Adolescent alcohol use, sociability and income as a young adult

Abstract

We use data from the National Longitudinal Survey of Adolescent Health (Add Health) to study how sociability and adolescent alcohol use impact personal income as a young adult. We find that factors which enhance not only individual sociability but also social interaction at the community level are positively linked to future earnings of adolescents. Adolescents whose friends and friends of friends have greater sociability reap long-term labor market rewards into adulthood. After adjusting for individual and community sociability, the effect of teenage alcohol consumption on labor market earnings as young adults is reduced. Our results suggest that earnings premiums associated with adolescent alcohol consumption may be partially explained by social network dynamics.

I. Introduction

Adolescent alcohol use may have long-term negative consequences. Notably, underage drinking may lead to poor labor market outcomes in the future. Estimating the effects of adolescent alcohol intake on future earnings is fundamental and timely in view of findings linking early alcohol use to increased risk of neurodegeneration and neurocognitive disorders (Zeigler et al., 2005) and to lower educational attainment (Renna, 2007; Staff et al., 2008).

To the best of our knowledge, only one study (Chatterji and DeSimone, 2006) investigated the impact of teen drinking on young people’s labor market outcomes and found a wage premium for adolescent drinkers. This finding is counter-intuitive, but it is consistent with adult drinking and income studies, which reveal a positive association between adult drinking and wages and/or income (French and Zarkin, 1995; Hamilton and Hamilton, 1997; Berger and Leigh, 1988; Zarkin et al., 1998; Barrett, 2002; van Ours, 2004; Bray, 2005). However, most of the published literature on alcohol use and labor market outcomes contain two empirical problems: (1) simultaneous equation bias and/or (2) omitted variables bias (Peters, 2004).

Instrumental variables (IV) estimation and fixed-effects modeling can at least partially address the first problem (Berger and Leigh, 1988; Heien, 1996; Hamilton and Hamilton, 1997; MacDonald and Shields 2004; Peters, 2004). Unfortunately, the strength and reliability of available instrumental variables in alcohol research are often debatable and may fail to remove bias in the single-equation estimates (French and Popovici, 2011). Fixed-effect modeling requires longitudinal datasets which are not always accessible or suitable to address the issue at hand. Also, fixed-effects models depend on within-person variation, and if predictor variables vary greatly across individuals, but have little variation within an individual over time, the fixed effects estimates will be imprecise.

The second empirical concern, omitted variables bias, relates to the possibility that some unobserved or unmeasured characteristic may correlate with drinking and thereby create bias in the estimated coefficients. Alcohol may be a marker for personality traits that contribute to higher earnings (Lye and Hirschberg, 2010). For example, alcohol is positively associated with sociability (Cook et al., 1998). Sociability, or the ability to create relatively weak, but numerous, relationship ties, contributes to employment success as well (Granovetter, 1983). Sociability may be one of the relevant unobserved factors that are associated with drinking, which distorts the true effect of drinking on earnings (Peters, 2004; van Ours, 2004; Chatterji and DeSimonne, 2006).

The primary objective of the current study is to develop a measure of sociability and determine whether this hard-to-measure characteristic is a significant variable in the alcohol use-labor outcome relationship. To measure sociability, we rely on the concept of social capital (Coleman, 1988; Lin, 1999), which is closely related to employment outcomes (Granovetter, 1983; Coleman, 1990; Montgomery, 1992; Putnam, 1995; Fukuyama, 1995). Unfortunately, the definition of social capital is not well conceptualized. Social ties, trust, and social participation are often referred to in the literature as indicators of social capital. (Coleman, 1990; Fitzpatrick and LaGory, 2000; Halpern, 2005; Rupasingha et al., 2006). Furthermore, social capital has been viewed as both an individual trait and as a community characteristic (Kawachi, 1999; Woolcock, 2001).

To operationalize sociability in our alcohol-earnings equation, we measure social participation as the number of social contacts over a fixed period of time (Lindstrom et al., 2003; Grootaert et al., 2004; Newton and Montero, 2007). Specifically, frequency of meeting socially with friends and relatives in informal settings will count as informal social participation. Frequency of interactions related to established social organizations and clubs will count as formal social participation (Folland, 2007). The number of social ties and social trust as measures of sociability will be added to the alcohol-income model as well. Individual and community level sociability characteristics will also enter into the empirical model.

To estimate the effects of adolescent drinking on labor market outcomes of young adults, we take advantage of the wealth of data offered by the National Longitudinal Study of Adolescent Health (Add Health), a nationally representative U.S. survey of seventh through twelfth grade students enrolled in 1995–1996 at Wave I and a subsequent follow-up in 2007–2008 at Wave IV. The remainder of the paper is organized as follows. The data and variable definitions are presented in Section II. The empirical model is introduced and explained in Section III. The fourth section reports the estimation results. The final section offers discussion and conclusions.

