Adolescent Mental Health

Neighborhood Context, Stress, and Mental Health

Neighborhood influences on individual behavior reflect an ecological approach to social phenomena. Such a bio-ecological framework (Bronfenbrenner & Evans, 2000, Dallaire et al., 2008) where demographic, family, personal, and neighborhood/community are assessed reflect a cumulative risk model to behavioral outcomes. Stress is a central component to the sociological models and sociomedical models for the social etiology of mental health (Aneshensel, Rutter, & Lachenbruch, 1991). In Pearlin’s (1989) discussion of the stress process, he identifies sources, mediators, and manifestations of stress—across various spheres—and argues that differences in exposure, access to resources, and coping strategies are central to examine. The stress process model continues to be a leading framework for understanding mental health (Aneshensel, 2009b; Cutrona, Wallace, & Wesner, 2006; M. Elliott, 2000; Hill, Ross, & Angel, 2005; Pearlin, 1999).

Neighborhoods are an important social context of the stress process (Aneshensel, 2009a). Generally, social environmental factors are thought of as stressors that can affect mental health directly or by diminishing protective factors. In a similar way that there is differential exposure to stress at the individual level, a neighborhood can also create different levels of stress for its inhabitants as well as differential access to resources that protect or mitigate stress. These chronic stressors can produce or exacerbate emotional or psychological distress (Matheson et al., 2006; Steptoe & Feldman, 2001). A stressful context can even produce physiological responses (such as hypertension) through chronic exposure (see Ross & Mirowsky, 2001). Moreover, environmental risks or stressors might activate an individual’s predisposition toward depression (Capsi et al., 2003).

Neighborhood characteristics—physical, structural, and social—are related to mental health outcomes (Wandersman & Nation, 1998). Certain ecological conditions, such as neighborhood economic disadvantage and residential instability, influence the type and level of stress exposure and available resources for coping. Following Massey (1996) and Wilson (1996), disadvantage and instability are important neighborhood conditions. Disadvantage can be characterized as providing environmental stressors and should be negatively related to emotional well-being. In reviews of the literature, significant association between mental health and neighborhood factors was reported in the majority of studies (Mair, Diez Roux, & Galea, 2008; Truong & Ma, 2006). Direct evidence of objective neighborhood economic disadvantage are associated with higher incidence or levels of depression (Galea, Ahern, & Vlahov, 2003; Kim, 2010; Matheson et al., 2006; Ross, 2000), psychological distress (Schulz et al., 2000), and mental illness (Goldsmith, Holzer, & Manderscheid, 1998) among adults.

Conversely, measures of environmental or neighborhood economic advantage are related to available resources and social capital (Coleman, 1988) that impact social networks, social cohesion, and collective efficacy (Sampson, Raudenbush, & Earls, 1997) and act to counter stress or simply are environments with less stress. Economic advantage is conceived to protect individuals and promote mental health through various mechanisms such as social capital and social support (Haines et al., 2011). Fundamentally, these positive circumstances are seen not simply as the obverse of stress but operate differently from stress to promote mental health (e.g., Aneshensel, 2009a, 2009b; Jessor, 1992). For example, researchers indicate that neighborhood resources are related to lowering delinquency and crime (Sampson et al., 1997), while the lack of social control and cohesion are linked to higher urban violence (Morenoff, Sampson, & Raudenbush, 2001). Gary, Stark, and LaVeist (2007) suggest that neighborhood social cohesion is associated with lower levels of anxiety and depression, but for Whites only.

Similarly, instability can be seen as a stress condition or also a general inability to form a sense of community that might be related to the Durkheimian sense of integration. Likewise, such factors are also linked to social disorganization, which are harmful to mental health (Latkin & Curry, 2003). Neighborhood residential mobility, capturing instability, appears associated with greater depression/distress among adult samples (Matheson et al., 2006; Ross, Reynolds, & Geis, 2000).

