Racial Inequalities in Adolescents’ Exposure to Racial and Socioeconomic Segregation, Collective Efficacy, and Violence

Abstract

In the United States, Black youth tend to grow up in remarkably less resourced neighborhoods than white youth. This study investigates whether and to what extent Black youth are moreover exposed to less resourced activity spaces beyond the home. We draw on GPS data from a large sample of urban youth in the Columbus, OH-based Adolescent Health and Development in Context study (2014-16) to examine to what extent Black youth experience non-trivial, disproportionate levels of exposure to more disadvantaged and segregated contexts in their daily routines compared with similarly residentially situated white youth. Specifically, we estimate Black-white differences in non-home exposure to concentrated disadvantage, racial segregation, collective efficacy, and violent crime. We find that Black youths’ activity spaces have substantially higher rates of racial segregation and violent crime, and substantially lower levels of collective efficacy compared to white youth, even after accounting for a host of individual- and home neighborhood-level characteristics. We find more modest evidence of differences in exposure to socioeconomic disadvantage. These findings have important implications for neighborhood-centered interventions focused on youth well-being and the contextual effects and segregation literatures more generally.

Despite declines in racial segregation, Black youth remain disproportionately exposed to socioeconomic disadvantage in their neighborhoods, with significant implications for inequalities in well-being (Reardon and Bischoff 2011; Sharkey and Faber 2014). Indeed, highly disadvantaged neighborhoods tend to be further characterized by heightened levels of health-relevant risk factors such as violent crime, as well as a reduced prevalence of protective factors such as collective efficacy, all of which influence youth well-being (Sampson 2012; Sharkey 2018). An emerging literature additionally calls attention to potential racial inequalities in youths’ activity spaces beyond the neighborhood, with recent research underscoring that youth spend little time in the neighborhood outside their home (Zenk et al. 2019; Browning, Calder, et al. 2021). Mounting evidence suggests that inequalities in activity space exposures have significant consequences for population health (Cagney et al. 2020), but little research has examined whether Black and white youth navigate racially segregated activity spaces. This omission is particularly important in light of the mixed success of residential mobility interventions focused on Black adolescents’ development and well-being (Chetty, Hendren, and Katz 2016). To the extent that these youth are disproportionately drawn to more disadvantaged activity spaces beyond the neighborhood, residential neighborhood-centered interventions may be excessively optimistic regarding returns for Black adolescents (Clampet-Lundquist et al. 2011; Graif 2015).

This study examines Black-white differences in the composition of activity spaces beyond the home using data from the Adolescent Health and Development in Context study (AHDC); a representative study of 1,405 youth aged 11 to 17 and their caregivers in Franklin County, OH. Five days of smartphone-based GPS data corroborated using a recall-aided space-time budget methodology were used to construct individual-level measures of non-home exposure compositions based on census block group aggregations. We focus specially on exposure to concentrated disadvantage, racial segregation, collective efficacy, and violent crime in light of the major relevance of these processes to youth well-being (Sharkey and Faber 2014).

BACKGROUND

Mirroring pervasive patterns of racial residential segregation in the United States, a growing literature on the demography of everyday mobility points to potential racial inequalities in activity spaces. Some of this research aligns with expectations of the prominent “geographic” or “social isolation” perspective on urban segregation, anticipating that activity space compositions largely reflect compositions of residents’ neighborhoods (Wilson 1987; Wang et al. 2018). Evidence is growing for a more dynamic “compelled mobility” perspective however, demonstrating that urban residents—and particularly those residing in disadvantaged neighborhoods—experience far more heterogeneity in exposure to neighborhood characteristics than would be predicted by the social isolation approach. Although urban exposures are likely more complex and heterogeneous than previously acknowledged, mobility dynamics are nevertheless expected to be characterized by substantial disparity in everyday exposures by race, even for Black and white youth residing in similar neighborhoods.

A variety of factors are expected to account for these heterogeneous but racially disparate patterns of exposure, including those that pull individuals out of their neighborhoods toward settings with resources and network ties, as well as those that push them away from some areas such as a reduced sense of safety. Pull factors are rooted in the segregation of people and resources characterizing most contemporary U.S. cities. On the one hand, activity locations are increasingly clustered within non-residential areas of cities, requiring residents of both disadvantaged and advantaged neighborhoods to leave their neighborhood to access resources (Frumkin 2002; Tana, Kwan, and Chai 2016). The outdated conception of neighborhoods as the center of urban life is likely least relevant for residents of disadvantaged and high proportion Black neighborhoods, however (Browning, Calder, et al. 2021). Such neighborhoods tend to contain fewer health-related organizations (Freeman Anderson 2017), nonprofits (Crubaugh 2021), schools (Owens 2020), and businesses (Small and McDermott 2006; Small et al. 2021), forcing residents of disadvantaged neighborhoods, in particular, to travel beyond the neighborhood—often to more advantaged areas—to access these resources. Yet, racial segregation in network ties and institutional affiliations likely contributes to racial segregation in activity space by driving youth to same-race dominated areas of cities, regardless of home residence (Small 2007; Small and Feldman 2012; Krysan and Crowder 2017). Indeed, research has long acknowledged the tendency toward racial homophily in interpersonal network ties (McPherson, Smith-Lovin, and Cook 2001; Small and Adler 2019). Krivo et al. (2013), for example, find that adult Black and Latino residents of Los Angeles tend to have routine activities in more disadvantaged census tracts relative to comparable white residents.

These pull factors suggest that while Black and white youth are exposed to more heterogenous environments than previously thought, there will still be Black-white differences in activity space exposures even for youth residing in the same neighborhood. For instance, a Black youth residing in a disadvantaged neighborhood would be expected to spend a non-trivial amount of time in more advantaged areas as they seek organizational resources. However, a white youth residing in the same neighborhood likely spend more time in these advantaged contexts based on the pull of both organizational resources and network ties. Similarly, white youth residing in more advantaged contexts may satisfy most organizational and social needs in comparably advantaged neighborhoods while Black youth living in advantaged neighborhoods would experience the pull of social ties and some institutional linkages in more disadvantaged neighborhoods.

