Prospective and cross-sectional associations of neighborhood socioeconomic disadvantage with drinking and heavy drinking in mid-adulthood in a US national cohort

Abstract

Background:

Understanding how neighborhood socioeconomic disadvantage contributes to risk for alcohol use and misuse has important implications for prevention and intervention.

Methods:

Data were from the National Longitudinal Survey of Adolescent to Adult Health. Past-month quantity-frequency of alcohol use and heavy drinking in mid-adulthood (ages 33–43) were prospectively and cross-sectionally predicted from neighborhood socioeconomic disadvantage measured in adolescence (ages 12–21) and in mid-adulthood (ages 33–43). Differences in exposure and vulnerability to neighborhood disadvantage by race/ethnicity, sex, and socioeconomic status were also examined.

Results:

There was a significant prospective, but not cross-sectional, association between greater neighborhood disadvantage and reduced alcohol use in mid-adulthood. Each additional standard deviation of disadvantage in adolescence was associated with a 14% (p < .001) decrease in the number of drinks (about three fewer drinks) and 20% (p < .001) reduced odds of heavy drinking in the past month in mid-adulthood and this was not affected by the neighborhood to which one was exposed in mid-adulthood. This relation was significantly moderated by race/ethnicity wherein the association between neighborhood disadvantage in adolescence and less alcohol use in mid-adulthood was significant among White but not among Hispanic, Black, or Asian/Pacific Islander individuals.

Conclusions:

These results suggest that prevention and intervention efforts should focus on exposure to neighborhood disadvantage in adolescence. Research is needed that can identify the mechanisms underlying the effect of neighborhood disadvantage on alcohol outcomes, especially mechanisms specific to race/ethnicity subgroups, accounts for exposure versus vulnerability, and can disentangle causal associations from differential selection into disadvantaged neighborhoods.

1. Introduction

The last three decades have seen a paradigm shift in epidemiology from focusing solely on characteristics of individuals in explaining health and disease to also including contextual factors such as socioeconomic characteristics of the neighborhood in which individuals reside (Diez-Roux, 1998; Link & Phelan, 1995; Robert, 1999). A similar shift has taken place in alcohol research (Mair et al, 2019). Research on the relation between neighborhood socioeconomic disadvantage and alcohol and other substance use was summarized in a systematic review of studies published from 1944 to 2007 (Karriker-Jaffe, 2011). The 34 studies identified and reviewed varied along many dimensions that may have led to the disparate findings observed. For example, studies conducted among adolescents and outside of the US were less likely to observe a significant association between neighborhood disadvantage and alcohol and other substance use than studies of adults and studies conducted in the US. Studies conducted in the US tended to focus on a specific location. Given the diversity of neighborhoods across the US (Kind et al, 2018; Acevedo-Garcia, 2020) the most generalizable evidence about the relation between neighborhood disadvantage and alcohol involvement in the US may come from a nationally representative epidemiologic survey. These studies are in Table 1. The earliest cross-sectional studies summarized in Table 1 observed a significant positive association between neighborhood disadvantage and alcohol use (Waitzman & Smith, 1998; Stimpson et al, 2007) whereas the later studies found that current drinking was negatively associated with neighborhood disadvantage (Karriker-Jaffee et al, 2012; Slutske et al, 2016).

Table 1.

Nationally representative epidemiologic surveys of neighborhood disadvantage and alcohol involvement among adults in the United States.

Authors, date	Study	N	Years, Estimated or actual birth years	Ages	Neighborhood disadvantage	Alcohol measure(s)	Results of associations of neighborhood disadvantage with alcohol involvementa
Cross-sectional studies
1) Waitzman & Smith, 1998	NHANES I	10,161	1971–1975 1897–1950	25–74	Tract-level 5-item composite at current age	Daily volume of alcohol consumed	Positive association	+
2) Stimpson et al, 2007	NHANES III	20,050	1988–1994 1898–1977	17–90	Tract-level 11-item composite at current age	5+ drinks almost daily	Positive association	+
3) Karriker-Jaffee et al, 2012	NAS	13,864	2000, 2005 1940–1987	18–60+	Tract-level 5-item composite at current age	Current Drinking	Negative association in full sample Positive association among Black, Hispanic, and male subgroups	- +
						Heavy drinkingb	NS in full sample Negative association among White subgroup, Positive association among Black subgroup	ns - +
						Consequencesb	NS in full sample Negative association among Hispanic subgroup	ns -
						Dependenceb	NS in full sample	ns
4) Slutske et al, 2016	AH	15,197	2001–2002 1974–1983	18–26	Tract-level 10-item composite at current age	Current drinking	Negative association	-
						Drink frequencyb	Negative association	-
						Heavy frequencyb	NS	ns
						Drunk frequencyb	Negative association	-
						Quantityb	Positive association	+
						Problemsa	Negative association	-
Longitudinal studies
5) Barr, 2018	AH	18,740	1994–2008 1974–1983	12–32	Tract-level 4-item composite from first wave at ages 11–18	Current drinking at 4 waves, at ages 12–32	NS intercept Negative slope NS quadratic	ns _ ns
						Monthly consumption at 4 waves, at ages 12–32	Positive intercept Negative slope Negative quadratic	+ _ _
						Heavy drinking at 4 waves, at ages 12–32	NS intercept Negative slope Negative quadratic	ns _ _
6) Karriker-Jaffee et al, 2019	NLSY79	4,226	1979–2012 1965–1969	14–52	County-level 3-item composite averaged across ages 14–19	Past month days 6+ drinks averaged across ages 25–35	NS in full sample, Negatively associated among White and female subgroups	ns _
						Past month days 6+ drinks averaged across ages 36–52	NS in full sample Positively associated among Black subgroup	ns +
7) Xia & Viera, 2025	ABCD	8,731	2016–2018 2006–2008	9–10	Tract-level ADIc averaged across five years	Maternal drinking during pregnancy	Negative association	_

Note:

^aresults from covariate-adjusted models are presented;

^bamong current drinkers;

