The Downward Spiral: Socioeconomic Causes and Consequences of Alcohol Dependence among Men in Late Young Adulthood, and Relations to Racial/ethnic Disparities

Abstract

Background:

While young adults are generally at highest risk for alcohol problems, not all age out of problem drinking. Evidence suggests that Blacks and Latinos age out more slowly than Whites, particularly among men. Targeting men, we investigated whether differences in lifecourse SES might explain racial/ethnic disparities in alcohol dependence in late young adulthood, along with how experiencing alcohol dependence at that life stage relates to subsequent SES.

Methods:

We used longitudinal, national data to 1) describe racial/ethnic disparities in late young adult alcohol dependence criteria (LYADC), 2) examine whether income trajectory in early young adulthood contributes to these racial/ethnic disparities, and 3) test whether LYADC reciprocally predicts income trajectory in early midlife. Data were from the 1979 National Longitudinal Survey of Youth (N=3,993), which measured LYADC in 1994 (mean age=33). Income trajectory classes were derived for early young adulthood (mean ages=21-31) and, separately, early midlife (mean ages=35-45). Analyses included negative binomial regressions and multinomial regression.

Results:

Both Black and US-born Latino men reported more LYADC than White men. Further, membership in the persistently low and slow increase (vs. stable middle) early young adult income trajectory classes was associated with more LYADC. Multivariate analyses suggested that Black-White disparities in LYADC were explained by early young adult income trajectories, whereas Latino-White disparities in the same were explained by both early young adult income trajectories and early education. In controlled models, more LYADC predicted a higher likelihood of membership in the persistently low (vs. stable middle) income trajectory class in early midlife.

Conclusions:

This study found that poorer SES in early adulthood contribute to alcohol dependence, which reciprocally contributes to poorer SES in early midlife. This cycle appears particularly likely to affect Black and US-born Latino men. Results underline the need to address socioeconomic factors in addressing racial/ethnic disparities in alcohol problems.

INTRODUCTION

Racial/Ethnic Disparities in Alcohol Use and Problems After Early Young Adulthood

Despite considerable research on the developmental course of alcohol use and problems, studies focusing on changes in drinking patterns and problems during the years following early young adulthood remain rare (Berg et al., 2013, Schulenberg and Maggs, 2008). Nonetheless, multiple studies using nationally representative data suggest that Black and Latino drinkers can have longer heavy drinking careers and more enduring alcohol problems than Whites, particularly among men. For example, analyzing the 1984 National Alcohol Survey (NAS) and 1992 follow-up, Caetano et al. reported greater persistence of heavy drinking in Black (vs. White) men (Caetano and Kaskutas, 1995) and greater persistence of “dependence-related problems” among both Black and Latino (vs. White) men (Caetano, 1997); no racial/ethnic differences emerged among women. Similarly, Dawson et al.’s (2005) retrospective analysis of the 2001-2 National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) revealed greater persistence of alcohol dependence among both Blacks and Latinos (vs. Whites) overall, and Grant et al. (2012), examining the 2001-2 NESARC and 2004-5 re-interview, reported greater onset and persistence of AUD among older (40+) US-born Latino men and US-born Latina women of all ages, vs. Whites, along with greater onset and persistence of alcohol dependence for specific Black and Latino subgroups.

A few studies have used multi-wave (panel) data to investigate racial/ethnic disparities in alcohol outcomes, offering a more detailed perspective on racial/ethnic disparities at specific ages. These include two sets of analyses, by our team, of the National Longitudinal Survey of Youth 1979 cohort (NLSY79), involving repeated (yearly or bi-yearly) assessments from adolescence (ages 14-21) through adulthood. One reported slower declines in heavy (6+) drinking frequency in the 20s among both Black and Latino (vs. White) men and women, along with a Latino-White crossover in men’s heavy drinking in the mid 30s (Mulia et al., 2017). The second reported higher rates of alcohol problem onset in the mid 30s among Blacks compared to Whites overall (Lui and Mulia, 2018). This latter finding extends prior studies of NLSY79, which were limited to earlier waves but also pointed to a crossover in alcohol problems for Blacks and Whites in the mid 30s (Cooper et al., 2008, Muthén and Muthén, 2000). Very little is known about the causal pathways underlying the above disparities, however, nor are the consequences of these disparities for functioning in later life well understood.

