Navigating Dangerous Streets: The Sources and Consequences of Street Efficacy

Patrick T. Sharkey Harvard University

The concept of street efficacy, defined as the perceived ability to avoid violent confrontations and to be safe in one’s neighborhood, is proposed as a mechanism connecting aspects of adolescents’“imposed” environments to the choices they make in creating their own “selected” environments that minimize the potential for violent confrontations. Empirical models using data from the Project on Human Development in Chicago Neighborhoods suggest that street efficacy is substantially influenced by various aspects of the social context surrounding adolescents. Adolescents who live in neighborhoods with concentrated disadvantage and low collective efficacy, respectively, are found to have less confidence in their ability to avoid violence after controlling for an extensive set of individual- and family-level factors. Exposure to violence also reduces street efficacy, although it does not explain the association between collective efficacy and individual street efficacy. Adolescents’ confidence in their ability to avoid violence is shown to be an important predictor of the types of environments they select for themselves. In particular, adolescents with high levels of street efficacy are less likely to resort to violence themselves or to associate with delinquent peers.

esearchers across the social sciences face Katz 1988). A key task for researchers studying Ran inherent difficulty in attempting to incor- adolescent development is to account for indi- porate the neighborhood context into individual- vidual agency and the variation in individual level models of adolescent behavior. Although behavior that occurs within social areas such as convincing arguments have been made about neighborhoods or schools, while also acknowl- why the places where adolescents live matter for edging the role that social contexts play in shap- their developmental trajectories (Brooks-Gunn ing developmental trajectories. et al. 1993; Gephart 1997; Wilson 1987), much To do so, it is necessary to make a concep- sociological research ignores the agency of indi- tual distinction between adolescents’“imposed” viduals as they navigate their social worlds and “selected” environments (Bandura (Bandura 1997; Bandura 2001; Coleman 1988; 1997:163), and to identify the relationship

Direct all correspondence to Patrick T. Sharkey, Harvard Proseminar on Inequality and Social Policy, Harvard University, William James Hall, 33 Kirkland ASR editor Jerry Jacobs, and anonymous ASR review- Street, Cambridge, MA 02138 (sharkey@fas. ers for helpful comments on an earlier draft of the harvard.edu). The John D. and Catherine MacArthur article. A different version of this article was pre- Foundation, the National Institute of Justice, and the sented at the annual meeting of the American National Institute of Mental Health provided fund- Sociological Association in August 2005, and the ing support for the collection of the data used in this author thanks all the participants and audience mem- research. The author thanks Robert Sampson, William bers in the session for their comments. Finally, the J. Wilson, and Christopher Winship for their thoughts author is particularly indebted to Robert Sampson and and comments throughout the development of the Steve Raudenbush for an extensive discussion on article. The author also thanks Thomas Cook, Jim the concept of street efficacy that took place early in Quane, Steve Raudenbush, participants in the the development of the manuscript.

AMERICAN SOCIOLOGICAL REVIEW, 2006, VOL. 71 (October:826–846) NAVIGATING DANGEROUS STREETS—–827 between the two. Adolescents are subject to expend in selecting environments and peer certain aspects of their physical and social envi- groups that minimize the potential for violent ronments over which they have little control, and confrontations. In this sense, social cognition is which may limit the range of choices or oppor- conceptualized as a primary mechanism con- tunities available to them. It is this imposed necting the imposed environment and the select- environment on which researchers studying ed environment, as depicted in Figure 1. “neighborhood effects” focus in their analysis To test the relationship between the imposed of adolescent development. However, within neighborhood context, street efficacy, and the the imposed environment, adolescents are able selected environment, several series of empiri- to exercise choice, whether that choice is con- cal models are estimated using data from the strained or not, and to select certain features of Project on Human Development in Chicago the environment that enable them to pursue Neighborhoods (Earls et al. 1994; Earls et al. their own unique set of goals and ambitions. In 1995). The first series examines the sources of this sense, individual agency plays a large role street efficacy and shows that adolescents’con- in determining how adolescents respond to their fidence in their ability to avoid violence is sub- social setting and the types of environments stantially influenced by aspects of the they select for themselves. neighborhood context, especially the level of An essential question is how aspects of the collective efficacy in the neighborhood. In a imposed environment affect the choices ado- second set of models, I find preliminary evi- lescents make as they select and create for them- dence to suggest that adolescents with high selves an actual, lived environment that will street efficacy take steps to select environments help shape their developmental trajectory. To that minimize the potential for violent con- answer this question, I draw on a social cogni- frontations. I find that adolescents with high tive theory of behavioral adaptation—self-effi- street efficacy are less likely to behave vio- cacy theory—to help explain choices relating to lently themselves or associate with delinquent violence that adolescents make as they grow up peers. However, adolescents confident in their in neighborhoods with varying levels of disad- ability to avoid violence are no less likely to take vantage and social organization (Bandura 1977; part in several types of unstructured activities Bandura 1997). In applying self-efficacy theo- associated with delinquency. Considered as a ry to the study of individual violence, I argue whole, the results provide support for a revised that efficacy should be considered a contextu- perspective on violence, one that emphasizes al concept, in that aspects of the imposed neigh- individual agency and the capacity of both indi- borhood context are likely to combine with viduals and collectivities to play an active role processes occurring in families and with char- in reducing the potential for violence in their acteristics of children to shape adolescents’ lives and communities. expectations about their ability to perform the actions necessary to achieve certain outcomes. THEORETICAL CONSIDERATIONS In particular, the concept of street efficacy, NEIGHBORHOODS AND INDIVIDUAL AGENCY defined as the perceived ability to avoid violent confrontations and find ways to be safe in one’s Neighborhoods vary with respect to the types neighborhood, is proposed as a tool for incor- of role models and peers they provide, the rel- porating the neighborhood context into the ative stock of various resources and support analysis of individual violent behavior. available to young people (e.g. the quality of I hypothesize that adolescents’ confidence schools, policing, health care systems, and pro- in their ability to find ways to avoid violence is grams for youth, as well as the informal over- likely to affect the creativity and effort they sight of public spaces by local residents), and