II. Methods

Data

The analysis uses data from Waves I and IV of the National Longitudinal Survey of Adolescent Health (Add Health). Add Health is a nationally representative survey of US adolescents conducted between 1995 (Wave I), and 2008 (Wave IV). Add Health participants were in grades 7 to 12, ages 12–18 years old, at the Wave I survey. The Add Health study used stratified sampling to enroll high schools that were representative of schools nationwide based on region of the country, urbanicity, school funding, and racial composition. The study enlisted corresponding middle school and junior high feeder schools for the participating high schools. Add Health was approved by the Institutional Review Board of the University of North Carolina-Chapel Hill.

All 7th through 12th graders at the 132 participating schools were invited to complete an In-School Survey. Students who responded to the In-School Survey (n=90,118) were then eligible to be randomly selected for a detailed in-home interview and parent survey (Wave I, n=20,745). Wave I was conducted from April 1995 to December 1995. At Wave IV, administrators re-contacted the original Wave I respondents between April 2007 and February 2008. At the time of the interview, Wave IV participants (n=15,701) were 24 to 32 years old.

About 5% of Wave IV respondents (n=779) did not answer the question on personal earnings and were excluded from the analysis sample. Another 2,956 subjects (18.8%) reported that they were not currently working for pay more than 10 hours per week or reported personal earnings of less than $500 per year and were also omitted from the analysis. The study sample is comprised of 11,966 (76.2%) participants who responded to the Wave IV earnings question, who were working for 10 or more hours per week, and who earned $500 or more in personal earnings in the past year.

Variable definitions

Table 1 presents definitions for the variables included in the core models and the sources of data from within the Add Health study. Variables are divided into several groups and discussed below.

Table 1.

Analysis Variables from the Add Health Survey

Variable	Description	Source
Outcome Variable
Ln(income)	Log of annual income as a young adult	Wave IV In-Home Survey
Alcohol Use as Adolescent
abstain	No alcohol use in past 12 months	Wave I In-Home Survey
binging	Days drinking five or more drinks in a row	Wave I In-Home Survey
Demographics
male	Male gender	Wave I In-Home Survey
age	Age	Wave I In-Home Survey
agesq	Age squared	Wave I In-Home Survey
grade	School grade	Wave I In-Home Survey
white	Non-Hispanic white race	Wave I In-Home Survey
black	African American race	Wave I In-Home Survey
nativeamer	Native American race	Wave I In-Home Survey
asian	Asian race	Wave I In-Home Survey
hispanic	White Hispanic race/ethnicity	Wave I In-Home Survey
Individual Characteristics
englishgpa	Grade in most recent English course	Wave I In-Home Survey
mathgpa	Grade in most recent Math course	Wave I In-Home Survey
delinquency	Score on 15-item delinquency scale	Wave I In-Home Survey
experience	Work experience in years	Wave IV In-Home Survey
Household Characteristics
momnograd	Mother did not graduate high school	Wave I In-Home Survey
momhsgrad	Mother high school graduate	Wave I In-Home Survey
momsomecollege	Mother attended college, did not graduate	Wave I In-Home Survey
momcollegegrad	Mother graduated college	Wave I In-Home Survey
dadnograd	Father did not graduate high school	Wave I In-Home Survey
dadhsgrad	Father high school graduate	Wave I In-Home Survey
dadsomecollege	Father attended college, did not graduate	Wave I In-Home Survey
dadcollegegrad	Father graduated college	Wave I In-Home Survey
ln(hh_income)	Natural log of household income	Wave I Parent Survey
Neighborhood Characteristics
censusincome	Median income in US census block	1990 U.S. Census*
safeneighborhood	Feeling safe in neighborhood	Wave I In-Home Survey
samehome	Same home as when child was born	Wave I Parent Survey
Social Characteristics
sports	Participation on sports teams	Wave I In-Home Survey
clubs	Participation in school clubs	In-School Survey
studentcouncil	Participated in student council	In-School Survey
friendnomination	Number of friend nominations received	In-School Survey
friendoffriends	Frequency that friends of friends meet	In-School Survey
hangoutwfriends	Frequency of hanging out with friends	Wave I In-Home Survey
phonemalefriend	Phoned male friend in past week	Wave I In-Home Survey
phonefemalefriend	Phoned female friend in past week	Wave I In-Home Survey
talkparentsfriend	Talked with parents about friends	Wave I In-Home Survey
videocomputer	Weekly hours of video games/computer	Wave I In-Home Survey
rollerblade	Frequency rollerblading or skateboarding	Wave I In-Home Survey

^*Add Health linked respondent addresses and/or GPS locations to U.S. Census data to provide community contextual files linked to Wave I survey responses

Personal income

At Wave IV, subjects replied to, ‘Are you currently working for pay at least 10 hours a week?’ and ‘Now think about your personal earnings. In the last year {2006/2007/2008}, how much income did you receive from personal earnings before taxes, that is, wages or salaries, including tips, bonuses, overtime pay, and income from self-employment?’ Due to the skewed nature of the income responses, we took the natural log of this measure and used the natural log of income in all models.