Residential racial/ethnic heterogeneity serves as a key conceptual condition in studies of urban life and social problems (e.g., Shaw & McKay, 1942) and is typically linked to neighborhood context as a stressor. Heterogeneity might indicate general community problems that are associated with more stress and depression (Gary et al., 2007; Latkin & Curry, 2003) or that represent the potential lack of integration and general stimulus overload as captured by classical sociologists (e.g., Durkheim, Simmel, and Wirth). Some studies show neighborhood racial/ethnic composition associations with depressive symptoms (Mair et al., 2010), while other studies of neighborhood context and depression among adults control for the diversity of population structure, including ethnic diversity, but report no neighborhood findings in multivariate analysis (Matheson et al., 2006).

Most research on mental health for adults and adolescents focuses on cumulative neighborhood disadvantage scores or indices or poverty-type measures. However, while we rely on the theoretical foundation of disadvantage to guide our analysis, we focus on indicators across several domains to identify specifically which features impact adolescent distress. We hypothesize that neighborhood context—such as indicators of economic disadvantage, instability, and heterogeneity—is posited to increase directly stress that increases risk of emotional distress, while positive aspects of the neighborhood—indicators of economic advantage—are expected to be associated with less distress.

Method

Measures

All outcomes and individual-level independent variables come from the RY High School Questionnaire (HSQ), a detailed self-report questionnaire capturing a range of youth behaviors, including emotional well-being, peer and family relations, and school behaviors. The HSQ is designed to use a minimal number of indicators to capture a broad range of factors associated with a set of diverse risk behaviors. Some scales in the questionnaire are truncated from their original version but have held up to validity and reliability analyses (Eggert, Thompson, & Herting, 1994).

Outcome

For this article, we rely on an emotional distress scale to tap into the general mental health of the youth in the sample. The distress scale is based on six questions about depressed affect derived from the CES-D (Center for Epidemiologic Studies Depression Scale) 20-item scale (Radloff, 1977), three questions capturing hopelessness, and three tapping anxiety. The six questions scale well (α = .85); in similar samples, this emotional distress scale correlated with the full CES-D scale at .73.

Individual background and risk/resource factors

We include race/ethnicity, sex, age, household/parent structure, and measures of SES in the individual-level equation. The race/ethnicity measure is based on self-report and school record data; we include categories of Black, Asian, Hispanic, and Other (Other includes individuals self-identified as of “mixed” or multiple ethnicity) and compare with Whites. Age and sex are self-reported information gathered at the time of the survey. Prior mobility is based on the reported number of middle and high schools attended prior to the baseline survey. Family structure, representing living with both biological parents, reconstituted stepparent households, single parent, and other, is included. Finally, the parent’s educational attainment and parent’s employment status based on the youth’s report are included. Using similar data from a previous study, we substantiate youth–parent agreement on current employment and education in line with prior research (Pu, Huang, & Chou, 2011). Given the nature of the households, we record the highest educational level of any present parent; in this manner, the variable represents the highest status of the household rather than the specific mother or father. Whether either parent works in a professional occupation is measured. We also include whether either parent is unemployed at the time of the survey but not a full-time caregiver.

We include five additional measures in the model to capture the known individual risk and protective factors related to distress: drug use, personal coping skills, family support functioning, deviant peer bonding, and positive school attachment/evaluation. Personal coping skills reflects a mean score based of three items tapping into positive coping abilities/strategies (α = .78); items reflect self-assessed ability to face problems directly and solve them (i.e., not to ignore problems). Family support functioning is based on five items capturing general support and communication satisfaction with family; it is derived from an established scale (α = .86). Drug use frequency is based on a sum of self-reported frequency of drug use in the past 30 days (across 10 types of drugs). Deviant peer bonding (D. S. Elliott, Huizinga, & Ageton, 1986) captures the number of close friends involved in six different delinquent behaviors (e.g., use drugs) and has a reliability of .83 and provides a control for peer group structure. Finally, a scale of the youth’s positive view of/attachment to school (e.g., assessment of school’s atmosphere) is included to control for the perceived context of school the individual attends (six items, α = .73).