Figure 1. The Relationship between Adolescents’ Imposed and Selected Environments 828—–AMERICAN SOCIOLOGICAL REVIEW the presence of poverty, drugs, weapons, crime, the resources upon which he draws. It is, in and violence (Anderson 1999; Jargowsky 1994; essence, the lived environment of the child. Wilson 1987). It is widely assumed that this Critics of the neighborhood effects literature variation affects the way children think about are correct in implying that the selected envi- their future, the types of people they look up to ronment is, in many ways, more relevant to as role models, the types of friends with whom adolescent development than the imposed envi- they associate, and the opportunities available ronment, yet one would be hard-pressed to make to them (Anderson 1999; Case and Katz 1991; the argument that the two are unrelated. To rec- Ellen and Turner 1997; Elliott et al. 1996; oncile these perspectives, a theoretical model is Sampson and Wilson 1995; Wilson 1987). needed to explain the way that adolescents trans- However, theories analyzing the ways that form the imposed environment into their own neighborhoods affect children have been criti- unique, selected environment. I propose a con- cized because they ignore the fact that children textualized version of self-efficacy theory as make choices about the people with whom they an appropriate model for this task. associate, the types of activities in which they engage, and the role models they choose for EFFICACY ON THE STREET themselves. This critique is best expressed in a Self-efficacy theory is based on the premise passage from Susan Mayer and Christopher that changes in behavior “achieved by different Jencks’review of the “neighborhood effects” lit- methods derive from a common cognitive mech- erature up to 1990, in which the authors ques- anism” (Bandura 1977:191). Bandura hypoth- tion the assumption that living in neighborhoods esizes that the effectiveness of a diverse array with deviant peers and poor role models should of psychological treatments can be explained by necessarily lead kids to negative outcomes: their effects on efficacy expectations, defined “Even in the poorest neighborhoods a teenag- as “the conviction that one can successfully er can find some friends who stay out of trou- execute the behavior required to produce [a ble, finish high school, go to college, and get given] outcome” (Bandura 1977:193).1 good jobs.|.|.|. As long as both kinds of role Support for the theory comes from a large models exist [good and bad] their relative num- body of experimental research demonstrating bers may not matter much” (Mayer and Jencks that efficacy expectations have a causal effect 1989:1441). The implication is that character- on subjects’ ambitions, the effort they expend istics of neighborhoods do not determine the to pursue these ambitions, and their persistence way a given child will respond to his or her in this effort in the face of obstacles (Bandura social setting. Instead, children respond to and 1997:54–61; Maddux 1995). Although the orig- shape their environment in very different ways, inal applications of self-efficacy theory focused and agency matters in any context. on the effects of psychological treatments The distinction between adolescents’ designed to manipulate efficacy expectations in “imposed” and “selected” environments allows controlled contexts, the theory has been extend- for the integration of these contrasting per- ed to help explain behavior in a wide variety of spectives on the role that neighborhoods play in settings. Research has shown that self-efficacy influencing adolescent development. The is a strong predictor for a range of social out- imposed environment consists of everything comes including health behaviors, the academ- that would surround the child if he or she were ic performance of students, the performance of an inactive object. This includes all the people, teachers and schools, and individual career all the physical structures, all the social inter- choices (Bandura 1997). actions, and all the resources that could poten- Considering its development as a theory of tially affect the child purely on the basis of behavioral adaptation in stressful environments, where that child lives. The selected environ- I argue that self-efficacy theory can be extend- ment is the imposed environment acted upon by the child. This includes all the people with whom the child actually comes into contact, 1 See Bandura (1997:10–15) for a comparison of the physical environments into which he places self-efficacy theory with related psychological mod- himself, the social interactions that he has, and els of behavior. NAVIGATING DANGEROUS STREETS—–829 ed naturally to help explain adolescents’ SOURCES OF STREET EFFICACY responses to the pressures that they confront in disadvantaged urban settings. Efficacy theory I conceptualize street efficacy as developing primarily from the interaction of individuals is consistent with the notion that a primary and their imposed environments (Bandura mechanism by which the neighborhood con- 1997:163). From this perspective, aspects of text affects adolescent development is through neighborhoods, families, and individual char- its impact on cognition (Rosenbaum, Reynolds, acteristics all may affect adolescents’ confi- and Deluca 2002; Ross, Mirowsky, and Pribesh dence in their ability to avoid violence. 2001; Wilson 1987; Wilson 1996). For instance, Beginning with characteristics of the child, I Wilson (1987) argues that one consequence of hypothesize that adolescents who lack key social the increasing social isolation of ghetto residents skills—for instance, those with low verbal abil- in the 1980s was the effect that these changes ity (Sampson, Morenoff, and Raudenbush 2005) had on children’s perceptions. In describing or a tendency toward impulsive behavior how a child might be affected by living in a (Farrington 1995, 1998; Gottfredson and Hirschi socioeconomically diverse neighborhood, 1990)—may be ill-equipped to handle poten- Wilson (1987:56) writes: tially violent social situations. This may in turn A perceptive ghetto youngster in a neighborhood reduce their confidence in their ability to avoid that includes a good number of working and pro- violence and be safe as they engage in public life fessional families may observe increasing job- in the neighborhood. lessness and idleness, but he will also witness Multiple aspects of the family environment many individuals going to and from work .|.|. , he may be cognizant of an increase in crime, but he also are likely to affect adolescent street efficacy. can recognize that many residents in his neigh- Adolescents in families with greater resources, borhood are not involved in criminal activity. both financial and social, are likely to have rel- atively high levels of street efficacy, especially Implicit in this observation is the idea that the if these resources are used by parents to provide effects of social environments on individual enriching and safe environments for their chil- behavior are at least partially mediated by inter- dren (Loeber and Stouthamer-Loeber 1986). nal cognitive processes. This idea is formalized Through cognitive processes of “modeling” in self-efficacy theory. In extending the theory (Bandura 1997), adolescents who witness fam- to help explain adolescent behavior in urban ily members engage in violence or crime, areas, it is important to treat disadvantaged whether it occurs in or outside of the home, are urban neighborhoods as unique ecological set- less likely to believe that they can avoid violence tings in which specific forms of efficacy are par- in their own lives (Farrington 1995). On the ticularly important in everyday life. In other hand, parents who monitor their children neighborhoods wherein violence structures pub- closely may help raise their confidence by mak- lic interactions, the ability to manage poten- ing them feel safe in the neighborhood or by tially violent situations is an essential skill for reducing the chance that they will find them- children (Anderson 1999). The concept of street selves in vulnerable situations. efficacy provides a theoretical tool that cap- Shifting the focus to life outside the home, tures the importance of adolescent cognition both structural characteristics of neighborhoods for understanding individual behavior, while and social processes occurring within neigh- placing cognition within a broader social con- borhoods are likely to have an impact on ado- 2 text. lescent street efficacy. First, high levels of violence in neighborhoods with concentrated disadvantage may reduce street efficacy simply by making it more difficult for adolescents to 2 Street efficacy is similar in some ways to avoid violent confrontations. Elijah Anderson’s Anderson’s (1990) concept of “street wisdom,” which (1990, 1999) work provides a clear picture of the emphasizes individuals’ understanding of accepted patterns of interpersonal interactions in the public role that interpersonal violence plays in struc- sphere. However, the street efficacy concept entails more than knowledge of common patterns of social interaction or an understanding of expected behav- ability to manage such public interactions while ior norms. It also involves a belief in one’s own avoiding violence. 830—–AMERICAN SOCIOLOGICAL REVIEW turing public life in the most disadvantaged of the neighborhood environment also may urban neighborhoods. Adolescents who grow up affect street efficacy indirectly by making it in such neighborhoods must expend a great more likely that adolescents will be personally deal more effort to avoid violence than adoles- victimized or exposed to violence. Two primary cents in less distressed areas of the city or sub- sources of efficacy expectations are past per- urbs. formance at achieving a given outcome and At the same time, it is not only the presence “modeling” processes, or “vicarious experi- of structural disadvantage or violence that is ences” (Bandura 1997). Clear successes and likely to affect adolescents’ perceptions, but failures offer subjects the most powerful evi- also the ways that community residents sup- dence of their current ability to perform the port and monitor local youth and the role that behaviors necessary to achieve an outcome the community plays in confronting violence (Ewart 1995; Maddux 1995:10). Similarly, if and maintaining social control. A growing lit- subjects see “models” with the same condition erature on neighborhood social organization as their own successfully accomplish a partic- emphasizes residents’shared expectations about ular behavior in circumstances similar to those their collective capability to achieve certain that they face, these vicarious experiences serve ends, such as the oversight of local youth and to increase the subject’s confidence in his or her the reduction of violence (Sampson, Morenoff, own ability to perform the behavior (Bandura and Earls 1999). This perspective is captured in 1997:86–92). Translation of these mechanisms the concept of collective efficacy, or the “dif- to the social world of urban neighborhoods ferential ability of neighborhoods to realize the leads to the hypothesis that adolescents who common values of residents and maintain effec- have been unable to avoid violence in the past tive controls” (Sampson, Raudenbush, and Earls and those who have witnessed frequent vio- 1997:918). lence will have doubts about their ability to Adolescents are likely to pick up cues from avoid violence and be safe in their neighbor- the community in which they live about their hoods. own ability to avoid violence. In communities where residents express a shared willingness STREET EFFICACY AND VIOLENT to support and monitor local youth and to inter- ENVIRONMENTS vene in public life to enforce common norms of behavior, adolescents intent on avoiding vio- To this point, adolescents’ environments have lence may feel more confident walking down been discussed as if they are fixed and imposed public streets, with the knowledge that they on individuals. Although the imposed environ- have advocates looking out for them. ment has some aspects over which adolescents Alternatively, in communities where residents have no control, a key element of efficacy the- retreat from public life and treat the presence of ory is the idea that environments are to some violence with resignation, adolescents may feel extent selected and created, in the sense that ado- that attempts to avoid violence are futile, and lescents can draw upon certain aspects of the that they are on their own in their attempts to do potential environments available to them on the so. In other words, the same processes that acti- basis of whether these selected aspects of the vate a community’s collective desire to monitor environment are conducive to their own goals and support local youth may bolster adoles- and ambitions (Bandura 1997:163; Maddux cents’personal sense of agency in managing the 1995:12–14). I hypothesize that an adolescent’s threat of violence in their own lives. confidence in his or her ability to avoid violence I thus hypothesize that after taking into plays a crucial role in the process by which the account characteristics of the child and his or her imposed neighborhood context is transformed family associated with violent behavior, aspects into the adolescent’s unique social world. of the imposed neighborhood context—specif- Psychological experiments have shown that ically the level of violence, disadvantage, and efficacy expectations affect the types of goals collective efficacy in the neighborhood—will people set for themselves, the effort they expend influence adolescents’street efficacy. Although to achieve those goals, and their persistence in this hypothesis implies that neighborhoods have pursuing their goals in the face of obstacles a direct effect on street efficacy, these aspects (Maddux 1995:12). Translated to the social NAVIGATING DANGEROUS STREETS—–831 worlds of adolescents in urban neighborhoods, how some subjects consider violence to be their these findings lead to the hypothesis that ado- only possible response to threats or perceived lescents with high street efficacy are more like- challenges, whereas others in similar environ- ly to expend greater effort and creativity to ments use multiple strategies to resolve their avoid violent confrontations, selecting social problems while avoiding a violent confrontation. settings, peer groups, and activities that provide Although Anderson does not provide a full them a better chance of doing so. For instance, theory to explain this variation, he does high- a young man intent on avoiding gang members light certain factors that help explain adoles- in his neighborhood may seek out after-school cents’ responses to dangerous environments. activities or organized sports leagues instead of First, primary emphasis is placed on the impor- hanging out on the street corner or the play- tance of a “decent” family as a buffer against the ground when school lets out, or he might devel- pressures of the street (Anderson 1999; Warr op a verbal script designed to defuse the 2004). Anderson (1990:3–4; 1999:145–146) potential for conflict in situations wherein he is also points to the presence of “old heads” and verbally confronted or insulted. “community mothers” in the neighborhood (i.e., I hypothesize that adolescents with high lev- older men and women willing to offer their els of street efficacy are more likely to select advice to youngsters) as potentially influential environments that reduce the potential for vio- role models who may inspire a commitment to lent confrontations. Because the selected envi- mainstream ideals among young men and ronment is unique to each child and women in urban ghettos. Despite the emphasis encompasses many different aspects of a child’s on the importance of a decent family, day-to-day life, it is difficult to operationalize Anderson’s fieldwork demonstrates that even fully with survey data. However, it is possible children from decent families often adapt to to develop several measures that may serve as the code of the street (Anderson 1999:99), and indicators of the potential for violence in the many examples of adolescents who fall prey to child’s everyday environments. The theoretical the violence in their social environment are perspective I outline suggests that the strength from families that Anderson would most likely of an adolescent’s commitment to avoiding vio- characterize as decent. This evidence suggests lence is reflected in his choices about the types that it is essential to understand the mecha- of peers with whom he surrounds himself, the nisms by which family members and neigh- types of situations and social settings into which borhood role models influence young adults in he places himself, and his willingness to resort disadvantaged environments, leading some to to violence. Accordingly, adolescents with high select environments characterized by violence, street efficacy should exhibit lower levels of but others to maintain a commitment to avoid- violence, should have relatively few delinquent ing violence. peers, and should spend less time in unstruc- Second, Anderson (1999:66) conceptualizes tured activities shown to be associated with violent social networks and strategic exhibi- individual deviant behavior (Birkbeck and tions of violent behavior as forms of “capital” LaFree 1993; Osgood et al. 1996). that adolescents use to maintain “respect” while There are numerous examples of these avoiding physical attacks in dangerous settings. processes at work in ethnographic studies of According to this argument, it often is necessary adolescents in disadvantaged urban neighbor- for youth living in violent environments to make hoods, yet none of these studies have advanced clear a willingness to resort to violence and to a full theory to distinguish the ways that ado- “run with” friends who command respect on the lescents respond to potentially violent environ- street to deter potential attackers (1999:72). ments (Anderson 1999; Irwin 2004; Jones Massey (1995) draws on this argument, but is 2004). In particular, Anderson provides sever- more direct, contending that violent behavior al case studies that bring into focus the crucial should be seen as a rational “adaptation” to distinctions in the ways that young people per- violent environments that provides protection for ceive and respond to the threat of violence in adolescents. For instance, Massey (1995:1216) their social environment (see the case studies of writes: “Robert” [pp. 290–310] and “Tyree” [pp. How does a person adapt to a harsh environment 80–103] in Anderson 1999). Anderson describes where violence is endemic, the odds of criminal 832—–AMERICAN SOCIOLOGICAL REVIEW

