Children’s Exposure to Community Violence: Implications for Understanding Risk and Resilience

Abstract

The 5 articles included in this special section are reviewed in this article. The studies encompassed were all focused on pre- or early adolescents, and samples were generally from inner-city areas, with 1 involving rural youth. Considered collectively, the results point to 3 major conclusions: Many children in America are regularly exposed to violence in communities; such exposure carries risk for psychopathology; and parents and other adults can provide valuable support but are limited in how much they can offset the effects of ongoing violence exposure. Intervention implications are, foremost, that community violence itself must be reduced and, second, that positive relationships with significant adults should be fostered to the degree possible among children living in high-risk, violence-prone communities.

The five articles on violence exposure within this special section reflect several strengths, including the focus on rural as well as urban children, the use of multiple methods and respondents in assessments, and the exploration of diverse potentially protective processes, including different parenting domains and high perceived support from various people in the child’s life. Considered in tandem, the results of these studies point to a series of three major conclusions.

Rates of Exposure to Violence Are High Among Children in the United States

Children living in American inner cities are both witness to, and victimized by, an alarming number of violent incidents. At each of three annual assessments of a sample initially assessed in the fifth and seventh grades, Gorman-Smith, Henry, and Tolan (this issue) found that approximately half of all students reported seeing someone beaten up and as many as one in five saw someone shot or killed. Across all three waves, 55% of students reported exposure to three or more violent events (Gorman-Smith et al., this issue). In the Ozer and Weinstein (this issue) study, three quarters of the seventh graders assessed said they had witnessed or been victimized by an act of violence in a period spanning just 6 months.

Whereas previous research has established high rates of violence exposure in American inner cities (e.g., Cooley-Quille, Boyd, Frantz, & Walsh, 2001; Schwab-Stone et al., 1995), the study by Sullivan, Kung, and Farrell (this issue) shows high rates in rural areas as well. In their sample of sixth graders from mostly low-income, agricultural communities, these authors found that lifetime prevalence of witnessing at least one violent act was 61%, and almost half the sample (45%) witnessed multiple violent acts.

The data on ethnic differences in violence exposure rates are mixed across the studies in this special section. Gorman-Smith and colleagues (this issue) found that, as compared to Hispanic youth, African Americans were more vulnerable in that they tended to live in neighborhoods with higher levels of violent crime and were also more likely themselves to be perpetrators of violence. On the other hand, Ozer and Weinstein’s (this issue) data reflect a more complicated picture of ethnic group differences. When witnessing different types of violent crime was the outcome, exposure levels were found to be sometimes higher among African Americans than Hispanics (e.g., seeing someone the person knows being beaten up) but the reverse in other instances (e.g., seeing stranger shot, stabbed, or killed). By contrast, when personal victimization was the outcome considered, Hispanics consistently reported higher exposure across several indicators, including being beaten up, chased or robbed, or shot at or stabbed. Also interesting are Ozer and Weinstein’s findings on Whites and Asian Americans, in that these groups tended to witness crime less often than did African Americans and Hispanics, but their personal victimization was as high and sometimes even higher. To illustrate, rates of being chased or robbed, for the four groups respectively, were 29% and 49% for Whites and Asian Americans, as opposed to 17% and 22% for African Americans and Hispanics, respectively (Ozer & Weinstein, this issue).

In the future, it would be useful for these authors, and others with similar accessible data, to examine the role of different protective and vulnerability processes separately in the different ethnic groups studied. In the regressions reported here, Ozer and Weinstein (this issue) controlled for ethnic differences at the outset and tested for three-way interactions (Ethnicity × Protective Factor × Violence Exposure), very few of which attained statistical significance. These three-way interactions, however, may have failed to detect ethnic differences in protective processes that did actually exist in reality, as such terms are notoriously unstable and almost always have minimal effect sizes. Furthermore, developmental psychopathologists have increasingly emphasized the value of conducting within-group analyses of subsets of children about whom little is known so far (e.g., Garcia Coll, Akerman, & Cicchetti, 2000; Luthar, 1999; Tucker & Herman, 2002); processes in these groups must be understood in their own right and not just in terms of how they statistically compare with processes in other groups. Of Ozer and Weinstein’s sample of 349 seventh graders, approximately 140 were Asian Americans, 100 Hispanic, and 75 African American. At least for these subgroups, sample sizes seem large enough for some potentially informative analyses of risk and protective processes highly prominent within each.

