Comparing two measures of neighborhood quality and internalizing and externalizing behaviors in the Adolescent Brain Cognitive Development Study

Abstract

Purpose:

There is widespread recognition of the importance and complexity of measuring neighborhood contexts within research on child psychopathology. In this study, we assessed the cross-sectional associations between two measures of neighborhood quality and internalizing and externalizing behaviors in preadolescence.

Methods:

Drawing on baseline data from the Adolescent Brain Cognitive Development Study (n = 10,577 preadolescents), we examined two multi-component assessments of neighborhood quality in relation to children’s internalizing and externalizing symptoms: the Area Deprivation Index (ADI), which measures socioeconomic adversity, and the Child Opportunity Index 2.0 (COI), which measures economic, educational, and environmental opportunity. Both measures were categorized into quintiles. We then used mixed-effects linear regression models to examine bivariate and adjusted associations.

Results:

The bivariate associations displayed strong inverse associations between the COI and ADI and externalizing symptoms, with a graded pattern of fewer externalizing behaviors with increasing neighborhood quality. Only the ADI was associated with externalizing behaviors in models adjusted for child and family characteristics. We did not observe a clear association between either measure of neighborhood quality and internalizing behaviors in bivariate or adjusted models.

Conclusion:

Neighborhood quality, as measured by the COI and ADI, was associated with externalizing behaviors in preadolescent children. The association using the ADI persisted after adjustment for family-level characteristics, including financial strain. Our results indicate that different assessments of neighborhood quality display distinct associations with preadolescent behavioral health. Future research is needed to assess the association between neighborhood quality and behavior trajectories and to identify place-based intervention strategies.

INTRODUCTION

There is widespread recognition of the importance and complexity of measuring neighborhood contexts within psychiatric epidemiology [1, 2] and public health more broadly [3], particularly in the field of child and adolescent health [4]. A variety of neighborhood-level indices have been proposed by governmental organizations and academic researchers to characterize the many ways that neighborhoods may affect health outcomes in a single metric, such as the Area Deprivation Index (ADI) and the Child Opportunity Index (COI). A growing number of studies use such indices to evaluate context-based risk for child psychopathology [5, 6]; however, very few studies have compared their performance.

Socioemotional behavior problems among adolescents in the United States, like that of many health disparities, are patterned by both social, demographic and geographic sources of disadvantage [7]. Ecological theories of disease distribution, which considered the nested influences of individuals within their family, community, and broader societal contexts, provide a compelling framework to understand these patterns [8-12]. Within these frameworks, neighborhood environments can be conceptualized as a key mediator through which economic, political, and societal forces coalesce to structure the opportunities for healthy development to which individual families and children have access. Thus, to develop interventions and policies that promote lifelong health equity, it is critical to advance knowledge about the neighborhood-level characteristics associated with adolescent behavioral health.

Prior research on adolescent mental health suggests that characteristics of neighborhood environments are associated with adolescent behavior using both longitudinal and cross-sectional data [13-16]. For example, place-based characteristics such as neighborhood poverty and composite measures of area-based disadvantage (which include a broader range of adversities in addition to poverty, such as receipt of public assistance and/or percentage of single-parent households) have been linked to poorer social, emotional, and behavioral outcomes [17]. For internalizing symptoms, such as fearfulness and social withdrawal, studies have shown inconsistent associations with neighborhood socioeconomic status (SES). In a recent systematic review of 234 studies, approximately half found null associations [18]. However, in studies considering neighborhood social toxicity (defined as negative processes that undermine child development, such as violence and crime), this systematic review noted a more consistent association, with greater social toxicity predicting higher levels of internalizing behaviors, especially among adolescent girls [18]. For externalizing symptoms, characterized by aggression and rule-breaking, these patterns differ. The same systematic review and a recent meta-analysis of 43 studies found that both measures of neighborhood SES and safety were strongly and consistently associated with adolescent externalizing symptoms [18-21].

Within theoretical models of child development, including adolescent behavioral health, the relevant dimensions of neighborhood environments have been conceptualized broadly, including SES, ethnicity-race, mobility-stability, crime, violence, incarceration rates, policing, social cohesion, green space, and community resources [18, 22, 23]. However, despite the existence of nuanced frameworks, most empirical studies in adolescence continue to focus on socioeconomic indicators, with only a few including social process variables (e.g., social cohesion, collective efficacy) [17, 18, 24]. Moreover, it is common for studies to use only a single, or small subset of metrics to characterize neighborhood contexts, rather than attempting to capture the complexity of the wide array of neighborhood features to which adolescents are simultaneously exposed. These methods thus align poorly with our theories, which instead suggest that comprehensive assessments of neighborhood context may facilitate a more informed understanding of the contribution of neighborhoods to health [25].

