Perceived neighborhood social cohesion moderates the relationship between neighborhood structural disadvantage and adolescent depressive symptoms

Abstract

There is a dearth of research exploring the moderating role of the social environment on neighborhood structural disadvantage and depressive symptoms, particularly among adolescents. Therefore, we examined if adolescent perceptions of neighborhood social cohesion and safety moderated the association between neighborhood structural disadvantage and adolescent depressive symptoms. This cross-sectional study used data from the National Longitudinal Study of Adolescent to Adult Health (Add Health). The study sample consisted of 12,105 adolescents enrolled in 9th–12th grades during the 1994–1995 school year across the United States (U.S.). Mixed effects multilevel modeling was used to determine if adolescent perceptions of neighborhoods moderated the relationship between neighborhood structural disadvantage and adolescent depressive symptoms. Results showed that perceived neighborhood social cohesion moderated the relationship between neighborhood structural disadvantage and adolescent depressive symptoms (p≤0.001). At higher levels of perceived neighborhood social cohesion, neighborhood structural disadvantage was associated with decreased depressive symptoms. Findings suggest that improving perceived neighborhood social cohesion may decrease adolescent depressive symptoms, particularly in neighborhoods with high disadvantage. This aspect of the neighborhood social environment may serve as a target for structural and other interventions to address the growing burden of depression among adolescents.

Introduction

Background

Depression is a common health condition experienced by adolescents and is becoming more common (Mojtabai, Olfson, & Han, 2016). The 12-month prevalence of major depressive episodes (MDE) in adolescents was 12.8% in 2016 compared to 8.7% in 2005 in the United States (U.S.) (Mojtabai, Olfson, & Han, 2016; National Institute of Mental Health, 2017). This is concerning because adolescent depression is a predictor of health risk behaviors and poor health outcomes later in life, including sexual risk-taking (Jackson, Seth, DiClemente, & Lin, 2015), sexually transmitted infections (STIs) (Jackson, Seth, DiClemente, & Lin, 2015), obesity (Schwartz et al., 2016), comorbid psychiatric disorders such as chronic depression and suicide (Goldston et al., 2016; Thapar, Collishaw, Pine, & Thapar, 2012), substance use (Maslowsky, Schulenberg, & Zucker, 2014), and criminal behavior (Anderson, Cesur, & Tekin, 2015). Depression not only affects the individual adolescent, but also the community at large. Depression among adolescents leads to increased expenditures for in- and out-patient costs in the public health sector (Wright et al., 2016).

Much of the research on depression among adolescents has focused on individual- and family-level predictors and correlates (Cairns, Yap, Pilkington, & Jorm, 2014). However, in recent years, research has begun to focus on how the neighborhood environment influences depression. Aspects of the neighborhood environment that have been increasingly studied are social cohesion, neighborhood safety, and neighborhood structural disadvantage. Social cohesion is defined as “neighbors knowing, helping, and trusting each other” and is a component of collective efficacy, which refers to the willingness of neighborhood residents to work together for the common good (Sampson, Raudenbush, & Earls, 1997). In contrast, neighborhood structural disadvantage refers to the lack of institutional, social, and material resources (Hill & Maimon, 2013).

Social cohesion, neighborhood safety, and neighborhood structural disadvantage have been shown to be linked to depression among adolescents. A longitudinal study conducted in Canada found that higher levels of social cohesion predicted adolescents having fewer depressive symptoms (Kingsbury et al., 2015). In addition, among a sample of urban Midwestern African American adolescents, increased levels of social cohesion were correlated with lower levels of depressive symptoms (Hurd, Stoddard, & Zimmerman, 2013). In a sample of Black youth, those who perceived their neighborhoods as unsafe had higher odds of major depressive disorder (Assari, & Caldwell, 2017). Furthermore, a study conducted in California found that adolescents who perceived their neighborhoods as unsafe were two times more likely than those who perceived their neighborhoods as safe to report serious psychological distress (Goldman-Mellor, Margerison-Zilko, Allen, & Cerda, 2016).

Findings related to the association between neighborhood structural disadvantage and depressive symptoms have been mixed (Simons et al., 2002; Xue, Leventhal, Brooks-Gunn, & Earls, 2005). A housing mobility intervention conducted among families in Baltimore, Boston, Chicago, Los Angeles, and New York, showed that boys residing in disadvantaged neighborhoods had increased risk of depressive symptoms (Kessler et al., 2014). Specifically, among African American adolescents, higher neighborhood disadvantage predicted greater internalizing symptoms (Hurd, Stoddard, & Zimmerman, 2013). However, another study among African American children did not find that neighborhood disadvantage was correlated with depressive symptoms (Simons et al., 2002). Neighborhood structural disadvantage was not associated with higher levels of depression severity among 5–11 year olds in Chicago after full adjustment for other covariates (Xue, Leventhal, Brooks-Gunn, & Earls, 2005).

The manner in which neighborhood structural disadvantage may exert its influence on depression is poorly understood, and there have been calls to examine potential pathways (Blair, Ross, Gariepy, & Schmitz, 2014). Studies examining potential mechanisms of neighborhood structural disadvantage on depressive symptoms have mainly focused on mediators. However, the findings of these studies have been inconclusive (Bassett, & Moore, 2013; Joshi et al., 2017; Lee & Liechty, 2015), and not all studies have found that social cohesion acts as a mediator. Studies exploring the moderating role of social cohesion and neighborhood safety on the relationship between neighborhood structural disadvantage and depression are nonexistent, particularly among adolescents. A better understanding of the underlying mechanisms is needed to develop effective multilevel interventions.