Neighborhood explanatory variables

Measures of neighborhood characteristics are designed to represent indicators of social/economic disadvantage, instability, and heterogeneity. We have basic measures from the U.S. Census and use separate indicators to represent these general stressors. We group the seven empirical indicators as follows: social-economic disadvantage/advantage (percent below U.S. poverty line, percent unemployed men, percent college educated), instability (percent mobile—people not present in the tract 5 years previously, percent vacant housing), and racial/ethnic heterogeneity (percent foreign born, racial heterogeneity scale). We also control for population density of the tract. In additional analyses, we used a Neighborhood Disadvantage scale constructed by Sampson and collaborators (1997) composed of four indicators of economic disadvantage at the census tract level: percentages of residents below the poverty level, households headed by a female, residents receiving public assistance, and unemployed residents aged 16 years or older (α = .82). The index in general did not perform as strongly as the separate indicators.

Some of the disadvantage indicators tap into relatively stable sources of stress/support (as captured by the poverty and the education measures) signifying the long-term nature of the neighborhood (i.e., as poor), while other indicators tap into the relatively responsive sources of stress/support (as captured by unemployment). Similarly, our measures of instability captures a characteristic of immediate visibility (e.g., vacant housing) and one that is more likely to be associated with the ability for the community to organize and present stable community institutions (e.g., mobility). Clearly, these factors will be correlated, but there is value in examining whether independent effects of these different mechanisms are present. Descriptive statistics of all individual- and neighborhood-level variables are reported in Table 1.

OPEN IN VIEWER

Table 1.

Descriptive Statistics of Neighborhood- and Individual-Level Variables.

	M/SD (n = 131)
Neighborhood level a
Disadvantage
Poverty	9.78/7.59
Unemployment	3.38/1.80
Advantage
College	27.95/9.55
Instability
Mobility	51.44/10.95
Vacancy	3.75/1.83
Heterogeneity
Racial diversity index	0.39/0.18
Foreign born	17.37/10.90
	M/SD (n = 2,006)
Individual level
Age	15.68/1.00
Sex (1 = female)	0.48/0.50
White	0.40/0.49
Black	0.21/0.41
Asian	0.15/0.35
Hispanic	0.05/0.21
Other	0.19/0.39
Parents of origin	0.45/0.50
Stepparents	0.14/0.35
Single parent	0.31/0.46
Other parent	0.10/0.30
Parent unemployed	0.21/0.41
Parent professional	0.51/0.49
Parent education	15.19/2.90
No. of school changes	0.28/0.71
High risk	0.72/0.45
High-risk peers	1.61/1.18
Drug use frequency	0.35/0.50
Family support	3.44/1.61
Personal control	4.20/1.29
School attachment	3.61/1.02

a.

All contextual measures are in percentages except the racial index of heterogeneity.

Statistical Approach

The analysis focuses on the influence of the surrounding neighborhood context for individual’s mental health as gauged by emotional distress. Multilevel techniques (hierarchical linear models [HLMs]) were used to assess the impact of neighborhood context on the distress scale. A hierarchical model explicitly incorporates variables at the individual level and at the aggregate level and accounts for the clustering of individuals in aggregate unit. HLM allows key parameters of interest at the individual level to vary across local contexts and our interest is to see whether this variation is systematically associated with neighborhood factors (Snijders & Bosker, 1999).

The approach was to first estimate a baseline model with effects of neighborhood factors in the model only and second to include individual-level variables. This strategy allows for a general assessment of whether neighborhood effects are associated with emotional distress and better captures the cumulative effects related to neighborhood context operating directly. Our second model is comprehensive, including the neighborhood-level variables and individual-level covariates, that include personal and family risk/protective factors and peer factors, and basic control variables (e.g., age, family structure). This final model controls for key features of the individual including factors that would explain away effects of neighborhood context.