A compelled mobility perspective further acknowledges that mobility patterns are a product not just of resource-seeking and network ties, but also of push factors that lead youth to avoid certain environments. For example, recent residential segregation research suggests that housing selection results from constrained access to information and perceptions of which neighborhoods may be less welcoming (Krysan and Crowder 2017). Networks and residential histories contribute to residents’ heuristics of place, shaping where movers seek to relocate net of influences of socioeconomic resources and overt discrimination. Crucially, this research underscores the contribution of tacit non-exclusionary discrimination ¹ and the anticipation of discrimination against minority home-seekers in shaping residential consideration sets (Krysan and Crowder 2017). The significance of anticipated discrimination likely extends to the activity patterns of minority youth, pushing them and their parents away from—or limiting their time in—some affluent areas in the course of their daily lives. For instance, anti-Black hate crimes are most numerous in low-proportion minority neighborhoods with high rates of informal social control (Lyons 2007), and Black urbanites are keenly aware of the heightened scrutiny and mistrust they are likely to encounter in affluent areas (Feagin 1991; Lee 2000; Krysan and Farley 2002; Anderson 2015). Black youth are moreover frequently overpoliced by law enforcement and residents within more affluent communities (Plant and Peruche 2005; Feagin 2010; Anderson 2015), with widespread consequences for their well-being (Geller et al. 2014; Sewell, Jefferson, and Lee 2016; Young 2018; DeAngelis 2021).

Some evidence that these factors lead to segregation in youths’ activity spaces comes from the Moving to Opportunity residential housing experiment (MTO), which randomized Black and Hispanic residents of high-poverty neighborhoods into treatment and control groups, the former of which involved relocation to lower poverty neighborhoods (Briggs, Popkin, and Goering 2010). Though beneficial effects were found for participants who moved at younger ages, adolescent participants experienced slightly negative effects early on and later in adulthood (Chetty et al. 2016; Schmidt, Krohn, and Osypuk 2018). An important insight has been that treatment group adolescents frequently faced adversities in their advantaged neighborhoods and schools, often leading to more time spend in (and relocation back to) disadvantaged neighborhoods (Briggs et al. 2008; Sampson 2008; Clampet-Lundquist et al. 2011). Treatment group males were particularly likely to struggle, reporting a heightened sense of scrutiny in their residential neighborhood in addition to continued reliance on social ties to more disadvantaged neighborhoods (Zuberi 2012; Boyd and Clampet-Lundquist 2019). These observations suggest that, even when residing in more advantaged neighborhoods, Black youth may attempt to avoid these settings, leading to disproportionate exposure to more disadvantaged activity spaces compared to similar white youth.

The Present Study

This study examines Black-white differences in exposure to disadvantaged activity space contexts. We hypothesize that, compared to white youth, Black youth will be exposed to higher levels of concentrated disadvantage and proportion Black residents in their activity spaces net of differences in home neighborhood conditions. We further expect Black youth will be disproportionately exposed to adverse health-related risk factors that typically cluster within more disadvantaged areas, focusing on lower collective efficacy and higher block group-level rates of violent crime (Sampson, Raudenbush, and Earls 1997). This focus is consistent with research highlighting the relevance of these measures to delinquency and victimization (Wikström et al. 2012) and physical and mental health (Ahern and Galea 2011) among youth.

We additionally consider whether inequalities in non-home exposures between Black and white adolescents vary by age, biological sex, levels of concentrated disadvantage in one’s neighborhood, or having recently moved addresses. We do so in light of evidence indicating that parents often allow older adolescent males more leeway to traverse neighborhoods, whereas younger and female youth are often more restricted in their independence, and in light of gender and age dependent findings from residential mobility interventions (Spilsbury 2005; Graif 2015). We assess racial disparities by neighborhood disadvantage and having recently moved to ensure that our findings generalize among residents of low-disadvantage areas and those more established in their neighborhood.

Data and Measures

The Adolescent Health and Development in Context (AHDC) study is a longitudinal data collection effort focused on the consequences of everyday contexts for health and well-being. The study was conducted in an urban and suburban area within Interstate 270—the Franklin County outer belt including the majority of the City of Columbus and numerous inner suburbs. Wave 1 of the AHDC is a representative sample of study area permanent residences with youth ages 11 to 17 and an English-speaking caregiver collected between 2014-2016. The sampling frame was based on a combination of a vendor-provided list of potentially eligible households and data from public school districts representing households in the study area. The AAPOR Response Rate 3, or the proportion of contacted households estimated to be eligible for inclusion that completed interviews, is 21.3%.² The AHDC sample is approximately representative of the population of youth in the study area with respect to race and household income (Boettner, Browning, and Calder 2019; Browning, Calder, et al. 2021). The Columbus, OH area is roughly average on key indicators of racial composition and segregation when compared with large metropolitan areas in the United States. In a recent analysis of 51 major U.S. metro areas, Columbus had a dissimilarity index score (62.2) comparable to the overall average (59) and Black prevalence (14.9%) nearly equivalent to the large metro area mean (15%) (Frey 2018). For more information on the AHDC sampling design and study area, see Boettner, Browning, and Calder (2019) and Browning et al. (2021).³

Data were collected in weeklong periods with day of study entry varying across respondents. First, an Entrance Survey was administered to both caregivers and a focal youth covering demographic and socioeconomic background, household composition, family structure and marital status, employment and income, health, social support, behavior, mental/physical health, schooling, family conflict, and legal troubles. The questionnaires include separate modules on the geographic coordinates of places to which the caregiver and youth are regularly exposed (e.g., school, friends’ houses). The Entrance Survey was followed by a seven-day smartphone-based Geographically Explicit Ecological Momentary Assessment (GEMA; Kirchner and Shiffman 2016) period for the youth, combining GPS tracking and ecological momentary assessment to examine youth perceptions, behaviors, and activity space locations across the study week. During the in-home interview, the interviewer provided a GPS-enabled smartphone to the youth with instructions to carry the phone continuously for the 7-day period. The GPS feature of the study facilitated collection of in-the-moment data on locations at which youth spend time through continuous tracking (with the exception of in-school hours). The phone app prioritizes spatial data from more accurate GPS satellites, logging location data every 30 seconds when connected. If no GPS satellite position has been saved in the last 10 minutes, location coordinates based on cell tower network position are collected every 60 seconds. If location services were turned off, the study application sent a prompt to remind the participant to turn services back on. The GPS data were uploaded to secure servers every hour.