^cArea Deprivation Index (Kind et al, 2018; NS = not statistically significant

NHANES = National Health and Nutrition Examination Survey (https://www.cdc.gov/nchs/nhanes/index.html); NAS = National Alcohol Surveys (https://arg.org/center/national-alcohol-surveys/); AH = National Longitudinal Study of Adolescent to Adult Health (https://addhealth.cpc.unc.edu/); NLSY79 = National Longitudinal Survey of Youth 1979 (https://www.bls.gov/nls/), ABCD = Adolescent Brain and Cognitive Development Study (https://abcdstudy.org/)

Race/ethnicities included in studies:

1) White, Black

2) Non-Hispanic White, Non-Hispanic Black, Mexican American, Other

3) White, Black/African American, Hispanic

4) White, Black, Hispanic, Asian, Native American

5) White, Black, Hispanic, Asian, Other

6) White, Black, Hispanic

7) White, Black, Hispanic, Asian/Pacific Islander, Alaska Native/Native American, Other

Nationally representative surveys can also reveal important individual differences in exposure and vulnerability (Hussein et al, 2018) to neighborhood disadvantage. For example, cross-sectional surveys conducted in 2000 and 2005 revealed that women and African American and Hispanic individuals were more likely to be exposed to greater neighborhood disadvantage than were men and White individuals, as were those who were unemployed and had lower incomes and educational attainment (Karriker-Jaffee et al, 2012). There were also race/ethnicity and sex differences in vulnerability to the influence of neighborhood disadvantage on heavy drinking. For instance, Black individuals living in more disadvantaged neighborhoods were more likely to engage in heavy drinking than were Black individuals living in less disadvantaged neighborhoods, whereas the opposite was true for White individuals, and there was no difference in heavy drinking for Hispanic individuals living in more or less disadvantaged neighborhoods.

The cross-sectional studies summarized in Table 1 were based on contemporaneous exposures to neighborhood disadvantage and did not consider earlier life exposures to disadvantage. Two longitudinal studies summarized in Table 1 prospectively linked early life exposure to neighborhood disadvantage to later drinking outcomes in adulthood but did not include later exposures to disadvantage (Barr, 2018; Karriker-Jaffe et al, 2019). Other studies focusing on general health (Vartanian & Houser, 2010) and mental health (Wheaton & Clarke, 2003) included earlier and later exposures to neighborhood disadvantage and found that early life neighborhoods were more important than current neighborhood context. No studies, to our knowledge, have examined multiple exposures to neighborhood disadvantage over the life course in order to examine whether early life exposure to neighborhood disadvantage is more strongly associated with alcohol involvement in adulthood than is current exposure, or whether the effect of earlier exposure to disadvantage is modified by later exposure to advantage or disadvantage.

The present study uses data from Add Health, a landmark study originally designed to understand the causes of health and behavior with special emphasis on the role of the environment (Harris et al, 2013). These data were used to examine the extent to which adolescent and mid-adult exposure to neighborhood socioeconomic disadvantage were associated with drinking and heavy drinking in mid-adulthood. We focused on mid-adulthood because: (1) this age period has been understudied compared to adolescence and early adulthood (Patrick et al, 2024), (2) alcohol use in mid-adulthood has been linked to accelerated aging and associated health risks (Topiwala et al, 2025; Wang et al, 2023), and (3) there have been recent reports of relative increases in heavy/binge drinking in this particular age group (e.g., Harris, 2023; Patrick et al, 2024). We also examined whether exposure and vulnerability to the influence of neighborhood disadvantage on drinking and heavy differed by race/ethnicity, sex, and SES (Levy, 2021; Robert, 1999), and whether early exposure in adolescence was moderated by later exposure to neighborhood disadvantage or advantage in mid-adulthood.

2. Method

2.1. Participants

Participants in the Add Health study have been followed for six waves to date, most recently in 2022–25 (Wave 6). The study was initiated in 1994–1995 when a random sample of schools was selected from all high schools in the United States, stratified by region, school size and type, and racial composition. From these schools, a random subsample of students in grades 7–12 was invited to complete an in-home interview (Wave 1; n = 20,745, age = 12–21 years, response rate [RR] 79.0%). The focus of the present study was alcohol use at Wave 5 predicted by neighborhood disadvantage at Waves 1 and 5. Implementation of Wave 5 in 2016–2018 was via a mixed-mode survey that assessed participants via the web (79.4%), in person (15.5%), by telephone (1.9%), or by mail (3.2%) (total n = 12,297, age = 33–43, RR = 71.8%). Wave 6 was not included because it had not yet been completed when this study was initiated and census-tract-level contextual data geocoded to residential addresses were not yet available.

Participants were born between March 1974 and July 1983; their mean age at Wave 5 was 37.26 years (range = 33–43) and 51% were female. The weighted racial/ethnic composition of the sample was 70.2% non-Hispanic (NH) White, 15.2% NH Black, 3.8% NH Asian/Pacific Islander, and 10.8% Hispanic (see Supplemental Table S1 for the ethnic compositions of the Asian/Pacific Islander and Hispanic groups); NH American Indian/Alaska Native and NH “Other” participants were excluded from this study because of small sample sizes leaving a sample size of 12,041. The participants resided in all 50 states and the District of Columbia, 1,106 counties, and 6,952 different census tracts (census tracts usually have a population size of between 1,200 and 8,000 people). Because Add Health used 2010 census geographic boundaries to create census tract, county, and state identifiers that were longitudinally consistent across the first 5 study waves, we were able to track consistencies and changes in regional area across waves. For example, most participants lived in the same state (74%) but in a different county (51%) and census tract (85%) in mid-adulthood than they lived in during adolescence.

2.2. Measures (for more details about measures see the Appendix)

2.2.1. Alcohol use.

We focused on the two alcohol use outcomes of past-month quantify x frequency of alcohol use and heavy drinking. Past-month quantity x frequency was a measure of the full spectrum of alcohol involvement, whereas heavy drinking was an alternate more extreme outcome. (Because quantity x frequency and heavy drinking were strongly correlated [r = 0.74, p < .001] detailed results for heavy drinking are presented in Supplemental Materials Text S1.)