Exposure to Violence Constitutes High Risk for Psychopathology

With one exception (Hammack, Richards, Luo, Edlynn, & Roy, this issue), studies in this special section underscore what has been increasingly reported in the empirical literature, that is, that children living in violent communities are at high risk for both internalizing and externalizing problems. In their sample of African American sixth graders, Hammack and colleagues found few links between violence exposure and depressive symptoms, as well as anxiety symptoms, over the preceding 2 weeks. On the other hand, Sullivan et al. (this issue) showed that sixth graders who had witnessed violence were about 1.4 times more likely to have initiated substance use than were others. Among 9- to 13-year-olds, Kliewer and colleagues (this issue) found that frequent violence exposure was significantly related to both internalizing and externalizing symptoms, with each of these operationalized by composite scores encompassing both caregivers’ reports and youth’s self-reports. Finally, Ozer and Weinstein (this issue) reported significant correlations with seventh-graders’ posttraumatic stress disorder symptoms and more modest ones with depressive symptoms. Considered collectively, these findings indicate that the effects of violence exposure may or may not show up in self-reported depressive or anxiety symptoms in the immediate past, but are more than likely to be evident when considering more long-standing problems across multiple psychopathology domains and as reported by different respondents (see also Margolin & Gordis, 2000).

In the Face of High Community Violence, Positive Family Processes May Have Modest Protective Potential

In many of the articles in this special section, researchers examined whether good family functioning might show “protective–stabilizing” influences—helping children to retain relatively good adaptation even as levels of risk exposure escalate—and, considered together, the findings provide modest support for this type of effect. Hammack and colleagues (this issue) found protective–stabilizing effects in about one sixth of all the instances examined. They tested a total of 32 models and, of these, 11 reached statistical significance; of the 11 significant effects, 5 suggested protective–stabilizing effects. In each of these five instances, children with the hypothesized asset did not show increases in psychopathology (anxiety or depression) despite increases in violence exposure (witnessing or victimization), whereas youth with low levels of the asset showed increased disturbance with greater violence exposure. The five effects reflecting this pattern included one that was longitudinal in nature, and this showed that girls who spent a great deal of time with family members evidenced little increase in anxiety over time even when violence was frequently witnessed, whereas those who spent little time with family reported increased anxiety. Four cross-sectional links suggested protective–stabilizing patterns, and these involved (a) maternal closeness and witnessing, in relation to anxiety as the outcome; (b) girls’ social support and witnessing, in relation to anxiety; (c) boys’ social support and witnessing, in relation to depression; and (d) boys’ social support and victimization, also in relation to depression.

Whereas these five interaction terms suggested some evidence in support of protective–stabilizing effects, the patterns inherent in the other six effects to attain statistical significance were mixed. In two cases, neither of the two slopes subsumed within the interaction effect were significantly different from zero. In two others, the patterns indicated that the hypothesized assets could actually connote heightened vulnerability: Contradicting the cross-sectional results, longitudinal analyses showed that girls with high support seemed to react to both witnessing and victimization with increases in anxiety symptoms whereas those with low support remained relatively stable. The last 2 of the 11 significant interactions showed what might be called “promotive but not protective” effects—promotive (Sameroff, Gutman, & Peck, 2003) because the attribute (social support) was beneficial at low levels of violence exposure and not protective because the benefits were clearly diminished when witnessing or victimization levels were high. Considering the nature of the 11 of 32 examined significant interactions, therefore, the findings were modest at best in supporting protective–stabilizing effects of strong family relationships in the face of community violence.

Ozer and Weinstein (this issue) also reported several instances of protective–stabilizing effects for high perceived support from different family members—mothers, fathers, and siblings. Interpretations of these findings too, however, are somewhat qualified by the fact that each of these potentially overlapping effects was examined in a separate regression equation. Simple correlations show modest positive associations among children’s perceived closeness to their mothers and to other family members. It is unclear, therefore, whether closeness to each family member would contribute significant unique variance to outcomes when effects of the other two are also considered in simultaneous regressions.

