Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Parental Problem-drinking and AdultChildren’s Labor Market Outcomes
Ana I. Balsa
a b s t r a c t
Current estimates of the societal costs of alcoholism do not consider theimpact of parental drinking on children. This paper analyzes theconsequences of parental problem-drinking on children’s labor marketoutcomes in adulthood. Using the NLSY79, I show that having a problem-drinking parent is associated with longer periods out of the labor force,lengthier unemployment, and lower wages, in particular for malerespondents. Increased probabilities of experiencing health problems andabusing alcohol are speculative forces behind these effects. While causalitycannot be determined due to imprecise IV estimates, the paper calls forfurther investigation of the intergeneration costs of problem-drinking.
I. Introduction
An important avenue of research in economics involves the identifi-cation and measurement of the economic costs of alcoholism. In most cases, thescope of this research has been limited to studying the consequences directly asso-ciated with the problem-drinking subject: for example, productivity losses, healthcare costs, and other costs directly resulting from the criminal activities of the sub-ject. Less is known, however, about indirect economic costs of alcoholism such as,for instance, the costs that one family member’s drinking inflicts upon other familymembers.
Ana I. Balsa is a visiting scientist with the Health Economics Research Group (HERG) at the Universityof Miami and a visiting professor at the University of Montevideo, Uruguay. Financial assistance forthis study was provided by the National Institute on Alcohol Abuse and Alcoholism (grant number R01AA13167). The author gratefully acknowledges comments and suggestions by attendants to the HERGresearch meetings, three anonymous referees, William Russell for editorial assistance, and Jamila Wadeand Colleen Trifilo for research assistance. The results, opinions, and positions presented in this paperreflect the views of the author alone and do not necessarily represent those of the University of Miamior the National Institute on Alcohol Abuse and Alcoholism. The data used in this article can beobtained beginning October 2008 through September 2011 from Ana I. Balsa, Ph.D., University ofMiami, Sociology Research Center, 5665 Ponce de Leon Boulevard, Flipse Building room 122, CoralGables, FL 33146-0719, U.S.A.; [email protected].[Submitted January 2006; accepted June 2007]ISSN 022-166X E-ISSN 1548-8004 � 2008 by the Board of Regents of the University of Wisconsin System
THE JOURNAL OF HUMAN RESOURCES d XLIII d 2
This paper takes advantage of the 1979 cohort of the National Longitudinal Surveyof Youth (NLSY79), a rich, nationally representative longitudinal data set, to analyzethe economic effects of parental problem-drinking on children. Specifically, the pa-per focuses upon labor market outcomes of adult children of alcoholics, includinglabor force participation, unemployment, and wages. Results are suggestive of inter-generation costs of problem-drinking, but the analysis cannot reach definite conclu-sions about the causality of the effects.
Alerting the public to the consequences of parental drinking has significant impli-cations for public policy. In general, the costs that children of alcoholics have to bearbecause of parental alcoholism are not considered when evaluating the economicconsequences of alcohol abuse, nor are they taken into account when designing rel-evant interventions. The current paper sheds new light on the economical signifi-cance of parental problem-drinking and emphasizes the importance of consideringthese costs when designing, evaluating, and funding interventions.
Section II surveys the previous literature on the effects of parental alcoholism inchildren and provides some insight on the pathways through which alcohol misuseby a parent could affect a child’s labor market outcomes at adulthood. A descriptionof the NLSY79 data, the variables used in the analysis, and the estimation methods isprovided in Section III. Section IV reports the results and checks for robustness usingalternate methodologies. In Section V, results are discussed and a conclusion isoffered.
II. Background and Significance
Most of the previous research on children of alcoholics has been per-formed in the fields of clinical psychology and family medicine. Findings suggestthat children of alcoholics are at higher risk than other children for depression, anx-iety disorders, problems with cognitive and verbal skills, and parental abuse or ne-glect during childhood (NIAAA 2000). The bulk of the research focuses on theshort- to medium-term sequels that parental alcohol misuse has on children. Al-though many studies relate parental alcoholism to higher likelihood of childrenengaging in problematic drinking, it remains unclear whether parental problem-drinking leaves a long-term economic imprint on children.
Connell and Goodman (2002) identify four mechanisms that relate parental alco-hol abuse to children’s adverse outcomes: (1) genetics; (2) complications during pre-natal development; (3) exposure to the parent’s behavior and knowledge; and (4)environmental stressors such as economic pressure, marital conflict, and disruptionthat are more common in a home with an alcoholic or problem-drinking parent.By affecting mental and physical health, levels of IQ, learning capabilities, substanceuse patterns, and attitudes in general, each of these mechanisms can shape a child’sfuture labor market success.
The higher propensity of children of alcoholics to develop drinking problems is afirst pathway that could relate parental drinking to children’s labor market outcomesat adulthood. A family history of alcoholism has been identified as a critical predis-posing factor of high risk for alcohol dependence at later ages (Jennison and Johnson1998; Windle 1996; Jennison and Johnson 2001). Findings of twin and adoption
Balsa 455
studies have enhanced the understanding of genetic susceptibility to alcohol depen-dence (McGue 1997). Prevalence of alcoholism among first-degree relatives of indi-viduals suffering from alcoholism is 3–4 times greater than it is among the generalpopulation, while the rates are even higher among identical twins (Schuckit 1999). Inaddition, research shows that children of alcoholics develop expectations about theeffects of alcohol by observing their parents’ drinking and may turn to alcohol asa means of alleviating other conditions (Ellis, Zucker, and Fitzgerald 1997). As withother behavioral conditions, the gene-environment interaction increases children ofalcoholics’ risks of developing alcohol abuse or dependence. This higher predisposi-tion to misuse alcohol may affect labor market outcomes through its negative effectson human capital formation and employment. Yamada, Kendrix, and Yamada (1996),Chatterji and DeSimone (2005), and Koch and McGeary (2005) find that drinking inhigh school significantly reduces the probability of high school graduation. Williams,Powell, and Wechsler (2003) show that alcohol consumption in college has a nega-tive effect on GPA, which works mainly through a reduction in the hours spent study-ing. Booth and Feng (2002) and McDonald and Shields (2001) find that heavydrinking significantly increases the probability of unemployment and reduces thenumber of weeks worked among those employed. Other work, however, finds littlerole for alcohol in educational attainment (Dee and Evans 2003; Koch and Ribar2001) or unemployment (Feng et al. 2001).
Prenatal exposure to alcohol is a second channel through which parental alcoholabuse can affect children’s productivity. The exposure of fetuses to high amountsof alcohol has been shown to have devastating effects on development. Fetal alcoholsyndrome (FAS) has been associated with structural abnormalities, growth deficits,and neurobehavioral anomalies. Among other problems, children with FAS face defi-ciencies related to activity, attention, learning, memory, language, motor, and visuo-spatial abilities (see Mattson and Riley (1998) for a review of this literature).
The effects of parental problem-drinking on a child’s health provide a third chan-nel of transmission. Parental alcoholism has been associated with reduced family co-hesion, increased odds of a single-parent household, and poor supervision ofchildren, all of which are linked to higher behavioral and psychological problemsamong children (Antecol and Bedard 2007; Amuedo-Dorantes and Mach 2002; Garis1998; Aizer 2004). Anda and colleagues (2002) find that depression is significantlyrelated to having adverse childhood experiences, which are more frequent in a homewith a parent who abuses alcohol. COAs are more likely to experience abuse or ne-glect that can affect their mental and physical health (Berger 2005; Markowitz andGrossman 1998). Moreover, the stress of being raised in a home with an alcoholicparent can make children vulnerable to certain health conditions, such as headacheor chronic pain (Dobkin et al. 1994). The harmful effect of parental problem-drink-ing on children’s health is underscored in the literature on children of alcoholics. Seo(1998) finds a statistically significant impact of maternal binge drinking on a child’snonnormative behavior and poor reading performance between 10–14 years of age.Jones, Miller, and Salkever (1999) provide evidence that parental alcohol misuseincreases children’s behavioral problems. Balsa (2006) finds that parental drinkingincreases children’s utilization of acute health care services, such as mental healthservices and hospitalizations. Although fewer studies have followed these childrenover time, mental health problems are likely to persist in adulthood and affect
456 The Journal of Human Resources
productivity. Depression and other health conditions may decrease labor force partic-ipation, reduce attendance to work for those employed, and affect productivity andwages.
