Urban Neighborhood Features and Longitudinal Weight Development in Girls

Abstract

Introduction

The literature on environment and obesity is characterized by studies that are often cross-sectional and lack racial diversity. This study examined associations between neighborhood features and BMI development over 6 years in an urban sample of 2,295 girls (56% African American; mean age at baseline, 11.2 years) in 2004. Analyses occurred in 2011–2015.

Methods

Girls, caregivers, and study staff completed annual neighborhood questionnaires. Linear mixed-effects modeling examined annual changes in neighborhood features and BMI and assessed whether baseline neighborhood features modified BMI growth over time.

Results

At baseline, 40% of participants were overweight/obese. Participants’ neighborhoods had few neighborhood problems, moderate levels of safety issues and inconvenient features, low levels of neighborhood disorder, few cases of loitering youth, and substantial traffic volume. Adverse neighborhood features were more common for African American than white participants. Neighborhood features were relatively stable over the follow-up period. African American girls with helpful neighbors had lower annual BMI growth (−0.09 kg/m²) than others. For white girls, BMI increased more for girls with helpful neighbors (+0.09 kg/m² annually). Regardless of race, living in a Census tract with low levels of educational achievement was linked with higher BMI growth (an additional 0.07 kg/m² annually). Girls living in Census tracts with high (versus low) levels of poverty gained an additional 0.08 kg/m² gain annually.

Conclusions

Social environment features are associated with BMI change in white and African American urban girls and may be helpful for identifying girls at risk for early adolescent weight gain.

Introduction

Although a considerable literature on the relationships between environment and obesity has emerged, substantial heterogeneity of the findings^1–5 makes the identification of policy implications challenging. The field has recently advanced into longitudinal studies^6–9; however, a preponderance of cross-sectional data^2,10–12 remains problematic for establishing causal relationships.

Studies examining neighborhood and body weight have tended to focus on white populations,^1,10 yet weight-related health risk is greatest among disadvantaged and minority individuals.¹³ Although a growing literature addresses race- or ethnicity-specific risk in youth,^13–16 longitudinal data remain limited. Lower-SES neighborhoods and those with minority populations often have fewer food and physical activity resources.⁵ Environmental features most strongly linked with body weight or physical activity in minority populations include supermarkets, places to exercise, sidewalks, traffic volume, and safety.^2,17

The following analyses capitalize on existing data from an urban, community sample of girls to better understand environmental influences on weight. Girls were followed through early adolescence (approximately age 11–16 years), over which time, BMI for girls of median height is expected to increase by approximately 3 kg/m²¹⁸ and physical activity tends to decline, particularly among African Americans.¹⁹

Methods

Study Sample

In 1999–2000, the Pittsburgh Girls Study (PGS) enrolled girls aged 5–8 years (approximately 25% of the sample per birth year, i.e., birth cohort).^20,21 A database of city housing units was generated from the city’s 911-dispatch system database (including street addresses and 1990 Census-based neighborhood designations) and a U.S. Postal Service database (including individual housing units). After excluding the downtown business district, GIS data divided the remaining neighborhoods into “assignment blocks” comprised of addresses along one side of a street between specified limits.

In the 23 disadvantaged Pittsburgh neighborhoods (≥25% of families living at or below poverty level per 1990 Census data), each assignment block was selected for enumeration. In each “non-disadvantaged” neighborhood (n=66), approximately 50% of assignment blocks were randomly selected for enumeration. In-person enumeration visits verified addresses, added addresses that the database omitted, and determined the demographics of residents’ children.

Forty-three percent of 9,485 assignment blocks were selected in non-disadvantaged neighborhoods, as were all 2,190 blocks in disadvantaged neighborhoods. Of selected blocks, 98% were fully enumerated in each neighborhood stratum. Overall, 103,238 households were enumerated. Enumeration procedures identified 3,241 girls (approximately 84% of girls aged 5–8 years identified by the 2000 U.S. Census). Of these, 2,876 were eligible (English-speaking and without developmental disabilities); 2,450 girls and their primary caretakers enrolled (85.2%). In-home interviews were conducted annually through 2014. PGS’s primary focus is on mental health development, so weight was not measured initially. An obesity sub-study was started during the fifth PGS assessment wave (2004–2005), which provided the baseline year for these analyses. These analyses used data from six annual assessments (Waves 5–10) for white or African American girls with Wave 5 BMI (n=2,295). Response rates were high: 89%–96% of girls completed each study assessment.