For example, effects of unemployment in the neighborhood may have simply captured the individual’s own status of having unemployed parents. Alternatively, some contextual effects may influence features of the individual’s background (e.g., level of family support), but once the individual feature is controlled, there would no longer be a direct effect of neighborhood on emotional distress. In this case, observing remaining effects of neighborhood on distress, net of these key individual effects provides more evidence that neighborhood contextual effects on distress are present than if such factors were left out of the models. In any spatially arranged data, concern for spatial autocorrelation is present; prior analyses and mapping of the data have not shown any significant spatial autocorrelation.

Results

The correlations between neighborhood context variables are as expected. Statistical associations among many of the neighborhood measures are at moderate to high levels (.5 or greater). While this makes the task of disentangling effects of these features difficult, it is reasonable to explore if separate effects can be discerned (see Table 2).

OPEN IN VIEWER

Table 2.

Correlation Among Neighborhood-Level Variables (n = 131). ^a

	1	2	3	4	5	6	7
1. Poverty	1.00
2. Unemployment	.62	1.00
3. College	−.52	−.44	1.00
4. Mobility	.59	.34	−.11	1.00
5. Vacancy	.57	.46	−.29	.55	1.00
6. Diversity	.64	.48	−.66	.19	.29	1.00
7. Foreign born	.54	.46	−.60	.10	.24	.74	1.00

a.

Correlation > |.16| is significant at p < .05.

As a first step in exploring neighborhood effects, we test whether there is significant between-census tract variation that would warrant exploring systematic factors measured at the neighborhood level. The emotional distress measure, net of the individual factors, showed significant variation across tracts (σ² = 0.0064, χ² = 157, df = 130, p < .05; without covariates: σ² = 0.0197, χ² = 186.32, df = 130, p < .01). Overall the amount of variance related to across neighborhoods is small; the intraclass correlation is .02. First, we estimate a two-level model with all neighborhood contextual variables included and no individual-level covariates (see Table 3, Model 1). We report p values, and in some places refer to 95% confidence intervals (CIs; Gelman & Stern, 2006), to give a more complete assessment of the level of neighborhood impact on emotional distress; we denote contextual effect coefficients by g and individual effects with b. This model shows neighborhood characteristics are significantly related to tract variation observed for the emotional distress scale.

OPEN IN VIEWER

Table 3.

Hierarchical Linear Models of the Effects of Neighborhood and Individual Factors on Emotional Distress Among Adolescents (Level 1, n = 2,006; Level 2, n = 131).

	Model 1			Model 2
	Coefficient	SE	p value	Coefficient	SE	p value
Neighborhood-level variables
Neighborhood disadvantage
Poverty	−0.008	0.006	.216	−0.001	0.005	.815
Unemployment	0.031	0.017	.076	0.028	0.016	.083
Neighborhood advantage
College	−0.012	0.004	.005	−0.008	0.003	.008
Neighborhood instability
Mobility	0.001	0.004	.888	−0.003	0.003	.420
Vacancy	0.028	0.016	.078	0.038	0.014	.008
Neighborhood heterogeneity
Racial diversity index	−0.252	0.201	.214	−0.150	0.167	.370
Foreign born	−0.007	0.003	.020	−0.005	0.003	.045
Population density	−0.008	0.006	.252	0.000	0.001	.909
Individual-level variables
Age				0.010	0.024	.684
Sex (1 = female)				0.235	0.040	.000
Race/ethnicity (White = reference group)
Hispanic				−0.072	0.110	.513
Black				−0.255	0.073	.001
Asian				−0.043	0.065	.505
Other				−0.037	0.060	.531
Personal control				−0.245	0.019	.000
Drug use frequency				0.073	0.054	.177
High-risk peers				0.109	0.021	.000
Family
Parents of origin				0.001	0.067	.992
Stepparents				−0.062	0.086	.469
Single parents				−0.049	0.071	.489
Parent unemployed				0.115	0.050	.023
Parent professional				−0.021	0.041	.608
Parent education				−0.010	0.006	.089
Family support				−0.013	0.016	.418
School
High risk (for drop out)				0.145	0.045	.002
High school changes				0.034	0.030	.263
School attachment				−0.117	0.025	.000
Constant	2.153	0.255	.000	1.827	0.216	.000
Deviance	5,724.4			5,264.8