Socioeconomic Status as Cause and Consequence of Alcohol Problems

Lower socioeconomic status (SES) may, theoretically, be both cause and consequence of drinking problems in adulthood, so SES is a natural focal point for further investigation. Low SES implies a restriction in access to and control over essential material, psychological, social, and health care resources over time, along with greater exposure to stressors and harms, and thus is likely to erode mental and physical health while reducing access to optimal avenues for coping (Link and Phelan, 1995, Mulia et al., 2008a, Myers, 2009). As a result, low and decreasing SES may contribute to the initiation and maintenance of problem drinking, among other high-risk behaviors (Mulia et al., 2008b). Conversely, problematic alcohol use may contribute to and intensify socioeconomic problems by interfering with education and job responsibilities (Berg et al., 2013) and by diverting critical financial resources toward alcohol purchase and the costs associated with alcohol harms.

Studies have repeatedly shown associations between SES and alcohol outcomes. Lower individual SES, measured as lower income, poverty status, lack of employment, and lower education, has been fairly consistently associated with greater abstinence, less frequent drinking, and lower alcohol volume among adults (Casswell et al., 2003, Dawson et al., 1995, Huckle et al., 2010), but also with heavier and more hazardous drinking on a given occasion (Casswell et al., 2003, Karlamangla et al., 2006, Tomkins et al., 2007) and with AUDs and alcohol-related consequences, hospitalizations, and mortality (Grittner et al., 2012, Herttua et al., 2015, Khan et al., 2002, Lee et al., 2013, Mulia et al., 2014, Zemore et al., 2013). Meanwhile a few longitudinal studies have supported associations between changes in SES and alcohol use and problems (Collins, 2016). For example, in a large study of the US Panel Study on Income Dynamics, Cerda et al. (2011) examined income trajectories in childhood/youth (ages 1-15 through 30-44) in relation to drinking patterns in adulthood (the 30s and 40s). They found that both stable low and decreasing medium (vs. stable high) income trajectories were associated with 2.14 and 1.72 times the odds, respectively, of heavy (vs. light-to-moderate) drinking. Another large study, here of a French cohort, found that men and women who experienced lifelong disadvantage or downward intergenerational mobility were at higher risk of premature mortality compared to those with a more favorable SES trajectory. These effects were stronger among men and partially explained, among men, by alcohol use and other factors (e.g., smoking, body mass index, and fruit and vegetable consumption) (Melchior et al., 2006).

Similarly, a few studies have examined the reciprocal effects of problem drinking on SES using longitudinal data. For example, a Finnish study, following participants from ninth grade through middle adulthood, reported that a steadily high (vs. moderate) heavy drinking trajectory from ages 16-42 was associated with greater socioeconomic disadvantage at age 42 among both men and women, even controlling for baseline SES (Berg et al., 2013). Another Finnish study, focusing on problem drinkers aged 18-34, found that early onset, persistent problem drinkers experienced a constant decline in employment with age, whereas early onset, limited-course problem drinkers showed improved employment trajectories, and late-onset problem drinkers experienced flat employment trajectories (Paljärvi et al., 2015). Last, a small US study modelling AUD trajectories in men from adolescence to age 29 found that persistent early-onset (vs. no) AUD was associated with greater financial problems at age 29 (Hicks et al., 2010). The above studies point to possible reciprocal effects of SES and alcohol problems in adulthood, but studies examining these pathways in conjunction with race/ethnicity are much needed to illuminate the causes and consequences of racial/ethnic disparities in alcohol problems in later adulthood—particularly because SES is a key dimension differentiating Blacks and Latinos from Whites (Myers, 2009, Singh et al., 2017, Williams and Purdie-Vaughns, 2016).

The Current Study

The current study investigates lifecourse SES as cause and consequence of racial/ethnic disparities in alcohol dependence specifically during late young adulthood. To address this aim, we capitalize on over 25 years of NLSY79 data (Bureau of Labor Statistics and U.S. Department of Labor, 2012, Mulia et al., 2017). Due to the high complexity of our research questions and analyses, we limited the paper’s scope, as follows. First, we focus on men because, as indicated, studies analyzing genders separately have found that racial/ethnic disparities in heavy drinking and alcohol problems in late young adulthood are largely limited to men (Caetano, 1997, Caetano and Kaskutas, 1995, Mulia et al., 2017), and because men are especially susceptible to alcohol-related problems (Dawson et al., 2015, Grant et al., 2015, Mulia et al., 2017). However, we conduct preliminary analyses to confirm that racial/ethnic disparities in alcohol dependence criteria in late young adulthood (our main outcome) are indeed limited to men in these data. Second, we focus on alcohol dependence criteria, and not other alcohol measures, due to the clear public health significance of alcohol dependence. Third, models target alcohol dependence criteria in the mid 30s (mean age=33) based on evidence for emergent racial/ethnic disparities during this life stage (Cooper et al., 2008, Lui and Mulia, 2018, Mulia et al., 2017, Muthén and Muthén, 2000). Finally, we focus on lifecourse income trajectories as our SES measure because income is a standard SES indicator and potent health correlate that evolves throughout the lifecourse (Collins, 2016, Singh et al., 2017). However, analyses also address early poverty and educational attainment.

