A Multilevel Analysis of Collective Efficacy,Neighborhood Disorder,and Individual Social Capital on Avoidance Behavior

Abstract

The principal aim of this multilevel study was to assess the impact of collective efficacy and disorder, as neighborhood characteristics, and individual social capital on an individual’s avoidance behavior, independent of the neighborhood composition. The theoretical backdrop to the present study integrates insights from social capital theory, collective efficacy theory, and broken windows theory. The multilevel model is based on an individual-level questionnaire of inhabitants (N = 2,730) and a neighborhood-level questionnaire of key informants in neighborhoods in Ghent, Belgium (N = 142). The results suggest small but significant neighborhood effects on an individual’s avoidance behavior. Individuals with lower levels of individual social capital and who live in neighborhoods with higher levels of disorder report more avoidance behavior.

Keywords

neighborhood avoidance behavior fear of crime social capital collective efficacy

Introduction

For more than 50 years, fear of crime has been a central topic in criminological research and debate (for an overview, we refer to Hale, 1996). Beside the research-related interest, there is a lot of attention for this topic in criminal policy at different levels. This attention is based, among others, on the relative stability of fear of crime, despite a significant crime drop (Rader, 2017). In this context, Furstenberg (1972) stated in an early stadium the existence of a “fear of crime paradox”: those who have the smallest risk of being victimized from an objective perspective tend to perceive their risk as the greatest (i.e., “fear of crime”). Nowadays, this “fear of crime paradox” is still a key finding in fear of crime studies (e.g., Hanley & Ruppanner, 2015; Rader, 2017); however, conclusions in this regard are not always straightforward, see, for example, the rationality debate on fear of crime (e.g., Jackson, 2004; Vanderveen, 2006), and the measurement issues and interrelated conceptual ambiguity (e.g., Farrall, Jackson, & Gray, 2009; Gabriel & Greve, 2003; Jackson, 2005).

Fear of crime is a concept with various definitions, conceptualizations, and measurements (Ferraro & LaGrange, 1987; Hardyns & Pauwels, 2010). However, two main conceptualizations can be distinguished: a narrow one and a broad one. The narrow conceptualization of fear of crime essentially corresponds with the definition of Ferraro (1995): “an emotional response of dread or anxiety to crime or symbols that a person associates with crime” (p. 4). Besides this emotional-affective dimension, the broad conceptualization of fear of crime distinguishes also a cognitive and a behavioral dimension (Fattah & Sacco, 1989; Ferraro & LaGrange, 1987; Gabriel & Greve, 2003; Greve, 1998). The cognitive component precedes the emotional-affective component and refers to a process that converts signals and stimuli which have to do with threat and danger into a risk assessment of personally becoming a victim of crime. The behavioral component of fear of crime, such as avoidance behavior, can be interpreted as a defensive reaction to an emotional state of mind when experiencing fear (Ferraro & LaGrange, 1987; Gabriel & Greve, 2003). This dimension of fear captures actual changes in human behavior and illustrates the overt effect of fear of crime in citizens’ everyday lives (Franklin, Franklin, & Fearn, 2008).

In this study, a behavioral dimension of fear of crime is examined from a predominant social disorganization perspective. The urban sociology, founded in the tradition of the Chicago School at the end of 19th century, noted that the particular neighborhood in which an individual resides has a significant impact on the individual’s (criminal) behavior. In the light of these observations, scholars have developed new theoretical frameworks, such as collective efficacy theory (Sampson, Raudenbush, & Earls, 1997) and broken windows theory (Wilson & Kelling, 1982). They further elaborated the social mechanisms that are responsible for the association between structural characteristics of neighborhoods and the concentration of crime and other related problems. In addition, key theoretical concepts from social capital theory (Coleman, 1988; Putnam, 1993) are integrated in our theoretical framework.

The aim of this multilevel study is to assess the impact of collective efficacy and disorder, as neighborhood characteristics, and individual social capital on an individual’s avoidance behavior, independent of the neighborhood composition. To our knowledge, this is the first study at this scale (i.e., all neighborhoods in a large Belgian city) that assesses the impact of individual- and neighborhood-level indicators of collective efficacy, disorder, and social capital on individual-level avoidance behavior in Belgium.

Theoretical Background

In this study, the integrated theoretical framework is based on social capital theory, collective efficacy theory, and broken windows theory. Collective efficacy theory stresses the importance of structural characteristics of spatial areas, such as economic disadvantage, ethnic heterogeneity, and residential mobility, which are negative characteristics of urbanized contexts (Bursik & Grasmick, 1993; Sampson et al., 1997; Shaw & McKay, 1942/1969). Collective efficacy can be seen as a buffer between these structural characteristics of the neighborhood and crime-related phenomena, such as fear of crime. From a broken windows perspective, we know that higher levels of disorder can indicate that there is “something wrong” in the neighborhood or that the inhabitants are giving up their neighborhood (Innes, 2004; Kelling & Coles, 1996; Wilson & Kelling, 1982).

In addition to collective efficacy on the neighborhood level, individual social capital offers personal resources and interpersonal trust, based on “the idea that relationships matter” (Field, 2003, p. 1). These research directions, as well as the question what the net effects of individual and neighborhood social processes are key questions in this study.

Social capital is generally used to refer to a myriad of aspects of the social context, ranging from, but not limited to, levels of social support, frequency of social contact with others, social cohesion, or generalized trust (Macinko & Starfield, 2001). It entails both quantitative (e.g., frequency of informal social contacts) and qualitative (e.g., levels of trust) aspects of the social context, which are, respectively, labeled as structural and cognitive components of social capital (Baum & Ziersch, 2003; Harpham, 2008). Notwithstanding the way in which social capital is operationalized, the central idea behind the concept is that being connected to others can provide people access to resources (tangible and/or intangible) they do not own themselves (Macinko & Starfield, 2001).

In literature, a distinction can be observed between two different conceptualizations of social capital (Kawachi, Kim, Coutts, & Subramanian, 2004). The first one, the “social cohesion school,” is strongly influenced by the seminal work of Putnam (Putnam, 1995; Putnam, Leonardi, & Nanetti, 1993) and defines social capital as cohesive processes which are the property of a group (e.g., social trust and informal social control in communities or neighborhoods). The second, a network school, sees social capital as the resources that are embedded within an individual’s social networks (e.g., social support of individuals) and mainly builds upon the work of Bourdieu (1986) and Lin (1999). However, this distinction in particular and the social capital debate in general are not without critiques.

However, a clear distinction between social capital at the individual and collective level is considered problematic because a negative association between levels of avoidance behavior and social capital at the neighborhood level might reflect both the clustering of inhabitants with specifically beneficial individual social capital in specific neighborhoods (“compositional effect”) and/or a “true” effect of neighborhood social capital (“contextual effect”; Lindstrom, Merlo, & Ostergren, 2002; Subramanian, Lochner, & Kawachi, 2003). Therefore, information on social capital at the individual level is needed to fully interpret the relationship between fear of crime outcomes and social capital at the collective level.