Measures

Primary caregivers reported their educational status, the child’s race, number of parents in the household, and household poverty (i.e., “receipt of public assistance,” including participation in the Women, Infants, and Children Program, food stamps, Medicaid insurance, and Temporary Assistance to Needy Families). Study staff measured height and weight via portable stadiometer and digital scale.

At annual assessments, the PGS administered two parent-reported neighborhood measures and field interviewers completed block observations of neighborhood characteristics following intensive training.²² Where the PGS used modified versions of scales validated in other populations, factor analyses with Wave 5 data were used to combine data. For dichotomous items, tetrachoric correlations were used compute the correlation matrix. This matrix was used in the factor analysis.

Caregivers assessed neighborhood problems using a 17-item “Your Neighborhood” questionnaire.²³ Respondents indicated whether each item was not a problem, somewhat of a problem, or a big problem. The questions are similar to items from a scale measuring social and physical incivilities.²⁴ An eight-item caregiver-completed questionnaire, the “Neighborhood Additional Questionnaire” (NQ), asked respondents to identify things they like about their neighborhood (from 12 options), reasons they might move (from 13 options), and questions on social cohesion.²⁵

A physical disorder index (PDI) was derived from five items (graffiti, bottles, cigarettes, litter, and abandoned cars) from interviewers’ block observations of the neighborhood. It has been linked with crime, firearm injuries, homicides and teen births.²² Ordinal responses (none [1], some [2], a lot [3]) for each item were averaged for a summary score. Three additional variables from the questionnaire were included as single-item questions, based on the obesity literature: loitering youth (groups of youth hanging out), single-family houses (whether buildings are predominantly single-family homes), and traffic volume, (moderate to heavy).

Survey measures were categorized as positive or negative factors. Because safety, but not urban density, is typically associated with BMI and physical activity in disadvantaged populations,^17,26,27 and single-family housing was expected to act as a marker for relatively low crime and higher aesthetic appeal compared with other neighborhoods, single-family housing was considered positive.

Five-year estimates (2006–2010) of Census tract–level data from the American Community Survey (ACS) were linked to PGS data for girls living in Allegheny County in each sample year. ACS variables included the percentage of the population with: (1) less than high school education; and (2) living below the poverty level.

Statistical Analysis

Racial differences in baseline sample characteristics were evaluated using t-tests and chi-square tests. Linear and generalized linear mixed-effects models with random subject effects and fixed effects for race and time were used to test if neighborhood characteristics changed over time.^28,29 Analyses were adjusted for fixed effects of caregiver’s education, single-parent households, household poverty, and birth cohort. A logit link was used for single item outcomes (loitering youth, single-family housing, and moderate/heavy traffic volume). Sampled neighborhood was fit as a random intercept and adjusted for an “administration” variable indicating a transition from paper to computerized survey instruments (using the same question wording) between Waves 6 and 7.

The same modeling approach, with the addition of a random slope for time, tested whether each baseline neighborhood feature was associated with different BMI growth trajectories. Each model initially included a three-way interaction among neighborhood feature, time, and race, and all two-way pairs of interactions between these variables. If the three-way interaction term was non-significant, it was dropped from the model to estimate associations for both racial groups combined. Interaction terms between neighborhood feature and time were examined to see whether annual BMI change differed with increasing level of each neighborhood feature. Models were adjusted for birth cohort, SES, and the survey administration variable. Each model was secondarily adjusted with a time-varying variable indicating when girls moved to a new address. All analyses were weighted to account for oversampling of disadvantaged neighborhoods.

To see whether neighborhood change was associated with BMI change, the BMI slope and neighborhood slope were calculated for each participant and linear regression analyzed the relationship between slopes.

This study was approved by the University of Pittsburgh IRB. Data were analyzed (2011–2015) using Stata, version 12.1 with two-sided testing (α=0.05).

