimization Survey. Furthermore, fear of crime scholars have used this question to measure fear recently (Robinson et al., 2003; Zhao, Lawton, & Longmire, 2010; Wyant, 2008).

 
2

Married and Cohabiting was combined given the nature of the survey question which asked if the person was living with an unmarried partner. These responses were combined with the married group in order identify individuals in the sample that were in some sort of a shared union.

 

Acknowledgements

The authors would like to thank Micheal O. Emerson and the Adele James access and support concerning the confidential PS-ARE dataset, which made portions of this analysis possible. However, all errors of fact or interpretation are solely those of the authors.

Copyright information

© Southern Criminal Justice Association 2011