The Current Study

In this study, we used baseline data from the Adolescent Brain Cognitive Development (ABCD) cohort to describe the association between neighborhood quality and externalizing and internalizing behaviors in a sample of pre-adolescents (ages 9-10), just before the onset of expected rapid developmental change in these behavioral symptoms [26]. Specifically, we examined two frequently-used multicomponent indices of neighborhood quality – the Area Deprivation Index (ADI) and the Childhood Opportunity Index 2.0 (COI) – as measures of neighborhood quality and considered differences in associations between these place-based measures. Conceptually, the ADI and the COI emphasize different domains of neighborhoods. The ADI provides a ranking of neighborhoods by socioeconomic disadvantage and includes 19 indicators related to neighborhood income, education, employment, and housing quality [27]. In comparison, the COI is a more expansive measure of diverse neighborhood features, with 29 indicators organized into the domains of education, health and environment, and social and economic variables [28] (see Table 1 for a comparison).

Table 1:

Comparison of Area Deprivation Index and Child Opportunity Index 2.0 Components

		ADI-only	Included in both the COI and ADI	COI-only
Social & Economic	Economic opportunity	Unemployment rate	Employment rate	Commute duration
	Resource availability	Poverty rate (<138% FPR) Income disparity Motor vehicle access Telephone access Complete plumbing	Poverty rate (<100% FPR) Median family income Single-headed households	Public assistance rate High skill employment
	Housing costs	Median home value Median gross rent Median monthly mortgage Housing crowding	Homeownership rate	Housing vacancy*
Education	Resource availability		Adult educational attainment	School poverty Teacher experience
	Early childhood education (ECE)			ECE centers High quality ECE centers ECE enrollment
	Elementary education			3rd grade math scores 3rd grade reading scores
	Secondary and postsecondary education			AP enrollment College enrollment HS graduation rate
Health & Environment	Resource availability			Health insurance
	Healthy environments			Access to healthy food Access to green space Walkability
	Toxic exposures			Hazardous waste sites Industrial pollutants Ozone concentration Airborne microparticles Extreme heat exposure

The categories used here to group the various indices included in the ADI and COI are roughly based on the subdomains outlined in the COI technical documentation. The only exception is the addition of “Housing costs” to the Social and Economic Domain. All COI components are grouped in an identical manner to their original categorization within the COI technical documentation except for “Housing vacancy,” which was originally grouped into the “Healthy environments” category within the Health and Education Subdomain, but for the purposes of this table was re-organized to fit within the “Housing costs” category within the Social and Economic Domain.

Abbreviations: ADI means Area Deprivation Index; COI means Child Opportunity Index 2.0.

In prior studies, the ADI has been linked with pediatric traumatic injury [29], pediatric cystic fibrosis outcomes [30], and poorer overall survival in children with leukemia [31]. Limited work has investigated the association between the ADI and pediatric mental and behavioral health outcomes. However, one study, also leveraging ABCD data, identified an association with both internalizing and externalizing outcomes in a subset 6,396 pre-adolescents, which persisted even after adjustment for race/ethnicity, parent marital status, and study site [32].

Similarly, a number of studies have examined the COI in relation to children’s physical health and health care service utilization [33-35], but less research has examined this measure in relation to behavioral outcomes. Notably, one recent study failed to find an association between the COI and anxiety and depression during the first year of the COVID-19 pandemic [36]. In this study, the authors leveraged secondary data from a larger randomized control trial for single-session interventions for adolescent depression. Among individuals who screened positive for depression, they identified 2,479 adolescents from across the US with both residential address and baseline self-reported depressive symptoms data available. However, the generalizability of these findings beyond this unique period and the clinical population under study is unknown. The dearth of research investigating the association between composite indices of neighborhood quality and child internalizing and externalizing symptoms highlights the need for the current study.

Based on prior findings and the robust sample size available through ABCD [18, 37], we hypothesized that lower neighborhood quality would be associated with elevated externalizing and internalizing behaviors among pre-adolescents. Further, we hypothesized that the COI, as a more comprehensive measure, would display larger and more precise associations with pre-adolescent behavior relative to the associations observed for the ADI, which is limited to information about neighborhood socioeconomic characteristics. We expected that these associations would hold even with adjustment for a variety of social and demographic covariates, including race/ethnicity, caregiver education, family structure, nativity, income-to-needs ratio, and financial adversity. However, we intentionally did not include family process variables, such as parental monitoring and parental attachment, in our hypothesized models. This decision was made because we view family processes as potential mediators of the relationship between neighborhood quality and child behaviors, given that these processes could be influenced by neighborhood context [38].