At the end of the 7-day period, the interviewer returned to the youth’s home for a follow-up Exit Survey, during which the adolescent participant completed a recall-aided interactive space-time budget covering Friday, Saturday, Sunday, and the two most recent weekdays. Consistent with the broader time use literature, this approach aims to adequately capture mobility occurring both during the school week and on the weekend when youth may have more spatial autonomy (Hofferth 2009). Prior to administering the space-time budget, the GPS data are processed using a convex hull-based binning algorithm that summarizes data points into stationary and travel periods. The space-time budget application takes the output of the convex hull processing of the raw GPS data and displays estimated locations to the respondent. Each location is combined with labels from nearby routine location self-reports from the Entrance survey along with Google Places search results; the respondent can then report whether each stable location was associated with a routine location, a Google Places result, write in other text, or change the location coordinates as needed for the corresponding 5 days of location data of the GEMA week. Our focus on 5 days of coverage aligns with recent research validating that as little as 1-6 days sufficiently captures between-person variability in activity spaces (Zenk et al. 2018).

Our analyses draw on these GEMA data to measure individual-level non-home activity space compositions. We employ location data from the space-time budget, geocoding coordinates from locations encountered over the 5-day period to census block groups. Home census tract and non-home block group census characteristics are constructed using the American Community Survey 2009-2013 five-year file (Manson et al. 2021). Individual-level “non-home” activity space measures are calculated by aggregating exposure data from the recall-aided space-time budget information provided by the youth over five days of the EMA/GPS week (Boettner et al. 2019). We define “non-home” locations as those the respondent identified as being separate from the home address, or as those at least 30 meters away from the home address in instances where the respondent did not identify a location. We calculate non-home individual-level mean exposure to block group characteristics across all locations within the study area of the I-270 Columbus outer belt boundary for the week, weighted by time spent in minutes at each location during waking hours.⁴ Home neighborhood measures are based on the census tract characteristics for the respondent’s self-reported primary address.^5,6

Activity Space and Neighborhood Compositions

Concentrated disadvantage is a scale averaging together the following block group-level characteristics: poverty rate, unemployment rate, percent female-headed households, and the percent of households receiving cash assistance. These items were similarly combined at the census-tract level to operationalize home neighborhood concentrated disadvantage. Proportion Black is the proportion of block group or census tract residents who are Black.

Collective efficacy is based on caregivers’ reports about their neighborhood and the areas surrounding their routine activity locations. Thus, in contrast to the conventional measures of collective efficacy that rely exclusively on residents to evaluate social environments, AHDC draws on non-resident visitors engaged in routine activities as collective efficacy informants. This approach is consistent with mounting evidence finding that levels of collective efficacy vary by land use, highlighting the need for measurement strategies incorporating non-residential areas (Corcoran et al. 2018; Wickes et al. 2019). The entrance survey of caregivers includes a “location generator” that prompts the caregiver to report on places they go during a typical week, including weekends, with the following list of possible location types: workplace, caregiver’s school/training, library, place of worship, grocery store, relative’s house, friend’s house, park/recreation center, restaurant, store/business, civic organization, neighborhood organization, and other. After selecting all that apply, the interviewer assists the caregiver in geolocating each place using a Google Maps interface embedded in the survey software; the interviewer can search for names of establishments or drop a pin to indicate the correct location. The addresses are then geocoded using the Google Maps API and the Google address, latitude and longitude are saved. Caregivers are able to report more than one location per type. The most commonly reported location types are grocery stores (90% of caregivers report at least one), child’s school (90%), workplace (67%), store/business (51%), restaurant (44%), and place of worship (43%). Location coordinates were then linked to census units using the R sf: Simple Features package (Pebesma et al. 2019).

For each reported routine location and neighborhood,⁷ respondents were asked to report to what extent they agree with the following statements: 1) whether people on the streets can be trusted (“trust”), 2) whether people are watching what is happening on the streets (“monitoring”), and 3) whether people would come to the defense of others being threatened (“norms toward intervention”). Responses options ranged from 1 (“strongly disagree”) to 5 (“strongly agree”), with items coded to indicate that higher values correspond to higher levels of informal social control. Respondents were asked to report on their residential neighborhood both during the day and at night, but provided summary evaluations for all other locations. We then estimated a block group-level aggregated measure combining reported trust (respondent n=1,257; report n=5,918; block group n=565), monitoring (respondent n=1,307; report n=7,592; block group n=578), and norms toward intervention (respondent n=1,300; report n=7,730; block group n=577) using a cross-classified linear model in which reports are clustered within respondents and within block groups (Raudenbush and Bryk 2002) using the lme4 package in R (Bates et al. 2015:4).⁸ A block group-level random effect was then recovered from this model for all block groups with at least one non-missing report of monitoring, trust, or norms toward intervention (respondent n=1,340; report n=21,411; block group n=580). To obtain corresponding estimates for unobserved block groups, this block group-level random effect was then spatially smoothed across the study area using a conditional autoregressive model proposed by Leroux (Leroux, Lei, and Breslow 2000), implemented in R using the CARBayes package (Lee 2013, 2020). This process yielded our measure of collective efficacy for all 615 Columbus block groups within the outer belt boundary. The correlation coefficient for the smoothed measure with the estimated random effects from the cross-classified multilevel models exceeds .99 indicating that our approach does not spatially smooth the estimated random effect except in block groups without location reports. This feature is desired as we did not want to spatially smooth across block groups with reports and blur real discontinuities in the collective efficacy process. In areas with limited data, however, spatial smoothing allowed the collective efficacy measure to be estimated. This procedure was replicated at the census tract-level to generate a measure of home neighborhood collective efficacy.