Results

At baseline, girls were aged 11 years on average, poverty was common, and 40% were overweight or obese (Table 1). African American girls had more overweight (19%) or obesity (26%) than white girls (16% and 17%, respectively), a larger increase in BMI over time, and more markers of low SES. Between Waves 5 and 10, 59% of girls (white, 40%; African American, 74%) moved to at least one new residence.

Table 1.

Baseline Sample Description^a and Average Annual BMI Change Over the Time-Frame of the Study

		Total sample (n=2,295)b	White girls (n=1,002)	African American girls (n=1,293)	p-value
Age		11.2 (1.3)	11.2 (1.3)	11.2 (1.3)	0.905
Body weight statusc					<0.001
	Underweight	40 ( 2%)	20 ( 2%)	20 ( 2%)
	Healthy weight	1,185 (59%)	560 (65%)	625 (54%)
	Overweight	358 (18%)	143 (16%)	215 (19%)
	Obese	440 (22%)	145 (17%)	295 (26%)
Average 5-year BMI change		4.40 (0.09)	3.91 (0.12)	4.75 (0.13)	<0.001
Caregiver’s education					<0.001
	>12 yr	1,793 (84%)	809 (88%)	984 (82%)
	< 12 yr	337 (16%)	114 (12%)	223 (18%)
Household status					<0.001
	Co-habiting parents	1,217 (57%)	719 (78%)	498 (41%)
	Single parent	913 (43%)	203 (22%)	710 (59%)
Receipt of public assistance					<0.001
	No	1,377 (64%)	765 (83%)	612 (51%)
	Yes	759 (36%)	159 (17%)	600 (49%)

Note: Boldface indicates statistical significance (p<0.05).

^aData are un-weighted. Continuous measures are depicted by mean (SE) and categorical as frequency (%).

^bDue to low levels of missing data, primarily due to non-completion of an annual survey, denominators differ slightly for the different variables. For white and African American girls, respectively: 8 and 7% of age data, 13%, and 11% of body weight data, 8% and 7% of education data, 8 and 7% of household status data and 8, and 6% of public assistance data are not available.

^cUnderweight: <5th percentile; Healthy weight: 5th to <85th percentile; Overweight: 85th to <95th percentile; Obese >95th percentile.⁴⁴

For the “Your Neighborhood” questionnaire,²³ initial exploratory analysis showed that all items belonged together in a single factor, representing 78% of total variance. Cronbach’s α was 0.95. A “Neighborhood Problem” score was therefore calculated as the mean response across the 17 items; higher scores indicate more problems. Based on factor analyses, responses from the NQ were combined into four subscales. The NQ-Appreciation subscale was calculated as the number of neighborhood features respondents indicated liking (e.g., friends, shopping, schools). Factor analysis showed factor loadings between 0.41 and 0.73. Two factors emerged from a series of responses about why respondents would move from their neighborhoods. The binary NQ-Unsafe subscale indicates any safety-related reasons why respondents would move (e.g., gangs, drugs, lack of safety); factor loadings were all >0.83. The NQ-Inconvenience subscale reflects any convenience concerns that could prompt respondents to move (e.g., for better access to shopping, public transportation, or work); factor loadings were all >0.75. The NQ-Community subscale’s four items are similar to components of the “informal social control” scale³¹ (e.g., How likely is it that one of your neighbors would do something about it: breaking into houses, selling drugs to kids, being beaten). Each item’s Likert response (very likely [4] to very unlikely [1]) were summed (construct range, 4–16); factor loadings were all >0.90.

At baseline, caregivers reported a moderate appreciation of neighborhood features (NQ-Appreciation mean, 7.12) and quite high expectations that neighbors would offer help when needed (NQ-Community, 13.70; Table 2). Most girls (74%) lived on a block of predominantly single-family houses. More positive neighborhood features were reported for white girls, whereas negative neighborhood features were more common among African Americans. Few neighborhood problems (construct mean, 1.33) were found. Safety or convenience concerns that could prompt a move from the neighborhood were reported by 28% and 22% of respondents, respectively. Low physical disorder levels (mean PDI, 1.23) and few instances of loitering youth (6%) were reported, but >25% of participants lived with substantial traffic volume. Changes in surveyed neighborhood features were similar for white and African American girls; most remained stable or improved modestly.