In addition, many of the studies examining neighborhood structural disadvantage, neighborhood perceptions, and depressive symptoms have focused on adults or young children (i.e., not adolescents). Focusing on adolescents is imperative given the potential implications that the structural and social environment may have on adolescent depressive symptoms (Hurd, Stoddard, & Zimmerman, 2013; Kingsbury et al., 2015). Studies on adolescents have been limited by small sample sizes, geographically restricted, and lacked racial/ethnic diversity. The studies on adolescents were conducted either outside the U.S. (i.e., Canada) or focused on U.S. subpopulations (i.e., African Americans) in select parts of the country. Despite this, study findings suggest that in the presence of neighborhood structural disadvantage, increased levels of social cohesion among adolescents may help to alleviate depressive symptoms, putting them at a lower risk of depression (Hurd, Stoddard, & Zimmerman, 2013; Kingsbury et al., 2015; Xue, Leventhal, Brooks-Gunn, & Earls, 2005).

Theoretical framework

The proposed study was guided by the Ecological Systems Theory of Human Development (Bronfenbrenner, 1994) and the theoretical framework linking neighborhoods to mental health outcomes, specifically depression, by Hill and Maimon (2013). These two theories were integrated due to their focus on youth development and neighborhood influence on mental health. The Ecological Systems Theory of Human Development posits that child development is influenced by the interaction between the individual and his/her environment. The environment can be divided into various subsystems including the microsystem, mesosystem, exosystem, macrosystem, and chronosystem. For the purposes of the proposed study, only the exosystem, which consists of the neighborhood-community context, was examined. The exosystem contains events that influence processes within the immediate environment in which a developing person lives. These processes can influence the psychological development of an adolescent in a positive or negative way, thereby impacting mental health. A theoretical framework proposed by Hill and Maimon (2013), helps elucidate the relationship between the neighborhood environment and depression. With its origins in social disorganization theory and grounded by empirical evidence, the framework shows how neighborhood structural disadvantage may influence depression itself but can be impacted by individual characteristics, such as perceptions of the neighborhood (Hill & Maimon, 2013). The theoretical framework shown in Figure 1 shows the adaptation of the theoretical framework by Hill & Maimon (2013) for the proposed study, which focuses solely on adolescents.

Figure 1.

Adapted theoretical framework from Hill & Maimon (2013) linking neighborhood structural disadvantage and adolescent depressive symptoms.

Study aims

The objective of the current study was to determine the moderating role of perceived neighborhood social cohesion on the association between neighborhood structural disadvantage and depressive symptoms among U.S. adolescents using our adapted theoretical framework. Moderation was examined due to our adapted theoretical framework drawing on the Ecological Systems Theory of Human Development (Bronfenbrenner, 1994) and the theoretical framework linking neighborhoods to mental health outcomes by Hill and Maimon (2013), which indicates that perceived neighborhood social cohesion may act as a potential moderator. We hypothesized that as perceived neighborhood social cohesion increases, the relationship between neighborhood structural disadvantage and levels of depressive symptoms among U.S. adolescents weakens. Percieved neighborhood safety was also examined as a potential moderator of interest. Findings from this study could help elucidate how the neighborhood environment impacts depressive symptoms and identify neighborhood characteristics that may serve as targets for multilevel interventions to prevent or decrease depressive symptoms among adolescents in the U.S.

Methods

Study design

This cross-sectional study used data from the 1994–1995 National Longitudinal Study of Adolescent to Adult Health (Add Health) to assess the moderating role of perceived neighborhood social cohesion on the association between neighborhood structural disadvantage and depressive symptoms among adolescents (Harris, 2009). Add Health is a nationally representative school-based study designed to determine the developmental trajectories of adolescents into adulthood. Additionally, it assesses adolescent perceptions in relation to their neighborhood. A detailed description of the Add Health methodology can be found elsewhere (Harris, 2011; Harris, 2013).

The study sample was restricted to adolescents who completed the in-home questionnaire, which contained information on neighborhoods needed to address the objective of the study (n=20,745). The sample was further restricted to the core sample due to weights not being available at the neighborhood level to account for unequal probability of selection, bringing the final sample size to n=12,105 (Chen & Chantala, 2014). The core sample is essentially self-weighting as not all students had an equal probability of being included in the study (Chen & Chantala, 2014). To obtain the core sample, U.S. schools were sampled based on region, urbanicity, size, type, and ethnic composition of the target population. For a school to be eligible, it had to have an 11th grade and more than 30 enrolled students. Students who completed the in-school questionnaire and students who did not complete the in-school questionnaire, but were still on the school roster, were eligible for selection. Students were stratified by grade and sex. About 17 students were randomly chosen from each stratum so that approximately 200 adolescents were selected from each of the 80 pairs of schools.

For the in-home questionnaire, data were collected at the respondent’s home using computer-assisted personal interviewing (CAPI) and computer-assisted self-interviewing (CASI). Additionally, parents of the core student sample were asked to complete a parent questionnaire. The mother or other female head of household was the preferred respondent because results from previous studies indicated that mothers tended to be more aware than fathers of their child’s schooling, health behaviors, and health status. However, parents could complete the questionnaire at a later time point if they were unavailable. Approval was obtained from the Florida International University Institutional Review Board prior to conducting the study.

Measures

Outcome

Depressive symptoms.