Exposure to violent crime is based on reported crime incidents in the Ohio Incident-Based Reporting System between 2014-2016. Incidents were geocoded to block groups based on x-y coordinates. These were then used to create counts at the day level for each block group, resulting in a day-block group level observation file that provided the total count of crimes that occurred at block group j on day t. Using this file, we then generated a 180-day rolling average for each block group. For example, the crime rate for block group j at day t would be equal to the sum of crimes that occurred at block group j between day t and day t–180, divided by the total population for block group j. We then used this 180-day rolling window to calculate time-weighted exposures to crime on a given day based on the proportion of time a participant spent in each location block group on that day. The violent crime rate combines homicide, robbery, aggravated assault and rape. Week-level measures of exposure to crime were then created for each respondent. Home neighborhood violent crime is measured using the 2014-2016 average violent crime rate, again combining incidents of homicide, robbery, aggravated assault, and rape.⁹

Control Variables

Respondent and family control variables are based on self-reported survey data. For this study, youth Race is a binary indicator with categories including non-Hispanic white (reference) and non-Hispanic Black. Youth Foreign born status is a binary indicator for whether the respondent was born in (reference) or outside the United States. Youth and caregiver Biological sex are binary measures where female is the reference category. Youth and Caregiver age are continuous measures of self-reported age. Household size is the caregiver reported number of occupants in the household. Household income is based on caregiver self-reported data, with categories including <$30,000 (reference); $30,001-$60,000, and >$60,001. Caregiver Marital status includes four categories: married (reference), cohabitating, single, and other. Caregiver education includes five self-reported categories (less than high school, high school/GED, some college, Bachelor’s degree, and graduate/professional degree). Home ownership is a binary indicator of whether the caregiver owns the place of residence. Years in neighborhood is a self-reported continuous measure of the number of years the caregiver has lived in their neighborhood (see footnote 8). Season is a four-category measure of during which season the youth participated in the study (winter (reference), spring, summer, and fall). Moved in the past two years is a binary caregiver-reported indicator. Block group-level population density is based on ACS-linked census data. We additionally control for the total number of minutes that a respondent spent outside the home over the course of the study week. Lastly, we created a three-category variable capturing the number of weekend days covered in the GEMA data by each respondent (0, 1, or 2 weekend days).¹⁰

Analytic Strategy

First, linear regression models are used to assess mean differences in non-home exposure to concentrated disadvantage, proportion Black, collective efficacy, and violent crime between all Black and white AHDC youth (n=1,180) net of controls only for respondent age and sex. For each dependent variable, the second model adds all previously mentioned individual-level controls for demographic factors and census tract characteristics. The third models then assess mean differences among Black and white AHDC participant youth who live in census tracts with participants of the other race (i.e. Black or white, n=674), and the fourth models assess Black-white differences for youth living in the same census tract by controlling for home census tract fixed effects. We then turn to models assessing whether Black-white differences in non-home exposures vary systematically by home neighborhood disadvantage, biological sex, age, or having recently moved (i.e. through statistical interaction). For each non-home exposure composition, we control for home neighborhood concentrated disadvantage given the significance of this measure within the neighborhood effects literature, as well as the respective census tract-level measure of the outcome (i.e., controlling for home tract collective efficacy when predicting non-home collective efficacy). We do not attempt to discern whether any observed Black-white differences in a given non-home exposure measure are due to Black-white differences in another home tract or non-home exposure measure, as these are all highly interrelated and likely to lead to problems of multicollinearity.¹¹ All presented regression models use cluster robust standard errors to account for clustering of respondents in census tracts of residence. An exception is made for models assessing cross-level interactions between respondent race and home neighborhood disadvantage, for which we use multilevel linear models with respondents clustered within census tracts to include a tract-level random slope for respondent race (Heisig and Schaeffer 2019).¹²

Results

Of the 1,405 youth in Wave I of AHDC, 1,258 self-identify their race-ethnicity as either non-Hispanic white or non-Hispanic Black. From this sample, 70 respondents are dropped for having no non-home time for the study week, and an additional 8 respondents are dropped for having no non-home time within the Columbus, OH study area (n=1,180). We retain respondents missing on control variables (driven mainly by missingness for household income; n=81) using multiple imputation by chained equations procedures with five imputed datasets in Stata 15, bringing our final analytic sample to 1,180 youth (StataCorp 2017; von Hippel 2020). In total, these youth have primary addresses in 178 of the 197 census tracts within the Columbus study area.

Table 1 displays descriptive statistics for study variables among the full sample of Black and white AHDC youth and those in the “fixed effects sample,” or respondents who live in a census tract with at least one respondent of the other race (i.e. Black or white). Distributions of home neighborhood concentrated disadvantage by race for both the full and fixed effects samples are displayed in Appendix Figures 1 and 2, respectively.

Table 1.

Means and proportions for study variables by analytic sample.

	Full Sample		Fixed Effects Sample
	Mean	SD	Mean	SD
Home neighborhood measures
Tract concentrated disadvantage	0.00	1.00	0.02	0.79
Tract proportion Black	0.32	0.30	0.31	0.26
Tract collective efficacy	0.00	1.00	−0.12	0.86
Tract Violent crime rate	18.88	18.90	20.43	18.48
Block group population density	5241.90	2933.05	5523.58	2940.54
Non-home exposure measures
Non-home concentrated disadvantage	0.00	1.00	0.09	0.97
Non-home proportion Black	0.25	0.24	0.26	0.22
Non-home collective efficacy	0.00	1.00	−0.14	0.93
Non-home violent crime rate	6.67	6.49	7.38	6.15
Adolescent measures
Total non-home time (minutes)	1758.91	930.42	1707.94	940.52
Black (vs. white)	0.48		0.51
Foreign born	0.02		0.02
Age	14.30	1.86	14.20	1.85
Male	0.46		0.46
Family Controls
Household size	4.65	1.61	4.63	1.60
Parent age	45.43	8.55	44.84	8.86
Parent male	0.12		0.12
Household income
Under $30,000	0.37		0.40
$30,001 - $60,000	0.24		0.26
$60,000 - $150,000+	0.40		0.34
Parent education
<High school	0.05		0.06
High school	0.16		0.17
Some college	0.36		0.42
College degree	0.25		0.21
Graduate degree	0.18		0.15
Parent marital status
Married	0.53		0.49
Cohabiting	0.10		0.12
Single	0.20		0.21
Other	0.17		0.18
Residence owned (vs. rented)	0.61		0.57
Years lived in current neighborhood	12.10	10.23	11.34	9.69
Moved in the last two years	0.16		0.16
Season
Winter (December-February)	0.23		0.24
Spring (March-May)	0.23		0.22
Summer (June-August)	0.29		0.28
Autumn (September-November)	0.26		0.26
Number of Weekend Days
0	0.02		0.02
1	0.03		0.04
2	0.95		0.95
N: Individual-level	1180		674