2.2.2. Region characteristics.

The main predictors in this study were neighborhood socioeconomic disadvantage measured at Waves 1 (adolescence) and 5 (mid-adulthood). Also included as covariates were the distance moved from Wave 1 to 5, census region (Northeast, West, Midwest, and South), urbanicity (metropolitan, micropolitan, small town, and rural), and a state alcohol policy scale score. Distance moved was included to determine whether leaving one’s adolescent neighborhood predicted drinking in mid-adulthood. Census region was included as a covariate to account for differences in regional drinking culture and norms. Similarly, urbanicity was covaried to account for effects associated with the urbanicity/rurality of the neighborhood census tract. The state alcohol policy scale was included as a covariate to account for state-level influences (such as excise taxes and density of outlets) on alcohol use.

2.2.3. Socioeconomic status (SES).

Three indicators of individual socioeconomic status (parental SES, household income, and educational attainment) were included as covariates to control for selection into neighborhoods (Wheaton & Clarke, 2003) and to determine whether neighborhood socioeconomic disadvantage predicted alcohol use independent of individual socioeconomic disadvantage.

2.3. Statistical Analysis

All analyses included a weighting variable, a clustering variable, and a post-stratification strata variable (“census region”) to account for clustering, stratification, selection, and attrition, and to obtain unbiased parameter estimates. There were no missing cases for the weighting and clustering variables and 2% of the cases were missing the post-stratification strata variable. Following Add Health study recommendations, these cases were excluded (Chen & Harris, 2020), leaving a sample size of 11,803. The sample was further restricted to participants with complete data on neighborhood disadvantage at Waves 1 and 5 and alcohol outcomes at Wave 5 to ensure that all models were based on the same sample of 9,561 participants (see Table S3 in Supplemental Materials).

Generalized linear models (GLMs) predicting drinking quantity-frequency from adolescent and mid-adult neighborhood disadvantage used gamma regression (gamma distribution, log link function), as appropriate for this highly skewed consumption variable that is bounded by zero (Neal & Simons, 2007); logistic regression models (binomial distribution, logit link function) were used to predict heavy drinking. All models controlled for the quantity-frequency of alcohol use at baseline (i.e., at Wave 1). Individual demographic (sex, race/ethnicity, household income, educational attainment, and parental socioeconomic status [SES] in adolescence) and geographic (distance moved, census region, urbanicity, and state alcohol policy score) covariates were also included in models.

Separate GLMs were used to test whether there were sex, race/ethnicity, or SES differences in vulnerability to the effect of neighborhood disadvantage at Wave 1 and Wave 5 on drinking quantity-frequency, that is, whether the neighborhood/drinking effect was moderated by sex, race/ethnicity, or SES; we also tested whether the effect of neighborhood disadvantage in adolescence was moderated by the effect of later disadvantage in mid-adulthood. Linear regressions were used to test whether each of the predictors were significantly associated with exposure to neighborhood disadvantage at Wave 1 (in adolescence) and Wave 5 (in mid-adulthood).

Post-estimation analyses of marginal means and average marginal effects (AMEs) were employed to facilitate interpretation of results and probe interaction effects (Long & Freese, 2014; Mize, 2019). The AME reflects the average instantaneous rate of change in the predicted outcome with respect to neighborhood disadvantage assuming the remaining covariates are held at their observed values. To illustrate these effects with respect to meaningful levels of disadvantage, we computed marginal predicted drinking outcomes at the neighborhood disadvantage mean and at one standard deviation above the mean.. Analyses were conducted using Stata Release 19 (Statacorp, 2025).

3. Results

Participants consumed 20.6 drinks in the past month on average and 13.9% had engaged in heavy drinking at Wave 5 in mid-adulthood (see Table 2 for descriptive statistics).

Table 2.

Descriptive statistics.

	Mean	%	SE	range
Alcohol outcomes
Wave 5 Quantity x frequency (drinks/past month)	20.64	---	0.705	0, 450
Wave 5 Heavy drinking (15+/8+ drinks/week for M/F)	---	13.87	0.500	0, 1
Neighborhood disadvantage
Wave 1 (adolescence)	24.78	---	0.838	5, 50
Wave 5 (mid-adulthood)	27.32	---	0.438	5, 50
Covariates
Wave 1 Quantity x frequency (drinks/past year)	104.60	---	7.850	0, 5078
Wave 1 Wave 5 distance moved	199.74	---	11.602	0, 9881
Wave 5 State alcohol policy scale	42.30	---	0.365	25.6, 67.9
Wave 1 Northeast census region	---	12.08	0.979	0, 1
Wave 1 West census region	---	15.97	1.342	0, 1
Wave 1 Midwest census region	---	33.68	2.575	0, 1
Wave 1 South census region	---	38.27	1.854	0, 1
Wave 5 Metropolitan	---	84.50	1.999	0, 1
Wave 5 Micropolitan	---	7.84	1.354	0, 1
Wave 5 Small town	---	3.86	0.779	0, 1
Wave 5 Rural	---	3.80	0.836	0, 1
Wave 1 Parental SESa	0.08	---	0.062	−5.0, 3.5
Wave 5 Household income (times ten thousand)	8.18	---	0.181	0.5, 20
Wave 5 Educational attainmentb	2.94	---	0.048	1, 6

Notes:

^aStandardized variable (mean = 0, SD = 1) derived from a factor analysis of four measures,

^bordinal variable with 1 = less than high school, 2 = high school graduate, 3 = completed vocational or technical training, 4 = college degree, 5 = master’s degree, and 6 = PhD, MD, or other advanced degree,

M/F = male/female, SES = socioeconomic status

3.1. Effects of Neighborhood Disadvantage in Adolescence and Mid-Adulthood, and Their Interaction, on Drinking in Mid-Adulthood