The following paragraphs further elaborate on the different theoretical concepts in our framework and discuss in detail how they impact one’s fear of crime. We do not further elaborate on the conceptualization of avoidance behavior, our dependent variable, as we explained this in the introduction.

Neighborhood Collective Efficacy

Collective efficacy theory underlines the importance of a neighborhood’s capacity to solve its commonly identified problems, such as concentrations of crime and fear of crime (Sampson, 2003, 2012; Sampson et al., 1997). This is in line with the aforementioned social cohesion school of social capital; however, there are some notable dissimilarities too, such as Waverijn, Groenewegen, and de Klerk’s (2017) description: “social capital places more emphasis on the value of social networks and the most commonly used measure of collective efficacy incorporates social control over deviant behavior” (p. 415). Collective efficacy is defined as “social cohesion among neighbors combined with their willingness to intervene on behalf of the common good” (Sampson et al., 1997, p. 918). This definition makes a clear link between two main dimensions of collective efficacy, which are social trust and informal social control. A neighborhood has collective efficacy when the combination of both social trust and informal social control is present. Social trust in a neighborhood is an essential condition that fosters informal social control, and thus the willingness to intervene for the common good. Therefore, neighborhoods characterized by high levels of collective efficacy might be more resistant to high local concentrations of disorder and crime and, therefore, show beneficial effects on the fear of crime of its residents (Sampson, 2012). These effects of collective efficacy on fear of crime can be understood as both direct and as indirect through the reduced neighborhood level of disorder and crime.

Neighborhood Disorder

Broken windows theory and Ross and Mirowsky’s condition-cognition-emotion theory emphasize the detrimental effects of social and physical disorder on several indicators, among which fear of crime (Innes, 2004; Kelling & Coles, 1996; Ross & Mirowsky, 2001, 2009; Skogan, 1990; Wilson & Kelling, 1982). Disorder may trigger a chain of negative events (or spiral of decline) that ultimately leads to higher crime rates in neighborhoods. As a consequence, it might be expected that neighborhoods that lack collective efficacy have to cope with more social and physical disorder which, in turn, directly increases the risk of higher levels of fear of crime among its residents (Sampson, 2003, 2012).

Individual Social Capital

Besides the social climate in the neighborhood, it is important to focus on the individual’s social network in relation to fear of crime. This is in line with the aforementioned network school of social capital studies (e.g., Lin, 1999). Social networks are linked to fear of crime because of the support they provide (Colvin, Cullen, & Vander Ven, 2002). In this study, individual social capital consists of two dimensions: the volume or size of the social network and individual social support. The volume or size of the social network represents an individual’s social contacts on a daily basis. Fu (2005) states that this single-item measure is simple, straightforward, and correlates highly to more complex network measures. Individual social support does not solely refer to local social ties that are embedded in the neighborhood but also to the support of social networks that are not bounded by the neighborhood. Researchers have found that an individual’s social support is beneficial for mental and physical health, for example, general well-being and the absence of the symptoms of diseases, and reduces the risk of feeling unsafe, depressive, or oppressive (Coyne & Downey, 1991; Sacco, 1993; Thoits, 1985, 1995, 2011; Wright & Cullen, 2001). Social support directly decreases an individual’s fear of crime by providing opportunities for interaction, help with practical tasks, and relieving feelings of loneliness (Ganster & Victor, 1988; Makarios & Livelsberger, 2013).

Individual Vulnerability

Regardless of the aforementioned theoretical framework, some individual-level factors—such as gender, age, and/or race—affecting the fear of crime, are known as the “vulnerability perspective.” This perspective distillates a number of individual demographical characteristics that are associated with fear of crime. The vulnerability thesis is based on the assumption that individuals with certain characteristics (i.e., proxy measures) perceive themselves as vulnerable to becoming a victim of crime and/or perceive that they are at a (mainly physical) disadvantage against potential assault (Cossman & Rader, 2011; Hindelang, Gottfredson, & Garofalo, 1978; Jackson, 2009; Wyant, 2008). These factors will be taken into account to control for compositional effects. Unfortunately, we cannot make any statement regarding the potential association between avoidance behavior and prior individual victimization because these data are not available for all waves within the Social capital and Well-being in Neighborhoods in Ghent (SWING) study.

Conceptual Model and Hypotheses

Prior studies conclude that specific neighborhood processes, such as collective efficacy and disorder, have an effect on the fear of crime: lower levels of collective efficacy and higher levels of disorder are associated with higher levels of fear of crime (Hardyns, 2010; Wikström & Dolmén, 2001). Furthermore, these studies prove that (components of) fear of crime clusters within socially disorganized neighborhoods, similar to the concentration of crime in the form of hot spots (Chataway, Hart, Coomber, & Bond, 2017; Wyant, 2008). In addition, higher levels of individual social capital are associated with lower levels of fear of crime and, consequently, lower levels of avoidance behavior. The hypothesized effects on avoidance behavior tested in this study were constructed stepwise according to the contextual model (see Figure 1). The hypotheses of this study are as follows:

Hypothesis 1: Significant variation in avoidance behavior can be attributed to the neighborhood level.

Hypothesis 2: The variation in avoidance behavior at the neighborhood level exists independently of the composition of the neighborhood (i.e., sociodemographic and socioeconomic characteristics of individuals).

Hypothesis 3: Neighborhood social trust and informal social control (i.e., dimension of collective efficacy) are associated with lower levels of avoidance behavior, independent of the neighborhood’s (sociodemographic and socioeconomic) composition.

Hypothesis 4: Neighborhood disorder is associated with higher levels of avoidance behavior, independent of the neighborhood’s (sociodemographic and socioeconomic) composition, and it reduces the effects of neighborhood social trust and informal social control.

Hypothesis 5: An individual’s social support and the volume/size of the social network (i.e., dimensions of individual social capital) are associated with lower levels of avoidance behavior, independent of the individual’s sociodemographic and socioeconomic background and neighborhood processes.

Figure 1.

Contextual model—Hypothesized effects on avoidance behavior.

Data and Method

The data collection for this study took place in Ghent. The data were collected for the purposes of the SWING study. Ghent is the second largest city of Belgium and is located in the southwest of Belgium (see Figure 2) with 257,945 residents in 2016 (Stad Gent, 2017) and a surface of 156 km² (±1,653 residents/km²).

Figure 2.

Selected neighborhoods.

SWING Study

The SWING study consists of three successive cross-sectional waves of data collection in neighborhoods¹ in Ghent in 2011, 2012, and 2013. In each cross-sectional wave, multiple methods of data collection were used. In what follows, the methodology and measurements are described per level of the multilevel model.