As was the case with the previously reported findings by Hammack et al. (this issue), other authors also uncovered instances of “promotive but not protective” effects. Sullivan and colleagues (this issue), for example, found that high family support and high levels of monitoring were each linked with low substance use when violence was witnessed rarely. When witnessing was frequent, however, the benefits of both these factors disappeared. Kliewer et al. (this issue) found that children with high felt parent acceptance showed few internalizing symptoms when violence exposure was low; but when exposure was high, the benefits of parent acceptance were no longer evident. Again, the common theme is that the family asset was in fact apparently helpful in the absence of violence exposure but seemed to do little when exposure levels were high (see also Miller, Wasserman, Neugebauer, Gorman-Smith, & Kamboukos, 1999).

For future studies of resilience, these findings are important in conveying the need for caution in labeling significant “main effects” as representing factors that are compensatory or protective against high-risk conditions. In contemporary research, hierarchical regression analyses are commonly used to examine links between hypothesized assets (a) in direct, main effect links with the outcome; and (b) in a block, including a series of interaction terms involving, in turn, each asset and the risk condition. In situations where the main effect link is statistically significant but the block of all interaction terms fails to achieve statistical significance, many investigators tend to drop from further consideration any of the individual interaction terms subsumed in the block (to guard against Type I errors). Based on the significant main effect finding for the asset in question, however, the convention has been to label this a compensatory or even a protective factor (Luthar, Cicchetti, & Becker, 2000), with the implicit suggestion that it ameliorates the negative effects of the adversity condition. The potential fallacy in this inference is evident in the many promotive but not protective effects identified in this special section—where the main effect arose not from compensation against risk but simply benefits among children with relatively benign life situations.

In future research, simple checks in this regard can be accomplished fairly easily, for example, by demar cating the children in a given sample whose risk scores fell in the top half or top third and then computing simple correlations between the hypothesized asset and each of the outcome variables in this group. If all these correlation coefficients were found to be close to zero, then researchers would be warned against applying labels such as “compensatory” or “protective” based on main effect findings in the sample as a whole. Conducting such checks is critical not only from the standpoint of rigor of research and theory on resilience but also from a pragmatic standpoint. Many clinicians and educators look to resilience research to learn about the specific domains that should be prioritized in most expediently preventing or ameliorating the ill effects of particular risk conditions (Luthar & Zelazo, 2003). It is therefore critical that resilience researchers take care to avoid creating false promise in the potential of particular assets based simply on main effect associations in their low- and high-risk samples combined.

The results discussed thus far in this article all derived from variable-based analyses, and the individual-based analyses implemented in one study (Gorman-Smith et al., this issue) yielded convergent evidence, again showing that positive parenting can be beneficial but that even the most effective parents can be limited in their power to protect against rampant community violence. In their high risk sample, Gorman-Smith and colleagues developed four clusters of families based on functioning across multiple domains; these included exceptionally functioning families, who had strong beliefs in the importance of family, high emotional cohesion, clear communication and organization, and effective parent discipline, monitoring, and involvement; task-oriented families, who had high discipline, monitoring, and structure but low warmth and beliefs about the family; struggling families, who were consistently low on all family assets; and moderately functioning families, who were adequate but not high on these assets. In comparison with youth in moderately functioning or normative families, results showed that children from exceptionally functioning families were significantly less likely to perpetrate acts of violence themselves. On the other hand, these youth were no less likely than the normative group to be personally exposed to violent incidents in their communities, and, as has been noted before, witnessing violence on a regular basis does constitute a significant risk for psychopathology.

As well-functioning families may help shield youth from some of the negative effects of community violence, individual-based analyses also established, conversely, that poor family function is very likely to exacerbate the risks. Gorman-Smith and colleagues’ (this issue) findings showed that youth from struggling families, who experienced consistently poor parenting and low emotional cohesion, were significantly more likely to be exposed to violence than the normative group. Overall, the findings point to the (unsurprising) conclusion that when children experience significant dysfunction in their proximal and their distal environments—of home as well as community—their risk for psychopathology becomes substantial.

Limitations of These Studies

In deriving conclusions based on studies in this special section, there are some limitations that must be considered. In one instance (Sullivan et al., this issue), all measurement was based solely on self-report. The authors correctly note that self-reports tend to be the single best route to assessing the particular constructs they assessed—early adolescents’ substance use, their exposure to violent events, and their perceptions of parents’ behaviors—but at the same time, the shared measurement in self-reports may have inflated some of the univariate associations found. At the multivariate level, this would constitute less of a problem because variance shared with other predictor variables is partialled out in testing of effects that are unique to each.