The percentage of individuals with college degrees or higher—an indicator of economic advantage or positive resources in the neighborhood—is significantly associated with reducing emotional distress (g = −.012, t = −2.92). Instability in the neighborhood also shows a positive impact as expected. The vacancy rate in the census tract appears related to distress net of these other context factors at the .10 level two-tailed test (g = .028, t = 1.77; CI = [−.003, .059]); the CI provides an additional assessment of this effect. Both indicators of diversity are negatively associated with distress though only percent foreign born is significantly related (g = −.007, t = −2.36). The coefficient for racial diversity is not significant. Without individual-level controls, we see that a factor from each of the general dimensions, economic disadvantage/advantage, instability, and heterogeneity, is related to adolescent emotional distress. A weak effect for unemployment is found. As hypothesized, unemployment in the neighborhood is positively related to distress at the p < .10 level two-tailed test. This suggests that individual distress (g = .031, t = 1.79; CI = [−.002, −.064]) increases with lower neighborhood economic viability as captured by unemployment or broadly economic disadvantage; this effect is net of the general level of poverty which is not significant.

In our second step, we add to the model the individual effects (see Table 3, Model 2) and again focus on the presence of key neighborhood features. The analysis of the individual-level model follows our general framework and controls for proximal family support functioning and family context, peer relations, personal coping skills, and basic demographic controls. For the most part, the neighborhood effects observed in the prior model hold once all relevant individual, family, peer, and school factors are included. Percent unemployment continues to be related to increased distress as expected albeit not significant using a two-tail test (g = .028, t = 1.74; CI = [−.003, .059]); this is net of the presence of the adolescent’s own family unemployment (i.e., whether any parent was unemployed in the household). Percent vacant housing—as indication of instability—is significant (g = .037, t = 2.70) in its positive effect on individual distress. Percent college educated maintains significance (g = −.008, t = −2.72) even with controls for individual’s parent’s education level and whether either parent is employed in a professional occupation. Percent foreign born remains negatively associated with the distress scale (g = −.005, t = 2.02) but the racial heterogeneity scale in the census tract remains unrelated in the model. Foreign born captures the general heterogeneous nature of the neighborhood context and has no direct counterpart at the individual level (i.e., we are unable to control for immigrant or generational status of the individual), although we do control for race/ethnicity of the youth.

At the individual level, results are consistent with research on adolescent depression and other indicators of emotional distress. Across all controls, we see female adolescents net of the controls continue to have greater level of distress compared with male adolescents (b = .23, t = 5.9) and individuals reporting high levels of self-efficacy/personal control show a negative effect on emotional distress (b = −.24, t = −13.2). Black adolescents self-report lower levels of emotional distress (b = −.25, t = −3.6) relative to White youth. Youth at risk of school dropout show higher levels of distress (b = .14, t = 3.2) as do youth with unemployed parents (b = .12, t = 2.2). Family support is negatively related—although does not achieve statistical significance—to emotional distress (b = −.01, t = −0.85) and reported associations with deviant friends is positively related with emotional distress (b = .11, t = 5.0). Effects of individual drug use are positive (b = .07, t = 1.3), but not significant at the p < .05 level. The individual’s positive assessment of school is negatively related to distress (b = −.12, t = −4.6). Overall, these results at the individual level correspond well to individual effects observed across a wide variety of studies.