The current study departs from Lui et al.’s (2018) prior analysis of alcohol problems in the NLSY79 in several important ways. Namely, the current study examines alcohol dependence criteria count overall among men in late young adulthood, rather than transitions into and out of alcohol problems (2+ criteria) among both genders together. In addition, we disaggregate US-born and foreign-born Latinos to ascertain how nativity may affect Latino-White disparities, as US nativity and higher acculturation have been repeatedly associated with more drinking among Latino men—though this literature is mixed (Alvarez et al., 2017, Zemore, 2007). Finally, our analysis addresses the reciprocal relationships between SES trajectories and alcohol dependence criteria.

Our core questions and hypotheses, all targeting men, are as follows:

1. Are there racial/ethnic disparities in alcohol dependence criteria in late young adulthood?

   H1: Both Black and US-born Latino men will report more dependence criteria in late young adulthood, vs. White men.

2. How do income trajectories in early young adulthood relate to racial/ethnic disparities in alcohol dependence criteria in late young adulthood?

   H2: A poorer early-adult income trajectory will predict more dependence criteria in late young adulthood, and this will help explain racial/ethnic disparities in dependence criteria at this same life stage.

3. Are there reciprocal relationships between income trajectories and alcohol dependence?

   H3: Experiencing more dependence criteria in late young adulthood will predict a poorer income trajectory in early midlife, even controlling for prior SES.

MATERIALS AND METHODS

Data Source

The data source was the NLSY79, which is an on-going study conducted by the US Bureau of Labor that, in 1979, used a stratified, clustered design to select a nationally representative sample of individuals born between 1957 and 1964. The sample included 6,111 non-institutionalized, civilian youths ages 14-21 and an oversample of 5,295 civilian Latino, Black, and economically disadvantaged youths. Respondents were re-interviewed annually from 1979 through 1994 and every two years hence, with respondents showing an age range of 7-9 years at each survey. The initial NLSY79 response rate was 90%, and retention was 90% or better during the first 16 waves, remaining above 80% in more recent waves (Bureau of Labor Statistics, 2012, Bureau of Labor Statistics and U.S. Department of Labor, 2012).

Excepting preliminary analyses, the analytic sample was limited to men. Analyses targeted survey years 1982-2006, spanning early young adulthood through prime earning years, and focused on alcohol dependence in 1994, or late young adulthood (mean age=33). Income trajectory classes were derived for the 10-year periods immediately prior to and following the 1994 assessment, encompassing early young adulthood, years 1982-1992 (mean ages=21-31), and early midlife, years 1996-2006 (mean ages=35-45). Men were excluded unless they participated in the 1994 survey and at least two follow-up surveys through 2006 (N=3,993). Data were weighted using NLSY79 custom weights (Bureau of Labor Statistics, 2016).

Key Measures

Alcohol Dependence Criteria: In 1994, the NLSY79 assessed past-year alcohol dependence criteria in all seven DSM-IV domains, including tolerance, withdrawal, drinking longer/larger quantities, trying to quit/control drinking, cutting down on activities to drink, spending a lot of time drinking/getting over drinking, and continuing to use despite health/emotional problems (American Psychiatric Association, 1994). Our key dependent measure was criteria count (0-7), as continuous measures offer greater precision and are more consistent with current conceptualizations of alcohol use disorder, compared to dichotomous measures (American Psychiatric Association, 2013).

Socioeconomic Status (SES): All surveys assessed total income from wages and salary in the past calendar year; total income is one of the most widely used economic indicators of SES (Duncan et al 2002), and was our main measure of SES. Income was categorized into $10,000 increments; to account for inflation, we standardized income for all years to the last year of data collection available at the time of our analysis (i.e., 2012). Data were used to capture income trajectories in early young adulthood, or 1982-1992 (mean ages=21-31), and separately early midlife, or 1996-2006 (mean ages=35-45).

Race/Ethnicity:

The 1979 survey assessed respondents’ racial/ethnic origin and place of birth. Responses were used to categorize respondents as follows: (1) non-Hispanic White, (2) non-Hispanic Black, (3) US-born Hispanic or Latino, (4) Foreign-born Hispanic or Latino, and (5) Other.