A final and more direct channel would consist in the need of an adult child to taketime from school or work to take care of a parent with a drinking problem.
In sum, children of alcoholics face a number of constraints that condition their la-bor market outcomes in adulthood: a higher predisposition to misuse alcohol, ahigher likelihood of congenital developmental problems, a stressful environment thatleads to mental and possibly other health problems, and the burden of having to takecare of a sick parent. The aim of this paper is to find evidence of the aggregate effectof these constraints on children’s labor market success.
III. Data and Methodology
A. The NLSY79
The data used for this analysis is the National Longitudinal Survey of Youth 1979(NLSY79). The NLSY79 is a nationally representative sample of 12,686 youngmen and women who were 14–22 years old when they were first surveyed in1979. These individuals were interviewed annually through 1994 and are currentlyinterviewed on a biennial basis. The NLSY79 cohort provides researchers with theopportunity to study a large sample of individuals representing American men andwomen born in the 1950s and 1960s and living in the United States in 1979. Labormarket information includes the start and stop dates for each job held since the lastinterview, labor market activities (looking for work, out of the labor force) duringgaps between jobs, hours worked, earnings, occupation, industry, benefits, and otherspecific job characteristics. In addition, the survey collects detailed informationabout personal and family characteristics, such as educational attainment, incomeand assets, health conditions that limit the ability to work, and use of alcohol andother substances.
B. Definition of Variables
1. Dependent variables
The analysis focuses on three indicators of labor market performance of children ofproblem-drinking parents at middle age: labor force participation, unemployment,and hourly wages. Rather than focusing on a single year, these measures are averagedacross a period spanning six years (1996–20021) and four interviews. This impliesthat for individuals who were 14 years old in 1979, labor market performance is mea-sured during the period when they were between 31 and 37 years old. For those in theother extreme (aged 22 in 1979), the analysis considers labor market outcomes whilein their late thirties and early-to-mid forties (39 to 45). Working with average meas-ures of outcomes across years has several advantages. First, it minimizes measure-ment error and the incidence of exogenous temporary shocks on labor market
1. Surveys were administered every two years in this period: 1996, 1998, 2000, and 2002.
Balsa 457
outcomes. We are interested in the permanent impact of parental problem-drinkingon the labor trajectories of their offspring in adulthood, and working with averagesachieves this aim more effectively than considering single years. Second, it substan-tially reduces problems of missing observations due to nonresponse. If an individualresponded to the survey in 1996 and 1998, but did not respond in 2002, we wouldstill have a measure of labor market outcomes. Third, averaging across years circum-vents the problem of not observing a reservation wage. Because a relatively low pro-portion of individuals remain unemployed for six years in a row, we are more likelyto observe the individual’s reservation wage. Also, we are more likely to observeindividuals who participate intermittently in the labor force.
Labor force participation is measured as the average number of weeks per year therespondent was out of the labor force between 1995 and 2001. Note that because ofthe biennial nature of the survey, the measure only considers the average number ofweeks of nonparticipation corresponding to the calendar years prior to the 1996,1998, 2000, and 2002 surveys.2 Unemployment is defined as the average numberof weeks per year the respondent was unemployed in 1995, 1997, 1999, and 2001.For the computation of hourly wages, only respondents that did not report self-employment in the 1996–2002 surveys are considered. Hourly wages in a particularyear are measured as the ratio of total wage earnings to total hours worked in thatyear, adjusted to 2002 values using the Bureau of Labor Statistics CPI. A measureof hourly wage is averaged across the years 1995, 1997, 1999, and 2001, and outliersshowing a wage per hour below $1 or above $300 are dropped from the regressions(76 observations less than $1 and eight observations more than $300). To normalizethe distribution, the logarithm of the average hourly wage is computed.
2. Parental Problem-drinking
In 1988, the NLSY respondents were asked whether they had ever had any relativeswho had been alcoholics or problem drinkers at any point in their lives and, if so, toindicate their relationship to each of these relatives. A dichotomous indicator forhaving had a problem-drinking parent is constructed on the basis of these questions.The measure considers both biological and nonbiological parents. The data offeredthe possibility of constructing separate indicators for a problem-drinking motherand a problem-drinking father. However, the small number of problem-drinkingmothers (5 percent in the female sample and 3 percent in the male sample) raisedconcerns about the power of the data to achieve statistically significant estimates.In addition, it was very hard to find instrumental variables that were relevant predic-tors of a problem-drinking mother.
3. Demographics
The analysis adjusts for age—nine categories ranging from 14 to 22 years old in1979—and for race/ ethnicity—White, Hispanic, Black, or Other. All analyses arerun separately for male and female respondents.
2. While NLSY has information on the cumulative number of weeks out of the labor force since the lastinterview, dealing with these variables is confusing because of the different rates of responses and differentinterview dates.
458 The Journal of Human Resources
4. State dummy variables
Thirty-nine different geographic dummies are constructed to indicate the state of res-idency of the respondent at age 14. Due to the low number of respondents in variousstates, a number of observations had to be aggregated into a single category. Includ-ing state dummy variables as controls contributes to isolate the effects of a problem-drinking parent from influences at the geographic level that can affect a respondent’sown propensity to drink as well as his/her human capital and labor market choices.
5. Other data
The relationship between a problem-drinking parent and an adult child’s labor mar-ket outcomes can be mediated by a number of factors, such as family and householdcharacteristics during childhood, investment in education, health status, drinking tra-jectories, and family choices, among others. While none of these measures is incor-porated in the main analysis due to the inherent endogeneity, a descriptive account ofsuch variables is presented in Table 5. Initial comparisons of these features amongchildren of problem-drinking parents and other children can contribute to build hy-potheses about the pathways linking parental problem-drinking to children’s labormarket outcomes.
C. Sample and Data Description
The sample used for this analysis excludes NLSY respondents from the military sam-ple, who were only surveyed up to 1984, as well as economically disadvantaged mi-norities of the supplemental subsample, who were not eligible for interview as of the1991 survey. A total of 9,986 individuals were eligible for interview in 1996–2002.Of these, 9,028 responded to at least one of the four surveys administered between1996 and 2002 (90 percent response rate). Among the remaining 9,028 observations,a total of 534 did not respond to the parental alcohol questions. The final sample has8,494 observations, including 4,199 for males and 4,295 for females. Rates of partic-ipation in the 1996–2002 interviews do not differ across children of problem-drink-ing parents and other respondents.
Table 1 compares the means of the dependent variables and demographics acrosschildren of problem-drinking parents and other respondents. The statistics are shownseparately for males and females. It is interesting to note that male respondents areless likely to answer the questions about parental drinking and less likely to report aproblem-drinking mother or father than female respondents. Nineteen percent ofmale respondents report having had a problem-drinking parent, while the rate is26 percent for females. This difference clearly indicates a reporting bias in the data.Regarding labor market outcomes, both male and female children of problemdrinkers are more likely to be out of the labor force. Daughters of problem drinkersearn slightly lower hourly wages than other females. Parental problem-drinking ismore prevalent in White families and less prevalent in African American families.