victimization are high, and the risk of death or [2001] for a full description of multiple impu- injury substantial? The most logical individual tation).3 This procedure is used to impute val- adaptation is for one to become violent oneself. By ues for individuals who are interviewed but do adopting a threatening demeanor, cultivating a not provide information on a given question, as reputation for the use of force, and selectively well as for individuals who leave the survey backing up that reputation with actual violence, one can deter potential criminals and increase the odds over the three-wave period. The method relies of survival. on the assumption that the data are missing at random, or that the probability of having miss- Although my analysis is not designed to pro- ing data on a given variable is unrelated to the vide a precise test of these arguments, it does true value of that variable (Allison 2001; Rubin shed light on the relationship between violent 1987; Schafer 1999). Procedures available in the behavior, peer delinquency, and adolescents’ HLM software (Raudenbush et al. 2001) are perceptions of their ability to avoid violence. By used to estimate multilevel models with multi- examining these relationships, it is possible to ply imputed data that fully account for the loss test whether youth who exhibit violent behav- of precision associated with imputation of val- ior or those who associate with delinquent peers ues. The relevant sample for all regression mod- feel more or less confident in their ability to els consists of 2,337 adolescent boys and girls engage in public life within their neighborhood and their caregivers living in 80 neighborhood safely. More generally, the theoretical diagram clusters across Chicago. Descriptive statistics for shown in Figure 1 expands upon several argu- key variables are displayed in Table 1. ments and observations in Code of the Street (Anderson 1999) and incorporates them into a THE MEASURE OF STREET EFFICACY formal theory designed to explain variation in adolescents’ responses to potentially danger- It is generally accepted that efficacy expecta- ous environments. tions “should be defined and measured in the context of relatively specific behaviors in spe- DATA cific situations or contexts” (Maddux 1995:8). Although in some circumstances a sense of effi- I use data from the Project on Human cacy in one area may extend to other areas of Development in Chicago Neighborhoods behavior or other situations, the extent to which (PHDCN), a longitudinal study of children and this occurs depends largely on the similarity of families living in a diverse set of Chicago neigh- the skills and behaviors necessary to achieve the borhoods in 1995 (Earls et al. 1994; Earls et al. outcome of interest in each situation (Bandura 1995). The PHDCN sampling design involved 1997:51; Maddux 1995:8–9). However, effica- two stages. First, all 343 Chicago neighbor- cy theorists argue against the use of generalized hood clusters (NCs) were stratified by racial/eth- scales developed to generate global measures of nic composition and socioeconomic status, and self-efficacy because these types of global meas- a random sample of NCs was selected within ures usually are poor predictors of behavior, strata (see Sampson et al. [1997] for a descrip- and because they imply that self-efficacy should tion of the construction of the 343 NCs). be conceptualized as a general personality trait Second, dwelling units were listed within each (Bandura 1997:47–50). Consistent with this NC. All households were listed, and age-eligi- approach, a measure of street efficacy is con- ble participants (household members within 12 structed that represents the mean value over months of age 0, 3, 6, 9, 12, 15, or 18 years) five survey items measuring adolescents’ per- were selected with certainty. Respondents and their caregivers were interviewed up to three times between 1995 and 2002. 3 For all survey items measuring violent behavior, The current analysis limits the sample to I use original values because the imputations pro- include only members of the 9-, 12-, and 15- duced in the multiple imputation procedure were not year-old cohorts and their caregivers. For all compatible with the assumptions of the Rasch model subject- and family-level variables, I use a mul- that is estimated. The Rasch model requires that each tiple imputation procedure suggested in van item in the scale be equally discriminating, an Buuren (1999) and implemented by Royston assumption that was not supported for certain sets of (2004) to impute missing values (see Allison imputations. NAVIGATING DANGEROUS STREETS—–833

Table 1. Descriptive Statistics for Key Variables

X Standard Variable N Mean Deviation Min Max Neighborhood-level Variables —Concentrated disadvantage 80 –.11 .80 –1.60 2.66 —Collective efficacy 80 3.89 .26 3.38 4.51 —Neighborhood violence 80 2.00 .33 1.41 2.74 Individual-level Variables —Violence item 1: hit someone you don’t live with in the past year 1646 17.3% .38 0 1 —Violence item 2: thrown objects like bottles or rocks at someone 1652 6.6% .25 0 1 —Violence item 3: carried a hidden weapon 1643 7.1% .26 0 1 —Violence item 4: maliciously set a fire 1653 .2% .05 0 1 —Violence item 5: attacked someone with a weapon 1651 2.1% .14 0 1 —Violence item 6: been in a gang fight 1652 3.2% .18 0 1 —Violence item 7: snatched someone’s purse or wallet 1651 .2% .05 0 1 —Violence item 8: chased someone with intent to hurt or scare 1623 6.9% .25 0 1 —Violence item 9: shot at someone 1652 .9% .09 0 1 —Violence item 10: purposefully damaged property 1647 5.7% .23 0 1 —Violence item 11: threatened to hurt someone physically 1650 5.2% .22 0 1 —Violence item 12: shot someone 1652 .4% .06 0 1 —Peer delinquency 2337 1.77 .54 1.00 3.67 —Unstructured activities 2337 .07 .69 –1.87 1.60 —Street efficacy 2337 3.09 .61 1.00 4.00 Note: See Table S1 in the online supplement to this article for descriptives on all variables. Means of all individual- level variables are based on imputed data, as described in text. Violence items do not use imputed values, and refer to activities in the year prior to the interview. ceptions of their ability to avoid violent con- incorporate both the level and the strength of frontations and find ways to be safe in their efficacy expectations (Bandura 1997:42–46). neighborhoods.4 The resulting scale ranges from 1 to 4, repre- For each survey item, respondents were asked senting the mean value of responses to the five to evaluate two statements and determine which sets of statements, and recoded so that high val- statement was a more accurate description of ues indicate high efficacy. how they felt in their neighborhood (see the online supplement for the survey items). They INDICATORS OF VIOLENT ENVIRONMENTS then were asked to decide whether the selected statement was “very true” or “sort of true.” Several measures are constructed to provide a These items follow Bandura’s prescriptions for description of the potential for violence in ado- the measurement of self-efficacy in that they lescents’selected environments. In particular, I focus on adolescent violent behavior, peer delin- quency, and routine activities as indicators for the types of environments into which adoles- 4 See the online supplement for a more detailed cents select. Although much research attempts description of the street efficacy scale, including the specific items comprising the scale of street effica- to identify the causal relationships among these cy and other key variables. The individual items pro- different aspects of the selected environment duce a scale with a reliability of ␣ = .56 at Wave 2 (e.g., the effects of delinquent peers or routine and ␣ = .62 at Wave 3. Because the reliability of the activities on violence) (Matsueda and Anderson scale is relatively low, I conduct all analyses on each 1998; Osgood et al. 1996), it is very likely that individual item in the scale of street efficacy as well the paths of causation are bidirectional. I avoid as the composite measure. I find that the results are the problem of trying to identify causal rela- extremely consistent across the items in the scale, although I report the few substantive differences tionships among these indicators by treating found when the individual items are used in notes violence, peer associations, and routine activi- throughout the text. ties as outcomes and as elements of the larger 834—–AMERICAN SOCIOLOGICAL REVIEW theoretical concept of the selected environ- poverty, the receipt of public assistance, unem- ment.5 ployment, female-headed households, and the Individual violent behavior is measured with density of children (see Sampson et al. [1997] self-reported responses to multiple survey items. for a detailed description of this scale). Sampson Subjects were asked whether they had com- et al. (1997) show this measure to be correlat- mitted any of 12 violent acts in the year before ed strongly with victimization rates and homi- the interview. Each dichotomous item is incor- cides. porated into a scale of violent behavior, as Measures of neighborhood collective effica- described in the “Results” section later. A sum- cy and neighborhood violence are constructed mary measure of violent behavior also is con- with data from the PHDCN Community Survey structed and used as a control variable in certain (Earls et al. 1995). The Community Survey is models. This measure represents the number an independent survey of 8,782 Chicago resi- of violent activities in which the subject has dents in all 343 neighborhood clusters in participated during the reference period. Chicago. Collected in 1995, data from the Previous research has shown that self-reported Community Survey are both temporally prior to survey items are reliable indicators of criminal and distinct from data collected in the or violent behavior (Thornberry and Krohn 2002). Longitudinal Cohort Study, the survey admin- Peer delinquency is measured as the mean istered to subject participants and their care- value of responses to several survey items ask- givers. Collective efficacy is based on a series ing subjects about delinquent activities of their of survey items asking about the extent of infor- friends (␣ = .78) (Huizinga, Esbenson, and mal social control and social cohesion/trust in Weihar 1991). In addition to violent behavior respondents’ neighborhoods. Sampson et al. and peer associations, the routine activities that (1997) show the scale to be a strong predictor occupy the adolescent’s everyday life represent of neighborhood-level violence as well as indi- a third component of the selected environment. vidual victimization. Neighborhood violence, a Although the analysis of routine activities is scale measuring residents’ perceptions about most often used to explain aggregate rates of the frequency of various violent events in the crime and violence (Birkbeck and LaFree 1993; neighborhood, is included to provide a measure Cohen and Felson 1979), Osgood et al. (1996) for the general level of violence in adolescents’ find evidence that certain forms of “unstructured neighborhoods. socializing” also predict individual delinquen- cy. I focus on a set of unstructured activities INDIVIDUAL AND FAMILY FACTORS RELATED TO shown to be associated with individual crime or VIOLENCE deviant behavior (␣ = .65). Several additional variables are included in MEASURES OF THE IMPOSED NEIGHBORHOOD regression models to capture theoretically rel- ENVIRONMENT evant characteristics of children and families, as well as adolescents’experiences with violence. A measure of neighborhood concentrated dis- The subject’s verbal ability is measured with a advantage is based on five variables construct- composite scale based on the results from two ed from the 1990 decennial census: the rates of tests given to subjects: the Wechsler Intelligence Scale for Children vocabulary test (Wechsler 1974) and the Wide Range Achievement Test, 5 A comprehensive portrait of the selected envi- which measures reading ability (Wilkinson ronment would need to consider at least four dimen- 1993). The subject’s impulsivity represents the sions: (1) people, (2) behavior, (3) activities, and (4) mean of several standardized responses to ques- places. My measures consider the first three dimen- tions asked of the caregiver from the Achenbach sions. The physical geography of adolescents’every- Child Behavior Checklist (␣ = .76) (Achenbach day lives is another important dimension of the selected environment, but information pertaining to 1993). The measures of both verbal ability and the actual locations where adolescents spend their impulsivity are shown to be associated with time is not available through the Project on Human individual violent behavior by Sampson et al. Development in Chicago Neighborhoods. (2005). NAVIGATING DANGEROUS STREETS—–835