The Level 1 measurement (i.e., sampling of inhabitants) is based on data which were collected by means of a representative survey in the form of face-to-face interviews with inhabitants. Both face-to-face structured questionnaires and a short self-administered questionnaire (for some possibly sensitive questions, for example, questions about income and substance use) were used to collect the data. For each neighborhood, a randomized stratified sample was drawn from the municipal registry. This sample was stratified by age, sex, and current nationality, therefore being representative of the composition of each neighborhood. The inclusion criteria to participate as a neighborhood inhabitant were as follows: (a) being older than 18 years, (b) not living in an institutional setting (e.g., a retirement home or prison), and (c) having sufficient knowledge of the Dutch language to complete the questionnaire. The intention was to gain the participation of 20 inhabitants in each of the 142 neighborhoods. In total, 2,730 neighborhood inhabitants were reached (Hardyns, Vyncke, Pauwels, & Willems, 2015).

The Level 2 measurement (i.e., sampling of key informants) is based on data which were collected by means of the key informant technique² using standardized questionnaires for the measurement of neighborhood social processes. This technique offers an alternative for simply aggregating the individual-level scores to neighborhood mean scores, to prevent an accumulation of the measurement error at the individual level. Hardyns et al. (2015) point out as follows: “In most studies, the community-level measures of social capital are simply the aggregates of the individual-level measures (using mean scores). However, independent measurement methods for neighborhood social capital should be preferred” (p. 3). In addition, Pauwels and Hardyns (2009) argue that “. . . the knowledge of well-chosen key informants about the social climate of an area is superior to the knowledge of the average inhabitant of that area” (p. 402). They also demonstrate that key informants—if chosen carefully—can fill the gap when measuring processes in local areas. They can provide high-quality data that represent community (dis)organizational processes. In the SWING study, the researchers strove for a heterogeneous set of 8 to 10 key informants per neighborhood. The inclusion criteria to participate as a key informant were as follows: (a) being older than 18 years, (b) having sufficient knowledge of the Dutch language to complete the questionnaire, and (c) being in a work position that presumes an above average knowledge of the social processes in one of the neighborhoods studied. In total, 1,400 key informants were reached (Hardyns et al., 2015).

Making a selection of neighborhoods was not necessary because data are available for the total population of (significant) inhabited neighborhoods. The aim of the study is to measure the association between individual and neighborhood characteristics, and therefore, neighborhoods with none or fewer inhabitants are not included. The SWING study selected only neighborhoods with a minimum population size of 200 adult inhabitants (see Figure 2); hence, 59 of 201 neighborhoods were excluded, resulting in 142 neighborhoods within this study.

Measures

The dependent variable was an individual-level, behavioral component of fear of crime: avoidance behavior. This concept was measured by asking the respondents to indicate how many times they exhibited avoidance behavior. These items were measured on a 5-point Likert-type scale ranging from “never” to “very often.” Avoidance behavior is operationalized by an additive index consisting of three items (see appendix for individual items and results of factor and reliability analysis).

The independent variables for the individual-level measures (Level 1) are used both for the measurement of individual social capital and as controls (i.e., sociodemographic and socioeconomic variables) to distinguish contextual effects from neighborhood compositional effects (Subramanian et al., 2003). Two types of background variables are included to describe the composition of the neighborhood: first, the sociodemographic variables age (as a standardized, continuous variable, in years), gender (0 = male, 1 = female), and nationality (0 = other nationality at birth, 1 = Belgian nationality at birth); second, the socioeconomic variables education, which refers to the respondents’ highest obtained degree, 1 = lower level of secondary education (similar to junior high school in the United States), 2 = highest level of secondary education (similar to high school in the United States), 3 = higher education/post-secondary education, and home ownership (1 = tenant in social rented housing, 2 = tenant in private rented housing, 3 = owner).

Individual social capital refers to the resources available in one’s network and is therefore believed to be a function of one’s network size and one’s access to specific resources (Van Der Gaag & Snijders, 2004). Regarding the network size, respondents are asked to estimate the number of people they had contact with on an average day. This single-item measure is considered to be a simple measure of network size and a valid indicator of individual social capital (Fu, 2005). This variable is incorporated in the model as a trichotomized measure, divided into tertiles: (a) ≤10 contacts, (b) 11 to 25 contacts, and (c) ≥26 contacts. The access to specific sources was measured by asking the respondents to indicate on several items how many friends, family members, or acquaintances would give practical support. These items were summed to create a scale with higher scores indicating higher levels of social support (see appendix for individual items and results of factor and reliability analysis).

We did not include structural neighborhood characteristics in this study because, as corroborated by prior research and as pre-methodological tests in the context of this study³ show, disorder and the social trust component of collective efficacy are strongly associated with socioeconomic disadvantage on the neighborhood level and, as a consequence, this will result in a multicollinearity problem. Because the aim of this study is to examine the effects of (individual and neighborhood) social processes on a behavioral dimension of fear of crime, these structural characteristics are beyond the scope of this contribution.

The independent variables for the neighborhood-level measures (Level 2) were constructed to measure neighborhood social processes through the key informant technique, as discussed earlier. A two-step approach was employed in constructing the neighborhood-level measures (Oberwittler & Wikström, 2009). First, the key informant data were summed to scales to measure social trust, informal social control, and disorder in the neighborhood at the individual level of key informants. Second, the individual scores on these scales were summed and aggregated to the neighborhood level for the respective neighborhood about which he or she was questioned: this after controlling for and with respect for the ecological reliability. All Level 2 variables (social trust, informal social control, and disorder) are scale constructs and entered into the model as standardized variables (based on z scores). In appendix, the individual items, as well as the results of factor analysis and reliability analysis of the Level 2 scale constructs can be found. These analyses reveal both high reliability values (Cronbach’s alpha) and high factor loadings (see appendix).

Analytic Strategy

SPSS Statistics (version 25) is used to perform a two-level hierarchical linear regression model⁴ with individuals at Level 1 and neighborhoods at Level 2. These multilevel analyses account for the nested data structure of people within neighborhoods and allows for estimation of (a) the effect of individual- and neighborhood-level variables on avoidance behavior (fixed part) and (b) the variation in avoidance behavior among neighborhoods that cannot be accounted for by the included independent variables (random part). In addition, multilevel analysis offers insight into the extent to which potential neighborhood differences in avoidance behavior are due to either compositional (individual-level) characteristics or contextual (neighborhood-level) characteristics.

First, an intercept-only model (so called “zero model”) is fitted, without any Level 1 or Level 2 predictors. Then, only individual-level sociodemographic and socioeconomic variables are included in Model 1 to determine to what extent differences in avoidance behavior can be explained as a compositional effect. Model 2 includes the two key dimensions of collective efficacy (Sampson et al., 1997): social trust and informal social control. In Model 3, disorder is added to test to what extent this variable at the neighborhood level is associated with avoidance behavior and to evaluate the net association of collective efficacy with avoidance behavior, both independent of neighborhood composition. In the final model (Model 4), two dimensions of individual social capital were included to verify whether these are associated with lower or higher levels of avoidance behavior, independent of the sociodemographic and socioeconomic background of individuals, and collective efficacy and disorder at the neighborhood level.

