Job Loss and Alcohol Dependence among Blacks and Whites in
a National, Longitudinal Survey
Sarah E. Zemore, Ph.D.
1
, Nina Mulia, Dr.P.H.
1
, Edwina Williams, M.P.H.
1
, and Paul Gilbert,
Ph.D.
2
1
Alcohol Research Group, 6475 Christie Ave., Suite 400, Emeryville, CA 94608-1010
2
University of Iowa, College of Public Health, Department of Community and Behavioral Health,
145 N. Riverside Dr., N414 CPHB, Iowa City, IA 52242
Abstract
We used the 1979 National Longitudinal Survey of Youth to test whether the association between
job loss and incidence of alcohol dependence differed across Blacks and Whites. Respondents
were interviewed annually from 1979–94; DSM-IV dependence was assessed in 1989 and 1994.
Analyses included only those employed in 1989 and involved lagged logistic regressions
predicting past-year dependence in 1994 from job loss during 1990–1993. Unexpectedly, results
showed stronger and more robust associations between job loss and dependence among Whites
(AOR=1.93, p<.05) than Blacks (AOR=0.82, nonsignificant). Findings diverge from prior
research, suggesting disparities may differ as a function of age and/or time.
Introduction
Unemployment affects many millions in the U.S., and Blacks much more than Whites. Data
from 2015 suggest an unemployment rate of 9.6% among Blacks, over double that among
Whites (at 4.6%). The Black-White differential in unemployment has remained remarkably
consistent over time, with rates among Blacks approximately 1.74–2.53 times that among
Whites from 1972 through 2015 (mean=2.18) (Bureau of Labor Statistics, 2016). Among the
unemployed, long-term unemployment (i.e., job loss lasting more than 6 months) is also
substantially higher among Blacks than Whites (Bureau of Labor Statistics, 2016;
Kosanovich & Sherman, 2015). Because unemployment has been associated with
significantly poorer mental and physical health (Catalano et al., 2011; Paul & Moser, 2009;
Strully, 2009; Wanberg, 2012), these differentials may contribute to the very large disparities
between Blacks and Whites in quality of life and health (Andrulis, Siddiqui, Purtle, &
Duchon, 2010; Centers for Disease Control and Prevention, 2011; Murray et al., 2006).
Job loss can result in a constellation of negative effects that may impact health, such as loss
of healthcare benefits and financial strain (Price, Choi, & Vinokur, 2002), reduced
relationship quality and social support (Backhans & Hemmingsson, 2012; Howe, Levy, &
Corresponding Author: Sarah E. Zemore, Ph.D., Alcohol Research Group, 6475 Christie Ave., Suite 400, Emeryville, CA 94608-1010,
szemore@arg.org, Tel: 510-597-3440; Fax: 510-985-6459.
Conflict of interest: None.
HHS Public Access
Author manuscript
J Ethn Subst Abuse
. Author manuscript; available in PMC 2018 July 01.
Published in final edited form as:
J Ethn Subst Abuse
. 2017 ; 16(3): 314–327. doi:10.1080/15332640.2016.1209144.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Caplan, 2004), increased psychological distress (Catalano et al., 2011; Flint, Bartley,
Shelton, & Sacker, 2013; Paul & Moser, 2009; Price et al., 2002; Wanberg, 2012), and
changes in health-related behaviors, including heavy drinking (Dooley & Prause, 1998;
Eliason & Storrie, 2009). Effects of job loss on alcohol use and problems are particularly
relevant to understanding, and intervening on, racial/ethnic health disparities because of the
large impacts of heavy and chronic alcohol use on health. However, rigorous research
examining racial/ethnic differences in the associations between job loss and alcohol
outcomes remains extremely limited.