Alcohol use

At the Wave I interview, students responded to ‘During the past 12 months, on how many days did you drink alcohol?’ and ‘During the past 12 months, on how many days did you drink five or more drinks in a row?’ For the purpose of the analysis, binge drinking was defined as having five or more drinks in a row. Responses ranged from 0 (never) to 6 (every day or almost every day) on an interval scale. These intervals were used to characterize frequency of binge drinking. We also included in our models an indicator variable for students who abstained from alcohol during the past 12 months.

Demographics and individual characteristics

The models included age, age squared, years of work experience, race/ethnicity, self-reported grade point average in most recent English class, self-reported grade point average in most recent Math class, and a 15-item delinquency scale score (Haynie, 2001).

Household characteristics

Mother’s and father’s education and household income were obtained in a Parent Survey at Wave I.

Neighborhood characteristics

Add Health linked 1990 U.S. Census block data to student addresses or GPS locations. Median income per census block was included in the analysis. In Wave I, students provided a yes/no response to ‘Do you usually feel safe in your neighborhood?’ Parents also indicated whether they were living in the same home as when their child was born. Each of these neighborhood variables were entered into the models.

Sociability: friendship nominations/social ties

Add Health respondents chose from a school roster the names of their 5 best male and 5 best female friends. Using the friendship survey data, we computed the total number of friendship nominations an individual received from the other students in their school.

A Kernel function identified how often friends, friends of friends, and friends of friends’ friends got together in the peer social network. The Kernel function evaluated the socializing pattern within three steps of friendship separation from the individual, where a directly nominated friend was characterized as one step away, a friend of a friend was classified as being two steps away, and a friend of a friend of a friend who could not be reached in one or two steps was described as being three steps away (three-step reach). Categorical responses within three-step reach to the question ‘During the past week, how many times did you just hang out with friends?’ were summed across the degrees of friendship separation, with a decay parameter of 0.5 for friends two steps away, and 0.25 for friends three steps away.

Sociability: trust

Trust was measured as the strength of relationships with friends and parents. We used student reports of talking with their friends on the phone in the past 7 days and talking with their parents about their friendships.

Sociability: social participation

Wave I respondents answered questions about their participation in informal social activities, such as getting together with friends outside of school. In addition, participants reported formal social involvement such as sports, clubs, and student council. Study participants also provided the number of hours per week they play video or computer games and number of times rollerblading/skateboarding/bicycling.

Empirical model

We estimated the effects of adolescent alcohol use on labor market outcomes using hierarchical linear models (HLM). Add Health respondents were sampled within schools. HLM separates between-school effects from within-school effects without assuming that individual observations are independent and that error terms are uncorrelated. As a result, HLM corrects parameter and standard error estimates to account for the multi-level structure of the data.

We estimated separate HLM models for men and women to avoid the confounding effects of gender and to allow the coefficients to differ for each group. The study used three HLMs to address our primary objectives. The parsimonious HLM estimated the effect of teen alcohol use in Wave I on natural log of income in Wave IV while controlling for a small set of demographic and background characteristics. This model provided a baseline for the degree to which Wave I adolescent alcohol use impacts earnings in Wave IV while controlling for school effects and individual characteristics. Specifically, we estimated the following equation:

• Ln(Income_ij) = α₀ + α₁ ABSTAIN_ij + α₂ BINGE_ij + α₃ X_ij + α₄ z_j + u_ij

where Income_ij is the individual’s Wave IV income, ABSTAIN_ij is an indicator for any alcohol use at Wave I, BINGE_ij is an interval variable for frequency of Wave I binge drinking, X_ij is a vector of individual covariates, z_j is a vector of school indicator variables, and u_ij is the error term.

To account for the influence of systematic differences in family background or neighborhood characteristics, the second HLM estimated the effect of adolescent alcohol consumption on young adult earnings with a number of parental and neighborhood characteristics held constant. Parental variables included mother’s and father’s education and natural log of household income. Neighborhood characteristics consisted of census block median income, neighborhood safety, and whether the family had changed homes since the child was born.

The second HLM takes the following explicit form:

• Ln(Income_ij) = α₀ + α₁ ABSTAIN_ij + α₂ BINGE_ij + α₃ X_ij + α₄ z_j + α₅ FAM_ij + u_ij

which is identical to the parsimonious model except that a vector of family and neighborhood characteristics, FAM_ij, is added to the model.

The fully augmented HLM incorporated social participation, social ties, and trust variables into the analysis. Social participation included both participation in formal activities, such as sports, clubs, and student council, and informal activities, such as frequency of getting together with friends outside of school. Trust included talking with a friend on the phone and talking with a parent about friends. Social ties consisted of number of friendship nominations from other students. A community sociability ‘friends of friends’ variable, constructed to tap into social contacts beyond dyadic relationships within three steps reach in a localized network of friends, was also added.