To evaluate the improvement of fit for each model, the deviance-statistic was used. This statistic was obtained by a likelihood ratio test. The deviance-statistic follows a chi-square distribution with degrees of freedom equal to the difference in the number of parameters estimated in the two models (Heck, Thomas, & Tabata, 2010; Hox, 2010). Besides, we checked the multicollinearity and none of the variables indicated a problematic Variation Inflation Factor; all values were below 1.6.

Because of the first law of geography, invoked by Waldo R. Tobler (1970)—“everything is related to everything else, but near things are more related than distant things”—we performed a Moran’s test to assess whether spatial dependencies are present in the avoidance behavior data on the neighborhood level. This is necessary because the statistical methods used assume (geographical) independent observations. We used ArcMap (version 10) to merge the spatial and statistical information and to compute Global Moran’s I. Within ArcMap, we chose a contiguity edges borders conceptualization with row standardization, given the shape of neighborhoods and the characteristics of the data. As we obtain a significant Global Moran’s I of 0.31 (z = 5.28, p < .001), we conclude that the spatial distribution of high values and/or low values in the variable avoidance behavior is more spatially clustered than would be expected if the underlying spatial processes were random. Because there is a significant spatial autocorrelation, this should be kept in mind when interpreting the results.

Results

The descriptive statistics of the 2,730 inhabitant-respondents and 142 neighborhoods are, respectively, listed in Table 1. The compositional variables show that gender is almost equally distributed because 48% of the sample are men and 52% of the sample are women. The mean age of the respondents is 48 years and ranges from 18⁵ to 95 years. Almost 10% of respondents had a non-Belgian nationality at birth. The majority of the sample is home owner (almost 70%). From the inhabitants of the sample who live in rental dwellings, 10% live in social rental housing and 20% live in private rental housing. Half of the sample has obtained a degree in higher education.

Table 1.

Descriptive Statistics (N = 2,730 Respondents in N = 142 Neighborhoods).

	M/percentage	SD	Minimum	Maximum
Avoidance behavior	5.81	2.99	3.00	15.00
Individual characteristics
Gender
Man	48.00%
Woman	52.00%
Age	47.97	18.76	18	95
Nationality
Belgian	90.30%
Other	9.70%
Home ownership
Owners	68.60%
Tenants in social rented housing	10.50%
Tenants in private rented housing	20.90%
Education level
Low	18.90%
Middle	33.80%
High	47.40%
Individual social support	17.25	6.32	0	28
Volume/size of social network
≤10	44.62%
11-25	25.27%
≥26	30.11%
Neighborhood characteristics
Social trust	14.22	1.34	10.33	17.25
Informal social control	18.37	2.20	12.00	24.50
Disorder	12.62	3.26	6.38	20.63

Avoidance behavior is an additive index consisting of three items, and as can be seen in Table 1, the full range of possible responses is met (3-15). The mean avoidance behavior (5.81) indicates that inhabitants with high levels of avoidance behavior are in the minority. Avoidance behavior has a significant positive skewed distribution (skewness = 1.03, standard error of skewness = 0.05).

The bivariate correlations between the neighborhood-level characteristics are presented in Table 2. As can be seen, there is a strong, significant negative association between neighborhood disorder and neighborhood social trust. Another observation worth mentioning is the correlation between the two dimensions of collective efficacy. In the well-known Chicago-study of Sampson et al. (1997), these two dimensions are significantly, strongly correlated (R = .80, p < .001). In general, there is a notable difference with the U.S. studies because, in the case of Ghent, social trust and informal social control do weakly correlate (R = .13), and this correlation is also not significant (p > .05). This corroborates earlier findings in a non-U.S. context (Hardyns, 2010; Zhang, Messner, & Liu, 2007; Zhang, Messner, Liu, & Zhuo, 2009). At the moment, we are unsure relating to the causes of this incoherence; however, this discussion is beyond the scope of this study.

Table 2.

Bivariate Correlations of Neighborhood-Level Variables (N = 142).

	1	2	3
1. Social trust	1
2. Informal social control	.13^ns	1
3. Disorder	−.57***	.09^ns	1

p < .05. **p < .01. ***p < .001.

The two individual-level parameters of different dimensions of social capital, that is, (a) individual social support and (b) the volume/size of the social network (both included as ratio variable), show a significant but small positive correlation (R = .20, p < .001). This small association indicates that both variables measure separate dimensions of individual social capital, indicating the existence of a multidimensional concept of social capital at the individual level.

Table 3 presents the results of five successive models for avoidance behavior, corresponding with the hypotheses as formulated earlier. Model 0 is an intercept-only model, without explanatory variables. Based on the findings in Table 3, it can be concluded that there is significant variation in avoidance behavior at the neighborhood level. The Intra-class Correlation Coefficient (ICC) is 8.06%, thus 8.06% of the observed individual differences in avoidance behavior can be explained by characteristics at the neighborhood level.

Table 3.

Estimates (and Standard Errors) of the Hierarchical Linear Models Explaining Avoidance Behavior (N = 2,730).

	Model 0B (SE)	Model 1B (SE)	Model 2B (SE)	Model 3B (SE)	Model 4B (SE)
Level 2: Neighborhood-level characteristics (fixed effects)
Social trust			−0.41*** (0.07)	ns	ns
Informal social control			ns	−0.15* (0.06)	ns
Disorder				0.53*** (0.08)	0.51*** (0.08)
Level 1: Individual-level characteristics (fixed effects)
Gender (ref. cat. = women)		−1.64*** (0.10)	−1.65*** (0.10)	−1.66*** (0.10)	−1.66*** (0.10)
Age		0.72*** (0.05)	0.73*** (0.05)	0.75*** (0.05)	0.57*** (0.06)
Nationality (ref. cat. = Belgian)		−0.44* (0.18)	−0.51** (0.18)	−0.57*** (0.18)	−0.70*** (0.18)
Education (ref. cat. = high)
Low		1.13*** (0.15)	1.11*** (0.15)	1.09*** (0.15)	0.90*** (0.15)
Middle		0.77*** (0.11)	0.78*** (0.11)	0.76*** (0.11)	0.61*** (0.11)
Home ownership (ref. cat. = owner)
Tenants in social rented housing		1.36*** (0.19)	1.27*** (0.19)	1.13*** (0.19)	0.97*** (0.19)
Tenants in private rented housing		ns	ns	ns	ns
Individual social support					−0.36*** (0.06)
Volume/size of social network (ref. cat. = ≥26)
≤10					0.37** (0.13)
11-25					ns
Intercept	5.79*** (0.09)	5.97*** (0.11)	6.00*** (0.11)	6.05*** (0.10)	6.00*** (0.13)
Random effects
Neighborhood	0.721*** (0.14)	0.543*** (0.11)	0.387*** (0.09)	0.221** (0.07)	0.224** (0.07)
Individual	8.232*** (0.23)	6.436*** (0.18)	6.433*** (0.18)	6.433*** (0.18)	6.280*** (0.18)
ICC	8.06%	7.78%	5.67%	3.32%	3.44%
Δ Deviance (Δdf)		832 (4)***	24 (2)***	33 (1)***	264 (2)***

Note. N (Level 1) = 2,730, N (Level 2) = 142. ICC = Intra-class Correlation Coefficient.

p < .05. **p < .01. ***p < .001.