One key limitation of existing research is that longitudinal studies of associations between
job loss and alcohol outcomes remain rare. Such studies are needed to support or disconfirm
causal associations between job loss and alcohol use and problems, particularly given that
selection effects (such as the overrepresentation of heavy drinkers among those who
experience job loss) may cloud the interpretation of effects. Indeed, reviews of the literature
have consistently called for better study designs that can address selection effects. For
example, McKee-Ryan et al. (2005) have argued, “Studies need to be designed to strengthen
causal inferences that can be drawn from them regarding the impact of unemployment on
employee well-being” (p. 69), and Wanberg (2012) has called for “more work designed to
carefully attend to selection issues” (p. 388).
Additionally, very little research targets racial/ethnic differences in associations between job
loss and alcohol outcomes. Multiple, recent reviews of the literature on job loss and health
have identified research on racial/ethnic minorities as a high priority (Paul & Moser, 2009;
Wanberg, 2012). Studies on unemployment have rarely examined racial/ethnic differences
and, where they have, have typically aggregated minorities—an extremely problematic
approach given that results may mask important differences between Blacks and other racial/
ethnic groups. Even so, existing evidence tentatively suggests potential racial/ethnic
differences in the effects of job loss. Supporting that point, Paul and Moser (2009) reported
in their systematic review that cross-sectional studies including higher percentages of
“minority” respondents produced stronger associations between unemployment and poorer
mental health, though this effect was only marginally significant. (Power limitations
prevented these authors from also examining moderator effects in findings from longitudinal
studies.) Further, a few individual-level studies have reported stronger associations between
job loss/unemployment and poorer mental/physical health among Blacks than Whites,
including studies of men experiencing the 1987 General Motors plant shutdowns (Hamilton,
Broman, Hoffman, & Renner, 1990) and men interviewed in 1976 and 1981 for the National
Longitudinal Surveys of Labor Market Experiences (Rushing, Ritter, & Burton, 1992).
Meanwhile, cross-sectional analyses of the 2010 National Alcohol Survey, a nationally
representative study conducted by the Alcohol Research Group, showed that associations
between job loss during the 2008–9 recession and both heavy drinking and alcohol
dependence symptoms were stronger among Blacks than Whites (Jones-Webb, Karriker-
Jaffe, Zemore, & Mulia; Zemore, Mulia, Jones-Webb, Lui, & Schmidt, 2013). Also relevant,
Lo et al. (2015), analyzing the 1979 National Longitudinal Survey of Youth (or NLSY, with
11 waves of alcohol measurement since recruitment in 1979), reported that associations
between weeks of unemployment and frequency of heavy drinking were stronger for Blacks
than Whites. These two studies suggest stronger associations between job loss and problem
Zemore et al.
Page 2
J Ethn Subst Abuse
. Author manuscript; available in PMC 2018 July 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
drinking among Blacks than Whites despite the fact that, overall, Blacks in the U.S. typically
(and historically) have reported similar or lower rates of any drinking, heavy drinking, and
alcohol use disorders compared to Whites (Grant et al., 2015; Kerr, Mulia, & Zemore, 2014;
Mulia, Ye, Greenfield, & Zemore, 2009; Zemore, Karriker-Jaffe, & Mulia, 2013). Yet,
individual-level studies on job loss and alcohol outcomes
per se
remain extremely sparse,
and to date there are no known longitudinal studies examining racial/ethnic disparities in
associations between job loss and alcohol dependence. Thus, racial/ethnic disparities in the
effects of job loss on alcohol outcomes remain unclear.
We address the above limitations by examining Black-White disparities in relationships
between job loss and alcohol dependence within a longitudinal context. The paper’s primary
objective is to elucidate possible differences in the impact of job on alcohol dependence
across Blacks and Whites. To address this objective, we directly explore (and analytically
address) the potential for selection biases that could affect the estimated effects, such as
differential selection into job loss by heavier drinkers. Following Lo et al. (2015), we adopt
the 1979 NLSY as our data source. These data offer a uniquely strong basis for examining
racial/ethnic disparities in the effects of job loss given the NLSY’s multi-wave assessment;
rigorous measures of employment and alcohol variables; and large oversample of Black
respondents. By addressing the above question, the current study should help to clarify the
societal costs of unemployment and identify populations at greatest risk for problematic
alcohol use and consequent health repercussions in times of economic downturn.