The fully augmented model takes the following form:

• Ln(Income_ij) = α₀ + α₁ ABSTAIN_ij + α₂ BINGE_ij + α₃ X_ij + α₄ z_j + α₅ FAM_ij + α₆ SOC_ij + u_ij

where SOC_ij, a vector of sociability variables, has been added to the model. All analyses used the PROC MIXED procedure in SAS.

III. Results

Table 2 presents descriptive statistics for all respondents in the analysis sample. Half of the participants are male. Over 40% of the sample are minorities, with 21% African-American and 14% Hispanic. Approximately 40% of the adolescents’ parents had attended college, with close to 30% of parents graduating from college. Respondents reported participating in sports roughly 2–3 times per week, on average. 20% of students belonged to a school club. The average number of friendship nominations from other students within the school was 4.6. Most students reported getting together with friends outside of school 3–4 times per week, and two thirds reported that they had talked with a friend on the phone in the past 7 days.

Table 2.

Descriptive Statistics of the Analysis Sample (n=11,966)

Variable	Mean	SD	Min	Max
Income
Income at Wave IV, past 12 months	40,950	46,976	500	999,995
Ln(Income)	10.342	0.787	6	14
Alcohol Use
Abstained from alcohol, past 12 months	0.523	0.499	0	1
Binge drinking frequency, past 12 months	0.631	1.276	0	6
Demographics
Male	0.501	0.500	0	1
Age, mean	15.655	1.717	12	18
Grade Level (%)
7th grade	0.130	0.337	0	1
8th grade	0.132	0.338	0	1
9th grade	0.178	0.382	0	1
10th grade	0.194	0.395	0	1
11th grade	0.193	0.395	0	1
12th grade	0.173	0.379	0	1
Race
Non-Hispanic white	0.567	0.496	0	1
Black	0.208	0.406	0	1
Native American	0.013	0.115	0	1
Asian	0.067	0.250	0	1
White Hispanic	0.144	0.352	0	1
Individual Characteristics
Grade point average in most recent English class	2.846	0.935	1	4
Grade point average in most recent Math class	2.681	0.991	1	4
Delinquency score	4.228	5.151	0	45
Years of work experience	7.809	3.142	0	15
Household Characteristics
Parental Education
Mom <HS graduate	0.166	0.385	0	1
Mom HS graduate	0.415	0.493	0	1
Mom some college	0.137	0.344	0	1
Mom college degree	0.282	0.456	0	1
Dad <HS graduate	0.219	0.428	0	1
Dad HS graduate	0.384	0.486	0	1
Dad some college	0.114	0.317	0	1
Dad college degree	0.283	0.451	0	1
Ln (household Income ($1,000))	3.619	0.741	0	7
Neighborhood Characteristics
Median income-US census block ($1,000)	35.514	8.183	15	61
Neighborhood safe	0.892	0.310	0	1
Same house since birth	0.201	0.401	0	1
Sociability
Sports	1.415	1.148	0	3
Clubs	0.196	0.397	0	1
Student council	0.069	0.253	0	1
Friendship nominations	4.577	3.100	0	32
Get together with friends	1.987	0.991	0	3
Talk on phone with male friend	0.664	0.472	0	1
Talk on phone with female friend	0.696	0.460	0	1
Talk with parent about friends	0.456	0.498	0	1
Friends of friends (three step reach) get together	4.285	12.013	0	124
Video or computer games, hours per week	2.671	6.126	0	99
Rollerblading/skateboarding/bicycling	0.623	0.933	0	3

Table 3 presents unadjusted means for the independent and dependent variables in the analysis, grouped by drinking status and gender. 52% of the males and 53% of the females in the sample abstained from using alcohol in the past 12 months, while 30% of male students and 23% of female students were binge drinkers. The unadjusted income data are consistent with the premise that there is an adolescent drinking effect on young adult incomes, with drinkers earning more than nondrinkers. The earnings premiums operated somewhat differently for males and females. For males, earnings were higher for alcohol drinkers than nondrinkers, and higher still for adolescent binge drinkers. For females, however, the drinking category that generated the highest earnings was any alcohol consumption, but not binge drinking.

Table 3.