Depressive symptoms were assessed using 19 out of the 20 items from the Center for Epidemiologic Studies Depression Scale (CES-D) used in Add Health. The 19-item scale has been found to have high reliability (Jacobson & Rowe, 1999; Wight, Botticello, & Aneshensel, 2006). The Cronbach’s alpha was calculated for the sample used for the current study and was 0.88. For each item, respondents had to choose how often each statement was true in the past week. Respondents could choose “never or rarely,” “sometimes,” “a lot of the time,” or “most of the time or all of the time.” A sample item is: “You were happy.” To obtain a depressive symptoms score, items were summed while taking reverse coding into consideration. The depressive symptoms score was treated as a continuous variable. Higher scores indicated higher levels of depressive symptoms.

Exposure

Neighborhood structural disadvantage index.

The index consisted of the following items: proportion of female-headed households with children aged <18 years, unemployment rate, proportion of households receiving public assistance, proportion of nonelderly residents with income below the poverty line, and proportion of African Americans (Burdette & Needham, 2012). Each item was assessed at the census-tract level taken from the Add Health Contextual I database. This database includes neighborhood variables from 19 sources such as the Alan Guttmacher Institute, National Center for Health Statistics, and U.S. Census Bureau (Billy, Wenzlow, & Grady, 1998). These variables are at multiple geographic levels (e.g. state, county census tract, and census block) for each participant. The specific geographic area for each participant was derived from geocoded addresses of participants. Scores for the neighborhood structural disadvantage index were obtained by conducting principal component analysis. Factor loadings from the first principal component were used as weights to arrive at a score for each census tract (or neighborhood). All values were standardized with a mean of zero and a standard deviation of one prior to conducting principal component analysis due to differences in units of measurement (Ringner, 2007). Scores were ranked into quartiles so that the first quartile (Quartile 1) represented the least disadvantaged neighborhoods to facilitate interpretation.

Moderators

Perceived neighborhood social cohesion.

Perceived neighborhood social cohesion was assessed by adolescent respondents indicating if the following 3 items were true: “In the past month, you have stopped on the street to talk with someone who lives in your neighborhood,” “You know most of the people in your neighborhood,” and “People in this neighborhood look out for each other.” Items were summed while taking into consideration reverse coding with higher scores indicating higher levels of perceived neighborhood social cohesion. The items have been found to have moderate reliability (Cronbach’s alpha=0.60) (Donnelly, 2015).

Perceived neighborhood safety.

Perceived neighborhood safety was assessed by adolescent respondents with the following item, “Do you usually feel safe in your neighborhood?” It was treated as a dichotomized categorical variable coded as no or yes.

Covariates

Age, gender, race/ethnicity, family structure, family income, parent occupation, and parent education were adjusted for in the analysis. Age was assessed as a continuous measure reported by the adolescent. Gender was assessed as the participants’ biological sex. It was treated as a dichotomized categorical variable coded as male or female. Race/ethnicity was assessed as a categorical variable and coded as non-Hispanic White, Hispanic, non-Hispanic Black or African American, non-Hispanic Asian or Pacific Islander, non-Hispanic American Indian or Native American, and non-Hispanic other. Family structure was assessed as a categorical variable coded as two-parent, one-parent, and other structured household. Family income was assessed as a continuous variable and reported by the parent in thousands of dollars as the total amount of income, before taxes, the family received in 1994. Parent occupation was assessed as a categorical variable reported by the adolescent for each parent based on U.S. Census Bureau classifications and were ranked as follows: 1) professional/managerial, 2) other professional, which included community/social services, education/training/library, and arts/design/entertainment/sports/media occupations; 3) sales, service, and administration; 4) manual/blue collar, 5) other (unspecified), and 6) not working. The highest occupation of either parent was used as the value for the variable. Parent education was taken as a categorical variable reported by the adolescent. It was coded as college graduate, some college, high school graduate, and less than high school. The highest educational level for either parent was used as the value for the variable.

Statistical analysis

Descriptive statistics included counts and percentages for categorical variables. For continuous variables, medians and interquartile ranges were calculated due to normality tests of residuals performed indicating a non-normal distribution (p<0.001; data not shown). Bivariate analyses were conducted to determine the association between neighborhood structural disadvantage, neighborhood perceptions, and adolescent depressive symptoms. Non-parametric tests Spearman’s rho for continuous variables and Kruskal-Wallis for categorical variables were performed. Correlations and chi-square statistics along with p-values are reported.

Mixed effects multilevel models were used to test the association between neighborhood perceptions and depressive symptoms as well as interactions between neighborhood perceptions and neighborhood structural disadvantage on depressive symptoms (SAS Institute, 2013a). The distribution of the depressive symptoms score, perceived neighborhood social cohesion, age, and family income was not normal; therefore, a square root transformation was used for these variables. Prior to modeling, transformations (i.e. log, log base of 10, square root, quadratic, and cubic) of the data were considered to as ways to approximate a normal distribution. Subsequent normality tests resulted in a significant p-value (p<0.001; data not shown), indicating non-normality. However, normality tests are generally conservative, and the modeling used is considered sufficiently robust to withstand a certain level of violations of assumptions. A square root transformation improved the distribution of the data compared to other transformations due to a skewness of 0.04 and a kurtosis of 0.21. Therefore, a square root transformation was performed on the data. All multilevel models were conducted using the transformed data.

Multilevel models were fitted in stages to evaluate individual- and neighborhood-level characteristics associated with adolescent depressive symptoms. The first model was the empty model, which included only the census tract random effect. The second included the census tract random effect and individual-level characteristics, including adolescent neighborhood perceptions. The third model included all the variables in the second model as well as the neighborhood structural disadvantage index. The fourth model additionally included the interaction terms. For any significant interactions, conditional beta estimates were calculated using the fully adjusted model for the square root of perceived neighborhood social cohesion at the median (highest level), median-3[interquartile range (IQR)] (intermediate level), and median-6IQR (lowest level) to provide a wide enough range of values to evaluate moderation. The median was the highest level at which to evaluate the interaction since the median for the transformed variable was at the highest possible value (median=2.45, range 1–2.45). Median scores and the interquartile range at which to evaluate significant interactions were used due to non-normal distribution of the transformed data. Graphs were generated for visual representation of the interactions. Estimates obtained from the models included the following: 1) intraclass correlation (ICC), 2) β (beta) estimates, 3) standard errors (SE), and 4) p-values. The alpha level applied to test significance was 0.05, including for the interaction terms.