Neighborhood disadvantage at Waves 1 and 5 (~23 years later) were significantly correlated (r = 0.38, t = 11.22, p < 0.001) even though only 15% of participants continued to live in the same census tract from Waves 1 to 5 (see Supplemental Table S4). Table 3 summarizes results from GLM models predicting Wave 5 alcohol quantity-frequency from neighborhood disadvantage measured at Wave 1 (Model 1), Wave 5 (Model 2), and both Waves 1 and 5 (Model 3) after controlling quantity-frequency of alcohol use at Wave 1. Wave 1 neighborhood disadvantage was significantly negatively associated with past month alcohol use in mid-adulthood (Model 1); Wave 5 neighborhood disadvantage was significantly associated with past month alcohol use in mid-adulthood (Model 2), but not in a model that also included Wave 1 neighborhood disadvantage (Model 3)¹. The mean ratio (MR) of 0.86 in Model 1 indicated that each standard deviation of disadvantage in adolescence was associated with a 14% (p < .001) decrease in the number of drinks consumed in the past month in mid-adulthood, which corresponded to an AME of −2.92 drinks (95% CI = −4.21, −1.63, p = .001). The marginal mean of past-month alcohol quantity-frequency was 20.48 drinks at the disadvantage mean and 17.56 at one standard deviation above the mean. Additional predictors of increased drinking in mid-adulthood included alcohol quantity-frequency in adolescence at Wave 1², higher parental SES in adolescence, household income in mid-adulthood, and male sex. Predictors of decreased alcohol consumption included living in the West census region in adolescence, higher educational attainment, and race/ethnicity (lower drinking in Asian/Pacific Islander vs. White participants). State alcohol policy scores and urbanicity were not associated with drinking in mid-adulthood.

Table 3.

Results from adjusted generalized linear models predicting alcohol quantity-frequency at Wave 5 from neighborhood socioeconomic disadvantage measured at Wave 1 (Model 1), Wave 5 (Model 2), and at Waves 1 and 5 (Model 3).

	Model 1: Wave 1 Disadvantage			Model 2: Wave 5 Disadvantage			Model 3: Waves 1 and 5 Disadvantage
Predictor	MR	95% CI	p	MR	95% CI	p	MR	95% CI	p
Wave 1 Disadvantage	0.86	0.80, 0.92	<.001	---	---	---	0.87	0.80, 0.94	<.001
Wave 5 Disadvantage	---	---	---	0.92	0.85, 1.00	.039	0.95	0.88, 1.02	.155
Wave 1–5 Distance moved	0.98	0.94, 1.02	.320	0.98	0.94, 1.03	.370	0.98	0.93, 1.02	.293
Wave 1 Alcohol Q-F	1.14	1.08, 1.21	<.001	1.13	1.08, 1.20	<.001	1.14	1.08, 1.20	<.001
Census region (ref = Northeast)
West	0.78	0.64, 0.94	.011	0.78	0.64, 0.95	.013	0.78	0.64, 0.95	.016
Midwest	0.89	0.74, 1.07	.212	0.86	0.72, 1.04	.115	0.90	0.75, 1.07	.231
South	0.91	0.75, 1.09	.286	0.87	0.72, 1.04	.127	0.91	0.76, 1.09	.301
Urbanicity (ref = Metropolitan)
Micropolitan	0.94	0.78, 1.13	.502	0.94	0.77, 1.14	.514	0.95	0.78, 1.14	.555
Small town	0.97	0.77, 1.22	.782	0.86	0.77, 1.21	.723	0.99	0.79, 1.24	.939
Rural	0.82	0.54, 1.25	.351	0.78	0.52, 1.19	.247	0.84	0.54, 1.28	.404
State Alcohol Policy Score	0.99	0.98, 1.00	.195	0.99	0.98, 1.00	.107	0.99	0.98, 1.00	.182
Parental SES	1.08	1.03, 1.13	.001	1.11	1.06, 1.17	<.001	1.08	1.03, 1.13	.002
Household Income	1.01	1.00, 1.03	.042	1.01	1.00, 1.03	.111	1.01	1.00, 0.03	.127
Educational Attainment	0.89	0.84, 0.95	<.001	0.89	0.83, 0.94	<.001	0.89	0.84, 0.94	<.001
Age	0.98	0.95, 1.01	0.188	0.98	0.95, 1.01	.248	0.98	0.95, 1.01	.199
Sex (ref = Female)	1.85	1.64, 2.08	<.001	1.87	1.66, 2.11	<.001	1.85	1.64, 2.08	<.001
Race/Ethnicity (ref = NH White)
Hispanic	1.01	0.83, 1.24	.905	0.93	0.76, 1.13	.437	1.01	0.83, 1.23	.909
NH Black	1.15	0.93, 1.42	.194	1.04	0.85, 1.27	.737	1.18	0.95, 1.46	.142
NH Asian/Pacific Islander	0.57	0.45, 0.72	<.001	0.53	0.42, 0.68	<.001	0.57	0.45, 0.71	<.001

Note: MR = mean ratio (exponentiated regression coefficient); CI = confidence interval; Q-F = quantity x frequency; NH = non-Hispanic.

For all three models: Unweighted sample size: 8,706; Weighted sample size: 14,523,764,

There was no evidence to suggest that the effect of neighborhood disadvantage in adolescence at Wave 1 was modified by the effect of later neighborhood disadvantage in mid-adulthood at Wave 5. That is, greater or lesser disadvantage in mid-adulthood did not amplify or buffer the influence of disadvantage in adolescence on drinking in mid-adulthood (Interaction F(1, 128) = 0.44, p = .51). For example, the AMEs of Wave 1 neighborhood disadvantage at one standard deviation above and at the mean of Wave 5 neighborhood disadvantage did not significantly differ (difference = 0.58 drinks/month, 95% CI = −0.68, 1.84, t = 0.91, p = .36).

3.2. Differences in Exposure to Neighborhood Disadvantage

Participant characteristics were significantly associated with exposure to neighborhood disadvantage at Wave 1 (in adolescence) and Wave 5 (in mid-adulthood) (Table 4). The strongest predictors of higher exposure to neighborhood disadvantage in adolescence and mid-adulthood were Black and Hispanic race/ethnicity, living in the South in adolescence, and living in a small town or rural census tract in mid-adulthood (Table 4). The strongest predictors of lower exposure to neighborhood disadvantage in adolescence and mid-adulthood were higher parental SES, household income, and educational attainment, and White race/ethnicity, living in the West or Northeast and in a metropolitan census tract (Table 4). Coupled with higher exposure to neighborhood disadvantage in adolescence and mid-adulthood, Black and Hispanic participants were also less likely to have moved and moved fewer miles between adolescence and mid-adulthood than did White participants (Supplemental Table S4).