Additional Measures

NLSY79 surveys assessed several sociodemographic variables that were used as controls. Control variables were assessed at a range of different time points to capture core developmental changes relevant to the development of alcohol use disorders. They included age (assessed at baseline), adolescent poverty (assessed in 1979, 1980, and 1981, and defined as proportion of these assessment that the respondents fell below U.S. federal poverty guidelines), educational attainment by age 25 (assessed following the respondent’s 25^th birthday and defined as less than high school, high school degree or equivalent, some college, or college degree or higher), and marital and parenting status (assessed in 1992, immediately prior to the measurement of alcohol dependence, with marital status defined as never married, married, or separated/divorced/widowed, and parenting status defined as number of children in the household). Surveys also assessed additional alcohol measures that were likewise used as control variables. These included positive history of alcohol problems in one’s biological family, including parents and siblings (assessed in 1998 only; yes vs. no), early onset of regular drinking (assessed in 1982, or mean age=21, and defined as initiating drinking at least 1-2 times a month in the past year before age 15; yes vs. no), and early heavy drinking (assessed in 1982 and defined as drinking 6+ drinks at least 4-5 times in the past month; yes vs. no). These three measures capture traditional risk factors for alcohol dependence (Hingson et al., 2006, Muthén and Muthén, 2000).

Analysis

We used Stata (StataCorp., 2017) for data management and all regressions; we used Mplus (Muthén and Muthén, 2013) for latent class growth analysis. The first set of analyses aimed to describe racial/ethnic disparities in alcohol dependence criteria in late young adulthood by gender. We deployed bivariate, negative binomial regressions modelling alcohol dependence criteria count in late young adulthood (mean age=33) as a function of race/ethnicity, separately by gender. These analyses confirmed the expected racial/ethnic disparities in alcohol dependence criteria among men, but not women (see Supplemental Table A). Thus, subsequent analyses were limited to men.

Subsequent analyses examined a) the role of SES in early young adulthood in explaining disparities in alcohol dependence in late young adulthood, and b) associations between alcohol dependence in late young adulthood and SES in early midlife. For these analyses, we began by defining income trajectories at two time periods: “early young adult” (mean ages=21-31) and “early midlife” (mean ages=35-45). We conducted latent class growth analysis (LCGA) (Nagin, 1999, Nagin, 2005), a semi-parametric longitudinal mixture modeling approach, to identify distinct patterns of SES, as defined by income, for men. LCGA attempts to identify clusters of individuals (defining the latent classes) who share common developmental trajectories over time across age while simultaneously estimating the growth parameters within each class (Muthén, 2001). We used survey year to index the growth process to ensure no overlap with alcohol dependence, captured in 1994. All models used full information maximum likelihood estimation, which provides flexibility to include respondents with incomplete data. Single-class growth curve models, including linear and quadratic growth factors, but no covariates or predictors, were estimated for income over time to assess adequacy of the growth functional form and work as the starting point for LCGA models. Subsequent models were estimated by successively increasing the number of classes to determine the best final models based on the data. Model selection was based on a) fit indices such as AIC, BIC, and sample-size adjusted BIC (with lower values indicating a better fit), b) mean posterior probability distributions (≥0.7 preferred), and c) theoretical justification, parsimony, and interpretability (Collins et al., 2010, Jung and Wickrama, 2008, Nagin, 2005, Nylund et al., 2007).

After determining the best-fitting income trajectory models, we used sequential multivariate regressions to examine our main hypotheses (see Measures for covariates). For these analyses, respondents were assigned to those income trajectory classes for which they had the highest probability of membership. We designated “stable middle” as the (low-risk) referent for all analysis, because no stable high class was identified for early young adulthood and because upward mobility has been associated with hazardous drinking, compared to other trajectories, making this a poor “low risk” referent group (Lui et al., 2015). To address Hypotheses 1 and 2, we applied negative binomial regression models regressing alcohol dependence criteria count in late young adulthood on race/ethnicity alone (Model 1) and including early adult income trajectory class (Model 2) and other covariates (Model 3). We expected that both race/ethnicity and income trajectory class would predict dependence criteria count, but also that effects for race/ethnicity would be diminished when accounting for income trajectory class, and that effects for income trajectory class would be robust in the full model. To address Hypothesis 3, we conducted a multinomial model regressing early midlife income trajectory class on dependence criteria count in late young adulthood and other covariates, here also including early young adult income trajectory class as a covariate. We expected that a higher dependence criteria count in late young adulthood would predict a poorer early midlife income trajectory class even controlling for prior income trajectory class and other covariates.