METHODS

Design and sample

We conducted a cross-sectional study using the ABCD cohort at baseline. ABCD is an ongoing longitudinal study of brain development and health, incorporating data from adolescents, their parents, and linked external databases [37]. Participants were recruited between the ages of 9 and 10 via stratified probability sampling within schools at 21 different sites across the United States. Recruitment was intentionally designed to approximate the demographic and socioeconomic make-up of the nation [39, 40]. Participants completed their baseline visits between October 1, 2016, and October 31, 2018. We accessed data for the current study via the 2021 ABCD Data Release 4.0 [41]. Our analysis was limited to participants with complete data for parent-report of child behaviors and a census-tract linkage (n = 10,577, 89.1% of the baseline cohort). In Figure S1 we present a flowchart to illustrate exclusion criteria and the demographic characteristics of excluded participants.

Measures

Neighborhood quality.

Neighborhood quality was operationalized using two different multicomponent indices, both collated by combining census tract-level metrics. The ADI includes 19 measures of socioeconomic disadvantage (see Table 1) [27] which are combined to assign a percentile score ranging from 1 to 100, with 100 representing the greatest deprivation. The COI includes 29 indicators, organized into domains of education, health and environment, and social and economic conditions, with the composite score designed to characterize compounding forms of opportunities for healthy development [28]. Once again, a percentile ranking score ranging from 1 to 100 is assigned to each census tract; this time, with 100 representing the greatest opportunity. To allow for comparison of ADI and COI estimates, we rescaled the ADI to match the direction of the COI by subtracting values from 100; accordingly, higher values on both the ADI and COI reflected higher quality neighborhoods. Consistent with prior COI and ADI research [34, 42, 43], we then categorized the continuous ADI and COI scores into ordered quintiles, with the first quintile (“Q1”) representing the lowest quality neighborhoods and the fifth quintile (“Q5”) representing the highest quality neighborhoods.

Externalizing and internalizing behaviors.

The primary study outcomes, externalizing and internalizing behaviors were assessed through parent-report using the Child Behavior Checklist (CBCL) [44]. The CBCL includes 119 items related to adolescent behavior problems, asking caregivers to rank them on a scale from 0 (not at all true) to 2 (very true or often true) [45, 46]. The externalizing scale provides a rating of the child’s symptoms related to hyperactivity, aggression, or non-compliance, whereas the internalizing scale describes the extent to which a child displays symptoms of depression, anxiety, or withdrawal. We used normed CBCL t-scores accounting for differences in age and sex as the primary outcome measures in all models. As sensitivity analyses, we explored two additional outcomes: (1) clinical diagnostic outcomes based on parent-report, defined as scores at or above the 90^th percentile (t-scores >= 64); and (2) normed t-scores from teacher-reports of adolescent externalizing and internalizing behaviors (available for 32.5% of participants, n = 3,434), measured via the 63-item Behavior Rating Inventory of Executive Function-Teacher Form (BRIEF-T) [47].

Social and demographic covariates.

We selected demographic variables reported by primary caregivers in a baseline questionnaire to use as covariates, consistent with prior research [26]. These include participant race/ethnicity, caregiver education, family structure, nativity, income-to-needs ratio, and financial adversity. Each one was operationalized in a manner consistent with prior ABCD studies [26, 48, 49].

Each child’s racial/ethnic group was reported by a parent or caregiver and children were categorized as White, Black, Hispanic, Asian, Native American/Alaskan Native, Multiracial, or Other. Participants whose parents indicated they were ethnically Hispanic were counted in the Hispanic category, regardless of other racial identifiers. Caregiver education was defined as the highest level of education attained by either the primary caregiver or their partner. Family structure was coded as a binary variable, differentiating between primary caregivers who were married or living with a partner versus those who were single. Family nativity status was also coded as a binary variable, categorizing each participant either as belonging to an entirely US-born nuclear family (participant, primary caregiver, and secondary caregiver all US-born) or belonging to a nuclear family with one or more foreign-born individuals. Income-to-needs ratio was calculated using federal poverty guidelines and caregiver-reported household size (following previously reported methods [49]). A measure of financial strain was calculated as a sum score of seven items on the Parent-Reported Financial Adversity Questionnaire [26], which asked questions about food insecurity, housing insecurity, and limited ability to afford essential goods and services (e.g., phone service, utilities, and medical care). This sum score was then coded as a binary variable, categorizing each participant as either endorsing no financial strain or endorsing one or more challenges.

Statistical analyses

First, we calculated the mean externalizing and internalizing outcomes and sociodemographic characteristics for the sample (1) overall and (2) stratified by ADI and COI quintiles. To better understand the resulting descriptive statistics, we also examined the relationships between the COI and ADI by looking at the correlation between the continuous neighborhood quality scores generated by the two measures, as well as a crosstab of how this translated to the assigned neighborhood quintiles used in subsequent analyses.

Next, to evaluate the association between neighborhood quality and pre-adolescent behaviors, we fit mixed-effects linear regression models, using the package lme4 version 1.127.1 in R. This approach accounted for nesting of participants by both family and recruitment site. We estimated bivariate associations and two sets of regression models for each combination of a neighborhood quality indicator and behavioral outcome: (1) partially adjusted models with basic sociodemographic covariates (race/ethnicity, caregiver education, family structure, and nativity); and (2) fully adjusted models using all covariates (i.e., covariates in Model 1 plus family-level economic variables, including income-to-needs ratio and financial adversity).