Parental supervision represents the mean of the dichotomous responses to questions asked a subset of standardized responses to survey of subjects about whether they had ever drunk items from the PHDCN Homelife Interview alcohol (“not just a sip or taste”) or smoked (Leventhal et al. 2004). Caregivers were asked marijuana in the year before the interview. The a set of questions about their involvement in var- measure of school violence is a scale based on ious aspects of the child’s life (e.g. knowledge seven questions asked of the subject about gang of friends, school activities) and the extent to activity, racial/ethnic problems, prevalence of which they monitor and enforce rules for the school fights, and other indicators of violence child (e.g., curfew, unsupervised time) (␣ = (e.g., metal detectors in the school) in the school .70). Similar measures of parental monitoring environment (␣ = .66). or supervision have been shown to be strong pre- Certain models control for a set of demo- dictors of violence and delinquency (Farrington graphic characteristics of the subject and care- 1995; Sampson and Lauritsen 1993:26). Family giver, as well as for several dimensions of family criminality represents the total number of fam- background and socioeconomic status. See the ily members identified by the caregiver as hav- online supplement to this article for detailed ing a criminal record. Domestic violence is descriptions of all variables. measured as the sum of dichotomous respons- es to nine survey items asking caregivers about RESULTS interactions with any current or previous domes- tic partner (␣ = .84). The validity of the scale THE DIVERSE SOURCES OF STREET EFFICACY is supported in Straus et al. (1996). The inclu- In the initial series of models, street efficacy is sion of the measures of family criminality and the dependent variable, and individual respon- domestic violence is based on a body of research dents are nested within neighborhoods (Bryk identifying criminality among family members and Raudenbush 1992). In the Level 2, or neigh- and violence occurring within the home as borhood-level models, measures of concentrat- important predictors of child delinquency ed disadvantage, neighborhood violence, and (Farrington 1998). These measures also allow collective efficacy predict variation in street for an examination of the effects that violent efficacy across neighborhoods in Chicago. In “models” have on adolescents’ perceptions of the Level 1, or person-level, models, individual- their ability to avoid violence. and family-level measures predict variation in The subject’s exposure to violence is the sum individual street efficacy within neighbor- of dichotomous responses to survey items ask- hoods.6 In all models, the independent vari- ing the subject whether he or she witnessed any ables are measured at Wave 1 of the survey, of several violent acts during the year before the 7 ␣ whereas street efficacy is measured at Wave 2. interview ( = .71), and victimization is the Results are presented in Table 2. sum of dichotomous responses to items asking In Model 1, I examine the sources of varia- the subject about any physical assaults or seri- tion in street efficacy across neighborhoods by ous threats of physical harm over the same peri- ␣ incorporating measures of concentrated disad- od ( = .55). Evidence of construct validity for vantage, collective efficacy, and perceived vio- the two scales is provided in Selner-O’Hagan et lence into the neighborhood-level model. The al. (1998). Because exposure to violence and three neighborhood-level variables in this model personal victimization may affect street effica- cy through two distinct pathways (“modeling” processes in the case of exposure to violence and processes of “performance” in the case of vic- 6 As with all models in the analysis, all continu- timization), I include them in empirical models ous variables, including the neighborhood-level meas- separately instead of incorporating the two con- ures, are centered around their grand means. Dichotomous and categorical variables are left uncen- structs into a single scale. tered. I also construct dichotomous measures of 7 Personal victimization and exposure to violence individual alcohol use and marijuana use on the are measured at Wave 2. Because these variables basis of research linking substance use with represent victimization or exposure in the year before violence (Kodjo, Auinger, and Ryan 2004; the interview, they can be thought of as temporally Miczek et al. 1993). These measures represent prior to the outcome. 836—–AMERICAN SOCIOLOGICAL REVIEW