A sensitivity analysis was performed to assess the potential issue of same source bias and reverse causality by excluding all participants with elevated depressive symptoms. The number of participants excluded with high depressive symptoms was 2,100 leading to a total sample size of n=10,005 for the sensitivity analysis. Elevated depressive symptoms were determined by using a clinical cutoff CES-D score of ≥18 following Mendle, Ferrero, Moore, & Harden (2013). The main analysis was repeated using the reduced sample to determine whether depressed adolescents were driving the cross-level interactions between the perceived neighborhood environment and depressive symptoms.

Multiple imputation was used to handle missing data. PROC MI was used to perform imputation and PROC MIANALYZE was used for pooling the estimates in SAS. Imputation methods used were linear regression and the discriminant function for continuous and categorical variables, respectively (Yaun, 2010). Twenty-five imputed datasets were created, following White, Royston, and Wood’s (2011) recommendation that the number of imputed datasets should equal the percentage of incomplete cases. The highest percentage of incomplete cases for a variable was for family income (percentage of incomplete cases was 24.2%, which was rounded up to the nearest whole number to obtain the number of datasets to be imputed). The percentage of missing data for all other variables was relatively low, ranging from 0.01-0.07%. All variables used in the analysis model, including the outcome, were used in the imputation model, as well as all interaction terms except for the census tract ID, to obtain adequate results (He, 2010). The sample size of each dataset was 12,105 after imputation. However, for the analysis the sample size decreased to n=11,977 due to missing census tract IDs for 128 participants. No weighting was applied to the data. Analyses were conducted using SAS v9.4 statistical software (SAS Institute, 2013b).

Results

Sample

The median age of the study sample was 16 [interquartile range (IQR)=3.0, range 11–21] (Table 1). The majority of participants were non-Hispanic White (61.4%), female (52.3%), and came from a two-parent family household (64.9%). The median perceived neighborhood social cohesion score was 6.0 (IQR=1.0, range 1–6). A high percentage (89.8%) of adolescents perceived their neighborhood as being safe. Adolescent depressive symptoms scores were in the mild range (median=10.0, IQR=10.0, range 0–54). Additional sample characteristics including individual- and neighborhood-level can be found in Table 1.

Table 1.

National Longitudinal Study of Adolescent to Adult Health (Add Health) sample characteristcs, 1994–1995 (n=12,105)^*ⱡ

	Median (IQR) or N (%)	Range
Individual characteristics
Age	16.0 (3)	11–21
Gender
Male	5780 (47.8)
Female	6324 (52.3)
Race/ethnicity
White, Non-Hispanic	7423 (61.4)
Other, Non-Hispanic	128 (1.1)
Native American or American Indian, Non-Hispanic	251 (2.1)
Asian or Pacific Islander, Non-Hispanic	508 (4.2)
Black or African American, Non-Hispanic	2328 (19.3)
Hispanic	1457 (12.1)
Parent educationa
College graduate	4023 (34.3)
Some college	1588 (13.5)
High school graduate	4354 (37.1)
Less than high school	1774 (15.1)
Parent occupationb
Professional/mangerial	1790 (15.2)
Other professionalc	2177 (18.5)
Sales, service, administration	2867 (24.3)
Manual/blue collar	1448 (12.3)
Other (unspecified)	2716 (23.0)
Not working	800 (6.8)
Family income, in thousands of dollars	40.0 (39.0)	0–999
Family structure
Two parents	7809 (64.9)
One parent	3530 (29.3)
Other	701 (5.8)
Perceived neighborhood social cohesion	6.0 (1)	1-6
Perceived neighborhood safety
No	1225 (10.2)
Yes	10793 (89.8)
Depressive symptoms	10.0 (10)	0–54
Neighborhood characteristic
Neighborhood structural disadvantage index
Quartile 1 (least)	2.15 (1.92)	0.72–14.56
Quartile 2	−0.16 (0.70)	−0.65–0.72
Quartile 3	−1.04 (0.25)	−1.28–(−0.65)
Quartile 4 (most)	−1.52 (0.35)	−2.30–(−1.28)

^*Totals and percentages may not add up to the total sample size and 100, respectively, due to missing data and rounding

^ⱡDue to rounding, values displayed may be zero

IQR=interqaurtile range

^aThe highest educational level for either parent was used as the value for the variable as reported by the adolescent.

^bThe highest occupation of either parent was used as the value for the variable as reported by the adolescent.

^cOther professional – community/social services, education/training/library, and arts/design/entertainment/sports/media occupations

Preliminary analyses

Normality tests of residuals were performed indicating a non-normal distribution for the depressive symptoms score, age, perceived neighborhood social cohesion, and family income (p<0.001; data not shown). Due to this, descriptive statistics included medians and interquartile ranges using the non-transformed variables for continuous variables as well as counts and percentages for categorical variables. For bivariate analyses, non-parametric tests Spearman’s rho for continuous variables and Kruskal-Wallis for categorical variables were performed.