Finally, we conducted three sensitivity analyses to test the robustness of our results. First, we used mixed-effect logistic regression to estimate odds ratios for clinical thresholds of the parent-reported behavioral outcomes. Second, we modeled an interaction term to test for any sex differences in our findings, given that other studies have noted gendered effects of neighborhood environments on adolescent mental health [26]. Third, we re-estimated all models using teacher-reported externalizing and internalizing behaviors for the subset of participants for whom this additional measure of behavior was available, to evaluate potential bias, as parent- and teacher- reports of child behaviors both may be biased in different ways and do not always agree [50-52].

Missing Data.

There was missingness for some of the covariates used in the models (caregiver education, family structure, nativity, income-to-needs ratio, and financial strain) (Table 2). To account for this, we used the package mice (version 3.14.0) to run multiple imputation (n = 25). Following prior studies using multiple imputation with ABCD baseline data [26], all of the variables included in the final mixed-effects models were also included in the imputation model, as were the additional variables of youth- and parent-perceived neighborhood safety, total behavior problems, age, and sex. Analyses were conducted in R version 4.1.2 by the first author (LB) and independently reviewed by the second author (RK), to optimize code quality and reproducibility [53].

Table 2:

Sample characteristics for the overall sample and stratified by Area Deprivation Index quintiles (n=10,577 preadolescents in the Adolescent Brain Cognitive Development Study)

	Full Sample	Area Deprivation Index Quintiles
		Quintile 1 n=1306 (12.4%)	Quintile 2 n=1140 (10.8%)	Quintile 3 n=1928 (18.3%)	Quintile 4 n=3518 (33.3%)	Quintile 5 n=2665 (25.2%)
	mean (SD)	mean (SD)	mean (SD)	mean (SD)	mean (SD)	mean (SD)
CBCL internalizing	48.43 (10.6)	49.14 (11.1)	48.76 (10.7)	48.84 (10.9)	48.39 (10.5)	47.71 (10.2)
CBCL externalizing	45.66 (10.3)	48.55 (11.4)	46.41 (10.6)	45.98 (10.3)	45.25 (9.9)	44.33 (9.6)
Age (months)	119.1 (7.49)	118.3 (7.33)	118.8 (7.52)	119.2 (7.48)	119.0 (7.53)	119.5 (7.48)
	n (%)	n (%)	n (%)	n (%)	n (%)	n (%)
Sex
Female	5016 (47.5)	637 (48.8)	552 (48.4)	930 (48.2)	1639 (46.6)	1258 (47.2)
Race/ethnicity
White	5498 (52.1)	262 (20.1)	441 (38.9)	970 (50.3)	2212 (64.9)	1613 (60.5)
Black	1477 (14.0)	637 (48.8)	291 (25.5)	250 (13.0)	193 (5.5)	106 (4.0)
Hispanic	2138 (20.3)	221 (16.9)	264 (23.2)	461 (23.9)	727 (20.7)	465 (17.4)
Asian	233 (2.2)	5 (0.3)	5 (0.4)	20 (1.0)	55 (1.6)	148 (5.6)
Native American	35 (0.3)	13 (1.0)	8 (0.7)	7 (0.3)	7 (0.2)	0 (0.0)
Multiracial	942 (9.0)	128 (9.8)	106 (9.3)	182 (9.4)	259 (7.4)	267 (10.0)
Other	60 (0.6)	12 (0.9)	5 (0.4)	16 (0.8)	12 (0.3)	15 (0.6)
Missing	174 (1.6)	28 (2.1)	20 (1.8)	22 (1.1)	53 (1.5)	51 (1.9)
Income-to-needs ratio
<100% FPL	1355 (12.8)	465 (35.6)	230 (20.2)	262 (13.6)	292 (8.3)	106 (4.0)
100-199% FPL	1562 (14.8)	330 (25.3)	277 (24.3)	375 (19.5)	410 (11.7)	170 (6.4)
200-399% FPL	2407 (22.8)	224 (17.2)	315 (27.6)	560 (29.0)	916 (26.0)	392 (14.7)
400-599% FPL	1560 (14.8)	38 (2.9)	92 (8.1)	254 (13.2)	641 (18.2)	532 (20.1)
600%+ FPL	2649 (27.8)	29 (2.2)	106 (9.3)	276 (14.3)	988 (28.1)	1250 (46.9)
Missing	1024 (9.7)	220 (16.8)	120 (10.5)	201 (10.4)	271 (7.7)	212 (8.0)
Financial adversity
0	8259 (79.0)	677 (51.8)	743 (65.2)	1462 (75.8)	2965 (84.3)	2412 (90.5)
1+	2193 (21.0)	606 (46.4)	381 (33.4)	451 (23.4)	520 (14.8)	235 (8.8)
Missing	105 (1.0)	23 (1.8)	16 (1.4)	15 (0.8)	33 (0.9)	18 (0.7)
Caregiver education
High school or less	1384 (13.1)	414 (31.7)	238 (20.9)	269 (14.0)	341 (9.7)	122 (4.6)
Some college	2668 (25.3)	575 (44.0)	422 (37.0)	620 (32.2)	733 (20.8)	318 (11.9)
College	2693 (25.5)	158 (12.1)	240 (21.1)	528 (27.4)	995 (28.3)	772 (29.0)
Masters/doctorate	3640 (34.5)	108 (8.3)	215 (18.9)	480 (24.9)	1405 (39.9)	1432 (53.7)
Missing	172 (1.6)	51 (3.9)	25 (2.2)	31 (1.6)	44 (1.3)	21 (7.9)
Family Structure
Married	7805 (73.9)	621 (47.5)	687 (60.3)	1377 (71.4)	2862 (81.4)	2258 (84.7)
Single	2673 (25.3)	658 (50.5)	438 (38.4)	535 (27.7)	640 (18.2)	402 (15.1)
Missing	79 (0.7)	27 (2.1)	15 (1.3)	16 (0.8)	16 (0.5)	5 (0.2)
Nativity
Foreign born	3472 (32.9)	241 (18.5)	300 (26.3)	589 (30.5)	1186 (33.7)	1156 (43.4)
US born	7079 (67.1)	1061 (81.2)	840 (73.7)	1338 (69.4)	2331 (66.3)	1509 (56.6)
Missing	6 (0.1)	4 (0.3)	0 (0.0)	1 (0.1)	1 (0.0)	0 (0.0)