Table 2. Models of Street Efficacy

Model 1 Model 2 Model 3 Neighborhood-level Model —Intercept 3.106 *** 3.246 *** 3.215 *** (.015) (.056) (.058) Concentrated disadvantage –.094 *** –.058 ** –.039 (.028) (.026) (.027) —Collective efficacy .268 *** .146 ** .147 ** (.061) (.056) (.058) —Neighborhood Violence –.061 .002 –.008 (.076) (.063) (.064) Person-level Model —Verbal ability .109 *** .109 *** — (.011) (.010) —Impulsivity –.061 ** –.043 (.027) (.026) —Parental supervision .090 ** .071 (.041) (.045) —Domestic violence .005 .006 (.010) (.010) —Family criminality –.019 –.012 (.014) (.014) —Marijuana use –.147 * –.028 (.077) (.081) —Alcohol use –.050 .023 (.042) (.043) —School violence –.024 –.011 (.017) (.017) —Personal victimization –.026 (.019) —Exposure to violence –.038 *** (.012) —Peer delinquency –.154 *** (.043) —Violent behavior –.022 * (.013) Subject race/ethnicity -Mexican –.170 ** –.167 ** (.062) (.063) —Puerto Rican –.134 ** –.108 * (.064) (.065) —Other Hispanic/Latino –.069 –.048 (.075) (.076) —African-American –.056 –.025 (.045) (.043) Subject immigrant status—1st generation .014 –.017 (.053) (.053) —2nd generation –.004 –.013 (.034) (.034) —Subject’s age .033 *** .043 *** ‘ (.008) (.008) Subject’s gender, 1 = male –.034 –.002 (.031) (.030) Note: Standard errors in parentheses; additional control variables included in Models 2 and 3 are: caregiver’s age and gender, caregiver’s total household income, education, and marital status, caregiver’s (and spouse’s) employ- ment and occupational status, caregiver’s welfare receipt, total household size, home ownership, length of resi- dence at current address, and subject’s body mass index. * p < .1; ** p < .05; *** p < .01 NAVIGATING DANGEROUS STREETS—–837 explain 75 percent of the variation in street effi- past year are less confident. Children of cacy across neighborhoods, indicating that these Mexican and Puerto Rican descent also have rel- three measures capture most of the key differ- atively low levels of street efficacy. Adolescents ences between neighborhoods.8 I find that both with parents who supervise them closely are concentrated disadvantage and collective effi- more likely to be confident in their ability to cacy are significantly associated with street avoid violence, although family criminality and efficacy. Adolescents in disadvantaged neigh- domestic violence have no effects in this model. borhoods are less likely to be confident in their The inclusion of these covariates accounts for ability to avoid violence, whereas those who live a portion of the association between the neigh- among neighbors who take an active role in borhood-level variables and street efficacy, enforcing common norms of behavior and main- although both neighborhood concentrated dis- taining social controls have relatively high street advantage and collective efficacy remain sig- efficacy. Interestingly, this model seems to imply nificant predictors in this model. that the perceived presence of violence in one’s In Model 3, I include measures of personal neighborhood is less important than other victimization, exposure to violence, prior vio- dimensions of disadvantage, and also less lent behavior, and peer delinquency to determine important than the role that the community whether the association between neighborhood- plays in regulating public activity and enforc- level factors and individual street efficacy is ing common social norms.9 explained by personal experiences with vio- Model 2 introduces a set of demographic and lence. I find that adolescents who have been family background variables along with sever- exposed to violence, those who engage in vio- al variables representing characteristics of indi- lent behavior, and those with delinquent friends viduals or families that are associated with have lower levels of street efficacy. The meas- violence. Results from theoretically relevant ure of concentrated disadvantage is no longer variables are shown in Table 2. I find that ver- significant in this model, suggesting that the bal ability and impulsivity are associated with association between neighborhood disadvan- street efficacy in the expected directions, sug- tage and street efficacy is largely explained by gesting that adolescents are differentially individual experiences with violence and the equipped with skills that help them avoid con- set of control variables included in the model. flict on the street. Older adolescents are more However, collective efficacy remains a signif- confident in their ability to avoid violence, icant predictor in this model.10 whereas those who have used marijuana in the Although the results from Model 3 suggest no gender differences in the level of street effi- cacy, it still might be the case that the sources 8 The proportion of explained Level 2 variance is of street efficacy are different for boys and girls. obtained by first estimating an unconditional model For instance, although ethnographers suggest with no predictors at the neighborhood or individual that the threat of violence in urban neighbor- level. In the unconditional model, about 6 percent of hoods is becoming more similar between boys the variation in street efficacy occurs between neigh- and girls (Anderson 1999; Jones 2004), a body borhoods. I subtract the Level 2 variance compo- of evidence shows substantial, persistent gen- nent in Model 1 from the Level 2 component in the der gaps in the prevalence of violence unconditional model, and divide this difference by the (Kruttschnitt 1993). Considering the mixed variance component in the unconditional model to results from research examining the link obtain the proportion of Level 2 variance explained. 9 To determine whether this finding is attributable to a deficient measure of neighborhood violence, I replicate the model using a measure of the prevalence 10 In separate models not shown, I replicate these of homicides in the neighborhood instead of the analyses using each individual item in the scale of measure of perceived violence in the neighborhood. street efficacy as an outcome. The results are extreme- The homicide measure uses aggregated counts of ly consistent across the items, with the exception of homicides in each neighborhood from 1990 to 1995 the item asking adolescents whether they can find available from the Chicago Homicide Dataset (Block ways to be safe when alone in their neighborhood (in and Block 1993). The results (available from the which concentrated disadvantage has a larger effect author) match those reported in Model 1, Table 2. than collective efficacy). 838—–AMERICAN SOCIOLOGICAL REVIEW between gender and violence, I estimate the quent prior violent behavior and association same model separately for boys and girls to with delinquent peers should not be regarded as determine whether the sources of street effica- protective for adolescents. One potential expla- cy vary by gender (results are shown in Table S2 nation for this finding is that Anderson’s and of the online supplement).11 I find minimal Massey’s arguments are applicable in only the gender differences. Adolescent boys living in most violent neighborhoods within the city. I test violent homes and those subjected to victim- this possibility by estimating Model 3 after lim- ization have lower levels of street efficacy. These iting the sample to include only respondents measures have no effects for girls. On the other who live in extremely violent neighborhoods.12 hand, impulsivity seems to reduce street efficacy I find that prior violent behavior has only a among girls, but not among boys. Although weak association with street efficacy in the other coefficients in the model have slightly most disadvantaged neighborhoods, although different magnitudes for boys and girls, none of peer delinquency continues to be negatively the other coefficients in the model vary signif- associated with street efficacy. icantly by gender. When these results are interpreted, it should As a whole, the results from this first set of be kept in mind that my measures of violence models provide strong evidence that adoles- and peer delinquency may be too blunt to cap- cents’ confidence in their ability to avoid vio- ture the nuances of the ways that youth use vio- lence is substantially influenced by their social lence and violent social networks to ward off context. Although individual characteristics potential attackers. However, these results make such as impulsivity and strong verbal ability clear that across all neighborhoods in Chicago, do affect street efficacy, several aspects of the adolescents who have taken part in violent activ- adolescent’s environment both in and outside the ities and who have formed affiliations with home also play a large role in influencing how delinquent peers find themselves less confi- children think about the potential for violence dent in their ability to avoid violence and to be in their lives. Especially interesting from a the- safe in their neighborhood. This result is con- oretical standpoint is the robust relationship sistent with the body of research showing that between neighborhood collective efficacy and violent offending is a strong predictor of violent individual street efficacy. This result suggests victimization (Lauritsen, Sampson, and Laub that adolescents’ perceptions about their own 1991; Sampson and Lauritsen 1993:30–34). ability to avoid violence are not based solely on Within the most violent neighborhoods, having the presence of disadvantage or violence in delinquent peers continues to be associated with their neighborhood, but rather on the role that low street efficacy, whereas violent behavior the community takes in responding to the poten- has no effect. Collectively, this set of findings tial for violence and in enforcing common should motivate future research designed to norms of behavior. clarify the relationship between violent behav- Finally, these models present a complicated ior, violent social networks, and adolescents’ picture of the relationship between prior violent perceived physical safety, and to provide a more behavior, delinquent social networks, and ado- precise test of Anderson’s and Massey’s argu- lescents’ perceptions of their ability to avoid ments than is possible in the current analysis. violence. Both Anderson (1999) and Massey (1995) argue that individuals in disadvantaged STREET EFFICACY AND VIOLENT environments use violence and violent social ENVIRONMENTS networks strategically in an effort to maintain status on the street and to ward off potential I now turn to the relationship between adoles- attackers. My models do not test this hypothe- cents’ perceptions of their ability to avoid vio- sis directly, but my results do suggest that fre- lence and the types of environments they select

11 I test for differences in the coefficients between 12 Specifically, I split the sample into thirds based boys and girls by estimating a fully interacted model on the level of perceived violence in the neighbor- in which all independent variables in the model are hood, keeping only adolescents who live in the most interacted with gender. violent neighborhoods in Chicago. NAVIGATING DANGEROUS STREETS—–839 for themselves. I begin by analyzing the rela- measure of street efficacy to test whether this tionship between street efficacy and violence in variable mediates the association between neigh- a series of three-level logistic hierarchical mod- borhood disadvantage and violence. In accor- els. In recent articles, multiple authors have dance with my hypothesis, I find that made convincing cases for the use of item adolescents with high levels of street efficacy response theory as a way to model individual are less likely to exhibit violent behavior. violence or crime using items from surveys of However, street efficacy only partially mediates self-reported delinquency or violence (Osgood, the association between neighborhood disad- McMorris, and Potenza 2002; Raudenbush, vantage and individual violence. Johnson, and Sampson 2003; Sampson et al. In Model 3, I include a set of demographic 2005). I follow closely the methodology of and family background variables to the person- Raudenbush et al. (2003) in applying a multi- level model. I find that adolescents from homes level Rasch model to the study of individual vio- with domestic violence or with family members lence. This model allows one to incorporate involved in the criminal justice system are more individual, family, and neighborhood variables into models predicting individual violence, likely to exhibit violent behavior, as are those while taking into account the relative frequen- who have used alcohol or marijuana. Results not cy of certain violent acts as compared with oth- shown in Table 3 show that boys are more like- ers, and the dependence of responses due to ly to be involved in violence than girls in the clustering of items within respondents and sample, as are older adolescents and African respondents within neighborhoods (Bryk and American sample members. First- and second- Raudenbush 1992; Sampson et al. 2005). generation immigrants are less likely to exhib- I estimate a three-level model in which indi- it violent behavior than subjects whose parents vidual survey items are incorporated into a scale were born in the United States. Although the of violence at Level 1.13 In the person-level inclusion of these controls accounts for a large model, characteristics of individual subjects portion of the association between neighborhood and their families are used to predict individual disadvantage and individual violence, these violence within neighborhoods, whereas in the variables account for only a small portion of the neighborhood-level model, measures of con- association between street efficacy and vio- centrated disadvantage, collective efficacy, and lence. perceived neighborhood violence are again used The most serious threat to these results is the to predict variation in individual violence across possibility that unmeasured characteristics of neighborhoods in Chicago (see Raudenbush et adolescents are biasing the results, or that there al. [2003] for a detailed description of the struc- are reciprocal effects of violence on street effi- ture of the three-level model of violence). All cacy that are not picked up in the models. I individual- and neighborhood-level variables attempt to address each of these possibilities by are measured at Wave 1 of the survey, with the including a measure of violence from Wave 1 exception of street efficacy, the mediating vari- of the survey as a control variable in Model 4. able, which is measured at Wave 2. Items com- prising the scale of violence are measured at This specification can be thought of as an Wave 3 to ensure proper temporal ordering. attempt to control both for the effects of past vio- In the first model (Table 3, Model 1), only the lent behavior on future violence and for any neighborhood-level measures are included as unmeasured tendencies toward violence that predictors. I find a positive and significant asso- are present by Wave 1 of the survey. The results ciation between concentrated disadvantage and from Model 4 show that adolescents with high individual violence, but collective efficacy and street efficacy remain significantly less likely neighborhood violence are not associated with to engage in violent behavior even after con- individual violence. Model 2 introduces the trolling for prior violence.14