The first model was the empty model, which included only the census tract random effect. From this model the ICC was calculated to be 0.03, although low is typical for observational studies (Killip, Mahfoud, & Pearce, 2004). The design effect was 1.18 and 1.03, for the mean and the median number of participants in a census tract, respectively. The mean and median were used to calculate the design effect due to the high number of census tracts with only 1 participant. Although the ICC and design effects were small, we were interested in level-2 as well as level-1 effects. Therefore, a multilevel model was considered appropriate for the study to control for the neighborhood environment in order to obtain unbiased estimates (Killip, Mahfoud, & Pearce, 2004; Lai & Kwok, 2015).

Bivariate analysis

Table 2 shows the bivariate analyses between variables of interest. Adolescent depressive symptoms scores were significantly associated with the neighborhood structural disadvantage index (p-value ≤0.001), perceptions of neighborhood social cohesion (p-value ≤0.001), and safety (p-value ≤0.001). All individual-level characteristics were significantly associated with adolescent depressive symptoms (p-value ≤0.001; data not shown). Due to our interest in effect modification, additional bivariate analyses between neighborhood perceptions and the neighborhood structural disadvantage index were conducted. Results showed that perceived neighborhood social cohesion (x²=27.26, p≤0.001) and safety (x²=349.20, p≤0.001) were each associated with neighborhood structural disadvantage.

Table 2.

Bivariate associations between depressive symptoms and neighborhood perceptions and structural disadvantage

	Spearman ρ coefficient or Kruskal- Wallis x2	p-value
Perceived neighborhood social cohesion	−0.10	≤0.001
Perceived neighborhood safety	291.46	≤0.001
Neighborhood structural disadvantage index quartiles	97.96	≤0.001

Associations of perceived neighborhood social cohesion and safety with depressive symptoms

The results of the multilevel models for perceived neighborhood social cohesion and perceived neighborhood safety are given in Table 3. Perceived neighborhood social cohesion was associated with depressive symptoms among adolescents after adjustment for individual characteristics and the neighborhood structural disadvantage index (p≤0.001) in Model 3. Every unit increase in the square root of perceived neighborhood social cohesion corresponded to a 0.24 unit decrease in the square root of depressive symptoms (SE=0.04). Also, perceived neighborhood safety was associated with depressive symptoms in adolescents (p≤0.001) in Model 3. Compared to adolescents who did not perceive their neighborhood as being safe, those who did perceive their neighborhood as being safe had a 0.47 unit (SE=0.04, p≤0.001) lower square root of depressive symptoms score. The neighborhood structural disadvantage index was not associated with depressive symptoms.

Table 3.

Beta coefficients from mixed effects multilevel models testing the interactions between neighborhood perceptions and neighborhood structural disadvantage on depressive symptoms^€ⱡ*

Random effect	Model 1		Model 2		Model 3		Model 4
	β Estimate	SE	β Estimate	SE	β Estimate	SE	β Estimate	SE
Intercept	0.05***	0.01	0.01**	0.00	0.01**	0.00	0.01**	0.00
Fixed effects
Individual characteristics
Age			0.45***	0.05	0.45***	0.05	0.44***	0.05
Gender
Male			-	-	-	-	-	-
Female			0.23***	0.02	0.23***	0.02	0.23***	0.02
Race/ethnicity
White, Non-Hispanic			-	-	-	-	-	-
Other, Non-Hispanic			0.03	0.10	0.03	0.10	0.02	0.10
Native American or American Indian, Non-Hispanic			0.24**	0.07	0.25**	0.07	0.25**	0.07
Asian or Pacific Islander, Non-Hispanic			0.34***	0.05	0.33***	0.05	0.33***	0.05
Black or African American, Non-Hispanic			0.10**	0.03	0.11***	0.03	0.11**	0.03
Hispanic			0.12***	0.04	0.13***	0.04	0.13***	0.04
Parent educationa
College graduate			-	-	-	-	-	-
Some college			0.02	0.04	0.02	0.04	0.02	0.04
High school graduate			0 19***	0.03	0 19***	0.03	0 19***	0.03
Less than high school			0.39***	0.04	0.40***	0.04	0.40***	0.04
Parent occupationb
Professional/mangerial			-	-	-	-	-	-
Other professionalc			0.02	0.04	0.02	0.04	0.02	0.04
Sales, service, administration			0.09*	0.04	0.09*	0.04	0.09*	0.04
Manual/blue collar			0.15***	0.04	0.16***	0.04	0.16***	0.04
Other (unspecified)			0.09**	0.04	0.10**	0.04	0.10**	0.04
Not working			0.05	0.06	0.06	0.06	0.06	0.06
Family income, in thousands of dollars			−0.01*	0.01	−0.01*	0.01	−0.01*	0.01
Family structure
Two parents			-	-	-	-	-	-
One parent			0.12***	0.03	0.12***	0.03	0.12***	0.03
Other			0.21***	0.05	0.21***	0.05	0.21***	0.05
Perceived neighborhood social cohesion			−0.24***	0.04	−0.24***	0.04	−0.11	0.10
Perceived neighborhood safety
No			-	-	-	-	-	-
Yes			−0.46***	0.04	−0.47***	0.04	−0.46***	0.06
Neighborhood characteristic
Neighborhood structural disadvantage index
Quartile 1 (least)					-	-	-	-
Quartile 2					0.03	0.04	−0.08	0.29
Quartile 3					0.03	0.04	0.41	0.30
Quartile 4 (most)					0.04	0.04	1.03***	0.30
Interaction terms
Neighborhood structural disadvantage index x perceived neighborhood social cohesion							0.14***	0.04
Neighborhood structural disadvantage index x perceived neighborhood safety							0.03	0.03

^*p<0.05

^**p≤0.01

^***p≤0.001

SE=standard error

^aThe highest educational level for either parent was used as the value for the variable as reported by the adolescent.