RESULTS

Sample characteristics

In this study of 10,577 pre-adolescents, the mean age was 9.9 years (SD = 7.49 months) and nearly half (47.5%) of participants were female. The majority (53.0%) of the sample was White, 20.6% was Hispanic, 14.2% was Black, 9.1% was Multiracial, 2.2% was Asian, 0.3% was Native American/Alaska Native, and 0.6% was categorized as Other or unknown race/ethnicity. The sample was skewed towards pre-adolescents residing in advantaged neighborhoods, according to both the ADI and COI.

Table 2 presents a complete summary of demographic characteristics for the sample overall and stratified by ADI quintile, and includes p-values for ANOVA and chi-square tests of significance differences across groups. In general, Black adolescents and adolescents belonging to families with a low income-to-needs ratio, documented financial adversity, lower levels of caregiver education, and a single primary caregiver were overrepresented in the lowest quality neighborhoods; whereas White adolescents and adolescents belonging to families with a high income-to-needs ratio, no documented financial adversity, higher levels of caregiver education, and married parents were overrepresented in the highest quality neighborhoods. In this sample, nuclear families with at least one foreign-born individual were more likely to be in higher quality neighborhoods than entirely US-born families. Similar patterns emerged when the data was stratified by COI quintile, with the only notable difference being a more even distribution of foreign-born and US-born families across neighborhood quality quintiles. This data is included in Table S1.

Using a simple correlation analysis, we find that ADI and COI continuous measures were highly correlated (R = 0.76, see Figure 1) [54]. Furthermore, when condensed into quintiles, the two measures produced highly overlapping categorizations of neighborhood quality (see Figure S2). For example, over 80% of participants in ADI Q4 and Q5 were categorized within COI Q4 and Q5 and vice versa.

Figure 1:

Correlation between Area Deprivation Index (ADI) and Child Opportunity Index (COI) scores for children in the Adolescent Behavior Cognitive Development Study (n=10,577)

Abbreviations: ADI = Area Deprivation Index; COI = Child Opportunity Index 2.0

Neighborhood quality and externalizing behaviors

Bivariate models looking at the association between neighborhood quality and preadolescent externalizing behaviors displayed strong inverse associations between both ADI and COI quintiles and externalizing symptoms. As shown in Figure 2A and 2B, stepwise patterns emerged, such that pre-adolescents in more advantaged neighborhoods exhibited fewer externalizing behaviors relative to those in lower quality neighborhoods. The estimated associations between externalizing behaviors and ADI neighborhood quintiles were larger than for COI neighborhood quintiles. For example, for the ADI, preadolescents living in the highest quality neighborhoods had externalizing symptoms that were 3.98 points lower (95% CI: 4.80 – 3.16) relative to preadolescents living in the lowest quality neighborhoods (as compared to an estimated differences of 2.57 points (95% CI: − 3.23 to − 1.91) for an identical comparison using the COI).