Variable Means by Adolescent Drinking Status (n=11,966)

	Males			Females
	Abstainer	Nonbinge	Binge	Abstainer	Nonbinge	Binge
		Drinker	Drinker		Drinker	Drinker
	(N=3,113)	(N=1,103)	(N=1,779)	(N=3,147)	(N=1,467)	(N=1,357)
Young Adult Income (Wave IV)
Income ($1,000)	45.0*	45.4*	49.2*	34.7	36.8	36.3
Ln(Income)	10.44***	10.48***	10.54***	10.17***	10.27***	10.21***
Demographic, Family, and Individual Characteristics (Wave I)
Age	15.3***	15.9***	16.5***	15.2***	15.8***	16.2***
Non-Hispanic White	.534***	.585***	.683***	.500***	.554***	.646***
African-American	.231***	.183***	.090***	.283***	.248***	.117***
Native American	.011	.012	.015	.016***	.005***	.023***
Asian	.082***	.072***	.053***	.072**	.050**	.052**
White Hispanic	.143	.148	.159	.130*	.143*	.161*
Grade in most recent English class	2.78***	2.67***	2.48***	3.11***	2.98***	2.86***
Grade in most recent Math class	2.69***	2.61***	2.50***	2.85***	2.72***	2.52***
Delinquency score	3.07***	5.12***	8.04***	2.18***	4.15***	6.00***
Years work experience	7.45***	8.15***	9.27***	7.00***	7.67***	8.46***
Mother high school graduate	.406	.424	.430	.402	.423	.427
Mother some college	.116***	.163***	.148***	.139	.142	.143
Mother college graduate	.306***	.291***	.254***	.283**	.290**	.244**
Father high school graduate	.367*	.381*	.405*	.373	.393	.410
Father some college	.109	.133	.117	.104	.127	.111
Father college graduate	.296**	.320**	.261**	.289*	.273*	.248*
Household income	43.7***	49.2***	49.0***	44.9**	47.4**	49.4**
Median income-census block	35.2*	36.0*	35.4*	35.0***	35.9***	36.6***
Neighborhood safe	.895	.914	.904	.874	.889	.896
Same house since birth	.214*	.183*	.193*	.205	.200	.187
Sociability (Wave I)
Sports	1.80***	1.65***	1.64***	1.14*	1.09*	1.04*
Clubs	.220	.216	.198	.304	.310	.272
Student council	.050**	.028**	.039**	.092	.094	.104
Friendship nominations	4.22***	4.46***	4.86***	4.44***	4.97***	5.03***
Get together with friends	1.90***	2.04***	2.28***	1.78***	2.01***	2.23***
Talk on phone with male friend	.667***	.729***	.772***	.527***	.694***	.749***
Talk on phone with fem. friend	.493***	.614***	.699***	.774***	.866***	.864***
Talk with parent about friends	.308***	.422***	.496***	.442***	.599***	.643***
Friends of friends (three step reach)	2.46***	2.83***	3.69***	2.24**	2.44**	2.89**
Video or computer games	4.54***	4.26***	2.99***	1.47***	1.22***	1.04***
Rollerblading/bicycling	.859***	.788***	.634***	.524***	.404***	.400***

^*denotes statistical significance at the 5% level

^**denotes statistical significance at the 1% level

^***denotes statistical significance at the 0.1% level

Importantly, there were many significant differences in individual characteristics and family backgrounds between the drinking groups, across both genders. In particular, Table 3 shows that both male and female drinkers and binge drinkers were older than nondrinkers, were more likely to be non-Hispanic whites, had lower grades, and were more likely to commit delinquent acts. Notably, for both males and females, binge drinkers were less likely to have parents who graduated from college, but their parents reported, on average, higher household incomes.

Table 3 also presents sociability characteristics by drinking groups and gender. Alcohol-using adolescents of both genders participated in fewer organized social activities―sports, clubs, and social council―than nondrinkers, but took part in more informal social activities and showed a greater sense of trust in their relationships. The drinkers’ and binge drinkers’ community of friends within three step reach also interacted with each other more frequently than did the friends of nondrinkers within three step reach.

Table 4 presents the results of the hierarchical linear models for males. As shown in the first panel, after adjusting for differences in demographic and other characteristics, young men who consumed alcohol as adolescents earned 1.8% more than nondrinkers, and an additional 3.1% with each increase in binge drinking frequency. Panel 2 of Table 4 added family background and neighborhood conditions to the model. After controlling for parental and neighborhood characteristics, the earnings premiums for adolescent drinking persisted. Moreover, binge-drinking adolescent males earned 3.3% more with each increase in binge drinking frequency than non-binge drinkers.

Table 4.

HLM Model of Wave IV Ln(Income) for Males (n=5,995)