13 Results from the Level 1 model show the rela- tive severity of each survey item in the scale of indi- 14 In results not shown, I replicated models pre- vidual violence, and are available from the author dicting individual violence using each individual upon request. item in the scale of street efficacy as predictors. The 840—–AMERICAN SOCIOLOGICAL REVIEW

Table 3. Models of Violent Behavior

Model 1 Model 2 Model 3 Model 4 Neighborhood-level Model —Intercept –1.630 *** –1.632 *** –1.973 *** –1.981 *** (.065) (.066) (.226) (.239) Concentrated disadvantage .465 *** .447 *** .155 ** .096 (.088) (.090) (.078) (.072) —Collective efficacy .193 .297 .052 .066 (.276) (.271) (.294) (.315) —Neighborhood violence –.378 –.403 –.376 –.292 (.270) (.271) .247 .259 Person-level Model —Street efficacy –.313 *** –.272 *** –.166 * (.086) (.093) (.089) —Verbal ability .011 –.002 — (.042) (.041) —Impulsivity .060 .036 (.072) (.078) —Parental supervision –.167 –.198 (.142) (.152) —Domestic violence .061 ** .066 *** (.026) (.023) —Family criminality .099 ** .073 *** (.042) (.041) —Marijuana use .586 *** .218 (.196) (.263) —Alcohol use .449 *** .128 (.164) (.176) —School violence –.017 –.034 (.062) (.062) Violent behavior—wave 1 .340 *** (.038) Note: Standard errors in parentheses; level 1 item severities are excluded from the table of results, but are avail- able from the author upon request; additional control variables included in Models 3 and 4 are: subject’s race/ ethnicity, age, gender, and immigration status, caregiver’s age and gender, caregiver’s total household income, education, and marital status, caregiver’s (and spouse’s) employment and occupational status, caregiver’s welfare receipt, total household size, home ownership, length of residence at current address, and subject’s body mass index. * p < .1; ** p < .05; *** p < .01

Because of the possibility of bias attributable on attempting to identify the magnitude of the to unmeasured variables, my primary focus has association. However, the results in Table 3 do been on providing evidence for the association allow for a rough estimate of the magnitude of the between street efficacy and violence rather than relationship between street efficacy and violence. On the basis of odds ratios not reported in Table estimated effects of each individual item on violence 3, the odds of exhibiting a given violent behav- were generally smaller than the composite measure, ior are between .76 (using Model 3) and .85 (using and were not always significant using the specifica- Model 4) times as high for an adolescent who is tion in Model 4. I also estimated a model in which “very” confident in his or her ability to avoid street efficacy is centered around the neighborhood- level mean rather than the grand mean. The purpose violence at Wave 2 than for an adolescent who is of this analysis is to determine whether adolescents “sort of ” confident.15 Considering the fact that within the same neighborhood who have relatively high levels of street efficacy are less likely to behave 15 The difference between an adolescent who is violently. The effects of street efficacy are found to “very” confident and one who is “sort of ” confident be very slightly smaller when examined within neigh- in his or her ability to avoid violence is equivalent to borhoods. a one-unit reduction in the scale of street efficacy. NAVIGATING DANGEROUS STREETS—–841

I am using a Wave 2 measure of street efficacy outcome across neighborhoods, and individual to predict violence at Wave 3 of the survey, and family-level characteristics of subjects pre- these estimates are conservative. In models not dict variation within neighborhoods. Results shown, using a measure of street efficacy from for selected models predicting both peer delin- Wave 3 of the survey, I find that the estimated quency and unstructured activities are present- association between street efficacy and vio- ed in Table 4. The full progression of models for lence is much stronger. each outcome is available in Tables S3 and S4 of the online supplement that accompanies this article. PEER DELINQUENCY AND UNSTRUCTURED In accordance with my hypothesis, the results ACTIVITIES in Model 1, Table 4 show that adolescents with I estimate models of peer delinquency and high street efficacy are less likely to select peers unstructured activities using two-level hierar- who are involved in various delinquent activi- chical linear models in which the neighbor- ties. I also find that the associations between hood-level measures predict variation in each each of the neighborhood-level variables and