Table 2.

Neighborhood Measures at Baseline^a and Change Over Time^b

Measure	Possible range	Interpretatio n of a higher score	Time-frame	Total sample (N=2,295)	White girls (N=1,002)	African American girls (N=1,293)
Positive features from survey data
NQ-Appreciatio n	0–12	More appreciated neighborhood features	Baselinec	7.12 (0.195)	8.10 (0.22)	6.02 (0.15)
			Change scored	1.65 (1.48, 1.82)	1.4 (1.18, 1.62)	1.88 (1.62, 2.13)
			Adjusted change scored	0.039 (−0.24, 0.32)	−0.16 (−0.50, 0.19)	0.25 (−0.16, 0.66)
NQ-Community	4–16	Neighbors likely to help in more situations	Baselinec	13.70 (0.11)	14.30 (0.10)	12.90 (0.11)
			Change scored	0.26 (0.12, 0.40)	0.15 (−0.020, 0.32)	0.35 (0.12, 0.57)
			Adjusted change scored	−0.0019 (−0.17, 0.17)	0.074 (−0.19, 0.33)	−0.083 (−0.34, 0.17)
Single family houses	0/1	Home block is primarily single-family	Baselinec	74%	89%	57%
			Change scored	2.92% (0.12%, 5.70%)	−0.67% (−3.81%, 2.48%)	6.9% (2.75%, 11.00%)
		homes	Adjusted change scored	2.82% (2.72%, 2.92%)	−0.68% (−0.75%, −0.61%)	8.7% (8.53%, 8.88%)
Negative features from survey data
Neighborho od Problems construct	1–3	More neighborhood problems	Baselinec	1.33 (0.02)	1.21 (0.02)	1.46 (0.03)
			Change scored	−0.048 (−0.067, −0.029)	−0.018 (−0.040, .0034)	−0.075 (−0.11, −0.045)
			Adjusted change scored	−0.047 (−0.072, −0.021)	−0.017 (−0.050, .015)	−0.078 (−0.12, −0.042)
NQ-Unsafe	0/1	Presence of safety concerns that could prompt moving from the neighborhood	Baselinec	28%	19%	39%
			Change scored	1.72% (−1.05%, 4.49%)	−0.13% (−3.79%, 3.54%)	2.65% (−1.35%, 6.65%)
			Adjusted change scored	1.93% (1.89%, 1.97%)	−0.37% (−0.392%, −0.35%)	5.49% (5.45%, 5.53%)
NQ-Inconvenience	0/1	Inconvenient features that may promptmoving from the neighborhood	Baselinec	22%	13%	31%
			Change scored	8.32% (5.67%, 11.00%)	4.37% (0.96%, 7.79%)	10.90% (6.95%, 14.80%)
			Adjusted change scored	9.00% (8.80%, 9.21%)	2.39% (2.30%, 2.47%)	13.60% (13.50%, 13.70%)
PDI	1–3	More markers of physical disorder	Baselinec	1.23 (0.018)	1.10 (0.010)	1.37 (0.017)
			Change scored	−0.034 (−0.053, −0.016)	0.030 (0.01, 0.05)	−0.085 (−0.12, −0.06)
			Adjusted change scored	0.038 (0.017, 0.059)	0.038 (0.01, 0.06)	0.038 (0.0003, 0.076)
Loitering youth	0/1	Presence of groups of loitering adolescents	Baselinec	6%	2%	10%
			Change scored	−1.1% (−2.62%, 0.41%)	−0.28% (−1.58%, 1.03%)	−2.44% (−4.98%, 0.10%)
			Adjusted change scored	−1.5% (−1.55%, −1.45%)	−0.12% (−0.13%, −0.12%)	−2.77% (−2.83%, −2.71%)
Heavy traffic	0/1	Presence of heavy traffic	Baselinec	28%	25%	32%
			Change scored	−0.62% (−3.42%, 2.19%)	2.24% (−2.04%, 6.52%)	−3.84% (−7.73%, 0.04%)
			Adjusted change scored	−2% (−2.01%, −1.99%)	−1.13% (−1.15%, −1.12%)	−2.66% (−2.67%, −2.65%)
ACS data
Percent with less than high school education	n/a	Less education	Baselinee	13.0 (6.62)	11.1 (6.38)	14.5 (6.42)
Percent below poverty	n/a	More poverty	Baselinee	24.3 (15.15)	17.2 (11.71)	29.6 (15.29)