Figure 2:

Bivariate Associations Between Neighborhood Quality and Internalizing and Externalizing Behaviors

In Model 1 (Table 3B), which adjusts for basic child and family demographic characteristics, higher neighborhood quality as measured by the ADI retained a graded association. This association between ADI neighborhood quality and externalizing symptoms was robust to further model adjustment for family income and financial strain (Table 3B ADI Model 2). Of note, however, the addition of these family-level economic covariates in Model 2 eliminated the stepwise pattern of reduced externalizing symptoms with each quintile improvement in neighborhood quality. Instead, residing anywhere outside of the lowest quality neighborhoods (Q1) was associated with a similar reduction in externalizing behavior problems.

Table 3:

Association between Area Deprivation Index and Child Opportunity Index 2.0 Quintiles and Parent-Reported Adolescent Internalizing/Externalizing Behaviors

A. Internalizing Symptoms
	ADI		COI
	Model 1	Model 2	Model 1	Model 2
Quintile	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)
Q1	Ref.	Ref.	Ref.	Ref.
Q2	−0.87 (−1.77, 0.03)	−0.52 (−1.41, 0.37)	0.39 (−0.42, 1.21)	0.60 (−0.20, 1.40)
Q3	−1.03 (−1.90, −0.16)	−0.36 (−1.22, 0.50)	−0.08 (−0.90, 0.74)	0.31 (−0.51, 1.12)
Q4	−1.40 (−2.27, −0.53)	−0.55 (−1.42, 0.33)	−0.22 (−1.01, 0.58)	0.43 (−0.37, 1.22)
Q5	−1.66 (−2.64, −0.68)	−0.66 (−1.64, 0.33)	−0.25 (−1.07, 0.58)	0.55 (−0.28, 1.37)
B. Externalizing Symptoms
Quintile	Model 1	Model 2	Model 1	Model 2
	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)
Q1	Ref.	Ref.	Ref.	Ref.
Q2	−1.71 (−2.57, −0.84)	−1.28 (−2.14, −0.42)	−0.51 (−1.29, 0.28)	−0.23 (−1.00, 0.54)
Q3	−1.83 (−2.67, −1.00)	−1.10 (−1.92, −0.27)	−0.82 (−1.61, −0.02)	−0.33 (−1.12, 0.45)
Q4	−2.04 (−2.88, −1.20)	−1.11 (−1.94, −0.27)	−0.96 (−1.73, −0.19)	−0.21 (−0.97, 0.56)
Q5	−2.54 (−3.48, −1.60)	−1.43 (−2.37, −0.49)	−1.20 (−1.99, −0.41)	−0.28 (−1.07, 0.51)

Model 1 includes the covariates race/ethnicity, family structure, nativity, and caregiver education.

Model 2 includes the variables in Model 1 plus income-to-needs ratio and financial adversity.

See Table 1 for details on the operationalization of each of these covariates. Consistent with prior studies in ABCD, financial strain was calculated as a sum score of seven items on a Parent-Reported Financial Adversity Questionnaire, and coded as a binary variable, categorizing each participant as either endorsing no financial strain or endorsing one or more challenges.

Abbreviations: ADI = Area Deprivation Index; COI = Child Opportunity Index 2.0

Both ADI and COI quintiles are ordered such that the first quintile (“Q1”) represents the lowest quality neighborhoods and the fifth quintile (“Q5”) representing the highest quality neighborhoods.

In models evaluating neighborhood quality as measured via the COI, partially adjusted models also showed an association with externalizing behaviors (see Table 3B COI Model 1). The magnitude of the effects for each quintile were smaller relative to the bivariate models; however, a graded reduction in symptoms persisted in the highest quality neighborhoods (COI Q4 and Q5). In contrast to ADI models, the COI was not associated with externalizing problems in the fully adjusted model (see Table 3B COI Model 2).

Neighborhood quality and internalizing behaviors

No clear bivariate relationship was found between neighborhood quality and internalizing behaviors, using either the ADI or COI as the measure of neighborhood quality (see Figure 2C and 2D). For the ADI, there was a significant difference in internalizing scores when comparing Quartile 1 to Quartile 5, only. Internalizing scores were roughly equivalent for children living in neighborhoods in Quartiles 1 through 3; there was a slight decrease for children living in neighborhoods in Quartile 4; and then a larger decrease for children living in neighborhoods in Quartile 5. This pattern was augmented after adjusting for basic child and family demographic characteristics (i.e., child race/ethnicity, caregiver education, family structure, and nativity), with fewer internalizing symptoms seen among children in highest quality neighborhoods (Q3-5) relative to those in the lowest quality neighborhoods (Q1 and Q2) (see Table 3A ADI Model 1). After additional adjustment for family-level economic variables including income-to-needs ratio and financial adversity, however, this association was attenuated (see Table 3A ADI Model 2). Also consistent with the bivariate analyses, identical models using the COI showed no association with internalizing symptoms in the partial or fully adjusted models (see Table 3A COI Models 1 and 2).