RESULTS

Consistent with Hypothesis 1, bivariate, negative binomial regressions found that dependence criteria count in late young adulthood (mean age=33) was significantly higher among both Black (IRR=1.25, p<.05) and US-born Latino (IRR=1.46, p<.01) men, compared to White men; White men, foreign-born Latino men, and men of Other race/ethnicity reported equivalent criteria counts. Dependence criteria count was lower among foreign-born Latina women (IRR=0.37, p<.01) than White women, while other racial/ethnic groups were equivalent to Whites among women. (See Supplemental Table A for full results of negative binomial regressions.)

Income trajectory classes were derived for all men across early young adulthood (mean ages=21-31) and early midlife (mean ages=35-45); see Figures 1 and 2. A four-class solution best fit the data in both cases, with early young adult classes characterized as persistently low, slow increase, stable middle, and rapid increase, and early midlife trajectories characterized as persistently low, stable middle, stable upper middle, and stable high. (Fit statistics are provided in Supplemental Tables B.1 and B.2.) Table 1 describes demographic differences across classes for both sets of trajectories. Income trajectory classes were associated with all covariates studied, except that early heavy drinking was not associated with early midlife income trajectory. Membership in the persistently low income class for both sets of trajectories (i.e., early young adulthood and early midlife) was associated with being Black or Latino (vs. White); younger age (mostly for the early young adult trajectory); lower in educational attainment by age 25; never married and separated/widowed/divorced (vs. married) in 1992; and a parent of fewer or no children in 1992. Those in the persistently low income classes were also generally more likely than those in the other classes to report a family history of alcohol problems and early onset of drinking, and reported more alcohol dependence criteria in late young adulthood (mean age=33) than all other classes.

Figure 1.

Income Trajectory Classes in Early Young Adulthood (Mean Ages=21-31, 1982-1992 Surveys) among Men.

Figure 2.

Income Trajectory Classes in Early Midlife (Mean Ages 35-45, 1996-2006 Surveys) among Men.

Table 1.

Descriptive statistics for income trajectory classes among men, with omnibus tests of association.

	Early Young Adult Income Trajectory Classes (Mean Ages=21-31, 1982-1992 Surveys)					Early Midlife Income Trajectory Classes (Mean Ages=35-45, 1996-2006 Surveys)
	Stable Middle	Persistently Low	Slow Increase	Rapid Increase	p	Stable Middle	Persistently Low	Stable Upper Middle	Stable High	p
weighted %	11.5	44.9	38.8	4.9		41.8	32.7	17.7	7.9
(unweighted n)	(360)	(2,183)	(1,452)	(145)		(1,542)	(1,651)	(580)	(220)
White	13.3	39.0	41.4	6.3	***	43.0	27.2	20.0	9.8	***
Black	4.6	67.4	26.8	1.2		31.6	57.9	9.1	1.4
US-born Latino	8.7	57.4	31.7	2.2		39.5	44.8	12.4	3.3
Foreign-born Latino	5.2	46.7	45.3	2.8		43.0	41.3	11.1	4.4
Other	12.9	44.5	38.8	3.8		45.8	29.1	18.0	7.1
Age in 1979 (years)	19.3	16.9	17.7	19.2	***	17.6	17.4	17.8	18.1	***
Adolescent poverty (prop. time)	0.04	0.21	0.90	0.60	***	0.10	0.23	0.06	0.07	***
Education by age 25 (%)
Less than HS	3.5	62.3	33.4	0.8	***	33.8	61.1	4.9	0.2	***
HS graduate	14.0	45.4	38.1	2.6		46.4	37.4	13.1	3.1
Some college	11.8	44.8	39.8	3.7		43.6	27.1	24.2	5.0
College or more	12.0	32.2	42.1	13.7		35.3	10.5	29.0	25.2
Marital Status in 1992 (%)
Never married	5.0	63.6	28.7	2.8	***	36.0	46.1	12.6	5.2	***
Married	14.9	34.0	44.5	6.7		44.1	23.2	22.2	10.5
Separated/Divorced/Widowed	13.2	48.8	35.7	2.3		44.7	43.7	9.4	2.2
Number children in 1992 (mean)	1.30	0.71	0.98	1.16	***	0.91	0.80	1.03	1.06	***
Alcohol Measures
Family history alcohol problems (%)	19.5	24.7	21.6	15.1	*	23.3	24.6	19.8	14.8	**
Early onset of drinking (%)	14.2	22.3	17.2	14.1	***	20.8	20.3	16.0	11.1	**
Early heavy drinking (%)	26.3	23.8	29.6	29.3	*	27.0	24.4	27.8	29.9
Dependence criteria count in late young	0.21	0.60	0.37	0.27	***	0.39	0.68	0.24	0.24	***

Notes:

^*p<0.05,

^**p<0.01,

^***p<.001. Significance tests were calculated using approximate F statistics based on the Wald test after multiple imputation.