Table 4.

Unadjusted associations between participant characteristics and exposure to neighborhood socioeconomic disadvantage measured at Wave 1 and Wave 5.

	Wave 1 Disadvantage			Wave 5 Disadvantage
Characteristic	r	95% CI	p	r	95% CI	p
Adolescent drinking
Wave 1 Alcohol Q-F	−.02	−.05, .01	.221	−.02	−.06, .01	.187
Census region
Wave 1 West	−.18	−.27, −.08	< .001	−.04	−.10, .02	.166
Wave 1 South	.26	.12, .40	< .001	.14	.06, .22	< .001
Wave 1 Midwest	−.03	−.20, .15	.764	−.06	−.15, .03	.214
Wave 1 Northeast	−.15	−.26, −.05	.003	−.07	−.13, −.02	.006
Census tract urbanicity
Wave 5 Metropolitan	−.13	−.21, −.05	.001	−.17	−.23, −.11	< .001
Wave 5 Micropolitan	.05	−.02, .12	.128	.09	.03 .15	.002
Wave 5 Small town	.08	.03, .12	.001	.13	.08, .19	< .001
Wave 5 Rural	.10	.02, .18	< .001	.06	.00, .11	.036
State
Wave 5 State Alcohol Policy Score	.08	−.02, .19	0.120	.06	−.01, .12	.089
Socioeconomic status (SES)
Wave 1 Parental SES	−.47	−.52, −.41	< .001	−.31	−.35, −.27	< .001
Wave 5 Household Income	−.28	−.32, −.24	< .001	−.41	−.44, −.38	< .001
Wave 5 Educational Attainment	−.24	−.29, −.18	< .001	−.29	−.32, −.25	< .001
Demographic
Age	.04	−.06, .14	.392	.02	−.04, .08	.553
Female sex	.03	.00, .06	.060	.02	−.01, .05	.269
White	−.36	−.45, −.27	< .001	−.23	−.28, −.19	< .001
Hispanic	.12	.03, .20	.007	.08	.04, .12	< .001
NH Black	.38	.30, .47	< .001	.26	.21, .30	< .001
NH Asian/Pacific Islander	−.04	−.09, .00	.057	−.05	−.09, −.01	.019

Note: CI = confidence interval; Q-F = quantity x frequency; NH = non-Hispanic

3.3. Differences in Vulnerability to Neighborhood Disadvantage

There was no evidence for differential vulnerability to the effects of adolescent and mid-adult neighborhood disadvantage on mid-adult drinking by sex, parental SES, or household income, but there was evidence for differential vulnerability by race/ethnicity (see Supplemental Table S5). Table 5 summarizes results from GLM models predicting Wave 5 alcohol quantity-frequency incorporating neighborhood disadvantage x race/ethnicity interaction terms and Figure 1 displays the marginal alcohol quantity-frequency means over disadvantage by race/ethnicity group. There was a significant reduction in drinking as a function of Wave 1 neighborhood disadvantage among White, but not among Hispanic, Black, or Asian/Pacific Islander participants (Figure 1, left panel). White participants living in neighborhoods one standard deviation above the mean in disadvantage as adolescents consumed 4.11 fewer drinks in the past month in mid-adulthood (95% CI = −5.52, −2.69, p < .001) than those living in neighborhoods at the disadvantage mean. This reduction was significantly greater than that observed among Hispanic (difference = −3.79, 95% CI = −7.23, −0.34, p = .03) and Black participants (difference = −4.28, 95% CI = −6.67, −1.89, p = .001) but did not significantly differ from Asian/Pacific Islander participants (difference = −1.87, 95% CI = −5.11, 1.37, p = .26).

Table 5.

Results from generalized linear models predicting alcohol quantity-frequency at Wave 5 from neighborhood socioeconomic disadvantage measured at Wave 1 and Wave 5 incorporating neighborhood disadvantage x race/ethnicity interaction terms.

	Model 1 Wave 1 Disadvantage			Model 3 Waves 1 and 5 Disadvantage
	MR	95% CI	p	MR	95% CI	p
Wave 1 Neighborhood Disadvantage	0.79	0.72, 0.87	<.001	0.87	0.80, 0.94	<.001
Wave 5 Neighborhood Disadvantage	---	---	---	0.94	0.86, 1.04	0.219
Wave 1–5 Distance moved	0.98	0.94, 1.02	0.317	0.98	0.93, 1.02	0.274
Wave 1 Alcohol Q-F	1.14	1.08, 1.21	<.001	1.14	1.08, 1.20	<.001
Census region (ref = Northeast)
West	0.79	0.65, 0.96	0.019	0.80	0.66, 0.97	0.026
Midwest	0.90	0.75, 1.09	0.279	0.91	0.76, 1.09	0.295
South	0.92	0.76, 1.10	0.355	0.92	0.77, 1.11	0.384
Urbanicity (ref = Metropolitan)
Micropolitan	0.93	0.77, 1.12	.449	0.93	0.78, 1.13	0.468
Small town	0.99	0.79, 1.26	.956	0.98	0.78, 1.23	0.834
Rural	0.86	0.55, 1.34	.508	0.84	0.55, 1.28	0.400
State Alcohol Policy Score	0.99	0.98, 1.00	.227	0.99	0.98, 1.00	0.150
Parental SES	1.08	1.03, 1.13	.002	1.08	1.03, 1.13	0.001
Household Income	1.01	1.00, 1.03	.065	1.01	1.00, 1.03	0.143
Educational Attainment	0.89	0.84, 0.95	<.001	0.89	0.84, 0.94	<.001
Age	0.98	0.95, 1.01	.203	0.98	0.95, 1.01	0.178
Sex (Ref = Female)	1.85	1.65, 2.08	<.001	1.85	1.65, 2.09	<.001
Race/Ethnicity (ref = NH White)
Hispanic	1.00	0.83, 1.22	.977	1.03	0.84, 1.25	0.798
NH Black	1.03	0.83, 1.27	.804	1.08	0.86, 1.35	0.523
NH Asian/Pacific Islander	0.58	0.44, 0.77	<.001	0.52	0.42, 0.64	<.001
Neighborhood Disadvantage x Race/Ethnicity
Neighborhood Disadvantage × Hispanic	1.24	1.03, 1.50	.024	0.88	0.70, 1.11	0.288
Neighborhood Disadvantage × NH Black	1.27	1.11, 1.46	.001	1.16	0.98, 1.37	0.087
Neighborhood Disadvantage × NH Asian/PI	1.02	0.70, 1.48	.937	0.83	0.66, 1.03	0.092