In Model 1, we added demographic control variables and reinterpreted the variation at the neighborhood level. This model shows that only a small percentage of the ecological variation is caused by the compositional effect because still 7.78% of the observed individual differences in avoidance behavior can be explained by characteristics at the neighborhood level. The estimates show that the women, the elderly, the less educated, and tenants of social rented housing have higher levels of avoidance behavior. This corroborates the vulnerability thesis. The results also show that non-Belgian inhabitants experience significant lower levels of avoidance behavior than Belgian inhabitants.

From Model 2, the neighborhood (Level 2) characteristics are introduced in the models. In Model 2, the two dimensions of collective efficacy, as described by Sampson et al. (1997), are added. Social trust has a significant negative effect on the levels of avoidance behavior. In other words, neighborhoods with higher levels of social trust have significant lower levels of avoidance behavior among the inhabitants. The impact of informal social control is not significant. The introduction of social trust has a significant decreasing effect on the ICC by a substantial reduction of the between-neighborhood variance in avoidance behavior. After addition of neighborhood social trust, 5.67% of the observed individual differences in avoidance behavior can be explained by the characteristics at the neighborhood level.

In Model 3, disorder is introduced as an independent variable at the neighborhood level. The net effect of disorder is in line with our hypothesis: higher levels of disorder are associated with higher levels of avoidance behavior. This association is significant (p < .001). This neighborhood indicator of incivilities has on its own a significant impact on the model because the ICC is reduced to 3.32%. The impact of the introduction of this variable on the other contextual characteristics is notable. By the introduction of disorder, the net effect of informal social control becomes significant and the net effect of social trust becomes nonsignificant.

In the final model, the two dimensions of individual social capital are added at the individual level. As shown in Table 3, both individual social support and the volume/size of the social network matter. Those who report higher levels of individual social support report lower levels of avoidance behavior. The association between a smaller size of the social network and avoidance behavior is, however, small, but significant. The group with 11 to 25 daily contacts, which can be seen as a group with a moderate size social network, has a nonsignificant association with avoidance behavior. The introduction of this individual-level characteristics results in the disappearance of the net effects of both dimensions of collective efficacy. The values of several compositional variables are changing due to introducing individual social capital. Noteworthy are the decreases of the estimates of age, education, and home ownership, and the increase of the negative estimate of avoidance behavior of Belgian inhabitants. A slight increase of the ICC is observed due to a decrease in the random effects on the individual level. All model improvements are significant, as can be seen by the deviance statistics in Table 3. As a result, we can conclude that Model 4, with all the variables presented, has the best model fit.

Discussion

This study corroborated some noticeable findings regarding the impact of individual social capital indicators and neighborhood characteristics on avoidance behavior. Although the effects are relatively small, these impacts do significantly matter. As hypothesized, significant variation in avoidance behavior can be attributed to the neighborhood level, as the unconditional model show that 8.06% of the observed differences in avoidance behavior can be explained by characteristics at the neighborhood level. Accounting for possible compositional effects resulted in a little decrease of 0.28% point, to 7.78%. This indicates that a significant variation in avoidance behavior exists independently of the composition of the neighborhood.

The included compositional variables are thus responsible for a small part of the observed neighborhood-level variation. Women, age (the elderly), (lower) levels of education, and tenants of social rented housing are significantly related to higher levels of avoidance behavior, in accordance with the vulnerability thesis. However, two limitations of our study have to be mentioned. First, we cannot test whether the vulnerability is associated with other (psycho)social processes (i.e., the “causes of the causes,” see Wikström, 2011), such as prior individual victimization, because these data are not available for all waves within the SWING study. Second, we cannot draw conclusions on the observation that non-Belgian inhabitants report lower levels of avoidance behavior because the “real” effects are indistinguishable from compositional effect. This is due to sampling bias: one of the inclusion criteria was having sufficient knowledge of the Dutch language to complete the questionnaire. As a result, the sample of non-Belgian inhabitants is not representative and, hence, problematic. For example, the average age of the total sample (N = 2,730) is 47.97 years, the average age of Belgian inhabitants is 49.06 years, and the average age of non-Belgian inhabitants is 37.86 years. Given the general positive association of age with levels of avoidance behavior, this could be a possible (compositional) explanation of the observed effect of nationality. On the other hand, a “true” effect would mean that ethnic majorities (i.e., Belgian inhabitants) experience “fear of the others” (i.e., non-Belgian inhabitants; e.g., Lupton, 1999).

Although the compositional factors are indeed important in understanding individual differences in avoidance behavior, let us now turn to the role of the neighborhood the inhabitants live in. When only the two dimensions of collective efficacy are included at the neighborhood level, social trust is important in the ecological variance in avoidance behavior. However, this strong effect diminishes when neighborhood disorder is added to the model. Remarkably, this effect of neighborhood disorder is accompanied by a significant negative effect of informal social control at the neighborhood level. This shows that neighborhood disorder does not eliminate all the other effects at the neighborhood levels and, hence, there also are significant social processes at work.

In addition, looking at the indicators of social capital at the individual level, our contribution shows that these social processes are important in explaining an individual’s avoidance behavior. Both dimensions of individual social capital are important concepts with significant effects. Higher levels of individual social support and a larger social network are associated with lower levels of avoidance behavior.

People living in neighborhoods with lower levels of collective efficacy and higher levels of disorder tend to report more avoidance behavior, independent of their demographic background. In this regard, we want to stress that the direct association of neighborhood disorder is stronger than the association of neighborhood social trust or informal social trust with avoidance behavior. From a policy point of view, this could indicate that diminishing the visual and lived triggers of disorder could be more effective to reduce inhabitants’ avoidance behavior. After all, the results show that the neighborhood contextual effects should not be overestimated and, at the same time, certainly not underestimated too because the ICC indicates there is a significant margin left to invest in the contextual effects. Besides neighborhood aspects of the social climate, it is important to mention the role of individual social capital. The social support an individual can count on and the volume/size of one’s social network does make a difference to an individual’s avoidance behavior. By introducing individual social capital in our model, the effect of informal social control of the neighborhood level diminishes, indicating that there is a micro-macro-interaction between these two-level dimensions of social capital.

One remarkable finding of this study is that the two dimensions of collective efficacy, social trust and informal social control, do not correlate at the neighborhood level. In line with prior findings of Carpiano (2006), Hardyns (2012), Pauwels and Hardyns (2010), and Hardyns, Vyncke, De Boeck, Pauwels, and Willems (2016) this is another indication that, dependent on the specific setting, the collective efficacy concept, as measured by key informants, should be split into both dimensions when studying contextual effects. Other researchers have emphasized that collective efficacy results from a specific international context and cannot simply be transferred and applied to other countries and settings (Reisig & Cancino, 2004; Zhang et al., 2007; Zhang et al., 2009). However, these differences could be due to different measurements; we think the questions used to measure collective efficacy (and especially informal social control) are specifically tuned to metropolitan cities and are not suited to large regional cities, such as the city of Ghent.