Average Racial Inequalities

Table 2 displays results from linear regression models with tract-level cluster robust standard errors for the four non-home exposure composition outcomes. We present reduced models focused on the Black (vs. white) coefficient of interest, but the full tables are displayed in Appendix Tables 2-9. Model 1 for z-score standardized non-home concentrated disadvantage indicates that, net of respondent sex and age, Black youth have an expected .902 (p < .001) standard deviations higher concentrated disadvantage in their non-home activity space than do white youth. Model 2 adds all the individual-level control variables and home census tract concentrated disadvantage. This model indicates that Black youth have an expected .166 (p < .05) standard deviations higher concentrated disadvantage in their non-home activity space compared to white youth. The third model reduces the analytic sample to white and Black respondents who live in a census tract with AHDC youth of the other race, with the coefficient for respondent Black (vs. white) race remaining positive (b = .117) but is statistically nonsignificant. Controlling for census tract fixed effects to compare youth who live in the same census tract in Model 4, the coefficient for respondent Black race remains statistically nonsignificant.

Table 2.

Linear regression models for non-home exposures with cluster robust standard errors.

Panel A. Non-home Concentrated Disadvantage and %Black
	Concentrated Disadvantage				Proportion Black
	Model 1	Model 2	Model 3	Model 4	Model 1	Model 2	Model 3	Model 4
Respondent Black (vs. white)	0.902*** (0.077)	0.166* (0.068)	0.117 (0.074)	0.093 (0.084)	0.305*** (0.017)	0.090*** (0.016)	0.094*** (0.018)	0.091*** (0.017)
Neighborhood-level measures
Tract concentrated disadvantage		0.407*** (0.047)	0.499*** (0.066)			0.019+ (0.012)	−0.004 (0.016)
Tract Proportion Black						0.357*** (0.040)	0.411*** (0.047)
Constant	−0.170 (0.226)	0.292 (0.443)	0.909 (0.587)	0.210 (0.743)	0.139** (0.048)	0.082 (0.086)	−0.022 (0.131)	0.321* (0.136)
Age and Sex	✓	✓	✓	✓	✓	✓	✓	✓
Control Variables		✓	✓	✓		✓	✓	✓
Home Tract Fixed Effects				✓				✓
N: Neighborhood-level	178	178	89	89	178	178	89	89
N: Individual-level	1180	1180	674	674	1180	1180	674	674
Panel B. Non-home Collective Efficacy and ln(Violent Crime).
	Collective Efficacy				ln(Violent Crime)
	Model 1	Model 2	Model 3	Model 4	Model 1	Model 2	Model 3	Model 4
Respondent Black (vs. white)	−1.062*** (0.086)	−0.286*** (0.063)	−0.277*** (0.074)	−0.228** (0.072)	1.435*** (0.123)	0.249*** (0.061)	0.226** (0.072)	0.171* (0.076)
Neighborhood-level measures
Tract concentrated disadvantage		−0.054 (0.043)	−0.025 (0.106)			0.135* (0.061)	0.176* (0.073)
Tract collective efficacy		0.473*** (0.059)	0.503*** (0.119)
ln(Tract Violent crime rate)						0.586*** (0.062)	0.654*** (0.053)
Constant	0.419+ (0.231)	−0.390 (0.402)	−0.478 (0.564)	0.263 (0.653)	−0.201 (0.293)	−0.521 (0.448)	−0.423 (0.593)	1.159 (0.893)
Age and Sex	✓	✓	✓	✓	✓	✓	✓	✓
Control Variables		✓	✓	✓		✓	✓	✓
Home Tract Fixed Effects				✓				✓
N: Neighborhood-level	178	178	89	89	178	178	89	89
N: Individual-level	1180	1180	674	674	1180	1180	674	674

^*p < .05

^**p < .01

^***p < .001; two-tailed tests.

Notes: The following variables are z-score standardized: Tract concentrated disadvantage, tract collective efficacy, non-home concentrated disadvantage, and non-home collective efficacy. Models control for all control variables described in the "Data and Measures" section. See Appendix for full tables.

The second set of fitted models are for the non-home proportion Black as the outcome (ranging from 0-1). Net of respondent age and sex, Model 1 indicates that respondent Black race is associated with about a .305 increase (p < .001) in expected non-home activity space proportion Black. Model 2 adds all individual-level demographic controls and home neighborhood-level concentrated disadvantage and proportion Black. Relative to white youth, Black youth are expected to have activity spaces that are .09 (p < .001) higher in proportion Black. Selecting on respondents who live in a census tract with a respondent of the other race in Model 2, respondent Black race is similarly associated with about an expected .094 increase (p < .001) in activity space proportion Black. The third model including census tract fixed effects again indicates that, relative to residentially comparable white youth, Black youth have an expected higher exposure to proportion Black activity spaces (b = .091, p < .001).

The third set of models are for the standardized non-home exposure to collective efficacy. The first model indicates that Black respondents are expected to be exposed to 1.062 standard deviations lower collective efficacy (p < .001) in their activity spaces compared to white youth when controlling for respondent age and sex. The magnitude of this coefficient drops to −0.286 (p < .001) in Model 2, net of individual-level demographic controls and home neighborhood-level concentrated disadvantage and collective efficacy. Selecting on youth residing in census tracts with respondents of the other race in Model 3, the coefficient for Black race remains similarly pronounced and is statistically significant (b = −.277, p < .001). The third model adds census tract fixed effects to control for all differences between census tracts of residence, with the coefficient for Black race indicating that these youth are expected to be exposed to .228 standard deviations lower collective efficacy (p < .01) than are white youth.

Finally, the fourth model set displays results for non-home exposure to violent crime on its natural log scale because of heavy positive skew in this measure. Net of respondent race, sex, and age, the first model indicates that Black youth have an expected exposure of about a (exp(1.435) = (4.20 - 1)*100 = 320, p < .001) 320% higher violent crime in their non-home activity space compared to white youth. Model 2 adds all the sociodemographic variables and natural log of violent crime in respondents’ home census tracts, indicating that Black youth are exposed to 28.27% more violent crime than white youth, net of these controls. Model 3 reduces the analytic sample to respondents living in census tracts with AHDC youth of the other race, and the coefficient for Black (vs. white) race indicates that these youth on average are exposed to 25.36% more violent crime in their activity space. Model 4 adds census tract fixed effects to explicitly compare youth who live in the same census tract, with the coefficient for Black race indicating that Black youth are expected to be exposed to about 18.65% more violent crime than are white youth.