Table 2 describes in more detail the labor market outcomes for males and females.More than one third of male respondents and half of female respondents report atleast one week out of the labor force during 1995–2001, and a few respondents staypermanently out of the labor force during the period (2.4 percent of men and 6.4 per-cent of women). Twenty-eight percent of both men and women experience some
Balsa 459
Ta
ble
1D
ata
Mea
ns,
by
Pa
ren
tal
Pro
ble
m-d
rin
kin
gS
tatu
sa
Mal
eR
esp
on
den
tsF
emal
eR
esp
on
den
ts
To
tal
N¼
9,9
86
b5
,03
44
,95
2P
erce
nta
ge
resp
on
din
gq
ues
tio
ns
abo
ut
par
enta
ld
rin
kin
g8
8.8
%9
0.6
%P
erce
nta
ge
rep
ort
ing
pro
ble
m-d
rin
kin
gp
aren
t1
9.0
%2
6.1
%P
erce
nta
ge
rep
ort
ing
pro
ble
m-d
rin
kin
gm
oth
er3
.4%
5.2
%P
erce
nta
ge
rep
ort
ing
pro
ble
m-d
rin
kin
gfa
ther
17
.3%
23
.0%
Pro
ble
m-
dri
nk
ing
par
ent
No
pro
blem
-d
rin
kin
gp
aren
t
Pro
ble
m-
dri
nk
ing
par
ent
No
pro
ble
m-
dri
nk
ing
par
ent
N8
51
3,6
18
1,1
69
3,3
15
Per
cen
tag
ein
terv
iew
edat
leas
to
nce
in1
99
6–
20
02
94
.7%
93
.8%
96
.5%
95
.5%
Lab
or
mar
ket
ou
tco
mes
Wee
ks
ou
to
fla
bo
rfo
rce
(ave
rag
e9
5,
97
,9
9,0
1)
7.4
08
4.8
89
**
*1
2.5
49
11
.57
6**
Wee
ks
unem
plo
yed
(ave
rage
95,
97,
99,
01)
2.2
92
1.9
81
1.6
52
1.6
51
Ln
ho
url
yw
age
(ave
rag
e9
5,
97
,9
9,
01
)2
.71
42
.74
52
.41
42
.48
0*
**
Co
ntr
ols
Ag
e1
97
9¼
14
0.0
99
0.0
92
0.0
67
0.0
84
**
Ag
e1
97
9¼
15
0.1
70
0.1
37
**
0.1
39
0.1
31
Ag
e1
97
9¼
16
0.1
27
0.1
43
0.1
24
0.1
45
**
Ag
e1
97
9¼
17
0.1
29
0.1
35
0.1
39
0.1
40
Ag
e1
97
9¼
18
0.1
20
0.1
46
**
0.1
30
0.1
35
Ag
e1
97
9¼
19
0.1
02
0.1
19
0.1
34
0.1
24
460 The Journal of Human Resources
Ag
e1
97
9¼
20
0.1
14
0.1
03
0.1
21
0.1
03
**A
ge
19
79¼
21
0.1
15
0.0
98
0.1
24
0.1
14
Ag
e1
97
9¼
22
0.0
24
0.0
27
0.0
21
0.0
24
No
n-H
isp
anic
wh
ite
0.5
28
0.4
84
**
0.5
26
0.4
85
***
His
pan
ic0
.17
70
.19
20
.18
60
.18
9B
lack
0.2
82
0.3
13
**
0.2
79
0.3
12
**O
ther
race
0.0
13
0.0
11
0.0
09
0.0
13
Note
:**
Dif
fere
nce
bet
wee
nre
sponden
tw
ith
pro
ble
m-d
rinkin
gpar
ent
and
resp
onden
tw
ith
no
pro
ble
m-d
rinkin
gpar
ent
stat
isti
call
ysi
gnifi
cant
at5
per
cent;
***
dif
fere
nce
stat
isti
call
ysi
gnifi
cant
at1
per
cent.
a.P
aren
tal
pro
ble
m-d
rinkin
g¼
1w
hen
resp
onden
tre
port
edei
ther
abio
logic
alor
nonbio
logic
alpro
ble
m-d
rinkin
gm
oth
eror
fath
er.
b.E
xcl
udes
NL
SY
79
resp
onden
tsfr
om
the
mil
itar
ysa
mple
and
econom
ical
lydis
adva
nta
ged
min
ori
ties
of
the
supple
men
tal
subsa
mple
.
Balsa 461
unemployment in 1995–2001. Wages are not observed for 9 percent of the male and12 percent of the female sample.
There are two sources of sample selection in the data: attrition and failure to an-swer the questions about alcohol problems in the family. Table 1A in the Appendixshows the mean differences between some of the variables in the sample used for themain analysis and variables in the other two samples. Those not interviewed in 1996–2002 are more likely to belong to White, highly educated families and less likely toreport a problem-drinking father at baseline, reducing concerns about attrition of themost severe cases. On the other hand, Hispanics and children raised in lower-incomefamilies are overrepresented among respondents who do not answer the parental al-cohol use questions. These individuals are more likely to be high school dropouts andshow more weeks out of the labor force. While their rate of alcohol dependence islower than that in the final sample, there is also a higher rate of nonresponse tothe DSM-IV questions. Judging from the profile of this sample, it could be possiblethat parental alcoholism was more severe among these individuals. If this were thecase, our results would be biased toward zero by failing to include the cases moreaffected by parental drinking.
Table 2Labor Market Outcomes for Respondents Interviewed at Least Once in 1996–2002
Males Females
N % N %
N interviewed at least once in survey years1996, 1998, 2000, 2002 with information onparental alcohol problems (N¼8,494)
4,199 100.0% 4,295 100.0%
Labor force participationPermanently in the labor force 2,559 60.9% 1761 41.0%Some time but not all out of labor force 1,499 35.7% 2251 52.4%Permanently out of the labor force 102 2.4% 275 6.4%Missing labor force participation 39 0.9% 8 0.2%
Working status and unemployment (among thosewith some labor force participation)Permanently working while in the labor force 2,883 68.7% 2830 65.9%Some time but not all unemployed 1,173 27.9% 1178 27.4%Permanently unemployed 2 0.0% 1 0.0%Missing unemployment 39 0.9% 8 0.2%
WagesWages observed 3,835 91.3% 3759 87.5%Wages not observed 364 8.7% 536 12.5%Missing because not working 185 4.4% 349 8.1%Outliers wage # $1 34 0.8% 32 0.7%Outliers wage > $300 6 0.1% 2 0.0%Missing because did not report wages 139 3.3% 153 3.6%
462 The Journal of Human Resources
D. Methodology
The equation of interest is of the form:
Yi ¼ a0 + a1Ai + Age9i a2 + Race9i a3 + State9i a4 + eið1Þ
where Yi is a labor market outcome and Ai is an indicator for a problem-drinking par-ent. Because family, household, and many individual characteristics are endogenous,only a set of purely exogenous dummies—age, race, and fixed effects for therespondent’s state of residency at age 14—are included as controls. The state dum-mies adjust the model for exogenous geographical influences that may shape an indi-vidual’s human capital and labor market opportunities. The choice of age 14responds to the belief that many attitudes, including the respondent’s own propensityto drink, are forged during adolescence by the geographical environment. The idea isto isolate this environmental effect from parental influences due to alcohol misuse.
In addition to estimating Equation 1 with ordinary least squares, the model is es-timated with alternative single equation techniques that better suit the particular out-come analyzed. In the case of count variables such as number of weeks out of thelabor force and number of weeks unemployed, the model is reestimated using neg-ative binomial regressions. When the outcome of interest is hourly wages, whichinvolves selection due to incidental truncation,3 the model is also estimated usinga Heckman two-step regression. In the first stage, I estimate the likelihood of observ-ing the wage among all those responding to the survey. The wage regression is esti-mated next, adjusting for the selection implicit in the first stage. The exclusionrestriction in the first stage is given by a variable constructed in 1979 that describedthe respondent’s level of impatience when completing the interview (as reported bythe interviewer).
One problem with using single-equation regression to estimate Equation 1 is thatomitted variables may be biasing the results. For example, exogenous negativeshocks to the family (for example, a death of a family member) may increase thelikelihood that a parent becomes a problem drinker and, at the same time, deterioratethe child’s mental health. Or an adverse economic period for the family could increaseparental drinking and at the same time reduce the child’s possibilities of graduatingfrom high school or attending college. These mental health limitations or economicadversities could have a long-term impact upon children’s productivity. It is also pos-sible that careless personalities could lead parents to drink rather than take care oftheir families. Inability to control for these or other latent variables might result inspurious associations between parental drinking and children’s long-term outcomes.