^aFor each measure, the table includes a range of possible values and information on interpretation, along with baseline value [mean (SE) or %, as appropriate]. All data are weighted. Neighborhood measures reflect several subscales of the Neighborhood Questionnaire (NQ), the Neighborhood Problems Questionnaire, and the Physical disorder index (PDI), along with single-question items on the presence of single-family houses, loitering youth or traffic volume.

^bFor each measure, the table includes the unadjusted 5-year change over time (difference between the baseline and final data waves of the six data waves were examined), as well as the adjusted change. The adjusted change score is calculated as the 5-year beta for linear change over time, estimated as the slope of a linear regression across the five years of follow-up. Models are adjusted for caregiver education, single parent households, household poverty and birth cohort.

^cMean (95% CI) or Frequency (%).

^dDifference (95% Confidence interval); since dichotomous data were averaged, for dichotomous measure, the outcome for such variables is a continuous proportion variable ranging from 0 to 1, and the unadjusted scores are the “mean proportion,” while the adjusted score is the mean proportion from the linear models.

^eACS data reflect pooled data spanning 2006–2010.

In each year, 91%–94% of the sample resided in Allegheny County. On average, in participants’ home Census tracts, 13.0% of the population had less than high school education and 24.3% lived below the poverty threshold, with lower education levels and more poverty near African American girls’ homes (p<0.001, Table 2).

BMI growth rates were consistently higher for African American girls than for white girls (Table 3). For example, African American girls whose caregivers appreciated many features of their neighborhood (third quartile NQ-Appreciation) grew on average by 0.80 (SE=0.031) kg/m² annually, whereas white girls in highly appreciated neighborhoods grew by 0.64 (SE=0.032) kg/m². Neighborhood appreciation did not alter BMI trajectory (two-way interaction, p=0.868), and this finding was not significantly different between the two racial groups (three-way interaction, p=0.054). The impact of helpful neighbors on BMI growth differed by race (for three-way interaction, p<0.001): African American girls with helpful neighbors (higher NQ-Community) showed 0.09 kg/m² less growth per year (p<0.001), whereas white girls with helpful neighbors showed 0.09 kg/m² more growth annually than those with less helpful neighbors (p=0.012). Girls’ BMI growth did not differ with home proximity to single-family housing.

Table 3.

Annual BMI Change Over 5 Years of Follow-Up From the Baseline for Girls With Varying Baseline Levels of Neighborhood Factors^a