Table 4. Selected Models of Peer Delinquency and Unstructured Activities

Peer Delinquency Unstructured Activities

Model 1 Model 2 Model 3 Model 4 Neighborhood-level Model —Intercept 1.757 *** 1.789 *** .075 *** –.070 (.021) (.077) (.020) (.072) —Concentrated disadvantage .038 –.013 .018 .012 (.027) (.034) (.042) (.036) —Collective efficacy .098 .027 .385 * .231 (.109) (.100) (.166) (.126) —Neighborhood violence –.016 .012 .141 .098 (.075) (.071) (.114) (.122) Person-level Model —Street efficacy –.081 *** –.062 ** .019 .006 — (.024) (.022) (.035) (.039) —Verbal ability –.013 –.003 — (.011) (.016) —Impulsivity .002 –.051 (.017) (.034) —Parental supervision –.046 –.076 (.064) (.050) —Domestic violence .016 ** .015 (.007) (.010) —Family criminality .034 *** .027 (.012) (.021) —Marijuana use .047 .136 * (.066) (.072) —Alcohol use .028 .099 (.047) (.071) —School violence .025 –.004 (.020) (.021) —Peer delinquency—wave 1 .260 *** (.046) Notes: Standard errors in parentheses; additional control variables included in Model 2 and Model 4 are: subject’s race/ethnicity, age, gender, and immigration status, caregiver’s age and gender, caregiver’s total household income, education, and marital status, caregiver’s (and spouse’s) employment and occupational status, caregiver’s welfare receipt, total household size, home ownership, length of residence at current address, and subject’s body mass index. * p < .1; ** p < .05; *** p < .01 842—–AMERICAN SOCIOLOGICAL REVIEW peer delinquency are not significant in this day lives. This is true in models using a com- model. As with the models of individual vio- posite scale of “unstructured activities” as an lence, however, unmeasured characteristics of outcome, and also in models using each of the adolescents may be biasing these results. Model individual items in the scale of unstructured 2 shows results from the full specification, in activities as outcomes. The absence of a rela- which the set of individual and family controls tionship between street efficacy and routine are added to the model, along with a lagged activities can be interpreted in multiple ways. measure of peer delinquency from Wave 1 of the One possibility is that adolescents’ confidence survey. I find that family criminality and domes- in their ability to avoid violence simply has no tic violence in the home are positively associ- effect on the types of activities in which they ated with peer delinquency. Street efficacy take part, which would run counter to my maintains a significant negative association hypothesis. with peer delinquency even after a lagged meas- A second possibility is that these measures of ure of peer delinquency is included in the unstructured activities are not capturing the model.16 potential for violence in adolescents’ everyday The magnitude of the association between lives. For instance, a young woman who indi- street efficacy and peer delinquency is rela- cates that she spends a substantial amount of tively small from a substantive standpoint. For time hanging out with friends may not neces- instance, a one-unit increase in street efficacy sarily be spending her time in unsafe environ- (e.g., moving from “sort of ” confident to “very” ments. The potential for violence depends on the confident) is associated with a reduction in the types of friends with whom she spends time, the peer delinquency scale of about .06 units (using places they go, and the way they carry them- results from Model 2) to .08 units (using results selves in those places. These types of factors from Model 3 in Table S3 in the online supple- may not be captured in the items used to meas- ment), in which a one-unit reduction in the peer ure routine activities. delinquency scale would mean adolescents A third possibility is that adolescents with report that “some” of their friends take part in high street efficacy may be better equipped to delinquent activities, as opposed to “all of handle unstructured or unsupervised activities, them.” However, in models not shown, I use a thus mitigating the potential for violence in measure of street efficacy from Wave 3 of the such environments. Indeed, these results may survey to estimate peer delinquency at Wave 3, indicate that youth who are confident in their and I find that the estimated association is sub- ability to avoid violence do not retreat from stantially larger. public life to avoid violence, but rather play an Unlike violent behavior and peer delinquen- active role in their communities while still effec- cy, I find that street efficacy has no effect on the tively managing potentially violent situations. routine activities that occupy adolescents’every- None of these interpretations can be validated with survey data, however. Instead, ethnographic work may well be necessary to develop a rich 16 In results not shown, I replicated these models picture of the ways that adolescents with vary- using each individual item in the scale of street effi- ing levels of street efficacy engage in public life cacy to predict peer delinquency rather than the com- and manage the potential for violence. posite measure. The results were extremely similar across items, although the effect size varied. The item asking whether adolescents can find ways to be SUMMARY safe within a few blocks of their home was the only Although street efficacy has no detectable rela- one not significantly associated with peer delin- tionship with the routine activities of adoles- quency in Model 2. I also estimated models in which cents, I find evidence suggesting that street efficacy is centered around the neighborhood- adolescents who are confident in their ability to level mean rather than the grand mean to determine whether variation in street efficacy occurring with- avoid violence are less likely to resort to vio- in neighborhoods leads to significant differences in lence themselves or to surround themselves peer delinquency. The relationship between street with delinquent peers. I also find that the rela- efficacy and peer delinquency is essentially tionship between neighborhoods, street effica- unchanged in these models. cy, and the selected environment appears to be NAVIGATING DANGEROUS STREETS—–843 more complicated than the diagram outlined in efficacy and either outcome without addition- Figure 1 would suggest. Specifically, although al tests of the theoretical model using alterna- neighborhood concentrated disadvantage is tive methods in contexts other than Chicago. strongly associated with individual violence, Experimental evaluations of programs that use street efficacy does not mediate this association. psychological treatments to manipulate effica- Furthermore, the various aspects of the neigh- cy expectations may ultimately be necessary to borhood environment have minimal effects on demonstrate a causal effect. unstructured activities and peer delinquency. These results make more sense if we consider DISCUSSION the tremendous heterogeneity among adoles- cents living within similar neighborhood envi- Results from the various sets of models provide ronments. For instance, in the models predicting support for a theoretical reconceptualization of street efficacy, I find that only 6 percent of the the ways that social contexts, particularly neigh- variation in street efficacy occurs between borhoods, influence adolescent behavior. At the neighborhoods, leaving 94 percent of the vari- heart of this reconceptualization is the recog- ation within neighborhoods. This does not mean nition that adolescents make choices which that the neighborhood environment is unim- influence their developmental trajectories. portant in explaining adolescents’confidence in Although these choices may be influenced by their ability to avoid violence—as we have seen, their social context, they are not determined by collective efficacy has a strong and robust rela- it. This idea is captured in the conceptual dis- tionship with street efficacy—but rather, that tinction between imposed and selected envi- characteristics of the neighborhood represent ronments. only one of multiple factors that influence how The street efficacy concept provides a theo- children interpret and respond to their environ- retical mechanism necessary for connecting the ment. Evidence of the diverse sources of street imposed environment to the decisions that ado- efficacy reinforces the idea that adolescents are lescents make as they create their own unique not simply passive recipients of the neighbor- lived environments. The concept also under- hood environment. Instead, various character- scores the point that disadvantaged urban neigh- istics of the child, his family, and his community borhoods form unique ecological contexts, combine to influence that child’s confidence in creating a need for specific theoretical con- his ability to engage in public life while avoid- structs suited for explaining social phenomena ing violence. In turn, the strength of an adoles- in such settings. In particular, my focus on street cent’s confidence appears to play an important efficacy reflects the central role that interper- role in influencing the character of the child’s sonal violence plays in the lives of youth living selected environment. in the most distressed urban areas. In summarizing these results, I again point out By developing a theory to explain the ways that my empirical models are limited in sever- that the imposed neighborhood context influ- al ways. The most serious threats to my findings ences adolescents’ beliefs about their ability to are the possibilities that unmeasured charac- avoid violence, it becomes possible to confront teristics of adolescents are biasing my results the critique that in analyzing the effects of neigh- and that there may be reciprocal effects of vio- borhoods, one is essentially ignoring the agency lence and peer delinquency on street efficacy of individuals. Indeed, my primary contention that are not picked up in my models. I attempt is that individual agency is essential to under- to address each of these possibilities by esti- standing the environments that adolescents cre- mating the association between street efficacy ate for themselves. However, the choices that measured at Wave 2 and social outcomes meas- adolescents make do not arise purely from with- ured at Wave 3 to ensure proper temporal order- in, but appear to be shaped at least in part by var- ing, and by including lagged measures of ious aspects of the imposed social environment violence and peer delinquency as control vari- that surrounds them. ables in the final models predicting each out- To help explain adolescents’ choices as they come. The results are found to be robust in navigate potentially dangerous streets, I expand these conservative models. Still, I am reluctant on Bandura’s project to remove self-efficacy to claim a causal relationship between street theory from the controlled context of the exper- 844—–AMERICAN SOCIOLOGICAL REVIEW imental setting (1997), and I apply the theory Harvard University. His research focuses on resi- to help explain variation in the way that ado- dential inequality in urban America and the conse- lescents respond to the potential for violence in quences for adolescents. His dissertation research is the neighborhood environment. In doing so, I aimed at understanding continuity and change in find strong evidence against the conception of the neighborhood environments of different racial and ethnic groups in America over the life course and efficacy as emerging and existing primarily across generations. within individuals or even within families. For instance, I find that adolescents exposed to vio- lence are less confident in their ability to avoid REFERENCES violence, which helps to explain the association Achenbach, Thomas M. 1993. The Young Adult Self- between concentrated disadvantage and indi- Report Revised and the Young Adult Behavior vidual street efficacy. Checklist. Burlington, VT: University of Vermont, It is not only the presence of structural dis- Department of Psychiatry. advantage that influences how adolescents think Allison, Paul D. 2001. Missing Data. Thousand Oaks, about the potential for violence in their neigh- CA: Sage Publications. borhood environment, but rather the way that the Anderson, Elijah. 1990. Streetwise: Race, Class, and Change in an Urban Community. Chicago, IL: community functions to monitor and support University of Chicago Press. local youth. Residents’ willingness and capac- ———. 1999. Code of the Street: Decency, ity to take an active role in regulating public Violence, and the Moral Life of the Inner City. interactions within a given neighborhood have New York, NY: W.W. Norton & Company. been shown to have important effects on the Bandura, Albert. 1977. “Self-efficacy: Toward a types of activities that occur within communi- Unifying Theory of Behavioral Change.” ties (Sampson et al. 1997), but these results Psychological Review 84:191–215. suggest that collective efficacy also may have ———. 1997. Self-Efficacy: The Exercise of an impact on processes of cognition occurring Control. New York: W.H. Freeman & Company. within adolescents in the neighborhood. ———. 2001. “Social Cognitive Theory: An Agentic Perspective.” Annual Review of Adolescents are more likely to believe that they Psychology 52:1–26. can engage in public life while avoiding violence Birkbeck, Christopher and Gary LaFree. 1993. “The when neighborhood residents share a set of Situational Analysis of Crime.” Annual Review of expectations regarding the collective oversight Sociology 19:113–37. of local youth and the enforcement of common Block, Carolyn R. and Richard L. Block. 1993. social norms (Sampson et al. 1999). A sense of “Overview of the Chicago Homicide Project.” Pp. confidence in one’s ability to avoid violence is, 97–122 in Questions and Answers in Lethal Non- in turn, found to be related to the types of envi- Lethal Violence: Proceedings of the First Annual ronments that adolescents carve out from what Workshop of the Homicide Research Working is available to them. In particular, I find that ado- Group, edited by C. R. Block and R. L. Block. Washington, DC: National Institute of Justice. lescents with high levels of street efficacy are Brooks-Gunn, Jeanne, Greg J. Duncan, Pamela Kato less likely to resort to violence themselves or to Klebanov, and Naomi Sealand. 1993. “Do associate with delinquent peers. Neighborhoods Influence Child and Adolescent These findings provide preliminary support Development?” American Journal of Sociology for a perspective on violence control that empha- 99:353–95. sizes individual agency, in which a sense of Bryk, Anthony S. and Stephen W. Raudenbush. 1992. efficacy can be thought of as flowing from con- Hierarchical Linear Models: Applications and cerned citizens to communities to individual Data Analysis Methods. Newbury Park, CA: Sage adolescents. This perspective needs to be test- Publications. ed empirically in contexts other than Chicago Case, Anne C. and Lawrence F. Katz. 1991. “The Company You Keep: The Effects of Family and neighborhoods, but it represents a promising Neighborhood on Disadvantaged Youth.” Working way to think about the role that individuals and Paper 3705. Cambridge, MA: National Bureau of collectivities play in controlling violence Economic Research. through informal sources and through individ- Cohen, Lawrence E. and Marcus Felson. 1979. ual efforts. “Social Change and Crime Rate Trends: A Routine Activity Approach.” American Sociological Review Patrick T. Sharkey is a Ph.D. candidate in the 44:588–608. Department of Sociology and Social Policy at Coleman, James S. 1988. “Social Capital in the NAVIGATING DANGEROUS STREETS—–845