Figure 1 provides a visual summary of the magnitude of coefficients for Black (vs. white) race from all the above noted Models 2-4 for non-home exposures, now with each outcome having been z-score standardized (mean=0, SD=1). For each outcome, the first bar and accompanying confidence interval corresponds to the Model 2s drawing on the full sample of Black and white AHDC youth. The second bar corresponds to models selecting on the “fixed effects” sample of youth, and the third bar corresponds to the models controlling for census tract fixed effects. For example, racial inequalities in non-home exposure to proportion Black are particularly sizable, with the second panel indicating that Black youth are exposed to about .4 standard deviations higher levels of proportion Black than are white youth, on average. Replications of figures for non-home proportion Black and the natural logarithm of the violent crime rate in their original, non-standardized metrics are presented Appendix Figure 3.

Figure 1.

The effect of respondent Black (vs. white) race on z-score standardized non-home exposure outcomes across models.

Racial Inequalities Interactions

We next examine whether racial inequalities in activity space exposures vary across the distribution of home neighborhood concentrated disadvantage, biological sex, age, and whether the respondent moved in the past two years given the importance of these moderators to residential mobility interventions and neighborhood effects more generally. Table 3 displays results from linear models for the full sample of Black and white youth as well as youth in the fixed effects sample, now with each model including a respective interaction between respondent Black race and neighborhood disadvantage, biological sex, age, or having recently moved. For each outcome, fitted models for the latter three interactions are linear regression models with cluster robust standard errors, while models for the interaction between Black and neighborhood disadvantage are two-level models including a random slope for Black race (Heisig and Schaeffer 2019). Across these models the only statistically significant interaction term is between respondent Black race and age with non-home collective efficacy (b = .043, p < .05) as an outcome, indicating that expected racial inequalities in this outcome are lower for older adolescents.

Table 3 Panel A.

Linear regression models for non-home exposures with cluster robust standard errors and interactions.

Panel A. Non-home Concentrated Disadvantage and Proportion Black
	Concentrated Disadvantage							Proportion Black
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7
Moved in the past two year	0.091 (0.066)	0.079 (0.075)	0.160 (0.098)	0.079 (0.075)	0.164+ (0.098)	−0.015 (0.097)	0.097 (0.132)	0.036* (0.014)	0.035* (0.014)	0.045* (0.018)	0.034* (0.014)	0.046* (0.018)	0.015 (0.019)	0.041 (0.028)
Age	−0.004 (0.013)	−0.003 (0.013)	−0.017 (0.015)	0.013 (0.016)	0.001 (0.021)	−0.004 (0.013)	−0.017 (0.015)	0.002 (0.003)	0.000 (0.003)	0.002 (0.003)	0.004 (0.003)	0.006 (0.004)	0.000 (0.003)	0.002 (0.003)
Male	−0.095* (0.044)	−0.068 (0.057)	−0.119 (0.080)	−0.081+ (0.045)	−0.158** (0.057)	−0.082+ (0.045)	−0.158** (0.057)	−0.017+ (0.010)	−0.005 (0.012)	−0.015 (0.016)	−0.017+ (0.009)	−0.033** (0.012)	−0.017+ (0.009)	−0.033* (0.013)
Respondent Black (vs. whi	0.176* (0.069)	0.179* (0.078)	0.132 (0.102)	0.656+ (0.365)	0.609 (0.513)	0.140+ (0.073)	0.074 (0.094)	0.089*** (0.015)	0.101*** (0.018)	0.109*** (0.019)	0.202** (0.077)	0.205+ (0.114)	0.084*** (0.017)	0.090*** (0.019)
Interaction terms
Black*Tract Disadvantage	−0.119+ (0.067)							−0.001 (0.014)
Black*Male		−0.027 (0.086)	−0.080 (0.115)						−0.023 (0.018)	−0.036 (0.023)
Black*Age				−0.034 (0.025)	−0.036 (0.035)						−0.008 (0.005)	−0.008 (0.008)
Black*Moved						0.147 (0.131)	0.104 (0.170)						0.030 (0.026)	0.007 (0.036)
Constant	0.177 (0.408)	0.282 (0.446)	0.174 (0.750)	0.054 (0.482)	−0.028 (0.763)	0.320 (0.440)	0.222 (0.736)	0.089 (0.087)	0.074 (0.086)	0.305* (0.138)	0.028 (0.089)	0.268+ (0.144)	0.088 (0.085)	0.322* (0.135)
Control Variables	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
Fixed Effects			✓		✓		✓			✓		✓		✓
N: Neighborhood-level	178	178	89	178	89	178	89	178	178	89	178	89	178	89
N: Individual-level	1180	1180	674	1180	674	1180	674	1180	1180	674	1180	674	1180	674

^*p < .05

^**p < .01

^***p < .001; two-tailed tests.

Notes: Models including an interaction for Black *Tract Disadvantage are two-level linear multilevel models clustering on census tracts, and include a random slope term for Black (vs. white) respondent. All other models are single-level linear regression models with cluster robust standard errors. The following variables are z-score standardized: Tract concentrated disadvantage, tract collective efficacy, non-home collective efficacy, and non-home concentrated disadvantage. See Appendix for full tables.

Supplemental Analyses

Non-school Exposures

Sensitivity analyses were conducted to ensure the robustness of these results. First, all analyses were replicated with non-home exposure dependent variables which exclude time spent at school. We conducted these analyses because of the tendency for adolescents’ non-home time to be dominated by time spent at school or engaged in school-based activities (Hofferth and Sandberg 2001; Hofferth 2009), and thus the potential for this time to also dominate our non-home exposure measures. Indeed, AHDC youth on average spent 44% of their non-home waking time at school (SD=.35). Results from replicated Table 2 Models 2-4 for each non-home exposure measure excluding school time are displayed in Appendix Table 10. Similarly, results from replicated interaction models in Table 3 for non-home exposure measures excluding school time are displayed in Appendix Tables 11-14. Two findings that contrast with the presented results are evident. First, there is consistent evidence that Black youth are exposed to more concentrated disadvantage in their non-home/school time compared with white youth. The most conservative estimate of this differences comes from the model controlling for home tract fixed effects, indicating that Black youth are exposed to about .218 standard deviations (p < .01) higher concentrated disadvantage. Second, there is some evidence that the Black-white difference in non-home/school exposure to concentrated disadvantage is largest among youth residing in neighborhoods with lower in concentrated disadvantage (see Appendix Table 11). Specifically, the interaction between respondent Black race and home tract concentrated disadvantage is negative and statistically significant (p < .05) when predicting non-home/school exposure to concentrated disadvantage.