Table 2 displays the results of negative binomial regressions predicting alcohol dependence criteria count in late young adulthood (mean age=33) from race/ethnicity, income trajectory class in early young adulthood (mean ages=21-31), and covariates, among men. Model 1 replicates the significant, bivariate effects for race/ethnicity described previously. Consistent with Hypothesis 2, Model 2 results reveal that poorer early young adult income trajectories were associated with more dependence criteria in late young adulthood: Membership in the persistently low and slow increase (vs. stable middle) classes was associated with significantly higher counts of dependence criteria. Further, including early young adult income trajectories in the equation rendered Black-White differences in dependence criteria non-significant. Differences between US-born Latino and White men were reduced but not eliminated in Model 2, and were nonsignificant in Model 3 (i.e., with comprehensive controls). Income trajectory class effects remained robust in Model 3.

Table 2.

Negative binomial regression of dependence criteria count in late young adulthood (mean age=33, 1994 survey) among men (N=3,993).

	Model 1			Model 2			Model 3
	IRR	95% CI	P	IRR	95% CI	P	IRR	95% CI	p
Race/ethnicity
White	Ref			Ref			Ref
Black	1.25	(1.03, 1.51)	*		1.05	(0.86, 1.28)	1.06	(0.84, 1.33)
US-born Latino	1.46	(1.15, 1.86)	**	1.34	(1.03, 1.74)	*	1.20	(0.92, 1.58)
Foreign-born Latino	0.79	(0.54, 1.16)		0.74	(0.50, 1.09)		0.85	(0.55, 1.33)
Other	0.84	(0.64, 1.10)		0.80	(0.61, 1.04)		0.80	(0.62, 1.04)
Early Young Adult Income Class
Stable middle				Ref			Ref
Persistently low				2.78	(1.90, 4.06)	***	2.02	(1.37, 2.96)	***
Slow increase				1.75	(1.19, 2.59)	**	1.54	(1.06, 2.25)	*
Rapid increase				1.27	(0.65, 2.48)		1.66	(0.83, 3.33)
All Other Covariates
Age							0.96	(0.92, 1.01)
Adolescent poverty							0.89	(0.65, 1.21)
Educational attainment by age 25
Less than high school							Ref
High school graduate							0.65	(0.51, 0.84)	**
Some college							0.48	(0.35, 0.65)	***
College or more							0.34	(0.24, 0.49)	***
Marital status in 1992
Married							Ref
Never married							2.01	(1.56, 2.60)	***
Separated/widowed/divorced							1.83	(1.35, 2.49)	***
Number children in 1992							1.08	(0.97, 1.19)
Family history alcohol problems							1.31	(1.05, 1.64)	*
Early onset of drinking							1.01	(0.79, 1.31)
Early heavy drinking							1.93	(1.59, 2.33)	***

Notes: IRR=Incidence rate ratios; CI=Confidence interval;

^*p<0.05,

^**p<0.01,

^***p<.001.

To better specify whether and how socioeconomic factors broadly contributed to the US-born Latino vs. White disparity in dependence criteria, we also ran a post-hoc regression (not shown) examining whether this disparity maintained when controlling only for socioeconomic factors: i.e., early young adult income trajectory class, adolescent poverty status, and educational attainment by age 25. In this model, similar to Model 3, the US-born Latino White difference became nonsignificant (OR=1.14, p=.32), while greater dependence criteria count was positively associated with membership in the persistently low (OR=2.59, p<.001) and slow increase (OR=1.75, p<.01) (vs. stable middle) income classes, and negatively associated with higher education (ORs for high school graduation, some college, and college plus=.62, .45, and .31 respectively, ps<.001) (vs. less than high school). Conversely, the effect for US-born Latinos vs. Whites remained significant when controlling for all Model 3 covariates not socioeconomic in nature (OR=1.35, p<.05). This suggests that, while early young adult income trajectories are predominant contributors to Black-White disparities, disparities across US-born Latinos and Whites are explained by both early young adult income trajectories and early education.

Last, Table 3 displays the results of our multinomial regression predicting income trajectory class in early midlife (mean ages=35-45) from dependence criteria count in late young adulthood (mean age=33) and covariates, again among men. This table shows only results comparing likelihood of membership in the persistently low (vs. stable middle) class, as dependence criteria count was not significantly associated with likelihood of membership in any other class in this multivariate model (see Supplemental Table C for full results). Supporting our prediction that experiencing more dependence criteria in late young adulthood would predict a poorer income trajectory in early midlife, we found that higher criteria count predicted higher likelihood of membership in the persistently low (vs. stable middle) class.

Table 3.