^bThe highest occupation of either parent was used as the value for the variable as reported by the adolescent.

^cOther professional – community/social services, education/training/library, and arts/design/entertainment/sports/media occupations

^€Multiple imputation was used to deal with missing data

^ⱡDue to rounding, values displayed may be zero

^*All analyses were conducted on transformed data

Moderation by perceived neighborhood social cohesion and safety

The interaction term between neighborhood structural disadvantage and the square root of perceived neighborhood social cohesion was significant (p≤0.001), but for perceived neighborhood safety it was not (Table 3, Model 4). Figure 2a shows the graphical representation of the structural disadvantage-social cohesion interaction adjusted for covariates and interactions from Table 3, Model 4. Conditional estimates for the neighborhood structural disadvantage quartiles by the square root of perceived neighborhood social cohesion at the median (highest levels), median-3IQR (intermediate levels), and median-6IQR (lowest levels) are also displayed in Figure 2a. At the lowest (median-6IQR) levels of the square root of perceived neighborhood social cohesion, neighborhood structural disadvantage was associated with a decrease in the square root of depressive symptoms as depicted by the slope and conditional estimate (β=−0.05, SE=0.07, p≥0.05). However, this association was not statistically significant. Similarly, the slopes and conditional estimates at intermediate (median-3IQR) (β=−0.13, SE=0.06, p<0.05) and higher levels (median) (β=−0.22, SE=0.07, p≤0.01) of the square root of perceived neighborhood social cohesion show that neighborhood structural disadvantage was associated with a decrease in the square root of depressive symptoms. The associations for the intermediate (median-3IQR) and higher (median) levels of the square root of perceived neighborhood social cohesion were statistically significant with the greatest decrease in depressive symptoms seen at the highest (median) levels of perceived neighborhood social cohesion as seen in the figure.

Figure 2a.

Interaction graph and conditional estimates of neighborhood structural disadvantage on depressive symptoms by levels of the square root of perceived neighborhood social cohesion for the full sample. Note: Conditional beta (β) estimates were calculated adjusted for covariates and interactions from Table 3 Model 4 and for the square root of perceived neighborhood social cohesion at the median=2.45, median-3IQR=1.82, and median-6IQR=1.19. Values of the square root of adolescent depressive symptom scores are at the median values depicted. Bold SE is standard error. Highlighted graph lines denote significance. Significance values: *p<0.05, **p≤0.01, ***p≤0.001

Sensitivity analysis

The sensitivity analysis excluding all participants with elevated depressive symptoms (n=10,005) revealed similar results for the variables of interest. However, estimates were attenuated. The model with individual- and neighborhood-level variables showed that both the square root of perceived neighborhood social cohesion and perceived neighborhood safety were significantly associated with a 0.14 (SE=0.04, p≤0.001) and 0.21 (SE=0.03, p≤0.001) decrease in the square root of depressive symptoms, respectively (see Appendix A, Model 3). These betas were smaller than those for the full sample (−0.24 and −0.47 respectively). As in the full model, the interaction term for the neighborhood structural disadvantage index and perceived neighborhood social cohesion was significant but the interaction term with perceived neighborhood safety was not.

The graphical representation of the interaction and the conditional estimates for the neighborhood structural disadvantage quartiles by the square root of perceived neighborhood social cohesion at the median (highest levels), median-3IQR (intermediate levels), and median-6IQR (lowest levels) for the sensitivity analysis using the restricted sample are displayed in Appendix B, Figure 2b. The figure shows that at the lowest (median-6IQR) (β=−0.01, SE=0.06, p≥0.05) and intermediate (median-3IQR) (β=−0.08, SE=0.06, p≥0.05) levels of the square root of perceived neighborhood social cohesion, neighborhood structural disadvantage was associated with a decrease in the square root of depressive symptoms as shown by the slopes and conditional estimates. However, these associations were not statistically significant. At higher (median) levels (β=−0.15, SE=0.07, p<0.05) of the square root of perceived neighborhood social cohesion, neighborhood structural disadvantage was associated with a decrease in the square root of depressive symptoms as depicted by the slope and conditional estimate, which was statistically significant. This differs from what was found for the full sample, in which neighborhood structural disadvantage was associated with a decrease in depressive symptoms for both the intermediate (median-3IQR) (β=−0.13, SE=0.06, p<0.05) and higher levels (median) (β=−0.22, SE=0.07, p≤0.01) of the square root of perceived neighborhood social cohesion.

Discussion

The current study examined relationships between neighborhood structural disadvantage, neighborhood perceptions, and adolescent depressive symptoms. Using a nationally representative sample of adolescents, we examined potential moderating factors that may mitigate the effects of living in a disadvantaged neighborhood on depression. To our knowledge, this is the first study that has conducted such an examination among adolescents, a population experiencing an increase in depression over the last decade. Our findings extend our knowledge of how neighborhood structural disadvantage and the social environment of communities may influence mental health among adolescents.