Instrumental variables, if relevant and valid, can address this problem by restrictingthe analysis to those components of the main explanatory variable (Problem-drinkingParent) that are uncorrelated with the error term in Equation 1. The instruments cho-sen to predict a problem-drinking parent are (1) an indicator of a problem-drinkinggrandfather on the father’s side, (2) cigarette and beer excise taxes in the family’sstate of residence at the time of the respondent’s birth, (3) cigarette excise taxesin the mother’s state of birth in 1947 (approximately 14 years prior to the respond-ent’s birth), and (4) a dummy variable indicating whether there were alcohol sales
3. Incidental truncation in this case is due to the fact that wages are only observed for people who work.
Balsa 463
controls in the father’s state of birth in 1947.4 The indicator of a problem-drinkinggrandfather was constructed on the basis of the same questions (1988 survey) usedto define the parental problem-drinking indicator. Excise taxes were obtained fromthe Book of States (1947–64).
Cigarette and beer taxes are commonly used instruments in this type of analysis(French and Maclean 2006; Mullahy and Sindelar 1996; Chatterji and Markowitz2001). Some researchers have criticized these instruments because they may be con-founded with attitudes at the geographic level. Such problem is minimized in this paperby including controls for the respondent’s state of residency at age 14 and by usingpolicy variables that were in place 14 or more years before at the respondent’s stateof residency and at the mother or father’s state of birth. Thus, we are exploiting the var-iation on excise taxes across time and across states (for those parents who moved).
Problem-drinking by other family members, such as grandparents, aunts, oruncles, are highly predictive of parental problem-drinking and have been used suc-cessfully in other related studies as instruments (French and Maclean 2006; Mullahyand Sindelar 1996). Problem-drinking by a grandfather is a valid instrument only if itaffects a child’s labor market outcomes through paternal or maternal alcohol prob-lems. It seems unlikely that this familial instrument would be correlated with con-temporaneous shocks or other omitted variables that could simultaneously affectthe children’s human capital and the parent’s drinking. It remains possible, however,that the influence of a problem-drinking grandfather on an adult grandchild worksindirectly through other channels. Our analysis therefore recognizes that such an in-strument can mitigate but probably not completely eliminate biases arising due toendogeneity. For this reason, some sensitivity analyses are conducted using only al-cohol policy variables as instrumental variables. Unfortunately, alcohol policy vari-ables tend to be weaker instruments when considered on their own.
Instrument relevance is tested using the F-test of the joint significance of the ex-cluded instruments in a linear probability model predicting a problem-drinking par-ent. For each outcome of interest, overidentifying restrictions are tested using theHansen J statistic (Hansen 1982).5 I also run tests of subsets of the exclusion restric-tions to ensure that each of the instruments, including State policies and the familydrinking history indicator, satisfy by themselves the orthogonality requirements. TheC-statistic or difference in Sargan test is used for this purpose (Hayashi 2000).
Endogeneity is addressed using generalized method of moments (GMM). Tests ofheteroskedasticity revealed that most of the models analyzed had heteroskedasticity
4. The mentioned alcohol and cigarette policies were selected from a more extensive data set that includedspirit, beer, wine, and cigarette taxes, as well as other policies such as state controls of alcohol sales. Thesepolicies were collected for several time periods and geographic locations: (a) at the family’s state of resi-dency by the time of the respondent’s birth; (b) at the father’s state of birth approximately 14 years prior tothe respondent’s birth; and (c) at the mother’s state of birth approximately 14 years prior to the respondent’sbirth. Taxes in dollars were adjusted to 2002 values using the BLS CPI. The final set of instruments in-cluded those policies that achieved jointly the highest level of significance in the prediction of aproblem-drinking parent.5. The Hansen J statistic differs from the Sargan’s statistic in that it tests overidentifying restrictions underthe assumption of heteroskedasticity. Note that failure to reject the hypothesis of no overidentification inany of these tests does not guarantee that the instruments satisfy the orthogonality requirements. Failureto reject could be due, for example, to low power in one or more of the instruments.
464 The Journal of Human Resources
of unknown form and GMM is more efficient than two stages least squares (2SLS) insuch situation. The GMM estimator solves:
�gðb̂Þ ¼ 0ð2Þ
where �gðb̂Þrepresents the L sample moments corresponding to the populationmoments giðbÞ ¼ Z9i ei ¼ Z9iðyi2XibÞ, and Z denotes the set of excluded instrumentssatisfying EðZieiÞ ¼ 0.
The C-statistic is used to test for the exogeneity of the main regressor, the indicatorof a problem-drinking parent.6 In the absence of endogeneity, single-equation esti-mation, such as OLS, is unbiased and more efficient than instrumental variables es-timation (either 2SLS or GMM). 7
All models are run separately for female and male respondents. For robustness, theGMM regressions are rerun using only policy variables as instruments. In addition,bivariate probit estimation is conducted to assess the effects of parental problem-drinking on the likelihood of participating in the labor force and on the likelihoodof being unemployed.
IV. Results
A. Parental Problem-drinking and Adult Children’s Weeks Out of the LaborForce, Weeks Unemployed, and Wages
Tables 3 and 4 display the core results for males and females respectively. Tables 3aand 4a describe the results of the different statistics used to test for heteroskedasticityand the first-stage statistics for the instrumental variables analysis, including rele-vance and validity of the instruments. Tables 3b and 4b show the estimation resultsfor the OLS, negative binomial, Heckman, and GMM models. For space reasons,only the coefficients, t-statistics, and marginal effects (when relevant) of the parentalproblem-drinking indicator are reported for each outcome. The last row in each tabledisplays the tests of exogeneity of the parental problem-drinking indicator. All mod-els adjust for age, race, and state fixed effects at age 14.
The instruments used to predict a problem-drinking parent are reported in Table3a. When estimating number of weeks out of the labor force and number of weeksunemployed the set of instrumental variables includes a problem-drinking grandfa-ther on the father’s side, cigarette taxes in the family’s state of residency at the timeof the respondent’s birth, and cigarette taxes at the mother’s state of birth in 1947.8
6. The C-statistic differs from the Durbin Wu Hausman test in that it works under the hypothesis of het-eroskedasticity.7. Failure to reject exogeneity could also be due to lack of power of the instruments. In such case, wewould not be able to say whether OLS estimates reflect causality.8. Beer taxes at birth and at the mother’s state of birth in 1947 were also significant in predicting a problem-drinking parent and had the right (negative) sign. However, cigarette taxes explained a higher fraction of theproblem-drinking-parent variance, and the F-statistic evaluating the joint significance of instruments decreasedwhen they were added together to the equation. Various studies have found significant relationships betweenthe prices of cigarettes and alcohol demand, although there are mixed results about the sign of the effect(Cameron and Williams 2001, Picone et al. 2004, Markowitz and Tauras 2006). Cigarette taxes present highercross-section variation (across states) relative to beer taxes and represent a higher fraction of the final price.These features could explain why cigarette taxes have better predictive power in the current setting.