	Model 1			Model 2			Model 3
Parameter	Beta	se	p-value	Beta	se	p-value	Beta	se	p-value
Alcohol Use
Abstained from alcohol	−0.018	0.024	0.438	−0.004	0.023	0.857	0.011	0.023	0.639
Binge drinking frequency	0.031	0.008	<.001	0.033	0.008	<.001	0.029	0.008	<.001
Demographics
Age	0.095	0.024	<.001	0.095	0.023	<.001	0.071	0.023	0.003
Age squared	−0.007	0.003	0.017	−0.008	0.003	0.013	−0.006	0.003	0.071
Race
Non-Hispanic white	0.043	0.033	0.188	−0.027	0.033	0.419	−0.021	0.033	0.519
African American	−0.189	0.037	<.001	−0.221	0.038	<.001	−0.218	0.038	<.001
Native American	−0.105	0.090	0.245	−0.144	0.090	0.109	−0.127	0.089	0.154
Asian	0.118	0.046	0.011	0.060	0.046	0.195	0.080	0.046	0.084
White Hispanic	.	.	.	.	.	.	.	.	.
Individual Characteristics
Grade in most recent English class	0.099	0.011	<.001	0.088	0.011	<.001	0.077	0.011	<.001
Grade in most recent Math class	0.049	0.010	<.001	0.042	0.010	<.001	0.038	0.010	<.001
Delinquency score	−0.004	0.002	0.018	−0.005	0.002	0.007	−0.006	0.002	<.001
Years work experience	−0.015	0.004	<.001	−0.008	0.004	0.032	−0.005	0.004	0.162
Household Characteristics
Mom, high school grad				0.031	0.028	0.265	0.019	0.028	0.490
Mom, some college				0.069	0.035	0.049	0.053	0.035	0.126
Mom, college grad				0.064	0.032	0.048	0.050	0.032	0.120
Dad, high school grad				0.063	0.025	0.013	0.058	0.025	0.022
Dad, some college				0.085	0.034	0.013	0.070	0.034	0.040
Dad, college grad				0.110	0.030	<.001	0.100	0.030	<.001
Ln(household income)				0.066	0.015	<.001	0.060	0.014	<.001
Neighborhood Characteristics
Median income ($1,000)				0.007	0.002	<.001	0.007	0.002	<.001
Safe neighborhood				0.077	0.032	0.016	0.058	0.032	0.068
Same house since birth				0.049	0.023	0.035	0.049	0.023	0.037
Sociability
Sports							0.044	0.009	<.001
Clubs							0.007	0.026	0.789
Student council							0.036	0.047	0.448
Number of friendship nominations							0.007	0.003	0.025
Get together with friends							−0.003	0.010	0.790
Friends of friends get together (three step reach)							0.002	0.001	0.025
Talk on phone with male friend							0.047	0.021	0.025
Talk on phone with female friend							0.030	0.020	0.141
Talk with parent about friends							0.117	0.020	<.001
Play video or computer games							−0.003	0.001	0.005
Rollerblading/skateboarding/bicycling							−0.020	0.009	0.033
Intercept	10.708	0.062	<.001	9.999	0.099	<.001	9.871	0.103	<.001
DF	5849			5839			5828

Finally, we added sociability variables to the model as shown in Panel 3 of Table 4. When controlling for sociability factors, the earnings premium for binge drinking was lower, although still statistically significant. For males, we estimate that adolescent binge drinking was associated with a 2.9% increase in adult earnings, which is a 12% reduction in the earnings premium compared to the model in Panel 2. Interestingly, the sign of the parameter associated with abstaining from alcohol changed from negative to positive with the addition of the sociability factors, although neither of the coefficients was statistically significant. Social participation variables contributing to the earnings premium among males were sports (p<.001), talking with parents about friends (p<.001), number of friendship nominations (p=.025), being part of a sociable community where friends and friends of friends get together (p=.025), and talking on the phone with a male friend (p=.025). For males, playing more video games or computer games (p=.005) and participating in rollerblading, skateboarding, or bicycling (p=.033) were negatively associated with earnings as a young adult.

Table 5 provides the hierarchical linear model results for females. As seen in Panel 1 of Table 5, young women who drank alcohol as adolescents earned 9.0% more income as young adults than nondrinkers, but each increase in binge drinking frequency as an adolescent was associated with a 1.9% decrease in young adult earnings. The second panel of Table 5 reports the results for females after adding individual and household characteristics. The earnings premium for female adolescent alcohol use was 7.8% after adjusting for these additional factors compared to females who abstained from drinking as adolescents. Finally, after adding sociability factors, we estimate that alcohol use as an adolescent was associated with a 5.1% increase in young adult earnings for females. The earnings premium was reduced by 35% from the model presented in Panel 2. Females who had a greater number of friendship nominations (p<.001), talked on the phone with a male friend (p<.001), participated in student council (p=.005), and who got together more often with friends (p=.050) earned higher incomes.

Table 5.