Overall, we found that in addition to individual-level factors, adolescent perceptions of higher neighborhood social cohesion and safety were associated with lower levels of depressive symptoms. Neighborhood structural disadvantage was associated with higher levels of depressive symptoms, but not in the moderation analysis. In the moderation analysis, when perceived neighborhood social cohesion was high, a reduction in depressive symptoms was seen in each quartile of neighborhood structural disadvantage with the greatest reduction in the most disadvantaged quartiles. Thus, there is a protective relationship between neighborhood structural disadvantage and depressive symptoms at high levels of perceived neighborhood social cohesion. Therefore, our hypothesis was partially supported as well as our adapted theoretical framework. These results may indicate that at high levels of perceived neighborhood social cohesion, there are no negative effects of neighborhood structural disadvantage on depressive symptoms. Similar results were found in a study conducted among adults in the Netherlands (Erdem, Van Lenthe, Prins, Voorham, & Burdorf, 2016). Significant interaction effects were found between neighborhood social cohesion and socioeconomic status. Individuals with financial deprivation living in neighborhoods with high social cohesion had lower psychological distress compared to those living in neighborhoods with low social cohesion. In addition, those who received disability, social assistance, or unemployment benefits and were living in high socially cohesive neighborhoods had lower psychological distress than those living in low socially cohesive neighborhoods. Despite this study being conducted among an adult population and examining psychological distress, it does lend support to our findings as it shows that neighborhood social cohesion may moderate the relationship between different aspects of disadvantage and mental health. Although perceived neighborhood safety was found to be associated with depressive symptoms, it was not found to moderate the association between depressive symptoms and neighborhood structural disadvantage. It is possible that the mechanism by which perceived neighborhood safety influences depression is different from that of perceived neighborhood social cohesion (i.e. mediation and not moderation) although it should be noted that there was only a small proportion of adolescents who perceived their neighborhood as not safe. It has been reported that internalized experiences with violence due to living in unsafe neighborhoods may influence depressive symptoms via perceptions of neighborhood disorder (Curry, Latkin, & Davey-Rothwell, 2008). Alternatively, it is possible that self-selection of families of adolescents with high depressive symptoms could have accounted for our findings (e.g., families of adolescents with high depressive symptoms moved to disadvantaged neighborhoods, perhaps due to a third variable such as low socioeconomic status). However, a sensitivity analysis revealed similar results when participants with high depressive symptoms were excluded.

The associations found between perceptions of the neighborhood social environment and adolescent depressive symptoms are in line with the existing literature. Previous studies have found that socially cohesive neighborhoods where there are strong social ties among residents support mental health among adolescents (Donnelly et al., 2016, Solmi et al., 2017, Lowe et al., 2014). A longitudinal study found that low social cohesion predicted higher odds of depressive symptoms at age 18 compared to adolescents living in highly cohesive neighborhoods (Solmi et al., 2017). Furthermore, elevated neighborhood crime and lack of neighborhood safety have been related to increased psychological stress and depression for fear of exposure to violence (Assari & Caldwell, 2017). In addition, among a sample of inner-city adolescent African American and Caribbean youth, a higher risk of major depressive disorder was found among males who perceived their neighborhood as being unsafe (Assari & Caldwell, 2017).

Our findings indicated that the neighborhood structural environment is important to adolescent mental health. We found that neighborhood structural disadvantage was associated with adolescent depressive symptoms, but not in the moderation analysis. The association between neighborhood structural disadvantage and adolescent mental health has been found in other studies. The ‘Moving to Opportunity’ housing mobility intervention conducted among families in Baltimore, Boston, Chicago, Los Angeles, and New York showed that boys residing in poor neighborhoods had increased risk of depressive symptoms (Kessler et al., 2014). In addition, a longitudinal study conducted in Canada found that living in poor disadvantaged neighborhoods predicted suicidal thoughts as well as suicide attempts in late adolescence (Dupéré, Leventhal, & Lacourse, 2009).

Limitations

These findings must be considered with caution in light of the study limitations. One limitation was our reliance on self-ratings of depressive symptoms, and we could not control for family history of depression because that information was not in the dataset. Moreover, due to the cross-sectional nature of the study, we cannot draw causal conclusions about the relationship between neighborhood structural disadvantage and adolescent depressive symptoms; nor were we able to assess mediation. It must be noted that we cannot rule out same-source bias; adolescents with depressive symptoms may have assessed their neighborhoods more negatively (Diez-Roux, 2007). Selection and non-response bias could have influenced our results. Non-response bias was addressed using multiple imputation. Although every effort was made to ensure that all participants’ addresses were geocoded, not every residence could be geocoded. Furthermore, defining a neighborhood by census tract may not have been the same as what participants perceived the boundary of their neighborhood is. Length of residence was not considered in the analysis. Other networks (e.g., peer groups) were also not considered due to our focus on perceptions of the neighborhood social environment. Perceived neighborhood social cohesion had a low Cronbach’s alpha for the study sample. Lastly, the data were collected in 1994–1995 and may not necessarily be representative of the present adolescent population.

Practical implications

Neighborhood structural disadvantage and neighborhood perceptions of social cohesion could serve as targets for the development of intervention strategies aimed at reducing depression, which has been suggested in the literature (Ahern & Galea, 2011; Fullerton et al., 2015). Ultimately, addressing neighborhood structural disadvantage and improving perceived neighborhood social cohesion along with perceived neighborhood safety, may help to reduce depressive symptoms and increase mental health service utilization among adolescents and subsequently depression risk and prevalence, thereby reducing the growing mental health burden among youth (Fleury, Ngui, Bamvita, Grenier, & Caron, 2014; Mmari, Marshall, Hsu, Shon, & Eguavoen, 2016).

Conclusions

This current study helps to advance the understanding of the associations between neighborhood processes and adolescent depressive symptoms. However, further studies are needed to validate our findings. Future studies should be conducted in other adolescent populations and use more recent available data. Even so, our study findings could help in the identification of neighborhood characteristics that impact depression among adolescents in the U.S. to advise the development of multilevel interventions. Perceived neighborhood social cohesion may be targeted in an intervention and thus increased by providing residents opportunities to engage in community activities that build social ties, solidarity, and trust (Chung et al., 2009). The neighborhood social environment may be more feasible as a multilevel intervention target than that of the structural environment due to the complexity of addressing such an aspect.