Another methodological finding has to be noticed. Neighborhoods were operationalized using administrative units in this study. It is not clear to what extent these units coincide with what the respondents perceive as “their” neighborhood and, as a result, to what extent the Modifiable Areal Unit Problem (MAUP) distorts the results (Dark & Bram, 2007). Although we followed the advice “smaller is better” (Oberwittler & Wikström, 2009) by selecting the lowest level administrative units available in Belgium, it is possible that neighborhood effects are underestimated due to heterogeneity in relevant factors within these geographical areas. Also related to the spatial context of our study, we found a significant spatial autocorrelation in the dependent variable. Although multilevel models inherently overcompensate for spatial autocorrelation, this is a notable implication that should be kept in mind.

Future Research

To conclude, we want to draw special attention to some recommendations for future studies on the relation between social capital and fear of crime. First, referring to the aforementioned methodological implications regarding the use of administrative units and the fact that there are many evidences about the usefulness of smaller units of analysis (e.g., Weisburd, 2015; Weisburd, Bernasco, & Bruinsma, 2009), the fear of crime should be measured and examined at units of analysis that are as small and homogeneous as possible. For that reason, it could be interesting to execute similar studies with units of analysis such as street segments, street-blocks (Weisburd, Bruinsma, & Bernasco, 2009), or grid-based units.

Another, second methodological point of particular interest should be the attention to reaching out to vulnerable groups, for example, non-native speakers, isolated people, and the homeless living in a neighborhood; however, they are often under-represented or even unregistered.

Third, it is likely that neighborhood effects on someone’s fear of crime are dependent on individual characteristics such as the length of residence of an individual in the neighborhood or prior individual victimization. For that reason, it is advisable to study the interaction effects of these determinants, to better reflect the construction of someone’s fear of crime.

Fourth, the combination of individual and neighborhood measures of social capital needs to be stimulated in future studies. Due to a continuously increasing geographic mobility of people, it can be expected that neighborhood dimensions of social capital will lose importance in favor of individual dimensions of social capital.

Footnotes

Appendix

Results of Factor and Reliability Analyses Regarding Scale Construction for Relevant Indicators.

Summative scale and individual items	Coding	Cronbach’s alpha/factor loadings
Individual characteristics
Avoidance behavior		.76
Does it ever happen that . . .
1. you avoid certain areas in your neighborhood because you think they are not safe?	never → very often^a	.64
2. you avoid opening the door for strangers because you think it is not safe?	never → very often^a	.76
3. you avoid leaving home after dark because you think it is not safe?	never → very often^a	.76
Individual social support		.81
How many friends, family members, or acquaintances . . .
1. understand your problems?	0 → >10^b	.72
2. would let you move into their house for a week if you temporarily could not stay at your house?	0 → >10^b	.72
3. would encourage you to go to the doctor if you experience health problems?	0 → >10^b	.73
4. make you feel good? (e.g., make you feel you are useful or make you feel they are glad to know you)	0 → >10^b	.70
Neighborhood characteristics
Social trust		.80
1. People around here are willing to help their neighbors	totally disagree → totally agree^a	.76
2. This is a close-knit neighborhood	totally disagree → totally agree^a	.76
3. People in this neighborhood can be trusted	totally disagree → totally agree^a	.62
4. Contacts between inhabitants in this neighborhood are generally positive	totally disagree → totally agree^a	.72
Informal social control		.84
How likely is it that you could count on neighbors intervening when . . .
1. children were skipping school and hang out on a street corner	very likely → very unlikely^a	.61
2. children were spray-painting graffiti on a local building	very likely → very unlikely^a	.73
3. children were showing disrespect to an adult	very likely → very unlikely^a	.69
4. a fight breaks out in front of their house	very likely → very unlikely^a	.72
5. children were making too much racket	very likely → very unlikely^a	.72
6. children are using soft drugs (smoking weed, hasj, etc.)	very likely → very unlikely^a	.63
Disorder		.85
1. Adolescents hanging around on street corners	never → very often^a	.66
2. Groups of adolescents harassing people to obtain money or goods	never → very often^a	.78
3. Men drinking alcohol in public	never → very often^a	.67
4. People selling drugs (hash, weed, etc.) on the streets	never → very often^a	.70
5. People being threatened on the streets with weapons or knives	never → very often^a	.65
6. Fights between adolescents on the streets	never → very often^a	.80

Note. Cronbach’s alphas in bold and the remaining values are factor loadings.

5-point Likert-type scale.

8-point scale.

Acknowledgements

The research team wishes to thank the city of Ghent for their help in facilitating this research project.

Authors’ Note

For information about the underlying research materials related to this paper, you can contact the corresponding author.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article. The entire research team wishes to express its sincere gratitude for the financial support provided by the Research Foundation—Flanders (FWO).

Notes

ORCID iD

Wim Hardyns

Author Biographies

Wim Hardyns is a professor of criminology at the Department of Criminology, Criminal Law and Social Law at Ghent University. He is member of the Institute of International Research on Criminal Policy (IRCP) since June 2015. He holds a PhD in criminology and has a master’s degree in criminology and business economics. His current interests are crime mapping and statistics, environmental criminology, crime prevention, new security technologies, big data, radicalization and terrorism, and sport-related crime. With reference to these topics, he has published numerous articles (and contributions to books) in national and international peer-reviewed journals and series.

Thom Snaphaan is an academic assistant and PhD researcher at the Department of Criminology, Criminal Law and Social Law at Ghent University. He is member of the IRCP since January 2018. He holds a master’s degree in safety sciences. His currents interests are environmental criminology, big data, new technologies, terrorism, and quantitative research methods.

Lieven J. R. Pauwels is a professor of criminology at the Department of Criminology, Criminal Law and Social Law at Ghent University and Director of the IRCP since June 2015. He received his PhD in criminology from Ghent University in 2006. His dissertation dealt with the topic of neighborhood contextual effects on adolescent offending: a test of social disorganization theory. He is interested in crime causation theories, unifying frameworks and empirical tests, philosophy of causation, and innovative methods in quantitative criminology.

Veerle Vyncke is a postdoctoral researcher at the Department of Family Medicine and Primary Health Care at Ghent University. She obtained her PhD in March 2015 with an exploratory research on social capital’s differential association with health in the context of health inequity. Her main research interests are social determinants of health, prison health, community oriented primary care, intersectoral care, social inclusion, and individual and community social networks.

Sara Willems is a professor of inequity in health care at the Department of Family Medicine and Primary Health Care at Ghent University. She leads the research group on equity in (primary) health care. This interdisciplinary research group leads or partners in numerous national and international research projects on the relation between health care organization and equity in access to care, treatment and outcomes of care, and the buffering effect of social capital in the relation between social status and health inequity.

References

Baum

Ziersch

A. M.

(2003). Social capital. Journal of Epidemiology and Community Health, 57, 320-323.