Method
Data Source
The 1979 National Longitudinal Survey of Youth (NLSY79) is an ongoing panel study
conducted by the Bureau of Labor Statistics of young men and women first recruited in 1979
(Bureau of Labor Statistics, 2006; Center for Human Resource Research, 1995).
Investigators used a stratified, clustered design to select a nationally representative sample,
achieving a baseline response rate of ~91%. Surveys focus on education and labor force
behavior, but also assess family income and assets, mental and physical health outcomes,
health behaviors, healthcare use, marital history, and other topics. Alcohol dependence
measures were administered in 1989 and 1994 only; hence, analyses target these time points.
The NLSY79 sample included 12,686 respondents aged 14–21 when first surveyed and 24–
31 in 1989. Respondents were interviewed, primarily in-person, annually through 1994.
Response rates at each wave through 1994 exceeded 90%. For the current analysis, we
selected only respondents who were employed (either part- or full-time) in 1989 and
surveyed in 1994. Because of limited sample sizes for individual racial/ethnic groups, we
also restricted the sample to those identifying as Black or White. Further, we excluded
oversamples of economically disadvantaged Whites and the military oversample from all
analyses, as these subgroups were not followed through 1994 (study N’s= 1,791 Blacks and
2,898 Whites). Our choice of measures and analysis capitalized on the longitudinal data
structure to ensure complete lagging between alcohol control variables (1989), job loss
(1990–93), and alcohol dependence (1994).
Zemore et al.
Page 3
J Ethn Subst Abuse
. Author manuscript; available in PMC 2018 July 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Measures
Employment Variables—All surveys assessed current (past-week) employment using
items and variable definitions consistent with those of the U.S. Census Bureau’s Current
Population Survey (Bureau of Labor Statistics, 2016). The current analysis adopted an
NLSY-created variable coding respondents, based on their employment activities, as
employed full/part-time, unemployed, or out of the labor force for each year. We also created
summary variables indicating any (vs. no) unemployment (job loss) from 1990–93, and
separately any period (vs. none) out of the labor force from 1990–93.
Alcohol Variables—
Past-year alcohol dependence
symptoms were assessed among
respondents who drank in 1989 and 1994 using measures based on the APA’s DSM-IV
(American Psychiatric Association, 1994). Twelve questions tapped symptoms in 7 domains,
including tolerance (2), withdrawal (2), drinking larger amounts/for longer than intended (2),
unsuccessful attempts to cut down/quit (2), neglect of important activities (1), excessive time
spent on drinking and recovery (1), and continued use despite physical/psychological
problems (2). Respondents were coded as positive on dependence if they reported 3+
symptoms at least once in the past year, and total count of positive domains in 1989 was
used as a covariate. Supporting measure validity, rates of abuse and dependence as measured
in the 1989 NLSY were highly similar to population rates obtained in the 1988 National
Health Interview Survey (Harford & Grant, 1994). Further, NLSY studies have linked
alcohol dependence in 1989 and 1994 to earlier age of onset and other theoretically relevant
variables (e.g., gender, age) as hypothesized (Dooley, Prause, Ham-Rowbottom, & Emptage,
2005; Grant, Stinson, & Harford, 2001). Additionally,
past-month frequency of binge
drinking
was determined for both 1989 and 1994 using responses to a quantity-frequency
measure assessing frequency of drinking 6+ drinks on one occasion in the past 30 days.
Respondents were coded into 3 categories: none, less than weekly, or weekly or more.
Demographic Variables—
S
elf-reported racial/ethnic origin in 1989 was used to
categorize respondents into White/Black/Other. Additional demographics, also assessed in
1989, included gender (male vs. female), age (continuous), marital status (never married and
separated/widowed/divorced vs. married), highest grade completed (continuous), and net
family income, coded using a quartile split and with missing as a separate category.