As is often the case in research with inner-city families, parental consent was not obtained for a substantial proportion of eligible children in these investigations, such that there could be some questions about sample representativeness. Considering the number of participants in relation to the total number of youth who could have participated, the recruitment rate was 58% at baseline in the Hammack et al. (this issue) study, and 60% in the Ozer and Weinstein (this issue) investigation. In the three other articles, recruitment was from a wide pool, but the authors did not specify the numbers of students who were invited to participate but did not do so. Kliewer et al. (this issue), for example, assessed 101 caregivers and their children living in six different inner-city neighborhoods, and Sullivan et al. (this issue) had a sample of 1,722 children drawn from eight different low-income schools, but in both cases, it is unclear how many of those invited refused to participate. Similarly, Gorman-Smith et al. (this issue) received high consent rates (92%) to screen participants eligible for their study, and of an identified “high-risk” subgroup, they randomly selected participants for the study; but here again, rates of refusal to participate are not reported.

Among the longitudinal studies, there was also some variability in attrition rates over time. Of the children initially assessed, the proportions remaining in the samples at subsequent assessments were as follows: 69% (Kliewer et al., this issue), 74% (Sullivan et al., this issue), 77% (Gorman-Smith et al., this issue), and 81% (Hammack et al., this issue). Furthermore, in the first two of these four studies, the children who dropped out of the sample were significantly more troubled, at baseline, than those who remained, indicating that the findings reported are not necessarily representative of processes among the subset of the most high risk of children in these settings.

Despite these limitations, the studies as a group yield results that are quite persuasive for at least two reasons: There is consistency of central themes across all the different samples, and the major results have high face validity. Scientific knowledge is inevitably cumulative in nature; whereas conventional significance testing makes no use of findings from other tests of similar hypotheses, the credibility of particular findings increases incrementally as they are replicated across different investigations (Henkel, 1986; Luthar & Ripple, 1994). As indicated earlier, the major themes that recur across all the studies in this special section (as well as others in the existing literature) are that exposure to violence constitutes a serious risk for child psychopathology and that parents and other adults may be able to help buffer these effects but only to a limited degree. At a purely intuitive level, these broad conclusions appear to be unassailable; their essence seems to be logical, reasonable, and, fundamentally, true to life.

Intervention Implications

From an intervention or prevention perspective, the most significant implication deriving from this body of work is that in prevention efforts, the primary focus must be on the environment rather than on the individuals experiencing the violence. There is broad consensus among the authors of these articles, as exemplified in Gorman-Smith et al.’s (this issue) concluding comments: “the real needed change is less community violence,” as well as those made by Kliewer et al. (this issue): “the first implication, of this study, to the extent that it is possible, is to reduce children’s exposure to community violence,” and Sullivan et al. (this issue): “interventions must focus on the source of the problem by reducing the prevalence of violence within communities.” Hammack et al. (this issue) conclude that ongoing violence exposure “overwhelms the individual, and vulnerability to developing symptoms is virtually inevitable.” These conclusions are also entirely resonant with those deriving from three decades of resilience research (Luthar & Zelazo, 2003); that is, that if children are faced with continuing and severe assaults from the external environment, then they simply cannot sustain resilient adaptation over time—regardless of how much they are helped to believe in themselves, how intelligent they are, or how well they learn to regulate their emotions (see Kliewer et al., this issue). Foremost, therefore, our responsibility as applied social scientists is to remain focused on changing the environments that carry such high levels of risk for children.

Within the environment, the community and family must both be targeted. Common sense dictates that it would generally be better to preempt traumatic events in the community than to help parents continue to be soothing or supportive as their children repeatedly witness acts of wanton destruction. Imperative, therefore, are national-level policy efforts to address the blight and disorganization rampant in poor urban communities and, in particular, the degree to which easy access to firearms facilitates random acts of violence (see Elliott, Williams, & Hamburg, 1998; Osofsky, 1997; Shahinfar, Fox, & Leavitt, 2000). Also important are community-level efforts to develop neighborhood social processes that might help mitigate violence (Cauce, Stewart, Rodriguez, Cochran, & Ginzler, 2003; Elliott et al., 1998; Gorman-Smith & Tolan, 2003; Osofsky, 1997; Sampson, Raudenbush, & Earls, 1997; Sullivan et al., this issue), for example, by empowering community members to take an active role in reporting and reducing crime, by providing good after-school programs and schools that provide “safe havens” for vulnerable youth, and by developing partnerships between the police force and community mental health centers (e.g., Marans & Adelman, 1997).