Importantly, the addition of individual-level effects does not strongly alter the previous observed neighborhood-level effects on the distress index and lends support to the effects of contextual factors on mental health status of adolescents. The variance explained in the between variation across neighborhoods by the eight neighborhood factors, controlling for the individual effects, is 19%. Consistent effects of vacancy (indicator of instability) suggest emotional distress is tied to these features of disadvantage in neighborhoods. Moreover, there is a weak effect for unemployment (indicator of disadvantage) that provides some support for the direct role of economic disadvantage in the form of unemployment rather than poverty. The negative effect of percent with college degrees (indicator of economic advantage associated with resources) on distress suggests that social capital and collective efficacy, as captured by higher education, might play a role at the neighborhood level in reducing distress for youth. That these effects are present net of individual-level controls for key individual resources and other proximal contexts, specifically the protective features of family and risk features of deviant peers. This suggests that the observe effects are not simply representing omitted individual factors but rather represent the potential impact of environmental context on adolescent emotional distress.

Additional analyses (not shown) explored how these key individual resources and proximal factors varied in their effects due to these same neighborhood factors. There was no strong evidence of moderation due to neighborhood characteristics for personal control, family support, peer behavior, or school context on emotional distress; this does not preclude their presence given the power to see such interaction effects is low.

Discussion

This article assessed the influence of neighborhood context on adolescent emotional distress. The findings advance our knowledge of the role of context for health and adolescence in key ways. First, this study confirms that neighborhood context plays a role in adolescent mental health. Second, the analysis demonstrates the impact of specific indicators related to three broad neighborhood constructs: disadvantage/advantage, instability, and heterogeneity. Third, these results remain significant in the presence of detailed, and complementary, factors at the individual level, including controls for immediate context of family, peers, and school; such multicontext models are uncharacteristic of this research.

The results from these analyses are largely consistent with work addressing neighborhood and other contextual findings on adolescent mental health as well as broader health behaviors (e.g., Ellickson, Bird, Orlando, Klein, & McCaffrey, 2003). In the data presented specific social and economic features of economic disadvantage (as measured by unemployment), residential instability (as indicated by rates of vacant housing), racial/ethnic heterogeneity (as assessed by percent foreign born), and economic advantage (as captured by high degree of college education) were related to emotional distress; these findings are over and above the contribution of individual characteristics.

The positive association between an indicator of instability (i.e., vacant housing) and emotional distress is also consistent with prior research. The relevance of this finding is related to perspectives about disinvestment, social disorder, and a lack of integration and stability (Wilson, 1996). This instability/vacancy may also be indicative of the inability to form stable social capital in local areas (i.e., instability leads to lower community integration and lowers the potential of community support and monitoring), which may be especially important for adolescent mental health. Alternatively, neighborhood stability has been associated with high levels of psychological distress (Ross et al., 2000), suggesting that less residential stability might indicate social isolation.

The negation association between an indicator of heterogeneity (i.e., foreign born) and emotional distress is somewhat surprising. The finding of percent foreign born—negatively associated with emotional distress—in our study may be capturing more of a distinctive community quality rather than a sense of heterogeneity. Given the observed “protective” nature (i.e., reducing distress) of these results, it is plausible that neighborhoods with a greater proportion of foreign-born residents are capturing a specific well-integrated but separated community. Research suggests that residential segregation leads to greater depression scores for Black men (Mair et al., 2010) and Mexican Americans (Lee, 2009). Neighborhood’s racial minority concentration was not associated with depression scores for youth but did interact with race/ethnicity—where there is a health disadvantage (depressive symptoms and general health status) for minority youth in neighborhoods with racial concentration (Abada et al., 2007). Alternatively, the nature of racial or ethnic neighborhoods may lead to either positive or negative health outcomes depending on the specific type of composition or demographic characteristics of the individual (Walton, 2009). Future research might incorporate more complex measures of racial and ethnic heterogeneity such as measures of dissimilarity or entropy or try to discern organized ethnic communities from simple concentrations of new groups.