Bourdieu

(1986). The forms of capital. In Richardson

(Ed.), Handbook of theory and research for the sociology of education (pp. 241-258). New York, NY: Greenwood.

Bursik

R. J.

Grasmick

H. G.

(1993). Neighborhoods and crime. New York, NY: Lexington.

Carpiano

R. M.

(2006). Toward a neighborhood resource-based theory of social capital for health: Can Bourdieu and sociology help? Social Science & Medicine, 62, 165-175.

Chataway

Hart

T. C.

Coomber

Bond

(2017). The geography of crime fear: A pilot study exploring event-based perceptions of risk using mobile technology. Applied Geography, 86, 300-307.

Coleman

J. S.

(1988). Social capital in the creation of human-capital. American Journal of Sociology, 94, S95-S120.

Colvin

Cullen

F. T.

Vander Ven

(2002). Coercion, social support, and crime: An emerging theoretical consensus. Criminology, 40, 19-42.

Cossman

J. S.

Rader

N. E.

(2011). Fear of crime and personal vulnerability: Examining self-reported health. Sociological Spectrum, 31, 141-162.

Coyne

J. C.

Downey

(1991). Social-factors and psychopathology: Stress, social support, and coping processes. Annual Review of Psychology, 42, 401-425.

10.

Dark

S. J.

Bram

(2007). The Modifiable Areal Unit Problem (MAUP) in physical geography. Progress in Physical Geography, 31, 471-479.

11.

Farrall

Jackson

Gray

(2009). Social order and the fear of crime in contemporary times. New York, NY: Oxford University Press.

12.

Fattah

E. A.

Sacco

V. F.

(1989). Crime and victimization of the elderly. New York, NY: Springer-Verlag.

13.

Ferraro

K. F.

(1995). Fear of crime: Interpreting victimization risk. Albany: State University of New York Press.

14.

Ferraro

K. F.

LaGrange

(1987). The measurement of fear of crime. Sociological Inquiry, 57, 70-101.

15.

Field

(2003). Social capital. London, England: Routledge.

16.

Franklin

T. W.

Franklin

C. A.

Fearn

N. E.

(2008). A multilevel analysis of the vulnerability, disorder, and social integration models of fear of crime. Social Justice Research, 21, 204-227.

17.

(2005). Measuring personal networks with daily contacts: A single-item survey question and the contact diary. Social Networks, 27, 169-186.

18.

Furstenberg

F. F.

(1972). Fear of crime and its effects on citizen behavior. Paper presented at the Symposium on Studies of Public Experience, Knowledge and Opinion of Crime and Justice, Washington, DC.

19.

Gabriel

Greve

(2003). The psychology of fear of crime: Conceptual and methodological perspectives. British Journal of Criminology, 43, 600-614.

20.

Ganster

D. C.

Victor

(1988). The impact of social support on mental and physical health. British Journal of Medical Psychology, 61, 17-36.

21.

Greve

(1998). Fear of crime among the elderly: Foresight, not fright. International Review of Victimology, 5, 277-309.

22.

Hale

(1996). Fear of crime: A review of the literature. International Review of Victimology, 4, 79-150.

23.

Hanley

Ruppanner

(2015). Understanding the effects of crime on women: Fear and well-being in the context of diverse relationships. Social Sciences, 4, 276-293.

24.

Hardyns

(2010). Social cohesion and crime. A multilevel study of collective efficacy, victimization and fear of crime (Doctoral thesis). Ghent University, Belgium.

25.

Hardyns

(2012). Social cohesion and fear of crime in the Flemish region: Looking back on the results and a preview of additional research. In Ponsaers

(Ed.), Social analysis of security: Financial, economic and ecological crime—Crime (in)security and (dis)trust—Public and private policing (pp. 167-202). The Hague, The Netherlands: Eleven International Publishing.

26.

Hardyns

Pauwels

(2010). Different measures of fear of crime and survey measurement error. In Cools

et al . (Eds.), Safety, societal problems and citizens’ perceptions: New empirical data, theories and analyses (Vol. 3, pp. 19-39). Antwerp, Belgium: Maklu.

27.

Hardyns

Vyncke

De Boeck

Pauwels

Willems

(2016). Are collective efficacy, disorder and social support associated with one’s quality of life? Evidence from the multilevel SWING study in Belgium. Applied Research in Quality of Life, 11, 739-756.

28.

Hardyns

Vyncke

Pauwels

Willems

(2015). Study protocol: SWING—Social capital and well-being in neighborhoods in Ghent. International Journal for Equity in Health, 14(1), Article 36. doi:10.1186/s12939-015-0163-1

29.

Harpham

(2008). The measurement of community social capital through surveys. In Kawachi

Subramanian

S.V.

Kim

(Eds.), Social capital and health (pp. 51-62). New York, NY: Springer.

30.

Heck

R. H.

Thomas

S. L.

Tabata

L. N.

(2010). Multilevel and longitudinal modeling with SPSS. New York, NY: Routledge.

31.

Hindelang

M. J.

Gottfredson

M. R.

Garofalo

(1978). Victims of personal crime: An empirical foundation for a theory of personal victimization. Cambridge, MA: Ballinger.

32.

Hox

J. J.

(2010). Multilevel analysis: Techniques and applications. New York, NY: Routledge.

33.

Innes

(2004). Signal crimes and signal disorders: Notes on deviance as communicative action. British Journal of Sociology, 55, 335-355.

34.

Jackson

(2004). An analysis of a construct and debate: The fear of crime. In Albrecht

H.-J.

Serassis

Kania

(Eds.), Images of crime II (pp. 35-64). Freiburg, Germany: Edition Iuscrim, Max Planck Institute.

35.

Jackson

(2005). Validating new measures of the fear of crime. International Journal of Social Research Methodology, 8, 297-315.

36.

Jackson

(2009). Bridging the social and the psychological in the fear of crime. In Lee

Farrall

(Eds.), Fear of crime: Critical voices in an age of anxiety (pp. 143-167). London, England: Routledge.

37.

Kawachi

Kim

Coutts

Subramanian

S. V.

(2004). Health by association? Social capital, social theory, and the political economy of public health—Commentary: Reconciling the three accounts of social capital. International Journal of Epidemiology, 33, 682-690.

38.

Kelling

G. L.

Coles

C. M.

(1996). Fixing broken windows. Restoring order and reducing crime in our communities. New York, NY: The Free Press.

39.

Lin

(1999). Building a network theory of social capital. Connections, 22(1), 28-51.

40.

Lindstrom

Merlo

Ostergren

P. O.

(2002). Individual and neighbourhood determinants of social participation and social capital: A multilevel analysis of the city of Malmo, Sweden. Social Science & Medicine, 54, 1779-1791.

41.

Lupton

(1999). Dangerous places and the unpredictable strangers: Constructions of fear of crime. The Australian & New Zealand Journal of Criminology, 32(1), 1-15.

42.

Macinko

Starfield

(2001). The utility of social capital in research on health determinants. The Milbank Quarterly, 79, 387-427.