HLM Model of Wave IV Ln(Income) for Females (n=5,971)

	Model 1			Model 2			Model 3
Parameter	Beta	se	p-value	Beta	se	p-value	Beta	se	p-value
Alcohol Use
Abstained from alcohol	−0.090	0.023	<.001	−0.078	0.023	<.001	−0.051	0.023	0.028
Binge drinking frequency	−0.019	0.010	0.062	−0.019	0.010	0.059	−0.023	0.010	0.026
Demographics
Age	0.107	0.025	<.001	0.108	0.024	<.001	0.088	0.024	<.001
Age squared	−0.004	0.003	0.199	−0.004	0.003	0.167	−0.002	0.003	0.530
Race
Non-Hispanic white	−0.062	0.036	0.082	−0.161	0.035	<.001	−0.167	0.035	<.001
African American	−0.095	0.038	0.012	−0.137	0.038	<.001	−0.135	0.038	<.001
Native American	0.018	0.085	0.833	−0.016	0.084	0.852	−0.020	0.083	0.806
Asian	0.098	0.051	0.056	0.009	0.050	0.855	0.013	0.050	0.803
White Hispanic	.	.	.	.	.	.	.	.	.
Individual Characteristics
Grade in most recent English class	0.130	0.012	<.001	0.110	0.012	<.001	0.103	0.012	<.001
Grade in most recent Math class	0.076	0.011	<.001	0.071	0.011	<.001	0.068	0.011	<.001
Delinquency score	−0.002	0.003	0.425	−0.002	0.003	0.482	−0.003	0.003	0.274
Years work experience	−0.039	0.004	<.001	−0.032	0.004	<.001	−0.032	0.004	<.001
Household Characteristics
Mom, high school grad				0.027	0.028	0.337	0.020	0.028	0.471
Mom, some college				0.100	0.035	0.004	0.087	0.035	0.012
Mom, college grad				0.119	0.033	<.001	0.105	0.033	0.001
Dad, high school grad				0.085	0.026	<.001	0.076	0.025	0.003
Dad, some college				0.098	0.035	0.006	0.085	0.035	0.016
Dad, college grad				0.124	0.031	<.001	0.110	0.031	<.001
Ln(household income)				0.074	0.015	<.001	0.078	0.015	<.001
Neighborhood Characteristics
Median income ($1,000)				0.011	0.002	<.001	0.011	0.002	<.001
Safe neighborhood				0.122	0.0301	<.001	0.117	0.030	<.001
Same house since birth				0.052	0.024	0.029	0.048	0.024	0.044
Sociability
Sports							0.000	0.009	0.971
Clubs							0.044	0.024	0.065
Student council							0.095	0.034	0.005
Number of friendship nominations							0.011	0.003	<.001
Get together with friends							0.019	0.010	0.050
Friends of friends get together (three step reach)							0.001	0.001	0.434
Talk on phone with male friend							0.076	0.021	<.001
Talk on phone with female friend							0.017	0.025	0.507
Talk with parent about friends							0.032	0.020	0.119
Play video or computer games							−0.003	0.003	0.331
Rollerblading/skateboarding/bicycling							−0.000	0.013	0.998
Intercept	10.689	0.064	<.001	9.752	0.100	<.001	9.599	0.104	<.001
DF	5826			5816			5805

Sensitivity Analysis

A supplementary analysis was conducted to examine the stability of the core findings. To test the eligibility criteria for inclusion in the analysis sample, we reconstructed the analysis sample by including all individuals with any positive amount of hours worked or earnings (n=13,853), and then re-estimated all of the models. The direction, magnitude, and significance of the key coefficients were identical or very similar to those when using the more restricted sample. In the fully-augmented model (i.e., including sociability factors), the binge drinking effect on young adult earnings among males fell by 17% and the negative effect on young adult earnings of abstaining from alcohol among females fell by 36% with the addition of sociability.

IV. Discussion

Much of the limited literature on the impact of adolescent drinking on young adult labor market outcomes has problems associated with omitted variables bias and/or unobserved heterogeneity. The present study attempts to overcome these empirical challenges by including hard-to-measure sociability characteristics into the analysis model. We take advantage of the recency and abundance of data provided by the nationally representative Add Health survey. The social participation responses, trust measures, and social ties variable, a unique feature of Add Health, allow us to measure adolescent sociability on an individual and community level.

The main study finding reveals that sociability as an adolescent is associated with higher earnings as a young adult and that, after adjusting for sociability, the effect of adolescent drinking on future earnings is somewhat reduced. These results are consistent with previous studies, which commented on sociability as an important omitted variable (Peters, 2004; van Ours, 2004; Chatterji and DeSimone, 2006).

Our results confirm the presence of gender differences in how sociability relates to labor market outcomes. For both males and females, the sociability construct relied on the number of social organizations the adolescents belonged to, the social ties they established, and on the degree of trust achieved in their relationships. For males, frequency of social contact through participation in sports and a greater number of friendship nominations added to the earnings premium in young adulthood. Furthermore, having stronger trust in their relationships, as measured by talking with friends over the phone in the past seven days and talking with parents about friends, was associated with greater income. Adolescent males who reported interest in more solitary pursuits such as skate-boarding, bicycling, and video or computer games had lower future earnings.