Multinomial regression of early midlife income trajectory class (mean ages=35-45, 1996-2006 surveys) among men, comparing likelihood of membership in the persistently low vs. stable middle class (N=3,993).

	RRR	95% CI	p
Dependence Criteria Count in Late Young	1.10	(1.01, 1.20)	*
All Covariates
Race/ethnicity
White	Ref
Black	1.55	(1.23, 1.95)	***
US-born Latino	1.10	(0.81, 1.48)
Foreign-born Latino	1.03	(0.68, 1.56)
Other	0.97	(0.72, 1.30)
Age	1.08	(1.03, 1.13)	***
Adolescent poverty	1.61	(1.15, 2.26)	**
Educational attainment by age 25
Less than high school	Ref
High school graduate	0.57	(0.41, 0.78 )	***
Some college	0.36	(0.25, 0.51)	***
College or more	0.18	(0.11, 0.29)	***
Marital status in 1992
Married	Ref
Never married	1.96	(1.49, 2.59)	***
Separated/widowed/divorced	1.72	(1.23, 2.40)	***
Number children in 1992	1.13	(1.01, 1.27)	*
Family history alcohol problems	0.91	(0.71, 1.16)
Early onset of drinking	0.84	(0.65, 1.09)
Early heavy drinking	0.85	(0.67, 1.09)
Early young adult income class
Stable middle	Ref
Persistently low	10.26	(6.24, 16.87)	***
Slow increase	1.81	(1.10, 2.98)	*
Rapid increase	0.90	(0.16, 5.00)

Notes: RRR=Relative risk ratios; CI=Confidence interval;

^*p<0.05,

^**p<0.01,

^***p<.001.

DISCUSSION

The current study advances understanding of racial/ethnic disparities in alcohol use across the lifespan in several ways. First, ours is the first known longitudinal study to examine racial/ethnic disparities in alcohol problems (here, alcohol dependence criteria) in late young adulthood by gender. Second, this is the first known study to examine changes in income over time as a contributor to racial/ethnic disparities in alcohol problems in late young adulthood, and to examine whether alcohol problems at this life stage may reciprocally affect income trajectories in later life.

Findings supported expectations for racial/ethnic disparities in alcohol dependence criteria in late young adulthood among men. Both Black and US-born Latino men reported significantly more past-year alcohol dependence criteria than White men in late young adulthood (mean age=33). Among women, just one racial/ethnic difference emerged, whereby foreign-born Latina women reported fewer dependence criteria than White women during these same ages. These findings complement and extend prior NLSY79 analyses revealing a Latino-White cross-over in men’s heavy drinking (Mulia et al., 2017) and higher onset of AUDs among Black (vs. White) respondents (Lui and Mulia, 2018), both in the mid 30s. Findings may help optimize alcohol interventions, as they emphasize a need for interventions addressing late young adulthood and Black and Latino men especially. This is particularly important because Black and Latino men experience significant alcohol-related disparities: Although total alcohol consumption and frequency of heavy drinking are generally similar across Whites, Blacks, and Latinos (Dawson et al., 2015, Grant et al., 2015), data show higher rates of alcohol-related cancers of the head and neck among Blacks (Caetano et al., 2014, Chartier et al., 2013) and alcohol-related liver cirrhosis among Latinos (Yoon and Chen, 2016, Yoon et al., 2011), vs. Whites.

Findings also partially supported expectations that poorer early young adult income trajectories would predict more dependence criteria in late young adulthood, and that this would help explain racial/ethnic disparities in dependence criteria count at that time. In multivariate regressions targeting men and adjusting for race/ethnicity, membership in the persistently low and slow increase (vs. stable middle) income trajectory classes during early young adulthood (mean ages=21-31) was associated with significantly more dependence criteria in late young adulthood (mean age=33). Further, including income trajectories in modelling dependence criteria rendered the Black vs. White difference nonsignificant (Model 1 IRR=1.25, p<.05, Model 2 IRR=1.05, ns), while the US-born Latino vs. White difference was reduced (Model 1 IRR=1.46, p<.01, Model 2 IRR=1.34, p<.05). In our final model, which also added early SES measures, the US-born Latino vs. White difference also became nonsignificant (Model 3 IRR=1.20, ns); post-hoc analyses suggested that the US-born Latino-White disparity was fully explained by income trajectories and education together. Results thus suggest that poorer early young adult income trajectories and lower education helped contribute to elevations in dependence criteria in late young adulthood for Black and US-born Latino men.