Census Tract Exposures

Our second set of sensitivity analyses replicates presented models for non-home exposure to concentrated disadvantage, proportion Black, and collective efficacy based on exposure to census tract-level features rather than census block group-level features.¹³ Results from replications of the analyses presented in Table 2 are presented in Appendix Table 15. Results from replications of the interaction analyses in Table 3 are presented in Appendix Tables 16-18. Two findings emerge from these analyses that contrast with results presented here. First, there is again evidence of Black-white inequalities in non-home exposure to concentrated disadvantage, with the most conservative estimate of this difference indicating that Black youth are exposed to about .182 standard deviations higher concentrated disadvantage (p < .05). Second, there is again some evidence that Black-white differences in non-home exposures are most evident among youth residing neighborhoods lower in concentrated disadvantage. Specifically, the interaction between Black race and tract concentrated disadvantage is negative and marginally significant (p < .1) when predicting non-home exposure to concentrated disadvantage, and positive and statistically significant (p < .001) when predicting non-home exposure to collective efficacy.

Resident-based Collective Efficacy

Our final set of sensitivity analyses replicate the presented models for non-home collective efficacy, but now measuring collective efficacy only with respondents’ reports about their home neighborhoods, and not about their routine activity locations. The correlation between collective efficacy measures based on all reports and only home neighborhoods=.79 at the block group level and .86 at the individual-level however, indicating notable consistency. Results from these analyses are displayed in Appendix Table 19. Consistent with the models discussed here, these results indicate that Black youth are exposed to lower levels of collective efficacy in their activity spaces compared to white youth. There is additionally evidence that this Black-white difference is larger among younger youth.

Discussion

Although research on immigration and residential mobility are cornerstones of demographic inquiry, the demography of everyday mobility remains in its infancy (Cagney et al. 2020; Browning, Pinchak, and Calder 2021). A growing literature finds evidence of racial segregation in activity spaces beyond the home among urban adults, but little research has examined whether or to what extent these patterns are apparent among adolescents (Krivo et al. 2013; Jones and Pebley 2014). Research suggests that segregation in activity space resources is consequential for population health inequalities above and beyond effects of one’s neighborhood (Cagney et al. 2020; Sharp and Kimbro 2021), and may moreover shed light on why Black youth relocating from high- to low-poverty neighborhoods experience some adverse outcomes (Graif 2015; Schmidt et al. 2018). Drawing on extensive smartphone-based GPS data from a large sample of urban youth, the present study examined demographic and resource compositions in naturally occurring mobility patterns. We found robust evidence that Black adolescents, relative to comparable white adolescents, are disproportionately exposed to activity spaces with lower levels of collective efficacy and higher levels of racial segregation, violent crime, and, to a less consistent degree, concentrated socioeconomic disadvantage. We additionally considered whether racial inequalities in activity space compositions vary by adolescent age, biological sex, home neighborhood concentrated disadvantage, or having recently moved, but found little evidence of this systematic variation.

The present study corroborates emerging evidence indicating that reliance on the residential census tract to approximate urban adolescents’ non-home exposures will lead to biased estimates, especially in studies focused on Black youth (Kwan 2009; Browning, Calder, et al. 2021). Our results suggest that this critique may extend to housing mobility programs focused solely on the composition of the immediate neighborhood area (Clampet-Lundquist and Massey 2008). Though neighborhood-centered interventions are no doubt important to reducing racial disparities in well-being, the anticipated returns to these programs may be inflated without attention to the daily mobility patterns and accessibility of resources experienced by disadvantaged racial minority residents (Clampet-Lundquist et al. 2011; Boyd and Clampet-Lundquist 2019). For example, extensive research documents heightened rates of adverse experiences faced by Black youth when navigating more affluent neighborhoods, pushing them away these areas (Feagin 1991; Lyons 2007; Sewell et al. 2016). Spatial patterns of racially segregated interpersonal and organizational network ties likely also disproportionately pull Black youth toward more disadvantaged neighborhoods (Small and McDermott 2006; Krysan and Crowder 2017; Small and Adler 2019). These processes motivated our expectation of racial segregation in activity space, but an investigation of how these processes comparatively contribute to the observed activity space compositions is beyond the scope of the present study. To this end, we urge researchers to consider how the spatial distribution of organizational resources, network ties, and discriminatory processes may differentially shape racial segregation in exposure to specific location types (e.g. schools, workplaces) or specific sections of cities (e.g. shopping districts). For example, racial segregation in exposure to commercial areas may be driven by personal or network-mediated reports of experiences with discrimination (Krysan and Crowder 2017), while racial segregation in school attendance may be more attributable to inequalities in available schooling options (Burdick-Will et al. 2020).¹⁴ The influence of these processes may furthermore vary across cities, motivating additional research considering how city-level social processes affect segregation in activity space compositions and everyday patterns of mobility (Massey and Denton 1993; Fenelon and Boudreaux 2019). Lastly, though we did not find that racial differences in activity space exposures vary by age, sex, home neighborhood disadvantage, or having recently moved, these interactions may yet be important for investigations of other everyday mobility processes—such as how Black youths’ strategies for safely navigating advantaged neighborhoods may depend on gender—and deserve continued investigation (Clampet-Lundquist et al. 2011; Browning, Pinchak, et al. 2021).