The protective role of economic advantage—individuals in the neighborhood with high education—is in line with some recent research. The relationship between neighborhood advantage and mental health is consistent with general arguments about social cohesion and collective efficacy (Aneshensel, 2009b; Sampson et al., 1997). However, others find that indicators of economic advantage do not relate to measures of mental health and that associations between neighborhood problems and health outcomes were independent of measures of social capital (Steptoe & Feldman, 2001). Perceived neighborhood cohesion—often linked to social capital—acted as a protective factor for general health status and depression symptoms (Abada et al., 2007). In general, measures of collective action and social cohesion are not easily represented by the census variables available in this study. However, as a proxy, level of education was consistently related and suggests specific measures of social capital/efficacy are worth investigating in relation to community attachment, social capital, and mental health (Caughy, O’Campo, & Muntaner, 2003).

Consistent with previous studies, the presence of economic disadvantage has some impact on emotional distress. Surprisingly, the unemployment measure was only weakly significant, but it is in the hypothesized direction and remained consistent when individual-level controls were added. A study by van Praag, Bracke, Christiaens, Levecque, and Pattyn (2009) reported that area-level unemployment rate in Belgium had a negative association with adult (aged 15 or older) depression. Also, Matheson and colleagues (2006) report that material deprivation—which included unemployment—is associated with depression among urban adults in Canada. Interestingly, the poverty measure was not significant in our analyses. The unemployment finding, as opposed to a poverty finding, may indicate that emotional distress is more tied to fluctuations in economic circumstances rather than general, often more stable, poverty levels. This could be consequential given current U.S. unemployment levels. Research also suggests that the influence of neighborhood disadvantage on adolescent developmental outcomes is mediated by specific organizational and cultural features of the neighborhood (Elliott et al., 1996).

In addition to these associations, neighborhood context can operate to moderate the influence of individual risk and protective factors (Cutrona et al., 2006). We explored for significant variation in the impact of family functioning, personal coping, peer relations, and school attachment at the individual level and observed weak variation across the tracts. Research also suggests that the impact of neighborhood context may differ by sociodemographic characteristics such as race, ethnicity (Lee, 2009), and sex/gender (Propper et al., 2005; van Praag et al., 2009). We explored possible sex effects but found no random variation in distress by sex across neighborhood units. So for this sample of adolescents, the environmental context that was assessed does not explain any part of the sex disparity on the distress index. Like this study, Matheson and colleagues (2006) explore possible sex-neighborhood cross-level interactions and found none.

Limitations

In applying contextual analysis to individual-level processes, we also face some limitations. There are general concerns about how neighborhoods are measured. Some studies call for smaller units such as block groups, while some research suggests no difference in estimates (Borrell, Diez Roux, Rose, Catellier, & Clark, 2004) or that there is no significant block-level variation for depression (Dupéré & Perkins, 2007). Others argue that the relationship between neighborhoods and mental health outcomes is not dependent on the neighborhood unit used (Sampson, Morenoff, & Gannon-Rowley, 2002). In studies of neighborhoods, one need to acknowledge the selection process in residential patterns. It is difficult to separate the impact of individual preferences and constraints on residential choices; though this effect is less salient given youth in the study are likely not making residential choices.

A major issue for neighborhood research is accurately assessing if reported neighborhood results are in fact reflective of neighborhood-level processes (Oakes, 2004). Although our research is limited in some of the same ways as neighborhood research in general, our analysis matches individual characteristics to key neighborhood variables in an attempt to better clarify the level—neighborhood or individual—of impact. While this process is not flawless, it does alleviate some of the concerns about mistaken attribution. Furthermore, we include a rich individual-level model—including psychosocial risk and protective factors—to control for key individual-level processes making it more difficult for neighborhood conditions to be evident. In sum, while we fully recognize that we cannot eliminate all of the effects of individual characteristics, we are confident that the observed neighborhood associations are not likely driven by left out individual-level variables.