Balsa 465
Ta
ble
3a
Test
so
fH
eter
osk
eda
stic
ity
an
dTe
sts
of
Rel
eva
nce
an
dV
ali
dit
yo
fIn
stru
men
tal
Va
ria
ble
s(M
ale
Sa
mp
le)
(1)
Wee
ks
ou
to
fth
ela
bo
rfo
rce
19
96
–2
00
2
(2)
Wee
ks
un
emp
loy
ed1
99
6–
20
02
(3)
Ln
ho
url
yw
age
19
96
–2
00
2
Tes
tsof
het
erosk
edas
tici
tyB
reush
-Pag
ante
stx
2 (52
)¼
64
8.5
70
x2 (5
2)¼
14
04
.53
0x
2 (50
)¼
12
2.9
40
H0
)M
od
elh
om
osk
edas
tic
p¼
0.0
00
p¼
0.0
00
p¼
0.0
00
Inst
rum
enta
lva
riab
les
Ex
clu
ded
inst
rum
ents
Pro
ble
m-d
rin
kin
gg
ran
dfa
ther
,C
igar
ette
tax
atb
irth
,ci
gar
ette
tax
atm
oth
er’s
stat
eo
fb
irth
in1
94
7
Pro
ble
m-d
rin
kin
gg
ran
dfa
ther
,C
igar
ette
tax
atb
irth
,ci
gar
ette
tax
atm
oth
er’s
stat
eo
fb
irth
in1
94
7
Pro
ble
m-d
rinkin
gg
ran
dfa
ther
Rel
evan
ceo
fin
stru
men
tal
vari
able
sF
-sta
to
fex
clu
ded
inst
rum
ents
F(3
,33
01
)¼1
2.3
7F
(3,3
30
1)¼
12
.37
F(1
,28
99
)¼1
4.8
9p¼
0.0
00
p¼
0.0
00
p¼
0.0
00
Fir
st-s
tag
eco
effi
cien
tsan
dst
andar
der
rors
of
inst
rum
enta
lva
riab
les
a
Pro
ble
m-d
rinkin
ggra
ndfa
ther
0.1
89***
0.1
89***
0.1
69***
(0.0
40
)(0
.04
0)
(0.0
44
)C
igar
ette
tax
atb
irth
20
.00
2*
*2
0.0
02
**
—(0
.00
1)
(0.0
01
)C
igar
ette
tax
atm
oth
er’s
stat
eo
fb
irth
in1
94
72
0.0
02
**
*2
0.0
02
**
*—
(0.0
01
)(0
.00
1)
466 The Journal of Human Resources
Val
idit
yo
fin
stru
men
tal
vari
able
sH
anse
nJ-
stat
.H
0)
Mo
del
isco
rrec
tly
spec
ified
x2 (2
)¼
2.7
37
p¼
0.2
54
x2 (2
)¼
2.7
38
p¼
0.2
54
Eq
uat
ion
exac
tly
iden
tifi
ed
Ort
ho
go
nal
ity
of
sub
sets
of
inst
rum
ents
–p-v
alu
eo
fC
stat
isti
cH
0)
Sel
ecte
din
stru
men
to
rth
og
on
alto
the
erro
rte
rm
Pro
ble
m-d
rin
kin
gg
ran
dfa
ther
p¼
0.2
79
Cig
aret
teta
xb
irth
Pro
ble
m-d
rin
kin
gg
ran
dfa
ther
p¼
0.4
14
Cig
aret
teta
xb
irth
Eq
uat
ion
exac
tly
iden
tifi
ed
p¼
0.7
29
p¼
0.1
05
Cig
aret
teta
xm
oth
erC
igar
ette
tax
mo
ther
p¼
0.1
02
p¼
0.4
51
Note
:**
Sta
tist
ical
lysi
gnifi
cant
at5
per
cent;
***
stat
isti
call
ysi
gnifi
cant
at1
per
cent
a.T
he
firs
t-st
age
esti
mat
ion
consi
sts
of
ali
nea
rpro
bab
ilit
ym
odel
pre
dic
ting
the
likel
ihood
of
hav
ing
apro
ble
m-d
rinkin
gpar
ent.
Inad
dit
ion
toth
ein
stru
men
tal
vari
able
ssp
ecifi
edin
the
table
,th
ere
gre
ssio
nad
just
sfo
rag
e,ra
ce,
and
stat
efi
xed
effe
cts
atag
e14.
Balsa 467
Ta
ble
3b
Eff
ects
of
aP
rob
lem
-dri
nki
ng
Pa
ren
to
nA
du
ltC
hil
dre
n’s
La
bo
rM
ark
etO
utc
om
es(M
ale
Sa
mp
le)
(1)
Wee
ks
ou
to
fth
ela
bo
rfo
rce
19
96
–2
00
2
(2)
Wee
ks
un
emp
loy
ed1
99
6–
20
02
(3)
Ln
ho
url
yw
age
19
96
–2
00
2
(1)
OL
Ses
tim
atio
nP
roble
m-d
rinkin
gpar
ent
2.4
83**
*0.4
17
*2
0.0
74
**
(0.5
71
)(0
.23
1)
(0.0
31
)(2
)N
egat
ive
bin
om
ial
esti
mat
ion
Pro
ble
m-d
rinkin
gpar
ent
0.5
38**
*0.2
73***
—(0
.09
7)
(0.1
09
)[2
.78
3]
[0.4
83
](3
)H
eck
man
esti
mat
ion
Pro
ble
m-d
rinkin
gpar
ent
(sel
ecti
on
equat
ion)
——
20
.14
2(0
.08
0)
Pro
ble
m-d
rinkin
gpar
ent
(wag
eseq
uat
ion)
——
20
.06
7*
*(0
.03
0)
Ath
rho
——
20
.07
4(0
.05
4)
(4)
GM
M(I
V)
esti
mat
ion
Pro
ble
m-d
rinkin
gpar
ent
5.8
18
22
.33
30
.03
9(4
.79
9)
(1.6
53
)(0
.90
7)
Exogen
eity
test
:C
-sta
tist
icH
0)
‘‘P
roble
m-d
rinkin
gpar
ent’
’is
exogen
ous
x2 (1
)¼
0.4
64
x2 (1
)¼
2.9
81
x2 (1
)¼
0.1
16
p¼
0.4
96
p¼
0.0
84
p¼
0.7
34
N3
,35
43
,35
42
,95
0
Note
:C
oef
fici
ents
,st
andar
der
rors
inpar
enth
eses
,an
dm
argin
alef
fect
sin
squar
edbra
cket
s.A
lles
tim
atio
ns
contr
ol
for
age,
race
,an
dst
ate
fix
edef
fect
sfo
rth
ere
spond-
ent’
sst
ate
of
resi
den
cyat
age
14.
***
Sta
tist
ical
lysi
gnifi
cant
at1
per
cent;
**
Sta
tist
ical
lysi
gnifi
cant
at5
per
cent;
*S
tati
stic
ally
signifi
cant
at10
per
cent
468 The Journal of Human Resources
Each excluded instrument predicts a ‘‘Problem-drinking Parent’’ with a statisticalsignificance of 5 percent or less. The F-statistics for joint significance of the instru-ment is above 12. The Hansen J statistic and C-tests of orthogonality of subsets ofinstruments cannot reject the null of no overidentification in the estimation of weeksout of the labor force and weeks unemployed. In particular, the C-statistic cannot re-ject the hypothesis of orthogonality between the problem-drinking-grandfather in-strument and the error processes of the outcomes.
When estimating hourly wages (Table 3a, Column 3), self-employed individualsare excluded from the sample. This exclusion reduces the predictive power of the al-cohol policy instruments, leaving the indicator for a problem-drinking grandfather asthe sole relevant instrument. The instrument has good predictive power in the first-stage (the F-statistic is 14.9), but its validity cannot be determined statistically be-cause the model is exactly identified.
Single equation results in Table 3a suggest that having a problem-drinking parentis associated with longer periods out of the labor force, lengthier unemployment, andlower hourly wages. According to OLS and negative binomial estimates, having aproblem-drinking parent increases a male respondent’s time out of the labor forceby around two and a half weeks (from an average of five weeks to an average ofseven and a half weeks) and increases the number of weeks of unemployment in halfa week per year (from a baseline of approximately two weeks for those without aproblem-drinking parent). OLS and Heckman estimates also show that hourly wagesof male children of problem drinkers are 7 percent below other respondents’ wages.
While single equation estimates are suggestive that parental problem-drinkingmay have some negative effects on adult children’s labor market outcomes, no con-clusions about causality can be derived from those results. Unfortunately, the GMMestimation, which was expected to shed light on the causality of the detected relation-ships, is not informative enough. None of the GMM estimates of the effect of a problem-drinking parent achieves statistical significance, and exogeneity of the main explanatoryvariable cannot be rejected at 5 percent significance for any of the analyzed out-comes. This failure to find statistically significant effects is most likely due to a lackof precision in the estimation, as reflected by the large standard errors of the GMMestimates.