Analysis
The analysis was conducted in two parts. First, we explored associations between job loss
from 1990–93 and demographic and alcohol variables as assessed in 1989, separately by
race/ethnicity. The purpose of this analysis was to inform the core analyses by establishing
key correlates of job loss, such as
prior
heavy drinking. Disaggregation by race/ethnicity
allowed us to examine whether predictors of job loss might vary across race/ethnicity. If so,
then standard approaches to statistical control would be inappropriate. Next, we explored the
effects of job loss from 1990–93 on subsequent drinking (i.e., our main question of interest)
in both bivariate and multivariate models. Specifically, we tested bivariate associations
between job loss from 1990–93 and alcohol dependence in 1994, and used preliminary
logistic regressions
incorporating interaction terns
to test whether associations between
changes in employment and alcohol dependence differed statistically across Whites and
Zemore et al.
Page 4
J Ethn Subst Abuse
. Author manuscript; available in PMC 2018 July 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript
Blacks. Our final analyses were informed by these preliminary tests and involved
hierarchical logistic regressions, conducted separately by race/ethnicity, first with 1)
employment variables only, then adding 2) demographic variables (assessing demographic
confounding), and finally adding 3) heavy drinking and dependence symptoms in 1989
(assessing selection into job loss by heavier drinkers). NLSY79 custom sample weights were
used for all analyses, which adjust for oversampling of racial/ethnic groups and differential
attrition over the 15-year study period. All analyses were conducted with Stata v.13.0 and
using survey commands and robust standard errors to account for clustering within primary
sampling units.
Results
Predictors of Job Loss
Descriptive statistics are shown in Table 1. Incidence of job loss from 1990–93 among
Blacks (at 18.4%, N=329) was about double that among Whites (at 9.9%, N=301). Yet,
exploratory analyses reveal a similar pattern of associations between job loss and baseline
characteristics across Whites and Blacks. Among both racial/ethnic groups, job loss was
significantly associated with unmarried status (vs. married status, p’s<.001), lower family
income (in linear fashion, p’s<.001), and lower educational attainment (such that among
Whites, those without a high school diploma were at greater risk than high school graduates
and those with any college, the latter two groups showing similar risk; among Blacks, all
those without a college education were at elevated risk compared to those with a college
education, p’s<.001). Age and gender were unrelated to job loss for both. Among Whites,
job loss was significantly associated with a higher likelihood of any past-month binge
drinking in 1989 and alcohol dependence in 1989 (p’s<.001); among Blacks, job loss was
significantly associated with binge drinking in 1989 only (p<.001), though prevalence of job
loss among those with vs. without alcohol dependence in 1989 was nonsignificantly greater
(at 24.6 vs. 18.2%).
Job Loss as a Predictor of Alcohol Dependence
Turning to our bivariate tests of associations between job loss and alcohol dependence (not
shown), among Whites, job loss during 1990–93 was significantly and strongly associated
with dependence in 1994 (with rates = 9.1% vs. 3.3% for those with vs. without job loss, p<.
01); unexpectedly, the bivariate association between 1990–93 job loss and dependence in
1994 among Blacks was nonsignificant, though again rates were in the expected direction
(i.e., 5.1% vs. 3.8% for those with vs. without job loss). Results from a logistic regression
predicting alcohol dependence in 1994 from race/ethnicity, any job loss in 1990–93, any
period out of the labor force in 1990–93, and interactions between race/ethnicity and the
latter variables showed significant interactions between race/ethnicity and both employment
variables (both
p
’s<.05). These interactions indicate significantly different associations
between job loss (as well as time out of the labor force) and alcohol dependence across
Whites and Blacks.
Last, Table 2 shows the results of our hierarchical logistic regressions predicting alcohol
dependence from employment variables and covariates, separately by race/ethnicity.
Zemore et al.
Page 5
J Ethn Subst Abuse
. Author manuscript; available in PMC 2018 July 01.
Author Manuscript Author Manuscript Author Manuscript Author Manuscript