Furthermore, the influence of neighborhood perceptions varied suggesting a need to modify interventions based on varying levels of neighborhood structural disadvantage. Even so, interventions aimed at increasing perceived neighborhood social cohesion for mental illness prevention have been found to be promising (Chung et al., 2009), and a randomized-controlled trial aimed at changing the neighborhood structural environment of participants was found to decrease depressive symptoms (Kling, Liebman, Katz, 2007; Leventhal & Brooks-Gunn, 2003). Even the reduction of low depressive symptoms among adolescents might help to prevent the progression of clinical depression since subclinical levels of depressive symptoms are associated with impaired functioning (Rodríguez, Nuevo, Chatterji, & Ayuso-Mateos, 2012). In addition, interventions with social cohesion as a target have been successful for other health outcomes such as HIV/AIDS and STDs (Bell et al., 2008; Carlson, Brennan, & Earls, 2012), suggesting that social cohesion is a malleable and potentially promising target for depression prevention interventions.

Highlights.

• Structurally disadvantaged areas associated with increased adolescent depression

• Perceived community social cohesion and safety associated with decreased depression

• Social cohesion moderates adolescent depression-structural disadvantage association

• Neighborhood social environment should be considered a target for interventions

Appendix A

Sensitivity analysis beta coefficients from mixed effects multilevel models testing the interactions between neighborhood perceptions and neighborhood structural disadvantage on depressive symptoms^€ⱡ*

Random effect	Model 1		Model 2		Model 3		Model 4
	β Estimate	SE	β Estimate	SE	β Estimate	SE	β Estimate	SE
Intercept	0.02***	0.00	0.01*	0.00	0.01*	0.00	0.01*	0.00
Fixed effects
Individual characteristics
Age			0.26***	0.04	0.26***	0.04	0.26***	0.04
Gender
Male			-	-	-	-	-	-
Female			0.06***	0.02	0.06***	0.02	0.06***	0.02
Race/ethnicity
White, Non-Hispanic			-	-	-	-	-	-
Other, Non-Hispanic			0.08	0.09	0.08	0.09	0.08	0.09
Native American or American Indian, Non-Hispanic			0.11	0.07	0.12	0.07	0.11	0.07
Asian or Pacific Islander, Non-Hispanic			0.24***	0.05	0.24***	0.05	0.24***	0.05
Black or African American, Non-Hispanic			Q 09***	0.03	0.10***	0.03	0.10**	0.03
Hispanic			0.08**	0.03	0.09**	0.03	0.09**	0.03
Parent educationa
College graduate			-	-	-	-	-	-
Some college			0.02	0.03	0.01	0.03	0.02	0.03
High school graduate			0.11***	0.02	0.11***	0.02	0.11***	0.02
Less than high school			0.25***	0.04	0.25***	0.04	0.25***	0.04
Parent occupationb
Professional/mangerial			-	-	-	-	-	-
Other professionalc			0.02	0.03	0.02	0.03	0.02	0.03
Sales, service, administration			0.09**	0.03	0.09**	0.03	0.09**	0.03
Manual/blue collar			0.14***	0.04	0.14***	0.04	0.14***	0.04
Other (unspecified)			0.08*	0.03	0.08*	0.03	0.08*	0.03
Not working			−0.02	0.05	−0.02	0.05	−0.02	0.05
Family income, in thousands of dollars			−0.01*	0.00	−0.01*	0.00	−0.01*	0.00
Family structure
Two parents			-	-	-	-	-	-
One parent			0.05	0.02	0.05	0.02	0.04	0.02
Other			0.09*	0.05	0.09*	0.05	0.09	0.05
Perceived neighborhood social cohesion			−0.14***	0.04	−0.14***	0.04	−0.03	0.09
Perceived neighborhood safety
No			-	-	-	-	-	-
Yes			−0.20***	0.03	−0.21***	0.03	−0.20***	0.05
Neighborhood characteristic
Neighborhood structural disadvantage index
Quartile 1 (least)					-	-	-	-
Quartile 2					0.02	0.03	0.05	0.27
Quartile 3					0.03	0.03	0.29	0.27
Quartile 4 (most)					0.01	0.03	0.74**	0.27
Interaction terms
Neighborhood structural disadvantage index x perceived neighborhood social cohesion							0.10**	0.04
Neighborhood structural disadvantage index x perceived neighborhood safety							0.02	0.03

^*p<0.05

^**p≤0.01

^***p≤0.001

SE=standard error

^aThe highest educational level for either parent was used as the value for the variable as reported by the adolescent.

^bThe highest occupation of either parent was used as the value for the variable as reported by the adolescent.

^cOther professional – community/social services, education/training/library, and arts/design/entertainment/sports/media occupations

^€Multiple imputation was used to deal with missing data

^ⱡDue to rounding, values displayed may be zero

^*All analyses were conducted on transformed data

Appendix B

Figure 2b.

Sensitivity analysis interaction graph and conditional estimates of neighborhood structural disadvantage on the square root of depressive symptoms by levels of the square root of perceived neighborhood social cohesion. Note: Conditional beta (β) estimates were calculated adjusted for covariates and interactions from the table in Appendix A Model 4 and for the square root of perceived neighborhood social cohesion at the median=2.45, median-3IQR=1.82, and median-6IQR=1.19. Values of the square root of adolescent depressive symptom scores are at the median values depicted. Bold SE is standard error. Highlighted graph lines denote significance. Significance values: *p<0.05, **p≤0.01, ***p≤0.001