Interventions at the school level are also important, as suggested by Ozer and Weinstein’s (this issue) findings that students who felt high teacher support showed good adaptive functioning even with high violence exposure, whereas those who felt little teacher support showed poorer functioning with increasing exposure. These results are consistent with arguments increasingly heard in the educational and developmental psychopathology literatures, asserting that close teacher–child relationships have great—and usually insufficiently harnessed—potential to provide critical support to at-risk children in their everyday environments (e.g., Becker & Luthar, 2002; Felner et al., 1993; Luthar & Zelazo, 2003; Pianta, 1999; Reddy, Rhodes, & Mulhall, 2003).

Even as there are community-level efforts at the various levels of policy, neighborhoods, and schools, attention to family processes is critical. As Luthar and Zelazo (2003) noted, all things considered and across diverse settings, it is the family that is the single most influential of external influences, being the earliest, the most proximal, as well as the most enduring of children’s social environments. Thus there must be concerted efforts to address parenting difficulties where they exist among families in high-crime areas (Gorman-Smith et al., this issue; Kliewer et al., this issue). Articles in this special section indicate that for parents of pre- and early adolescents in particular, there are several domains of parenting that, if facilitated, could help, to some degree, reduce the effects of violence exposure on children; these include monitoring, closeness to and support from the primary caregiver, and, more generally, emotional cohesion in the family. Kliewer and colleagues also emphasize explicitly the need to attend to mental health problems of the primary caregiver. Various preexisting manualized interventions can be considered to address parenting concerns for children at different developmental stages ages as well as parents’ own psychological distress, such as those developed by Cicchetti, Toth, and Rogosch (1999); Heinicke et al., (1999); Lieberman, (1992); Luthar and Suchman (2000); and Webster-Stratton and Reid (2003).

Summary

To summarize, the following major conclusions can be drawn from the articles in this special section. First of all, rates of violence exposure among children in contemporary America are high, in the inner cities as well as, potentially, in rural settings. Second, the effects of violence exposure on preadolescents are pronounced, in domains ranging from internalizing and externalizing symptoms to early initiation of substance use. Third, among families living in conditions of poverty, positive parenting—encompassing high monitoring, support, and cohesiveness—can help children maintain adequate levels of adjustment, but even the “best” of these families will be limited in shielding their children when living in neighborhoods where violence is a constant fact of life.

For future resilience research, this collection of articles yields an important message regarding the empirical identification of protective processes. Statistical links among all hypothesized protective factors and the different outcomes must be scrutinized specifically in the high-risk group before presenting any conclusions about compensation or protection against risk based simply on overall main effects links.

For interventions, the findings underscore the need to control violence in communities, with efforts at both national policy and community levels addressing issues such as gun control, neighborhood cohesiveness, empowering communities, and safety in schools. As such interventions evolve, it is also critical to work with the primary caregivers of children facing neighborhood violence, for the experience of recurrent, unpredictable trauma in the everyday environment is inimical for children’s adaptation and that of the adults charged with their care. The challenges of the task involved are vividly conveyed in vignettes such as the following:

The second-grade class in a Boston school takes a field trip that includes a lunch stop at the local McDonald’s restaurant. As the children are getting seated at the picnic tables outside, gunfire erupts. There is much screaming and chaos; several minutes later, the children see two young men lying on the ground in the parking lot. Both are bleeding profusely. The teachers have grabbed the children near them and pulled them to the ground. As soon as they can, they gather the children and return to the school. Police officers are trying to get names and information from the children in the hope of finding an eyewitness to the crime. (Groves & Zuckerman, 1997, p. 191)

With an 8-year-old child who has witnessed an event such as this—and has seen or heard about others such as this on a regular basis in the community—it is clear that even the most effective of parents and teachers would be seriously challenged in being able to effectively erase these memories. The first priority, then, must be to reduce children’s exposure to violence in their everyday environments.