This relates to broader issues about causality (Oakes, 2004). As is true of cross-sectional neighborhood research, we are also not able to assess casual or accurately assess issues of causal ordering. The proposed analysis moves us closer to understanding the role that neighborhood context plays in adolescents’ emotional distress by offering a reasonably well-specified model of individual- and neighborhood-level factors and helps inform future research to better assess causal mechanisms.

The timing and location of the data also have to be considered. The data were compiled between 1998 and 2003. While we may prefer to estimate similar models on more recent data, there is no current comparable data available. However, even if newer data are preferable, the timing of the data should not fundamentally change the relationship between neighborhood conditions and emotional distress. Moreover, as this is theory-driven research trying to get at the processes of neighborhood characteristics and emotional distress, the exact timing of the data is less important.

The data come from one metropolitan area—which is true of much of the neighborhood research—limiting the generalizability of the current analysis. The strength of our data set is that it is a large sample of high school–aged youth with racial, ethnic, and class diversity, and considerable detail at the individual level which is often missing from other studies. Future research, using data from multiple cities, might be able to assess specific city effects that we are unable to do with only one city.

Finally, this analysis did not include measures of perceived social disorder or measures of social capital—social cohesion—that are not easily captured with the objective census data and, if present, might shed more light on the relationship between neighborhoods and mental health among adolescents. Much research points to perceived neighborhood conditions as a mediator of objective neighborhood conditions on mental health outcomes (Aneshensel & Sucoff, 1996; Kim, 2010; Kruger, Reischl, & Gee, 2007; Ross, 2000; Ross & Mirowsky, 2009; Ross et al., 2000). Future research needs to explore perceptions alongside objective neighborhood conditions. Despite these limitations, this study contributes to the growing research linking neighborhood context to health-related behaviors for adolescents, and given the detailed specification helps assure these contextual factors are, in fact, likely present.

Conclusion and Future Research

Overall, this research suggests that greater exposure to stress—in terms of the unemployment and vacancy rate—is associated with emotional distress. The level of economic advantage, in terms of college-educated residents, might reflect a more generalized ability for the community to organize resources available to youth, rather than some general level of knowledge. These findings are consistent with identified mechanisms specific to mental health that indicate how context may generate stressors, exacerbate stressors, impact social capital and social networks, and reduce protective resources used to combat stressors. Future research needs to more fully model and test possible causal mechanisms such as stress exposure, social networks, access, and effectiveness of sociopsychological resources (Aneshensel, 2009b; Cutrona et al., 2006).

Adolescent stress and mental health are important public health concerns and are linked to other youth risk behavior, such as suicide risk (Brent, 1995), sexual risk (Chen, Stiffman, Cheng, & Dore, 1997), social adaptation (Aneshensel & Sucoff, 1996), and long-term health in adulthood (Turner & Butler, 2003; Wickrama, Conger, Wallace, & Elder, 2003). Arguably understanding the links of neighborhood to adolescent mental health is compelling given the importance of local environment to adolescent networks and activities, the extant evidence of these results, and the potential for prevention efforts at the community level (Crosnoe & Johnson, 2011; Leventhal, Dupere, & Brooks-Gunn, 2009; Ruel, 2012). Recent research in the United Kingdom suggests that neighborhood context has implications for intervention effectiveness and needs to be considered in design and implementation (Chiu & West, 2007). Mental health disparities are well documented and certain neighborhood conditions might amplify such disparities that could be vital to community prevention interventions. Aneshensel (2009b) discusses the significant stratification of mental health disparities and argues for the need to identify “leverage points for interventions” (p. 387), which might include multiple contexts and spheres. Additional work needs to be done to better understand the pathways between neighborhoods, other social contexts, and mental health. Meso-level conditions, such as neighborhood context variables assessed here, serve to link individuals to the social structural macro-level conditions that may be central for redefining interventions. This is vital for neighborhood interventions and public awareness (Cutrona et al., 2006). That we observe findings of neighborhood suggest that place for adolescents matters and are not fully buffered by family or simply reflects peer and school context.