Table 4a shows the first-stage test-statistics for females. As with male respondents,all models are heteroskedastic, instruments are relevant (the F-statistic is higher than18), and satisfy the exclusion restrictions at 5 percent significance, even when con-sidering subsets of them. The instrumental variables used in the estimation offemales’ labor market outcomes differ from those used with men. This was not un-expected given the difference in the prevalence of parental problem-drinkingreported by males and females. The instruments used in the case of females are: aproblem-drinking grandfather, beer excise taxes in the family’s state of residencyat the time of the respondent’s birth, and state controls of alcohol sales at the father’sstate of birth in 1947.9
9. In this case, cigarette taxes were also significant in explaining the likelihood of a problem-drinking par-ent, but less so than beer taxes. It is hard to make comparisons between the instruments in the female andmale sample. Differences in the likelihood of reporting a problem-drinking parent across genders result in abigger pool of problem-drinking parents for female respondents than for male respondents.
Balsa 469
Ta
ble
4a
Test
so
fH
eter
osk
eda
stic
ity
an
dTe
sts
of
Rel
eva
nce
an
dV
ali
dit
yo
fIn
stru
men
tal
Va
ria
ble
s(F
ema
leS
am
ple
)
(1)
Wee
ks
ou
to
fth
ela
bo
rfo
rce
19
96
–2
00
2
(2)
Wee
ks
un
emp
loy
ed1
99
6–
20
02
(3)
Ln
ho
url
yw
age
19
96
–2
00
2
Tes
tso
fh
eter
osk
edas
tici
tyB
reu
sh-P
agan
test
x2 (5
0)¼
99
.56
1x
2 (50
)¼
13
86
.72
0x
2 (50
)¼
74
.22
5H
0)M
odel
hom
oske
dast
icp¼
0.0
00
p¼
0.0
00
p¼
0.0
15
Inst
rum
enta
lva
riab
les
Excl
uded
Inst
rum
ents
Pro
ble
m-d
rinkin
ggra
ndfa
ther
,B
eer
tax
atb
irth
,A
lco
ho
lco
ntr
ols
atfa
ther
’sst
ate
of
bir
thin
19
47
Pro
ble
m-d
rinkin
ggra
ndfa
ther
,B
eer
tax
atbir
th,
Alc
ohol
con
tro
lsat
fath
er’s
stat
eo
fb
irth
in1
94
7
Pro
ble
m-d
rin
kin
gg
ran
dfa
ther
,B
eer
tax
atb
irth
,A
lco
ho
lco
ntr
ols
atfa
ther
’sst
ate
of
bir
thin
19
47
Rel
evan
ceo
fin
stru
men
tal
vari
able
sF
-sta
to
fex
clu
ded
ixF
(3,3
33
9)¼
18
.68
F(3
,33
39
)¼1
8.6
8F
(3,2
44
6)¼
19
.23
p¼
0.0
00
p¼
0.0
00
p¼
0.0
00
Fir
st-s
tag
eco
effi
cien
tsan
dst
and
ard
erro
rso
fin
stru
men
tal
vari
able
sa
Pro
ble
m-d
rin
kin
gg
ran
dfa
ther
0.2
34
**
*0
.23
4*
**
0.2
87
**
*(0
.03
5)
(0.0
35
)(0
.03
8)
470 The Journal of Human Resources
Bee
rta
xat
bir
th2
0.0
06
**
*2
0.0
06
**
*2
0.0
06
**
(0.0
02
)(0
.00
2)
(0.0
02
)A
lco
ho
lco
ntr
ols
atfa
ther
’sst
ate
of
bir
thin
1947
20
.03
32
0.0
33
20
.04
3
(0.0
22
)(0
.02
2)
(0.0
25
)V
alid
ity
of
inst
rum
enta
lva
riab
les
Han
sen
J–
stat
isti
cx
2 (2)¼
1.2
68
x2 (2
)¼
3.4
42
x2 (2
)¼
1.7
43
H0
)M
od
elis
corr
ectl
ysp
ecifi
edp¼
0.5
30
p¼
0.1
79
p¼
0.4
18
Ort
hogonal
ity
of
subse
tsof
inst
rum
ents
–p
valu
eo
fC
stat
isti
c
H0)
Susp
ect
inst
rum
ent
ort
hog
on
alto
the
erro
rte
rmP
roble
m-d
rinkin
ggra
ndfa
ther
Pro
ble
m-d
rinkin
ggra
ndfa
ther
Pro
ble
m-d
rinkin
ggra
ndfa
ther
p¼
0.2
79
p¼
0.6
19
p¼
0.4
73
Bee
rta
xb
irth
Bee
rta
xb
irth
Bee
rta
xb
irth
p¼
0.2
92
p¼
0.4
80
p¼
0.2
07
Co
ntr
ol
fath
erC
on
tro
lfa
ther
Co
ntr
ol
fath
erp¼
0.7
77
p¼
0.0
75
p¼
0.5
73
Note
:**
Sta
tist
ical
lysi
gnifi
cant
at5
per
cent;
***
Sta
tist
ical
lysi
gnifi
cant
at1
per
cent.
a.T
he
firs
t-st
age
esti
mat
ion
consi
sts
of
ali
nea
rpro
bab
ilit
ym
odel
pre
dic
ting
the
likel
ihood
of
hav
ing
apro
ble
m-d
rinkin
gpar
ent.
Inad
dit
ion
toth
ein
stru
men
tal
vari
able
ssp
ecifi
edin
the
table
,th
ere
gres
sion
adju
sts
for
age,
race
,an
dst
ate
fixe
def
fect
sat
age
14.
Balsa 471
Ta
ble
4b
Eff
ects
of
aP
roble
m-d
rinki
ng
Pa
rent
on
Adult
Chil
dre
n’s
Labor
Mark
etO
utc
om
es(F
emale
Sam
ple
)
(1)
Wee
ks
ou
to
fth
ela
bo
rfo
rce
19
96
–2
00
2
(2)
Wee
ks
un
emp
loy
ed1
99
6–
20
02
(3)
Ln
ho
url
yw
age
19
96
–2
00
2
(1)
OL
Ses
tim
atio
nP
rob
lem
-dri
nk
ing
par
ent
1.1
19
*0
.09
72
0.0
60
*
(0.6
39
)(0
.17
8)
(0.0
31
)(2
)N
egat
ive
bin
om
ial
esti
mat
ion
Pro
ble
m-d
rin
kin
gp
aren
t0
.13
0*
*0
.16
5—
(0.0
56
)(0
.10
7)
[1.4
98
][0
.23
8]
(3)
Hec
km
anes
tim
atio
nP
rob
lem
-dri
nk
ing
par
ent
(sel
ecti
on
equ
atio
n)
——
0.0
31
(0.0
60
)P
rob
lem
-dri
nk
ing
par
ent
(wag
eseq
uat
ion
)—
—2
0.0
49
(0.0
35
)A
thrh
o—
—2
0.4
36
(0.2
69
)(4
)G
MM
(IV
)es
tim
atio
nP
rob
lem
-dri
nk
ing
par
ent
26
.12
02
1.1
72
0.1
82
(4.5
89
)(0
.99
5)
(0.1
65
)E
xo
gen
eity
test
:C
-sta
tist
icH
0)
‘‘P
roble
m-d
rinkin
gpar
ent’
’is
exogen
ous
x2 (1
)¼
0.2
63
x2 (1
)¼
1.6
97
x2 (1
)¼
2.3
19
p¼
0.1
05
p¼
0.1
93
p¼
0.1
28
N3
,39
23
,39
22
,49
9
Note
:C
oef
fici
ents
,st
andar
der
rors
inpar
enth
eses
,an
dm
argi
nal
effe
cts
insq
uar
edbra
cket
s.A
lles
tim
atio
ns
contr
ol
for
age,
race
,an
dst
ate
fix
edef
fect
sfo
rth
ere
spond-
ent’
sst
ate
of
resi
den
cyat
age
14.
***
Sta
tist
ical
lysi
gnifi
cant
at1
per
cent;
**
Sta
tist
ical
lysi
gnifi
cant
at5
per
cent;
*S
tati
stic
ally
signifi
cant
at10
per
cent.