Results of sensitivity analyses

We conducted three sets of sensitivity analyses to examine the robustness of our results. First, we examined the exposures of interest in relation to clinical thresholds for internalizing and externalizing symptoms. As shown in Table S3 the associations were null for internalizing symptoms using both the COI and the ADI. For externalizing symptoms, associations were also null for the COI. On the ADI, the odds of having clinically significant behavioral problems were 0.65 (95% CI: 0.46, 0.92), 0.67 (95% CI: 0.50, 0.91), and 0.69 (95% CI: 0.50, 0.94) times lower for preadolescents in quartiles Q5, Q4, and Q2 of the ADI, respectively, relative to those in quartile Q1 in partially adjusted models (Model 1). However, these associations were not sustained after accounting for economic variables (Model 2). Second, considering potential effect modification by sex, interaction terms modeled between both ADI and sex and COI and sex were not significant for either internalizing or externalizing symptoms. Finally, in models using teacher-reported behavioral symptoms (n = 3,434; see Table S4 for a comparison to the full sample), results were generally consistent with the primary analysis of parent-reported symptoms using the full sample (see Table S5).

Post-hoc Analysis

To further explore the null findings for internalizing behaviors, we conducted a post-hoc analysis to see if these patterns held for all sub-domains of internalizing behaviors captured on the CBCL (Somatic Complaints, Withdrawn Symptoms, and Anxious/Depressive Symptoms) [55]. While the association between neighborhood quality measured via the ADI and behavior remained null for both somatic and anxious/depressive symptoms, a significant difference in withdrawal symptoms was found between children living in neighborhoods in Quartile 1 relative to all other quartiles, such that children in higher quality neighborhoods had fewer withdrawal symptoms. Conversely, no association was found between the COI and either withdrawn or anxious/depressive symptoms, but children in higher quality neighborhoods had worse somatic symptoms than children in the lowest quality neighborhoods in this sample. All results are presented in Table S2.

DISCUSSION

In this large, cross-sectional study of pre-adolescents, we examined associations between two multicomponent indices of neighborhood quality, the ADI and the COI, and adolescent behavioral health measured via parent report. In bivariate and partially adjusted models accounting for race/ethnicity, caregiver education, family structure, and nativity, both the ADI and the COI were associated with externalizing symptoms in a stepwise pattern. Contrary to our initial hypothesis, however, in fully adjusted models that additionally incorporated family economic strain, this association was completely attenuated in models using the COI (i.e., the more comprehensive measure of neighborhood quality) and the stepwise pattern was eliminated in models that used the ADI (i.e., a more limited set of neighborhood-level socioeconomic characteristics), such that higher levels of externalizing symptoms were only evident among children residing in the lowest quality neighborhoods. Neither index showed robust associations with preadolescent internalizing symptoms. These results were not moderated by sex and were reproduced with teacher-report data in a subset of the sample.

The associations we report between neighborhood quality and adolescent socioemotional behavior problems are consistent with the existing literature. Specifically, the lack of an association between either the ADI or the COI with internalizing symptoms in adjusted models is consistent with prior studies that do not find a clear association between neighborhood socioeconomic characteristics and fearfulness or social withdrawal [18]. Likewise, the observed associations between the ADI and the COI and externalizing behaviors, mirrors a large body of research that has consistently reaffirmed a link between rule-breaking and aggressive behaviors and a wide range of neighborhood quality and safety measures [18]. Our study complements and extends on this work using a current and diverse national sample and well-validated psychiatric assessments conducted via both teacher- and parent-report, and by documenting the sensitivity of associations to composite indices of neighborhood quality.

Consistent with current theories of child development [25], we had hypothesized that the COI, which incorporates wide-ranging tract-level measures of education, health and environment, and social and economic conditions, would be a more robust predictor of adolescent behavior than the ADI, which is limited to information about neighborhood socioeconomic characteristics. While the measures generally displayed similar associations for internalizing problems, for externalizing behaviors, neighborhood quality measured via the ADI displayed a more robust association in the fully adjusted model relative to that observed for models using the COI. The ADI and the COI were highly correlated, yet the difference in what the instruments measure appears to have been meaningful in this case.