For females, social contact through student council and a greater number of social ties, as measured by friendship nominations, was associated with higher income. Speaking on the phone with a friend in the past seven days, which may be viewed as a higher degree of trust in the friendship, was also linked to earnings premiums. Interestingly, prior research indicates that women attain higher earnings when exhibiting conscientiousness and openness to others, personality traits that may be linked to student council participation (Mueller and Plug, 2006). A greater number of friendship nominations has also been associated with alcohol consumption (Balsa et al., 2011), which is in line with our claim that sociability is a key missing variable because it is correlated with both alcohol use and earnings.

Data limitations preclude us from investigating whether the strength of social relationships (i.e. trust), the number of social contacts through organizations, the number of social ties, or all of these factors combined contribute to the earnings premium in adulthood. This topic should be explored in future investigations. In view of the growing popularity of on-line social network activities (e.g. Facebook, Twitter), future studies may also wish to explore whether social contacts through digital medium also contribute to adolescent sociability and have an impact on young adult labor market outcomes.

We also find evidence that factors which enhance sociability on a community level are linked to future earnings of adolescents. Notably, our measure of community social contacts, defined here as frequency of social contact among friends within three step reach, is a significant determinant of young people’s market labor outcomes. This finding suggests that adolescents whose friends and friends of friends have greater sociability reap long-term earnings rewards in adulthood. This result is in line with medical studies showing that health behaviors and outcomes (e.g. smoking, drinking, obesity) spread through social networks (Christakis and Fowler, 2007). To the best of our knowledge, our study is first to uncover social network effects beyond dyadic relationships that influence future earnings of adolescents. This empirical verification of social network effects on labor market outcomes is important for future investigations of wages and earnings as well for educational programs and policies.

The persistence of a differential gender effect of adolescent alcohol use on future earnings, although somewhat mitigated by the introduction of sociability factors, is an interesting phenomenon. Several studies in the economics literature have found that gender differences are present in the relationship between alcohol use and labor market outcomes (Berger and Leigh, 1988; Dave and Kaestner, 2002; Kenkel and Ribar, 1994; Mullahy and Sindelar, 1991, 1996, 1997). In addition, drinkers in general, and sometimes heavy drinkers, tend to have higher earnings compared to abstainers or light drinkers (Berger and Leigh, 1988; Bray, 2005; French and Zarkin, 1995; Hamilton and Hamilton, 1997; Heien, 1996; Lye and Hirschberg, 2010; Zarkin et al., 1998). Several explanations are possible. First, female workers may consume alcohol as a way to connect with their male colleagues and enhance career growth. Second, many females stop drinking entirely or cut back considerably during years of family formation. These individuals also tend to work less and earn less. Third, regardless of gender, an income effect may be present as individuals consume more alcohol when their earnings rise because alcohol is a normal good. Finally, young males who consume relatively more alcohol may enter the labor market sooner, acquire more work experience, and thereby have higher earnings compared to those who drink less, pursue additional education, and have less labor market experience.

Limitations

Our study has several caveats. First, alcohol use is self-reported, although self-reported drinking is generally considered to be a valid measure (Babor et al., 2000). Second, young adult income is self-reported. Individuals are more likely to refuse to answer or misreport income relative to most other variables, but we have no reliable means of verifying income or wealth of the study participants. It should be noted, however, that the Add Health survey administrators were well-trained and experienced, which probably minimized measurement error in all areas. Third, while the study design benefits from the use of longitudinal data, it is not possible to conclusively identify the direction of causality between alcohol use and sociability. Sociability is generally perceived as a character trait, but it is possible that social skills are developed simultaneously with alcohol use behaviors. Fourth, the sociability variables are contemporaneous in nature, and the possible cumulative effect of social skills development over time is not clear. Finally, the study results are potentially affected by sample selection and attrition biases. Subjects were excluded from the analysis if they were not working or could not be located at the time of the Wave IV survey. However, over 75% of eligible participants completed Wave IV in-home interviews, making it unlikely that the results suffer from significant response bias. Nonresponse has been investigated by the Survey Research Unit at the University of North Carolina and findings showed that bias for measures of health and risk behaviors rarely exceeded 1% (Kalsbeek et al., 2001). It is also possible that individuals who choose to participate in the formal labor market are significantly different from those who choose not to participate. A Heckman sample selection model could identify and correct selection bias among labor market participants. However, the Add Health survey does not contain any viable measures that significantly predict the labor force participation decision without also being correlated with earnings. In prior research, limiting study samples to employed individuals did not significantly bias the estimated wage premiums (Zarkin et al., 1998; Peters, 2004).

Conclusion

Operationalizing sociability and then adding these constructs to income models is a new and meaningful contribution to the labor economics literature. We tested these models with data on adolescents and young adults to determine whether adolescent drinking was significantly related to future earnings. Our findings suggest that any alcohol consumption as well as binge drinking can have long-term effects on labor market outcomes. In addition, sociability is a significant moderating factor in these relationships, which vary by gender. Subject to data availability, we encourage future studies on this topic to consider including these and other measures of sociability in their empirical models.