Note: MR = mean ratio; CI = confidence interval; Q-F = quantity x frequency; NH = non-Hispanic, PI = Pacific Islander

Model 1: Wave 1 Neighborhood Disadvantage × Race Adjusted Wald test = F (3, 126) = 4.90, p = .003

Model 3: Wave 5 Neighborhood Disadvantage × Race Adjusted Wald test = F (3,126) = 3.67, p = .01

For both models: Unweighted sample size: 8,706; Weighted sample size: 14,523,764

Figure 1.

Predicted marginal effects from adjusted gamma regressions predicting past month drinks in mid-adulthood from neighborhood disadvantage in adolescence (Wave 1) and mid-adulthood (Wave 5) and the interactions with race/ethnicity in a nationally representative US sample. Neighborhood disadvantage is standardized so that −1 represents one standard deviation below the mean, 0 is at the mean, and +1 represents one standard deviation above the mean level of neighborhood disadvantage within each wave. Both models included Wave 1 past year drinks, age, sex, parental SES (in adolescence), current household income, educational attainment, distance moved from Wave 1 to Wave 5, census region, urbanicity, and state alcohol policy scores as covariates. The Wave 5 model also covaried Wave 1 neighborhood disadvantage.

Although there was not a main effect of Wave 5 neighborhood disadvantage on alcohol quantity-frequency, there was a significant moderating effect of race/ethnicity (Table 5, Model 3). There was a significant inverse relation between drinking and Wave 5 neighborhood disadvantage among Asian/Pacific Islander, but not among White, Hispanic, or Black participants (Figure 1, right panel). Asian/Pacific Islander participants in neighborhoods one standard deviation above the mean in disadvantage in mid-adulthood consumed 2.41 fewer drinks in the past month (95% CI = −4.31, −0.52, p = .01) than those living in neighborhoods at the disadvantage mean.

The results for the more extreme alcohol outcome of heavy drinking were similar to the results for the full spectrum of use captured by alcohol quantity-frequency (see Text S1 and Tables S6 and S7 in Supplemental Materials).

3.4. Sensitivity Analyses Excluding Lifetime Alcohol Abstainers

There are alternative approaches to dealing with alcohol abstainers in observational alcohol studies. We elected to include all participants, including lifetime alcohol abstainers, and to code their past month alcohol use as zero. Although only 6.9% of the sample were lifetime alcohol abstainers, this differed significantly by race/ethnicity (Hispanic: 8.5%, NH Black: 13.5%, NH Asian/Pacific Islander: 9.7%, and NH White: 5.0%, F (2.78, 355.27) = 22.80 p < 0.0001). Table S8 in the Supplemental Materials summarizes the reanalysis of the adjusted GLM model predicting Wave 5 alcohol quantity-frequency from neighborhood disadvantage measured at Wave 1 excluding lifetime alcohol abstainers. The results based on analyses that included (Table 3, Model 1) and excluded (Table S8) lifetime alcohol abstainers were nearly identical.

Discussion

There are profound differences between US neighborhoods in their level of socioeconomic disadvantage (Kind et al, 2018; Acevedo-Garcia, 2020) and these disparities have been increasing (Jargowsky, 2013; Singh, 2003; Wei & Knox, 2014). Understanding how neighborhood disadvantage contributes to risk for alcohol use and misuse has important implications for prevention and intervention. Results from nationally representative US epidemiologic surveys on the association between neighborhood disadvantage and alcohol involvement, however, have been mixed, with studies reporting a positive association, a negative association, and no association.

Of the nationally representative epidemiologic surveys of alcohol involvement that also included geocoded neighborhood characteristics, the present study was based on the most recently born cohort that longitudinally followed participants into mid-adulthood. Results showed that exposure to one additional standard deviation of neighborhood disadvantage in adolescence was prospectively associated with a 14% decrease in the number of drinks consumed (about three fewer drinks) and having 20% lower odds of heavy drinking in the past month in mid-adulthood, about 23 years later. This was independent of alcohol consumption, parental SES, and census region of residence in adolescence and household income and educational attainment, rural versus urban residence and state alcohol policy score in mid-adulthood, which were controlled in the models.

Neither of the previous longitudinal studies (Barr, 2018; Karriker-Jaffe et al, 2019) included a measure of neighborhood disadvantage beyond adolescence, leaving open the question of whether a more proximal measure of neighborhood context might be a better predictor of drinking in mid-adulthood. The present study yielded evidence that exposure to neighborhood disadvantage in adolescence, but not contemporaneous exposure in mid-adulthood, predicted drinking and heavy drinking in mid-adulthood. Furthermore, exposure to neighborhood disadvantage in mid-adulthood did not moderate the influence of exposure in adolescence on mid-adult drinking and heavy drinking. That is, there were negligible differences between individuals who consistently resided in a disadvantaged neighborhood, who consistently resided in an advantaged neighborhood, who moved from a disadvantaged neighborhood in adolescence to an advantaged neighborhood in mid-adulthood, or who moved from an advantaged neighborhood in adolescence to a disadvantaged neighborhood in mid-adulthood on their drinking in mid-adulthood. This is a novel and surprising finding that runs counter to several life-course models of neighborhood disadvantage (Levy, 2021) but is consistent with a previous report based on Add Health data showing that metabolic syndrome at Wave 4 was associated with neighborhood disadvantage at Add Health Wave 1 but not Waves 3 or 4 (Martin et al 2019). This supports the premise that adolescence is a sensitive period in the genesis of the drinking career and further suggests that subsequent neighborhood changes may be unlikely to disrupt the course of the drinking career.