	High and low levels of each neighborhood feature		p-value for 3-way interactionb	Mean (SE) annual change in BMI for girls living in neighborhoods with high and low levels of each neighborhood feature							p-value for neighborhood x wave interaction
	Low levelc	High levelc		Race	Low-level		High-level		Δslopesd
Positive features
NQ – Appreciation		11	0.054	White	0.64 (0.036)		0.64 (0.032)		0.004		0.868
				African American	0.80 (0.032)		0.80 (0.031)
NQ – Community	12	16	<0.001	White	0.59 (0.038)		0.68 (0.035)		0.09		0.012
				African American	0.82 (0.032)		0.73 (0.031)		−0.09		<0.001
Single family houses	0	1	0.057	White	0.66 (0.045)		0.65 (0.030)		−0.01		0.792
				African American	0.80 (0.035)		0.79 (0.037)
Negative features
Neighborhood problems	1	1.52	0.279	White	0.64 (0.030)		0.66 (0.034)		0.02		0.408
				African American	0.78 (0.034)		0.80 (0.030)
NQ – Unsafe	0	1	0.215	White	0.64 (0.031)		0.66 (0.042)		0.02		0.642
				African American	0.79 (0.030)		0.80 (0.039)
NQ – Inconvenience	0	1	0.844	White	0.64 (0.032)		0.66 (0.038)		0.01		0.650
				African American	0.79 (0.031)		0.80 (0.037)
Physical disorder (PDI)	1	1.4	0.150	White	0.64 (0.031)		0.66 (0.032)		0.02		0.292
				African	0.79		0.80
				American	(0.032)		(0.032)
Loitering youth	0	1	0.556	White	0.65 (0.030)		0.70 (0.072)		0.05		0.414
				African American	0.80 (0.033)		0.85 (0.060)
Traffic volume	0	1	0.650	White	0.67 (0.030)		0.61 (0.039)		−0.06		0.083
				African American	0.82 (0.032)		0.76 (0.041)
Census variables
Percent less than HS	7.9	17.6	0.989	White	0.64 (0.033)		0.71 (0.035)		0.07		0.014
				African American	0.75 (0.030)		0.81 (0.034)
Percent below poverty	11.1	33.1	0.052	White	0.64 (0.032)		0.72 (0.043)
				African American	0.73 (0.032)		0.81 (0.033)		0.08		0.007

Note: Boldface indicates statistical significance (p<0.05).

^aThe table displays betas from linear regression models for girls with high versus low levels of each neighborhood features at baseline (i.e., girls at the first and third quartile of each neighborhood measure); models are adjusted for birth cohort, SES, and the survey administration variable; each neighborhood feature is modeled separately.

^bInteraction between neighborhood, data collection wave and race.

^cLow and high levels are defined by the first and 3rd quartiles for continuous variables, and by the two levels of binary variables.

^dDifference in slopes of BMI change over time for girls living in neighborhoods with high versus low levels of the neighborhood factor under consideration (high level – low level); because the 3-way interaction term was dropped when non-significant, the difference in slopes is not race-specific for those models.

Negative neighborhood features had similar effects on BMI change across racial groups. BMI growth did not differ with any of the negative neighborhood features. Adjusting the longitudinal models for girls’ moves to new home residences impacted the findings minimally and did not alter the model interpretations (data not shown).

Regardless of race, girls living in Census tracts where at least 18% of residents had less than high school education gained an additional 0.07 kg/m² each year, compared with girls in areas where low education levels were uncommon (Table 3). Likewise, living in a Census tract with high (versus low) levels of poverty was linked with 0.08 kg/m² additional BMI gain annually.

The slope of BMI change was not correlated with the slope for the change in any of the continuous neighborhood measures (correlation, −0.06 to 0.06). Slopes for the binary neighborhood variables could not be calculated owing to insufficient variation.

Discussion

In this female sample from disproportionately disadvantaged urban neighborhoods, neighborhood features were relatively stable over 6 sample years. Social environment features, but not physical environment features, were associated with BMI change. African American girls in neighborhoods where parents felt that neighbors would help out if difficult circumstances arose had lower annual BMI growth than those living near less helpful neighbors. Conversely, white girls living near helpful neighbors gained more BMI annually than those with less helpful neighbors. Girls living in Census tracts with low educational attainment had greater BMI growth than those in more-educated locales. Likewise, girls in poor Census tracts had higher BMI growth rates than those in more-affluent areas. However, annual BMI change was not associated with changes in survey-based neighborhood features.

The lack of associations between physical environment features and BMI are consistent with prior research showing environmental factors to be less reliably associated with obesity or physical activity among low-income or minority individuals versus other populations.^10,17,27 This may reflect different interactions among land use, infrastructure, and social environment in urban versus suburban populations.³⁰ Our findings suggest that social environment features may be more salient for girls in predominantly disadvantaged urban settings.