This study draws on extensive GPS-derived mobility data in a large contemporary sample of urban youth, but is not without methodological limitations. Importantly, while the present study is motivated in part by findings from the MTO experiment, our data are not capturing activity space disparities in the context of experimentally induced moves. Even so, examining these patterns as they naturally occur sheds light on neighborhood effects research findings and has implications for understanding and implementing interventions focused on adolescents’ residential neighborhood environment. In addition, though GPS data make possible unprecedented examinations of youths’ everyday exposures, these data are still subject to error and ambiguity in the precise determination of location (Boettner et al. 2019; Browning, Pinchak, et al. 2021). Our focus on GPS coverage spanning the weekend and three weekdays is bolstered by research finding between-person variability in activity spaces can be adequately capture with 1-6 days of GPS data (Zenk et al. 2018). Nevertheless, five days of coverage is evidently less suited to capture within-person variability in activity spaces, and thus mobility data collected over longer periods of time may be necessary to make claims regarding racial inequalities in mobility at more fine-grained levels (Zenk et al. 2018). The data analyzed here are also cross-sectional rather than longitudinal in nature, and thus we urge future studies to consider how inequalities in activity space resources may change over time.

It is also important to note that our measures of non-home exposure to concentrated disadvantage and proportion Black are capturing exposure to residential populations rather than ambient populations, such as to the composition of residents and visitors in an area simultaneously (Vallée 2018; Hall, Iceland, and Yi 2019). This focus is consistent with a voluminous literature finding that exposure to neighborhoods of differing residential compositions are consequential to well-being, particularly among Black youth (Lyons 2007; Winkler 2012; Anderson 2015). Non-home exposures based on ambient populations may nevertheless contrast with the present findings, and we thus urge researchers to assess this as well as how ambient and residential population processes may differentially shape youth well-being. Inequalities in public and personal transportation access or in routine distance traveled beyond the home—such as to workplaces or schools—were not considered in this study, but may illuminate the observed results and warrant more attention in the everyday mobility literature (Tana et al. 2016; Anderson and Galaskiewicz 2021).¹⁵ Finally, though recent studies indicate that the Columbus, OH study area is representative of U.S. metro areas with respect to racial segregation (Frey 2018; Hess et al. 2019), we acknowledge that our results may not be reflective of youth residing in other cities or among youth facing housing instability. Mobility data collection efforts spanning multiple cities and countries and targeting more residentially mobile respondents thus remain necessary to fully inform the generalizability of this study.

This study has implications for the growing literature on activity spaces, and the study of “contextual effects” more generally. On the one hand, our results suggest that disproportionately “disadvantaged” activity space exposures among Black youth may contribute to racial disparities in life course well-being. However, the literature underscores that not all ecological resources necessarily benefit Black youth, in particular. Indeed, neighborhood rates of informal social control can be positively associated with racial hate crimes (Lyons 2007). Some research additionally finds evidence for ecological relative deprivation processes (Sharp, Denney, and Kimbro 2015; DeAngelis 2021; Pinchak and Swisher 2022), suggesting that heightened exposure to “advantaged” areas may confer some adverse consequences for disadvantaged youth. These findings urge researchers to consider how neighborhood and activity space resources additively and multiplicatively work together to shape youth well-being.

Table 3 Panel B.

Linear regression models for non-home exposures with cluster robust standard errors and interactions.

Panel B. Non-home Collective Efficacy and ln(Violent Crime).
	Collective Efficacy							ln(Violent Crime)
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7
Moved in the past two years	−0.052 (0.059)	−0.045 (0.059)	−0.070 (0.078)	−0.045 (0.058)	−0.070 (0.078)	0.079 (0.097)	−0.119 (0.124)	0.012 (0.065)	0.002 (0.072)	0.034 (0.097)	0.002 (0.072)	0.032 (0.098)	0.070 (0.129)	0.201 (0.185)
Age	−0.009 (0.011)	−0.010 (0.012)	−0.010 (0.012)	−0.030* (0.015)	−0.004 (0.017)	−0.009 (0.012)	−0.011 (0.012)	0.034** (0.012)	0.033** (0.012)	0.014 (0.015)	0.041* (0.017)	0.017 (0.021)	0.034** (0.012)	0.015 (0.015)
Male	0.114** (0.040)	0.090 (0.061)	0.079 (0.089)	0.127** (0.040)	0.125* (0.057)	0.128** (0.041)	0.125* (0.057)	−0.034 (0.044)	−0.042 (0.071)	−0.088 (0.092)	−0.029 (0.049)	−0.013 (0.062)	−0.029 (0.049)	−0.017 (0.062)
Respondent Black (vs. white)	−0.273*** (0.060)	−0.323*** (0.076)	−0.272** (0.093)	−0.899** (0.312)	−0.050 (0.395)	−0.252*** (0.068)	−0.242** (0.080)	0.238*** (0.066)	0.237** (0.074)	0.099 (0.089)	0.486 (0.295)	0.262 (0.396)	0.267*** (0.065)	0.221* (0.085)
Interaction terms
Black*Tract Disadvantage	0.053 (0.062)							−0.034 (0.069)
Black*Male		0.077 (0.080)	0.089 (0.115)						0.025 (0.087)	0.148 (0.104)
Black*Age				0.043* (0.021)	−0.013 (0.027)						−0.017 (0.021)	−0.006 (0.028)
Black*Moved						−0.194 (0.121)	0.078 (0.154)						−0.106 (0.147)	−0.276 (0.203)
Constant	−0.133 (0.371)	−0.360 (0.404)	0.303 (0.658)	−0.092 (0.430)	0.180 (0.671)	−0.426 (0.403)	0.272 (0.652)	−0.534 (0.407)	−0.512 (0.454)	1.226 (0.896)	−0.638 (0.470)	1.117 (0.930)	−0.541 (0.450)	1.126 (0.878)
Control Variables	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
Fixed Effects			✓		✓		✓			✓		✓		✓
N: Neighborhood-level	178	178	89	178	89	178	89	178	178	89	178	89	178	89
N: Individual-level	1180	1180	674	1180	674	1180	674	1180	1180	674	1180	674	1180	674

^*p < .05

^**p < .01

^***p < .001; two-tailed tests.

Notes: Models including an interaction for Black * Tract Disadvantage are two-level linear multilevel models clustering on census tracts, and include a random slope term for Black (vs. white) respondent. All other models are single-level linear regression models with cluster robust standard errors. The following variables are z-score standardized: Tract concentrated disadvantage, tract collective efficacy, non-home collective efficacy, and non-home concentrated disadvantage. See Appendix for full tables.