As shown in Table 1, the ADI includes several socioeconomic metrics that the COI does not (e.g., unemployment rate, income disparity, motor vehicle access, telephone access, complete plumbing, median home value, median monthly mortgage/rent costs, and housing crowding). On the flipside, the COI incorporates variables reflecting environmental health, healthcare access, and education that are absent from the ADI. Our finding of stronger associations between the ADI and externalizing symptoms compared to the COI suggests that externalizing problems may display the strongest associations with socioeconomic characteristics of a neighborhood in the preadolescent period. The combined results from across the multicomponent indices also suggest two other complimentary possibilities: first, that variables unique to the ADI (such as income disparity or crowding) may have a strong association with preadolescent externalizing behaviors. Second, it is possible that some of the variables included in the COI may be extraneous to our outcome of interest. Each of these hypotheses is based on the idea that particular elements in a multicomponent index may augment or dilute the measured association between neighborhood quality and health. Different indices thus capture the impact of neighborhoods more meaningfully for different outcomes, pointing to the importance of careful measure selection and potentially to the design of more impactful interventions. For example, the COI has previously been linked to disparities in pediatric hospitalizations, child mortality, and to adolescent cardiometabolic risks [33, 35, 56], and has been shown to be a stronger predictor of diabetes mellitus mortality, physical pain, and positive emotions in adults relative to the ADI [57]. However, in the case of pre-adolescent externalizing symptoms, our results indicate that the COI is not strongly associated with these behavioral outcomes at age 9–10. Instead, the ADI, with its narrower focus on neighborhood socioeconomic conditions, is a stronger predictor. Further research is needed to understand whether this finding extends to other behavioral and physical health outcomes and to the outcomes we have studied using a longitudinal research design.

There are several limitations to our analyses. First, despite the large size of the ABCD baseline cohort and the careful efforts to recruit a diverse set of children and families, the sample is not representative of the United States population [40]. Notably, the participants included in this study had parents with higher educational attainment relative to national estimates [58], which may have implications for the generalizability of our findings. Second, due to the cross-sectional nature of our analysis, we were only able to estimate the associations between each measure of neighborhood quality and behavior symptoms at a single point in time during preadolescence. This design introduces a risk of residual confounding and limits our ability to infer causality, to study temporality between our exposures and outcomes, and to examine behavior trajectories over time.

In addition, given ABCD Study policies designed to protect participant privacy, it was not possible to nest mixed- effects models by census tract. To account for this, ABCD site location was used instead in nested models, as has been described in prior studies with this data [19]. As a result of this limitation, the models may underestimate the width of the standard errors for regression effect estimates. Related, the metrics for neighborhood quality used, which rely on information pooled at the census-tract level, may not perfectly map onto how neighborhood environments are experienced by individual children and their families. Although census-tract measures are widely used and considered a meaningful degree of spatial resolution in the field, they are an imperfect tool [59, 60]. We also lacked access to information on residential mobility, a potential confounder that may be associated with both neighborhood quality and child behavior [61]. Future studies with the ability to consider recent child moves would strengthen the evidence presented here.

There was missing data across many of the covariates included in our models. We addressed this via multiple imputation to minimize any potential bias introduced. Multiple imputation techniques rely on the assumption that data is missing at random. In this analysis, as is the case in many social epidemiologic studies, it is likely that variables such as income are not missing at random, such that families with low incomes are more likely to choose not to respond to financial questions than families with higher incomes. However, given that there was low missingness overall, and given that we had multiple covariates available in our imputation models to triangulate family socioeconomic position, multiple imputation remained a reasonable choice to address missingness in the sample, consistent with what is commonly done in ABCD [26].

This research can be extended in several important ways. First, this cross-sectional analysis provides a launching point for future longitudinal studies. Particularly because adolescence is a critical period in behavioral development, it is essential to consider the associations between neighborhood contexts and both (1) the starting points at which children enter pre-adolescence (as we have done here), as well as (2) trajectories in behavioral symptoms. Longitudinal modeling thus provides an important opportunity to further this work. For example, future work could investigate whether the disparities in externalizing behaviors for individuals living in the lowest quality neighborhoods persist throughout adolescence, or if these gaps narrow or widen with time.

Additionally, when considering these findings in the context of an eco-social theory of disease distribution [10], more research is needed into both the upstream factors that shape neighborhood contexts, as well as the downstream mechanisms through which neighborhood environments influence child development. Upstream, research is needed to demonstrate the impact of historical and contemporary policies and political geographies on neighborhood resource distribution. Downstream, much work remains to be done to examine potential causal mechanisms through which neighborhood level adversity is biologically embedded. This is an active area of investigation. Several recent studies have identified associations between neighborhood environments and neurologic, immune, metabolic, and microbiologic biomarkers, which may all play a role in the embodiment of structural adversity [5, 6, 62, 63]. In addition, links have been established between neighborhoods and family process variables, such as parental attachment and parental monitoring, which in turn have been associated with child behaviors [38, 64, 65]. Finally, in future research it will be important to consider temporality, including both length of time each participant is exposed to their neighborhood environment and developmental periods across the life course.

In conclusion, our findings have several important implications for future work. First, they highlight the utility of using multicomponent indices to investigate the interconnected features of neighborhoods in relation to preadolescent behaviors. Second, by documenting cross-sectional associations in pre-adolescents, results from this study help build a foundation for future research into the longitudinal influences of neighborhoods on health and development using causally-informed designs, and attention to mediating mechanisms. This line of research holds the potential to inform the development of multilevel interventions to reduce disparities in externalizing behaviors in adolescents and their adult consequences.