472 The Journal of Human Resources
Table 4b shows the effects of a problem-drinking parent on women’s labor out-comes. Single equation regression reveals a statistically significant and positive as-sociation between a problem-drinking parent and women’s number of weeks outof the labor force. The marginal effect is smaller than that identified for men: itstands between 1.1 and 1.5 weeks per year. No statistically significant and robusteffects are identified in the estimation of weeks of unemployment or wages. As be-fore, the exogeneity of the parental problem-drinking variable cannot be rejected inany of the models and the GMM estimates are statistically insignificant. In particular,both the estimates and the standard errors in the GMM estimation are large relative tothe single equation ones, raising again concerns about the ability of the instrumentsto estimate the effects of interest with precision.
In sum, while the single equation estimates are indicative of detrimental effects ofparental problem-drinking on children’s labor market outcomes, the IV estimates arenot precise enough to determine whether omitted variables bias has been successfullydealt with.
B. Sensitivity Analysis Using only Policy Variables as Instruments
For robustness, I repeat the instrumental variables estimation using only policy var-iables as instruments. For males, the instruments used are cigarette taxes at birth andcigarette taxes at the mother’s state of birth in 1947. These variables have an F-statisticof joint significance of 7.4, a value below the rule of thumb of ten generally acceptedfor satisfactory IV estimation, but still reasonable to check for robustness. This set ofinstrumental variables satisfies also the exclusion restrictions. The results obtainedwith these instruments are qualitatively similar to those described in Section IV.A,although the magnitudes of the standard errors raise even more concerns about theefficiency of the estimation. The exogeneity of the ‘‘Problem-drinking Parent’’ in-dicator cannot be rejected at 5 percent for weeks out of the labor force or weeksof unemployment. As in the main analysis, the GMM estimators of the effect of aproblem-drinking parent do not achieve statistical significance.10 In the case ofhourly wages, the alcohol policy instruments were too weak to run an alternativeGMM estimation (the F-statistic for joint significance is barely above three).
For females, only one variable serves as relevant instrument: beer excise taxes atbirth. The F-statistic for such an instrument in the prediction of a problem-drinkingparent ranges between seven and eight, depending on the outcome. Exogeneity of theparental problem-drinking indicator cannot be rejected at 5 percent significance forany of the labor market measures analyzed but is rejected at 10 percent significance forweeks out of the labor force. GMM estimates are again statistically insignificant.11
C. Bivariate Probit Analysis
Rather than analyzing the number of weeks out of the labor force or unemployed, Irun bivariate probit models that estimate simultaneously the probability of having a
10. For males, GMM estimates of the effect of parental problem-drinking (and standard errors) when usingonly policy variables as instruments are: -2.021 (8.606) for weeks out of the labor force and -4.768 (3.618)for weeks unemployed. Standard errors are twice as big as the GMM standard errors in the core results.11. When using beer tax as the only instrument, GMM point estimates (and standard errors) of the effect ofparental problem-drinking on women’s number of weeks out of the labor force and unemployment are, re-spectively: -20.215 (14.399) and 1.148 (3.579).
Balsa 473
problem-drinking parent and the probability of being out of the labor force (or un-employed). Each model estimates the correlation between the error terms in the firstand second regression. To identify one equation from the other, I use the same set ofinstruments as in the main analysis.
Results from this estimation are displayed in Tables 5 and 6. For men (Table 5),both probit and bivariate probit estimates show a positive and statistically significanteffect of having a problem-drinking parent on the likelihood of being out of the laborforce. The correlation term (rho) is different from zero at a statistical significance of10 percent. Moreover, the sign of the bias is as expected; marginal effects are smallerin the bivariate probit estimation. A problem-drinking parent increases the likelihoodthat a son does not participate in the labor force by six percentage points in the bi-variate probit estimation, compared to nine percentage points in the probit model. Inthe case of unemployment, the probit coefficient for a problem-drinking parent ispositive and statistically significant but neither rho nor the bivariate probit coefficientis statistically significant. For women (Table 6), there is a positive and statisticallysignificant association between parental drinking and failure to participate in the la-bor force when using single equation regression, but no statistically significanteffects of interest are found in the bivariate probit estimation.
V. Discussion and Conclusion
Using NLSY79, this paper studies the association between parentalproblem-drinking and children’s labor market outcomes at adulthood. Labor marketoutcomes are measured while the respondents are in their late 30s to mid-40s, a stagein which return to school or early retirement are less likely to be behind labor marketdecisions. Results from single equation regressions are suggestive of important inter-generation costs of problem-drinking. For men, having a problem-drinking parent isassociated with longer periods out of the labor force, lengthier unemployment spells,and lower wages. For women, an association is identified between parental problem-drinking and time out of the labor force. While the analysis attempts to address endo-geneity, instrumental variables (GMM) estimation is not precise enough to concludeabout the causality of these effects. On the other hand, results from bivariate probitestimation, a technique that addresses endogeneity but is less likely to be affected bylow-powered instruments, support a negative effect of parental problem-drinking onmale children’s labor force participation.
Although the paper did not seek to identify concrete mechanisms of transmissionof the effect of parental drinking on adult children’s labor market outcomes, it is in-teresting to conjecture about possible channels of influence. A simple comparison ofmeans presented in Table 7 shows that children of problem drinkers are, in effect,disadvantaged in several aspects. Families with problem-drinking parents are morelikely to be nonintact and of lower socioeconomic status. Regardless of gender, chil-dren of problem drinkers have lower educational attainments: they are more likely todrop out of high school and less likely to achieve college or graduate education. Theyare also more likely to experience depressive symptoms at middle age and havehigher odds of experiencing health conditions that limit the type or amount of workthey can do. Relative to children of sober parents, children of alcoholics are more
474 The Journal of Human Resources
likely to start drinking by age 15 and more likely to be alcohol dependent ten and15 years after the baseline interview. Most or all of these variables are endogenousand no causality can be inferred from the comparisons. However, we can at leastspeculate that the impacts on health conditions and alcohol use trajectories mediatesome of the effects of problem-drinking parents on children’s labor force participa-tion and wages. Future research needs to address in more depth these links.
There are several limitations to the analysis. In the first place, around 5 percent ofthe sample that completed at least one of the 1996–2001 interviews did not respondto the questions about problem-drinking in the family. An initial mean comparisonshowed that these respondents belong to a fraction of the population with a higherrisk of having alcoholic parents. The omission of these individuals from the finalsample could be biasing our estimates toward zero.
A second problem may result from reporting bias. The data shows that the thresh-olds that males use to define an alcohol problem are different from those used byfemales. This may not be important if the extent to which an individual is affectedby parental alcoholism depends on his or her personal or subjective threshold.In other words, an individual may have a lower threshold when defining ‘‘parentalproblem-drinking’’ because she or he is by nature more readily affected by such aproblem. However, if thresholds are objective, male underreporting of parental problem-drinking will bias the results toward zero. Another type of reporting bias may result
Table 5Effects of parental problem-drinking on the likelihood of being out of the labor forceand of being unemployed (Male sample)
(1)Any weeks out
of the labor force1996–2002
(2)Any weeksunemployed1996–2002
(1) Probit estimatesProblem-drinking parent 0.231*** 0.168***
(0.057) (0.060)[0.089] [0.058]
(2) Recursive bivariate probit estimatesa
Problem-drinking parent 0.861** 20.168(0.339) (0.399)[0.055] [-0.012]
Rho 20.361* 0.192(0.197) (0.229)
Note: Coefficients, standard errors in parentheses, and marginal effects in squared brackets. All estimationscontrol for age, race, and state fixed effects for the respondent’s state of residency at age 14. *** Statisti-cally significant at 1 percent; ** Statistically significant at 5 percent; * Statistically significant at 10 percent.a. Instruments used to identify reduced form problem-drinking equation are cigarette tax at state of birth,cigarette tax at mother’s state of birth in 1947 and problem-drinking grandfather. Chi2 test of joint signif-icance of instruments ¼ 44.29 (p¼0.000)
Balsa 475
from sicker people having a more distorted recall of their past family problems. Hav-ing objective quantitative information on parental drinking would probably improvethe consistency of the estimates and provide better information for policy purposes.At this point, I am unaware of any data set that can provide better information thanthe NLSY on parental drinking and on children’s long term labor market outcomes ata nationally representative level.