We replicated the well-established finding that Black and Hispanic individuals were more likely to be exposed to neighborhood disadvantage than were White individuals (Denney et al, 2018; Jargowsky, 2013; Karriker-Jaffee et al, 2012) and to extend this by showing they were also less likely to have moved and moved fewer miles by mid-adulthood. Although Black and Hispanic individuals were more likely to have been exposed to neighborhood disadvantage than were White individuals, they were not more vulnerable to the effect of neighborhood disadvantage on alcohol use. Another example of invulnerability in the context of greater exposure to neighborhood disadvantage comes from a national prospective US study of all-cause mortality in which greater exposure to neighborhood disadvantage was coupled with lower vulnerability (no increased risk of mortality) among Black and Hispanic individuals; in contrast, lower exposure was coupled with higher vulnerability (increased risk of mortality) among White individuals (Denney et al, 2018).

Although this study was not designed to explain why exposure to neighborhood disadvantage in adolescence predicted drinking and heavy drinking 23 years later in mid-adulthood, it provided several clues. The overall negative relation observed between neighborhood disadvantage and drinking and heavy drinking narrows down the list of possible explanations. Previous studies have posited that the effect of neighborhood advantage might be explained by the community and peer alcohol use norms characteristic of more affluent neighborhoods (Barr, 2018; Karriker-Jaffe, 2011). Additionally, adolescents from more advantaged neighborhoods may be more likely to attend college (Barr, 2018), which has been associated with higher prevalences of drinking and alcohol use disorders (Slutske, 2005).

An alternative explanation for the negative relation between neighborhood disadvantage and drinking and heavy drinking may be differential selection by the parents into more advantaged neighborhoods (Heckman & Landorsø, 2022). Consistent with this is the large correlation of −0.47 between parental SES and exposure to neighborhood disadvantage, both measured in adolescence. Alcohol use appears to be distinct from other health-risk behaviors such as smoking and physical inactivity that tend to be positively associated with neighborhood disadvantage (e.g. Jokela, 2014). Those who drink more tend to be more affluent and to live in more affluent neighborhoods than people who drink less, whereas those who smoke (Diez Roux et al, 2003; Xia & Vieira, 2025) or are physically inactive (Boone-Heinonen et al, 2011) tend to be less affluent and to live in less affluent neighborhoods. A nice illustration of this comes from the Adolescent Brain and Cognitive Development (ABCD) study in which prenatal exposure to alcohol was greater among youth born to mothers residing in less disadvantaged neighborhoods. These youth were more likely to be White, were from families with higher incomes and educational attainment, and were less likely to be living in poverty than those not exposed to alcohol in utero (Xia & Vieira, 2025; see Table 1). In contrast, prenatal exposure to tobacco was greater among youth born to mothers residing in more disadvantaged neighborhoods, were less likely to be White, were from families with lower incomes and educational attainment, and were more likely to be living in poverty than those not exposed to tobacco in utero.

We may be the first to draw attention to the positive associations between neighborhood socioeconomic disadvantage and alcohol involvement reported in earlier nationally representative US studies (Waitzman & Smith, 1998; Stimpson et al, 2007) and the negative associations reported in more recent studies (Barr, 2018; Slutske et al, 2016; Xia & Vieira, 2025). Building on evidence for birth cohort effects for alcohol involvement (Keyes, 2022), attitudes towards alcohol use (Livingston et al, 2023), and the changing demographics of the US (Singh. 2003), we speculate that the nature of the association between neighborhood disadvantage and alcohol use may vary by birth cohort.

4.1. Limitations

There are limitations of this study. First, it is not clear whether these results will generalize across nations (Karriker-Jaffe, 2011) or time. There is greater income inequality in the US than in other industrialized countries (Gottschalk & Smeeding, 2000), and neighborhood inequalities within the US have been changing over time (Jargowsky, 2013; Singh, 2003; Wei & Knox, 2014). Second, the focus on socioeconomic disadvantage omitted many neighborhood characteristics, such as alcohol outlet density (Campbell et al, 2009), neighborhood disorder (Barr, 2018), income inequality (Galea et al, 2007), crime rates (Churchill et al, 2025), residential greenspace (Martin et al, 2025) and other components of the social exposome (Xia & Vieira, 2025) that may be associated with drinking in mid-adulthood. Third, because we focused on drinking in mid-adulthood the results may not generalize to drinking during other developmental periods, although a previous study based on Add Health Wave 3 reported negative associations between a contemporaneous measure of neighborhood disadvantage and four out of six indicators of alcohol involvement in early adulthood (when participants were 18–26 years of age; Slutske et al, 2016; see Table 1). Fourth, we focused on the effect of neighborhood disadvantage at Wave 1 when participants were 12–21 years old and at Wave 5 when they were 33–43, and their interaction. Future research might benefit from a more complete characterization of neighborhood disadvantage across all five Waves of Add Health. Fifth, the study design assumed that the neighborhood environment was impacting upon the individual’s use of alcohol and did not consider the alternate interpretation that this might reflect the impact of the alcohol use of the individual on her choice of environment (e.g., Buu et al, 2007). Sixth, the earliest available neighborhood information came from the years when the participants were adolescents and potentially important earlier exposures were unavailable. Seventh, American Indian/Alaska Native participants were excluded due to an inadequate sample size for moderation and subgroup analyses. The sample sizes of the included race/ethnicity groups may also have been inadequate to detect significant within-subgroup effects.

4.2. Summary, Conclusions, and Future Directions

In a nationally representative US cohort, exposure to neighborhood socioeconomic disadvantage in adolescence was prospectively associated, whereas exposure to neighborhood socioeconomic disadvantage in mid-adulthood was not cross-sectionally associated, with less drinking and with lower odds of heavy drinking in mid-adulthood. There was no evidence to suggest that greater or lesser disadvantage in mid-adulthood amplified or buffered the influence of exposure to disadvantage in adolescence on drinking in mid-adulthood. The associations between neighborhood socioeconomic disadvantage and drinking and heavy drinking did not differ by sex or individual SES but there was a significant difference by race/ethnicity. These results have implications for prevention and intervention. Because adolescence appeared to be the critical developmental period during which neighborhood disadvantage had the greatest impact, this is when prevention and intervention efforts should be focused.