The current paper extends another published analysis of the NLSY79 that focused on occupational trajectories over time. In that paper, Meyer et al. (2014) modelled trajectories of work “substantive complexity,” defined as decision latitude, active learning, and ability to use and expand one’s abilities at work, across from 1882-2002; they then investigated whether trajectory membership predicted AUD in 1989 and 1994. Among both men and women, membership in lower-complexity work trajectories was associated with higher odds of AUD at both surveys, with effects being strongest among women (ORs 1.42-3.13). Meyer et al.’s results reinforce the importance of SES trajectories to AUD in late young adulthood, but their paper did not address racial/ethnic disparities in AUD; race/ethnicity was included only as a covariate. Our findings—which suggest that income trajectories and early education make independent, and critical, contributions to racial/ethnic disparities in AUDs in late young adulthood—underline the need to address socioeconomic disparities among Black and US-born Latino men in order to address health disparities in these populations.

Finally, results supported the hypothesized reciprocal relationships between dependence criteria and income trajectories. Alcohol dependence criteria in late young adulthood (mean age=33) contributed to worsening income trajectories in later life (mean ages=35-45) independently of prior SES: In the fully adjusted model, each additional dependence criteria predicted an increase of 10% in the relative risk of membership in the persistently low (vs. stable middle) income trajectory class. These findings again are consistent with Meyer et al.’s (2014) paper on substantive work complexity in the NLSY79. In further analyses, the authors examined trajectory class after 1994, and found that incident AUD in 1994 was strongly associated with membership in downward (vs. upward/stable) trajectories for men originally in the highest-complexity class. Interestingly, effects for AUD incidence on work complexity after 1994 were even stronger and more pervasive among women.

We acknowledge several limitations. First, this study represents just one birth cohort (i.e., that born between 1957 and 1964), and results may not generalize to others. Our team recently compared heavy drinking trajectories in the NLSY97 cohort to those of the NLSY79 cohort, and concluded that lifecourse heavy drinking patterns differed across cohorts (Williams et al., 2018). Still, the NLSY79 data remain the only data suitable for examining racial/ethnic disparities in alcohol problems and income past the early 30s, and information about the NLSY79 cohort remains relevant to policy planning. Second, our examination of alcohol dependence criteria was limited to a single time point, partly due to limited assessment. Ideally, future research would capture broader patterns of AUD onset, persistence, and desistance across time. Third, we were limited to DSM-IV alcohol dependence in measuring alcohol problems. It is not known how results would compare when using the DSM-V measure. Fourth, because study population ranged in age from 14-21 years at recruitment, some measures (i.e., adolescent poverty and early heavy drinking) captured varying time points: For example, adolescent poverty was defined for some across ages 14-16, and for others across ages 21-24. This introduces unavoidable error into our covariate adjustments, though the impact should be minor as these variables were not focal. Fifth, we cannot know, given the present analyses, whether poorer income trajectories would predict higher odds of transitioning into alcohol dependence, higher odds of persistent alcohol dependence, or both; similarly, we cannot know whether reporting more dependence criteria would predict higher odds of transitioning out of higher income classes, higher odds of staying in the lowest income class, or both. Finally, due to sample sizes, we coded all those indicating a racial/ethnic group other than White, Black, or Latino as “Other.” Future research should include detailed examination of other racial/ethnic minorities, and particularly disadvantaged groups such as American Indians. Future research should also examine why Black and Latina women do not show greater susceptibility to alcohol use disorders in late young adulthood despite their relatively low SES and despite confirmed associations between poorer SES trajectories and higher odds of alcohol dependence (Meyer and Mutambudzi, 2014).

Albeit with limitations, the present paper suggests powerful, reciprocal relationships between income trajectories and alcohol problems in late young adulthood, pointing to a downward spiral in SES and alcohol misuse that particularly affects Black and US-born Latino men. SES has been under-explored in the U.S. alcohol literature, and alcohol interventions have largely neglected the socioeconomic conditions that contribute to drinking problems and alcohol-related disparities. Our results argue that interventions will be maximally effective overall and in addressing disparities when targeting alcohol and SES together. Interventions could be improved by addressing both the direct and indirect effects of financial stress. For example, screening, brief intervention, and referral to treatment (SBIRT) programs could be adapted to also screen for and intervene on financial stress, offering referral to social services agencies (for job placement) as well as mental health treatment (for addressing associated mental health problems), as appropriate. Public treatment programs could also be encouraged/mandated to provide employment training and mental health services, both of which can be lacking: 61% of publicly-funded programs report providing no employment counselling or training for clients, and 32% provide no mental health services (Substance Abuse and Mental Health Services Administration, 2018). It is hoped that the current results will contribute to the growing recognition of the complex and multi-faceted nature of alcohol problems and alcohol-related disparities, which call for ongoing, coordinated care of the entire person and his or her financial, mental health, and physical health needs.