43.

Makarios

M. D.

Livelsberger

(2013). Social support and crime. In Cullen

F. T.

Wilcox

(Eds.), The Oxford handbook on criminological theory (pp. 160-185). Oxford, UK: Oxford University Press.

44.

Oberwittler

Wikström

P.-O. H.

(2009). Why small is better: Advancing the study of the role of behavioral contexts in crime causation. In Bruinsma

G. J. N.

Weisburd

Bernasco

(Eds.), Putting crime in its place: Units of analysis in geographic criminology (pp. 33-58). New York, NY: Springer.

45.

Pauwels

Hardyns

(2009). Measuring community (dis)organizational processes through key informant analysis. European Journal of Criminology, 6, 401-417. doi:10.1177/1477370809337878

46.

Pauwels

Hardyns

(2010). Does it really have to be “an urban village”? A multilevel analysis of fear of crime. In Pauwels

Hardyns

Van de Velde

(Eds.), Social disorganisation, offending, fear and victimisation: Findings from Belgian studies on the urban context of crime (pp. 177-197). Den Haag, The Netherlands: Boom Juridische Uitgevers.

47.

Putnam

(1993). Making democracy work: Civic traditions in modern Italy. Princeton, NJ: Princeton University Press.

48.

Putnam

(1995). Bowling alone: America’s declining social capital. Journal of Democracy, 6, 65-78.

49.

Putnam

Leonardi

Nanetti

(1993). Making democracy work: Civic traditions in modern Italy. Princeton, NJ: Princeton University Press.

50.

Rader

(2017). Fear of crime. In Oxford Research Encyclopedia of Criminology. Retrieved from http://criminology.oxfordre.com/view/10.1093/acrefore/9780190264079.001.0001/acrefore-9780190264079-e-10

51.

Reisig

M. D.

Cancino

J. M.

(2004). Incivilities in nonmetropolitan communities: The effects of structural constraints, social conditions, and crime. Journal of Criminal Justice, 32, 15-29.

52.

Ross

C. E.

Mirowsky

(2001). Neighborhood disadvantage, disorder, and health. Journal of Health and Social Behavior, 42, 258-276.

53.

Ross

C. E.

Mirowsky

(2009). Neighborhood disorder, subjective alienation, and distress. Journal of Health and Social Behavior, 50, 49-64.

54.

Sacco

V. F.

(1993). Social support and the fear of crime. Canadian Journal of Criminology, 35, 187-196.

55.

Sampson

R. J.

(2003). The neighborhood context of well-being. Perspectives in Biology and Medicine, 46(3), S53-S64.

56.

Sampson

R. J.

(2012). Great American city: Chicago and the enduring neighborhood effect. Chicago, IL: The University of Chicago Press.

57.

Sampson

R. J.

Raudenbush

S. W.

Earls

(1997). Neighborhoods and violent crime: A multilevel study of collective efficacy. Science, 277, 918-924.

58.

Shaw

C. R.

McKay

(1969). Juvenile delinquency and urban areas: A study of rates of delinquency in relation to differential characteristics of local communities in American cities. Chicago, IL: The University of Chicago Press. (Original work published 1942)

59.

Skogan

W. G.

(1990). Disorder and decline. New York, NY: The Free Press.

60.

Stad Gent. (2017). Gent in cijfers [Ghent in numbers]. Available from https://gent.buurtmonitor.be/

61.

Subramanian

S. V.

Lochner

K. A.

Kawachi

(2003). Neighborhood differences in social capital: A compositional artifact or a contextual construct? Health & Place, 9, 33-44. doi:10.1016/S1353-8292(02)00028-X

62.

Thoits

P. A.

(1985). Social support processes and psychological wellbeing theoretical possibilities. In Sarason

G. I.

Sarason

B. R.

(Eds.), Social support theory, research and applications (pp. 51-72). The Hague, The Netherlands: Martinus Nijhof.

63.

Thoits

P. A.

(1995). Stress, coping, and social support processes: Where are we? What next? Journal of Health and Social Behavior, 35, 53-79.

64.

Thoits

P. A.

(2011). Mechanisms linking social ties and support to physical and mental health. Journal of Health and Social Behavior, 52, 145-161.

65.

Tobler

W. R.

(1970). A computer movie simulating urban growth in the Detroit region. Economic Geography, 46, 234-240.

66.

Van Der Gaag

M. P. J.

Snijders

T. A. B.

(2004). Proposals for the measurement of individual social capital. In Flap

Volker

(Eds.), Creation and returns of social capital (pp. 199-218). London, England: Routledge.

67.

Vanderveen

(2006). Interpreting fear, crime, risk and unsafety. Den Haag, The Netherlands: Boom Juridische Uitgevers.

68.

Waverijn

Groenewegen

P. P.

de Klerk

(2017). Social capital, collective efficacy and the provision of social support services and amenities by municipalities in the Netherlands. Health and Social Care in the Community, 25, 414-423.

69.

Weisburd

(2015). The law of crime concentrations and the criminology of place. Criminology, 53, 133-157.

70.

Weisburd

Bernasco

Bruinsma

G. J. N.

(2009). Putting crime in its place: Units of analysis in geographic criminology. New York, NY: Springer.

71.

Weisburd

Bruinsma

G. J. N.

Bernasco

(2009). Units of analysis in geographic criminology: Historical development, critical issues, and open questions. In Bruinsma

G. J. N.

Weisburd

Bernasco

(Eds.), Putting crime in its place: Units of analysis in geographic criminology (pp. 3-31). New York, NY: Springer.

72.

Wikström

P.-O. H.

(2011). Does everything matter? Addressing the problem of causation and explanation in the study of crime. In McGloin

J. M.

Sullivan

C. J.

Kennedy

L. W.

(Eds.), When crime appears: The role of emergence (pp. 53-72). London, England: Routledge.

73.

Wikström

P.-O. H.

Dolmén

(2001). Urbanisation, neighbourhood social integration, informal social control, minor social disorder, victimisation and fear of crime. International Review of Victimology, 8, 121-140.

74.

Wilson

J. Q.

Kelling

G. L.

(1982, March). Broken windows: The police and neighborhood safety. The Atlantic, 249, 29-38.

75.

Wright

J. P.

Cullen

F. T.

(2001). Parental efficacy and delinquent behavior: Do control and support matter? Criminology, 39, 677-705.

76.

Wyant

B. R.

(2008). Multilevel impacts of perceived incivilities and perceptions of crime risk on fear of crime. Journal of Research in Crime and Delinquency, 45, 39-64.

77.

Zhang

L. N.

Messner

S. F.

Liu

J. H.

(2007). A multilevel analysis of the risk of household burglary in the city of Tianjin, China. British Journal of Criminology, 47, 918-937.

78.

Zhang

L. N.

Messner

S. F.

Liu

J. H.

Zhuo

Y. A.

(2009). Guanxi and fear of crime in contemporary urban China. British Journal of Criminology, 49, 472-490.