We propose four directions for future research. First, given the modest size of the effect of neighborhood disadvantage on alcohol use, a larger study (potentially with oversampling of race/ethnicity subgroups) will provide greater clarity on potential heterogeneity of the neighborhood disadvantage effect. Second, there needs to be more research that can identify the mechanisms underlying the effect of neighborhood disadvantage on alcohol outcomes, especially mechanisms specific to race/ethnicity subgroups, and research that accounts for exposure versus vulnerability and can disentangle causal associations from differential selection. Third, a new systematic review incorporating the more recent research and including birth cohort as a moderator may confirm the hypothesis that different results obtained in earlier versus later studies may be explained by birth cohort effects. Fourth, given the shift in epidemiology (Diez-Roux, 1998; Link & Phelan, 1995; Robert, 1999) and alcohol research (Mair et al, 2019) towards emphasizing the role of contextual factors, future national alcohol surveys should incorporate geocoding into their assessment protocol (Bluemke et al, 2017). This will provide the critical information needed to better understand the role that socioeconomic differences between US neighborhoods have on alcohol use and misuse.

Highlights.

• Exposure to neighborhood socioeconomic disadvantage in adolescence was associated with less drinking 23 years later, in mid-adulthood.

• Current exposure to neighborhood socioeconomic disadvantage was not associated with more or less drinking in mid-adulthood.

• The effect of adolescent exposure to neighborhood disadvantage on mid-adult drinking was not buffered or amplified by the level of neighborhood disadvantage experienced in mid-adulthood.

Appendix. Study measures

Alcohol outcomes	Wave/Ages	Description
Quantity x frequency	Wave 5 33–43 years	Past-month quantity-frequency of alcohol use was the product of typical drinking quantity and drinking frequency. Past-month quantity of use was assessed with the question “Think of all the times you have had a drink during the past 30 days. How many drinks did you usually have each time? A ‘drink’ is a glass of wine, a can or bottle of beer, a wine cooler, a shot glass of liquor, or a mixed drink,” with open-ended responses. This variable was winsorized (Ruppert, 2006) at 15 drinks per occasion (there were 75 individuals [0.6% of the sample] who reported 16–80 drinks per occasion whose response was recoded to 15). Past month frequency was assessed with the question “During the past 30 days, on how many days did you drink alcohol (beer, wine, or liquor)?” with 7 response categories ranging from ‘none’ to ‘every day or almost every day’. These categories were used to create a continuous frequency measure by using the midpoint of the category bin.
Heavy drinking	Wave 5 33–43 years	Heavy drinking is defined by the Centers for Disease Control as 15 or more drinks/week for men and 8 or more drinks/week for women (Boersma et al, 2020) was also examined. Heavy drinking was a dichotomous variable created by dividing past-month quantity-frequency by four and imposing sex-specific cut-offs.
Quantity x frequency	Wave 1 12–21 years	This was the product of past-month typical drinking quantity and past-year drinking frequency.
Region characteristics
Neighborhood Disadvantage	Wave 5 33–43 years	Census-tract-level indicators were extracted from the 2014–2018 census. Belsky et al (2019) created a measure of neighborhood socioeconomic disadvantage based on the following five indicators extracted from census data at Wave 1: proportions of female-headed households, individuals living below the poverty threshold, individuals receiving public assistance, adults with less than a high school education, and adults who were unemployed. Each of the indicators were converted to deciles and these were summed to create the neighborhood disadvantage composite. The same process was used to create the same measure of neighborhood disadvantage for Wave 5. At both Waves, the measure of neighborhood disadvantage was a continuous measure ranging from 5 to 50, these were standardized into z-scores (M=0, SD=1) for analyses. At both Waves 1 and 5, the five indicators were all significantly correlated with each other and loaded onto a single common factor.
Neighborhood Disadvantage	Wave 1 12–21 years	This was based on census-tract-level indicators extracted from the 1990 census.
Census region	Wave 1 12–21 years	This is a post-stratification variable that was created to adjust the sample weights. It can also be used to identify participants who were living in different census regions (Northeast, West, Midwest, and South) at the inception of the Add Health study.
Distance moved	Waves 1–5 12–43 years	This was the distance in meters based on geocodes of home addresses from Waves 1 and 2. This was standardized into a z-score for analyses.
Urbanicity	Wave 5 33–43 years	This was based on Rural-Urban Commuting (RUCA) Codes (U.S. Department of Agriculture Economic Research Service, 2020) that was developed using data from the 2010 decennial census and the 2006–10 American Community Survey (US Census Bureau). RUCA codes 1–2 were classified as “metropolitan,” codes 3–6 were classified as “micropolitan,” codes 7–8 were classified as “small town,” and codes 9–10 as “rural.”
State alcohol policy score	Wave 5 33–43 years	This is an efficacy- and implementation-weighted composite of 29 state-level alcohol policies for 2018 that provides an index of the relative restrictiveness of the alcohol policy environment (Blanchette et al, 2020). The policies receiving the highest efficacy ratings were beer sales tax, state monopoly on sales, and outlet density (Naimi et al, 2014). The raw APS score for each state was divided by the maximum possible score and multiplied by 100 to rescale it within a theoretical range of 0 to 100. The SPS has only been linked to Wave 5 participant data (Gaydosh et al, 2024). However, most participants (74%) lived in the same state in adolescence as they did in mid-adulthood. In addition, because “policies geared toward excessive alcohol consumption remained virtually unchanged” from 1999 to 2018 (p. 63, Blanchette et al, 2020), this single alcohol policy score likely has relevance for participants before mid-adulthood.
Socioeconomic status (SES)
Parental SES	Wave 1 12–21 years	Parental SES was based on parent reports obtained at the Wave I interview of parental educational attainment, parental occupational attainment, household income, and household receipt of public assistance (Belsky et al, 2018).
Household income	Wave 5 33–43 years	This was past-year household income.
Educational attainment	Wave 5 33–43 years	Highest level of education achieved was recoded into a six-level ordinal educational attainment variable with the following categories: 1 = less than high school, 2 = high school graduate, 3 = completed vocational or technical training, 4 = college degree, 5 = master’s degree, and 6 = PhD, MD, or other advanced degree.