Creation of Human Capital.” American Journal of Conflict and Violence in the Inner City.” The Sociology 94:S95–S120. Annals of the American Academy of Political and Earls, Felton J., Jeanne Brooks-Gunn, Stephen W. Social Science 595:49–62. Raudenbush, Albert J. Reiss, Jr., and Robert J. Katz, Jack. 1988. Seductions of Crime. New York: Sampson. 1995. Project on Human Development Basic Books. in Chicago Neighborhoods (PHDCN): Community Kodjo, Cheryl M., Peggy Auinger, and Sheryl A. Survey, 1995. Ann Arbor, MI: Inter-university Ryan. 2004. “Prevalence of, and Factors Associated Consortium for Political and Social Research. with, Adolescent Physical Fighting While under the Earls, Felton J., Jeanne Brooks-Gunn, Stephen W. Influence of Alcohol and Drugs.” Journal of Raudenbush, and Robert J. Sampson. 1994. Project Adolescent Health 35:e11–e16. on Human Development in Chicago Kruttschnitt, Candace. 1993. “Gender and Neighborhoods (PHDCN): Longitudinal Cohort Interpersonal Violence.” Pp. 293–377 in Study, Waves 1–3 1994–2003. Ann Arbor, MI: Understanding and Preventing Violence, Volume 3, Inter-university Consortium for Political and Social edited by A. J. Reiss and J. A. Roth. Washington, Research. DC: National Academy Press. Ellen, Ingrid G. and Margery A. Turner. 1997. “Does Lauritsen, Janet, Robert J. Sampson, and John Laub. Neighborhood Matter? Assessing Recent 1991. “The Link between Offending and Evidence.” Housing Policy Debate 8:833–66. Victimization among Adolescents.” Criminology Elliott, Delbert S., William J. Wilson, David Huizinga, 29:265–92. Robert J. Sampson, Amanda Elliott, and Bruce Leventhal, Tama, Mary Beth Selner-O’Hagan, Jeanne Rankin. 1996. “The Effects of Neighborhood Brooks-Gunn, Jeffrey Bingenheimer, and Felton J. Disadvantage on Adolescent Development.” Earls. 2004. “The Homelife Interview from the Journal of Research in Crime and Delinquency Project on Human Development in Chicago 33:389–426. Neighborhoods: Assessment of Parenting and Ewart, Craig K. 1995. “Self-Efficacy and Recovery Home Environment for 3- to 15- Year-Olds.” from Heart Attack: Implications for a Social Parenting: Science and Practice 4:211–41. Cognitive Analysis of Exercise and Emotion.” Pp. Loeber, Rolf and Magda Stouthamer-Loeber. 1986. 173–96 in Self-Efficacy, Adaptation, and “Family Factors as Correlates and Predictors of Adjustment: Theory, Research and Application, Juvenile Conduct Problems and Delinquency.” Pp. edited by J. E. Maddux. New York: Plenum Press. 29–149 in Crime and Justice, edited by M. Tonry Farrington, David P. 1995. “The Development of and N. Morris. Chicago, IL: University of Chicago Offending and Antisocial Behavior from Press. Childhood: Key Findings from the Cambridge Maddux, James E. 1995. “Self-Efficacy Theory: An Study in Delinquent Development.” Journal of Introduction.” Pp. 3–27 in Self-Efficacy, Child Psychology and Psychiatry 360:929–64. Adaptation, and Adjustment: Theory, Research, ———. 1998. “Individual Differences and and Application, edited by J.E. Maddux. New York: Offending.” Pp. 241–68 in The Handbook of Crime Plenum Press. and Punishment, edited by M. Tonry. New York, Massey, Douglas S. 1995. “Getting Away with NY: Oxford University Press. Murder: Segregation and Violent Crime in Urban Gephart, Martha A. 1997. “Neighborhoods and America.” University of Pennsylvania Law Review Communities as Contexts for Development.” Pp. 143:1203–232. 1–43 in Neighborhood Poverty, Volume I: Context Matsueda, Ross L. and Kathleen Anderson. 1998. and Consequences for Children, edited by J. “The Dynamics of Delinquent Peers and Brooks-Gunn, G. J. Duncan, and J. L. Aber. New Delinquent Behavior.” Criminology 36:269–308. York: Russell Sage Foundation. Mayer, Susan E. and Christopher Jencks. 1989. Gottfredson, Michael R. and Travis Hirschi. 1990. A “Growing Up in Poor Neighborhoods: How Much General Theory of Crime. Stanford, CA: Stanford Does It Matter?” Science 243:1441–445. University Press. Miczek, Klaus A., Joseph F. DeBold, Margaret Haney, Huizinga, David, Finn-Aage Esbenson, and Anne W. Jennifer Tidey, Jeffrey Vivian, and Elise M. Weerts. Weihar. 1991. “Are There Multiple Paths to 1993. “Alcohol, Drugs of Abuse, Aggression, and Delinquency?” The Journal of Criminal Law and Violence.” Pp. 377–570 in Understanding and Criminology 82:83–118. Preventing Violence, Volume 3, edited by A. J. Irwin, Katherine. 2004. “The Violence of Adolescent Reiss and J. A. Roth. Washington, DC: National Life.” Youth & Society 35:452–79. Academy Press. Jargowsky, Paul A. 1994. Poverty and Place: Ghettos, Nakao, Keiko and Judith Treas. 1994. “Updating Barrios, and the American City. New York: Russell Occupational Prestige and Socioeconomic Scores: Sage Foundation. How the New Measures Measure Up.” Jones, Nikki. 2004. “‘It’s Not Where You Live, It’s Sociological Methodology 24:1–72. How You Live’: How Young Women Negotiate Osgood, D. Wayne, Barbara J. McMorris, and Maria 846—–AMERICAN SOCIOLOGICAL REVIEW

T. Potenza. 2002. “Analyzing Multiple-Item Crime: A Multilevel Study of Collective Efficacy.” Measures of Crime and Deviance I: Item Response Science 277:918–24. Theory Scaling.” Journal of Quantitative Sampson, Robert J. and William J. Wilson. 1995. Criminology 18:267–96. “Toward a Theory of Race, Crime, and Urban Osgood, D. Wayne, Janet K. Wilson, Patrick M. Inequality.” Pp. 37–54 in Crime and Inequality, O’Malley, Jerald G. Bachman, Lloyd D. Johnston. edited by J. Hagan and R. Petersen. Stanford, CA: 1996. “Routine Activities and Individual Deviant Stanford University Press. Behavior.” American Sociological Review 61: Schafer, Joseph L. 1999. “Multiple Imputation: A 635–655. Primer.” Statistical Methods in Medical Research Raudenbush, Stephen W., Anthony S. Bryk, Yuk Fai 8:3–15. Cheong, and Richard Congdon. 2001. HLM5: Selner-O’Hagan, Mary Beth, Daniel J. Kindlon, Hierarchical Linear and Nonlinear Modeling. Stephen L. Buka, Stephen W. Raudenbush, and Lincolnwood, IL: Scientific Software International. Felton J. Earls. 1998. “Assessing Exposure to Raudenbush, Stephen W., Christopher Johnson, and Violence in Urban Youth.” Journal of Child Robert J. Sampson. 2003. “A Multivariate, Psychology and Psychiatry 39:215–24. Multilevel Rasch Model with Application to Self- Straus, Murray A., Sherry L. Hamby, Sue Boney- Reported Criminal Behavior.” Sociological McCoy, and David B. Sugarman. 1996. “The Methodology 33:169–211. Revised Conflict Tactics Scale (CTS2): Rosenbaum, James E., Lisa Reynolds, and Stefanie Development and Preliminary Psychometric Data.” Deluca. 2002. “How Do Places Matter? The Journal of Family Issues 17:283–316. Geography of Opportunity, Self-Efficacy and a Thornberry, Terence P. and Marvin D. Krohn. 2002. Look inside the Black Box of Residential “Comparison of Self-Report and Official Data for Mobility.” Housing Studies 17:71–82. Measuring Crime.” Pp. 43–94 in Measurement Ross, Catherine E., John Mirowsky, and Shana Problems in Criminal Justice Research: Workshop Pribesh. 2001. “Powerlessness and the Summary. Washington, DC: National Academies Amplification of Threat: Neighborhood Press. Disadvantage, Disorder, and Mistrust.” American van Buuren, Stef, Hendriek C. Boshuizen, and Dick Sociological Review 66:568–91. L. Knook. 1999. “Multiple Imputation of Missing Royston, Patrick. 2004. “Multiple Imputation of Blood Pressure Covariates in Survival Analysis.” Missing Values.” Stata Journal 4:227–41. Statistics in Medicine 18:681–94. Rubin, Donald B. 1987. Multiple Imputation for Warr, Mark. 2004. “Making Delinquent Friends: Nonresponse in Surveys. New York: John Wiley Adult Supervision and Children’s Affiliations.” and Sons. Criminology 43:77–105. Sampson, Robert J. and Janet L. Lauritsen. 1993. Wechsler, David. 1974. Wechsler Intelligence Scale “Violent Victimization and Offending: Individual, for Children—Revised. New York: Psychological Situational, and Community-Level Risk Factors.” Corporation. Pp. 1–114 in Understanding and Preventing Wilkinson, Gary S. 1993. WRAT-3: Wide Range Violence, Volume 3, edited by A. J. Reiss and J. A. Achievement Test. Administration Manual. Roth. Washington, DC: National Academy Press. Wilmington, DE: Wide Range, Inc. Sampson, Robert J., Jeffrey D. Morenoff and Felton Wilson, William J. 1987. The Truly Disadvantaged: Earls. 1999. “Beyond Social Capital: Spatial The Inner City, the Underclass, and Public Policy. Dynamics of Collective Efficacy for Children.” Chicago, IL: The University of Chicago Press. American Sociological Review 64:633–60. ———. 1996. When Work Disappears: The World Sampson, Robert J., Jeffrey D. Morenoff, and Stephen of the New Urban Poor. New York: Alfred A. W. Raudenbush. 2005. “Social Anatomy of Racial Knopf. and Ethnic Disparities in Violence.” American Winship, Christopher S. and Stephen L. Morgan. Journal of Public Health 95:224–32. 1999. “The Estimation of Causal Effects from Sampson, Robert J., Stephen W. Raudenbush, and Observational Data.” The Annual Review of Felton Earls. 1997. “Neighborhoods and Violent Sociology 25:659–707.