The differential effect of helpful neighbors by race was unexpected, but both effects are supported by prior literature. The NQ-Community measure is based on questions used to operationalize informal social control (i.e., the capacity of a group to regulate its members to achieve collective goals), which is closely related to social cohesion.^31,32 As safety concerns may impact physical activity or body weight in urban settings,² and perceived safety may mediate associations between neighborhood quality and children’s physical activity,^3,33 we hyothesized that girls whose neighbors were willing to intercede would be more likely to be physically active and therefore gain less BMI. However, other literature focusing on social networks and social support shows that obesity can spread through social ties,^34,35 success in pursuing lifestyle change and weight loss are influenced by social contacts, and attempts to establish healthy eating or physical activity behaviors may be perceived as conflicting with cultural norms.^36–41 It is possible that neighbors who are willing to intervene for community goals act as a marker of social pressure to conform to unhealthy eating and physical activity standards. It is unclear why the effect of neighbors may differ by race, but social cohesion has been linked with health outcomes²⁴ and there are racial variations in social cohesion, body image, and perceived discrimination.^42–44 This finding warrants follow-up to clarify its mechanisms.

The single geographic region may limit generalizability. For example, after Pittsburgh required state fiscal oversight to avoid bankruptcy in 2004, the local economy rebounded more successfully than some other U.S. cities,^45,46 so neighborhood changes over the study’s timeframe may be atypical. However, PGS obesity prevalence is consistent with national estimates.⁴⁷ Furthermore, the PGS provides an opportunity to understand the effects of the urban environment in a particularly high-risk young urban population for whom national-level data are typically unavailable. These analyses do not address environments outside the home neighborhood (e.g., school), or the full range of neighborhood features that may impact BMI (e.g., proximity to parks, grocery stores, or restaurants).

Although PGS data reflect the perceptions of caregivers and trained interviewers, the girls’ perceptions are not recorded and are likely to become increasingly relevant as participants approach adulthood. Another limitation is the subjective nature of survey measurements. These analyses leverage a unique opportunity to examine detailed and relatively long-term longitudinal associations between environment and body weight. Although objective environmental measures are often preferred owing to accuracy and reliability concerns, perceived environmental attributes can convey complementary information.^5,48 Environment perceptions at times more accurately reflect resource accessibility (e.g., the discounting of potential resources because of safety concerns)⁴⁹ and facilitate assessment of social environment factors (e.g., community support).⁵⁰ GIS data are more easily obtained and compared with data at other locations, but the areal units of Census data are without consistent geographic size¹⁰ and may inappropriately aggregate data from qualitatively distinct locales. Although PGS parents may have considered different spatial areas in their assessments, parental perception of the extent their home environs is likely to reflect family norms more closely than do official neighborhood definitions.

Neighborhood change was not associated with BMI change. This may suggest that environmental intervention is not a useful weight management approach for urban girls. Indeed, prior work indicates that environmental approaches for promoting healthy lifestyles may be particularly challenging in female populations.^8,51,52 Alternatively, the relatively small observed range of neighborhood change may be insufficient to detect an effect on BMI trajectory, and smaller than what could be achieved with a well-delivered intervention. Furthermore, the effect of a given neighborhood change may vary with social context. For example, over this study’s timeframe, foreclosure and demolition of low-income Pittsburgh housing sites led to the relocation of hundreds of tenants.^53–55 Such an imperative for relocation may influence youth differently than, for instance, a parent acquiring a more lucrative job and choosing to move to a more affluent neighborhood. As the study of environment and body weight moves into longitudinal analyses, these findings underscore the importance of considering the complexity that neighborhood change can comprise, including alterations in the neighborhood itself and relocation of individuals, in the context of financial or social constraints, neighborhood selection factors, and evolving behavior.^8,11,51

Many participants moved to a new home during the follow-up period, yet the relationships between baseline neighborhood features and BMI change were minimally impacted by adjustment for home relocations. Because prior home locations predict subsequent neighborhood characteristics,⁵⁶ one possible explanation is that relocated girls’ neighborhood features may often have been quite stable throughout the follow-up period.

Although the potential for environmental change to alter weight trajectories is unclear from these observational data, these findings focus attention on neighborhood aspects that could form the basis of intervention studies. In addition, these findings can be used to identify girls at high risk for early adolescent weight gain and help them to access resources supporting healthier lifestyles. Such implications may be of particular relevance among urban African American girls, given the minority girls’ greater likelihood of both living in adverse neighborhood conditions and of going on to develop weight-related health problems such as diabetes and heart disease.