Third, it is possible that results are being led by problem-drinking fathers, conceal-ing other interesting effects. Around 20 percent of respondents reported a problem-drinking father while only 4 percent admitted having a problem-drinking mother.The small number of problem-drinking mothers in the data does not provide enoughpower to achieve statistically significant effects. If effects worked across gender lines(fathers affecting sons and mothers affecting daughters), as some literature and pre-liminary analyses of the data appeared to suggest, the weaker findings for femalesmight be due to poor power rather than to genuine effects.
Fourth, the methodology in the main analysis relies on the use of an indicator of aproblem-drinking grandfather as an instrumental variable. While the instrument sat-isfied formally the exclusion restrictions, its exogeneity could be questionable on the-oretical grounds. Thus, by using such variable, the expectation was to at leastmitigate, but not necessarily eliminate biases due to endogeneity.
Table 6Effects of parental problem-drinking on the likelihood of being out of the labor forceand of being unemployed (Female sample)
(1)Any weeks out
of the labor force1996–2002
(2)Any weeksunemployed1996–2002
(1) Probit estimatesProblem-drinking parent 0.157*** 0.070
(0.046) (0.048)[0.061] [0.023]
(2) Recursive bivariate probit estimatesa
Problem-drinking parent 20.145 20.362(0.338) (0.419)
[-0.014] [-0.036]Rho 0.158 0.268
(0.199) (0.250)
Note: Coefficients, standard errors in parentheses, and marginal effects in squared brackets. All estimationscontrol for age, race, and state fixed effects for the respondent’s state of residency at age 14. *** Statisti-cally significant at 1 percent; ** Statistically significant at 5 percent; * Statistically significant at 10 percent.a. Instruments used to identify reduced form problem-drinking equation are beer tax at state of birth,whether there were alcohol controls at father’s state of birth in 1947 and problem-drinking grandfather.Chi2(3) test of joint significance of instruments ¼ 61.21 (p¼0.000)
476 The Journal of Human Resources
Ta
ble
7D
iffe
ren
ces
inB
ack
gro
un
dC
ha
ract
eris
tics
,Ed
uca
tio
n,H
ealt
hS
tatu
s,A
lco
ho
lU
se,a
nd
La
bo
rM
ark
etE
xper
ien
ceb
etw
een
Ch
ild
ren
of
Pro
ble
m-d
rin
kin
gP
are
nts
aa
nd
Oth
erR
esp
on
den
ts
Mal
eR
esponden
tsF
emal
eR
esponden
ts
Pro
ble
m-d
rinkin
gp
aren
tN
op
rob
lem
-dri
nk
ing
par
ent
Pro
ble
m-d
rin
kin
gp
aren
tN
op
rob
lem
-dri
nk
ing
par
ent
Indiv
idual
char
acte
rist
ics
and
fam
ily
bac
kgro
und
(in
19
79
or
atag
e1
4)
AF
QT
39
.58
34
0.1
01
38
.07
23
8.5
50
Mar
ried
in1
97
90
.05
40
.04
50
.15
70
.11
2**
*F
ore
ign
lan
gu
age
atag
e1
40
.21
00
.23
20
.22
40
.23
2N
ot
reli
gio
us
19
79
0.1
58
0.1
16
**
*0
.10
60
.07
6**
*A
tten
ded
reli
gio
us
serv
ices
19
79
0.3
62
0.4
32
**
*0
.43
10
.54
3**
*S
ing
lep
aren
tfa
mil
yat
age
14
0.2
32
0.1
66
**
*0
.22
30
.16
2**
*S
tep
par
ent
fam
ily
atag
e1
40
.15
90
.06
1*
**
0.1
42
0.0
60
***
Oth
ern
on
inta
ctfa
mil
yst
ruct
ure
atag
e1
40
.07
30
.05
5*
**
0.0
91
0.0
59
***
Nu
mb
ero
fsi
bli
ng
s1
97
93
.95
53
.82
24
.05
53
.83
9**
Ln
aver
age
fam
ily
inco
me
19
79
/19
80
10
.45
61
0.5
91
**
*1
0.3
07
10
.50
6*
**
Ln
pov
erty
19
79
0.2
52
0.2
34
0.2
80
0.2
52
**R
ecei
ved
pu
bli
cas
sist
ance
19
79
0.1
81
0.1
32
**
*0
.19
90
.14
3**
*W
hit
eco
llar
fath
er0
.21
90
.24
50
.20
70
.25
4**
*A
du
ltfe
mal
ein
ho
use
ho
ldw
ork
ed0
.54
90
.51
6*
*0
.57
10
.51
8**
*H
igh
est
edu
cati
on
fam
ily
:n
oh
igh
sch
oo
l0
.33
80
.32
2*
*0
.37
30
.34
0**
Hig
hes
ted
uca
tio
nfa
mil
y:
hig
hsc
ho
ol
0.5
32
0.5
12
**
0.5
14
0.4
98
Hig
hes
ted
uca
tio
nin
fam
ily
:co
lleg
eo
r+
0.1
30
0.1
66
**
0.1
13
0.1
61
***
Balsa 477
Fam
ily
had
ali
bra
ryca
rd(a
tag
e1
4)
0.7
15
0.6
76
**
0.7
22
0.7
19
Liv
edin
sam
ere
sid
ence
sin
ceb
irth
(by
19
79
)0
.38
70
.47
9*
**
0.4
12
0.4
77
**
*U
rban
resi
den
cyat
age
14
0.8
40
0.7
84
**
*0
.81
10
.79
2C
ou
nty
of
resi
den
cycr
ime
rate
in1
97
55
6.2
20
53
.59
0*
*5
4.6
80
54
.27
8C
ou
nty
un
emp
loy
men
tra
tein
19
79
6.4
20
6.2
35
**
6.2
17
6.2
41
Educa
tion
No
hig
hsc
ho
ol
or
GE
D0
.15
10
.12
4*
*0
.11
30
.08
9*
**
Hig
hsc
ho
ol
gra
du
ate
0.4
70
0.4
50
0.4
29
0.4
09
So
me
coll
ege
0.2
21
0.2
11
0.2
71
0.2
66
Co
lleg
eg
rad
uat
e0
.10
80
.11
90
.10
50
.13
2*
**
Gra
du
ate
stu
die
s0
.05
10
.09
5*
**
0.0
82
0.1
03
**
Hig
hes
tg
rad
eco
mp
lete
d1
2.7
44
13
.12
9*
**1
3.0
98
13
.38
9**
*H
ealt
hD
epre
ssiv
esy
mp
tom
s1
99
21
1.9
39
10
.82
4*
**1
3.6
40
12
.27
9**
*M
ajor
dep
ress
ion
1992
0.1
27
0.0
93***
0.1
76
0.1
43***
Any
hea
lth
lim
itat
ion
s1
99
6–
20
02
0.1
76
0.1
26
**
*0
.23
50
.18
3*
**
Alc
ohol
use
traj
ecto
ryA
ge
star
ted
dri
nki
ng
15
.92
91
6.3
31
***
16
.67
71
7.0
77*
**
Sta
rted
dri
nki
ng
bef
ore
age
15
0.3
21
0.2
50
**
*0
.19
60
.13
9*
**
Alc
oh
ol
dep
end
ent
in1
98
9o
r1
99
40
.21
00
.13
5*
**
0.0
75
0.0
51
**
*F
amil
ych
arac
teri
stic
s1
99
6–
20
02
Mar
ried
19
96
–2
00
20
.41
10
.48
0*
**
0.4
49
0.4
93
**
*S
epar
ated
19
96
–2
00
20
.23
70
.17
8*
**
0.2
06
0.1
88
Lab
or
mar
ket
exp
erie
nce
(by
19
94
)L
abo
rm
ark
etex
per
ien
ce(y
ears
)1
2.0
15
12
.45
8*
*1
0.0
54
10
.51
6**
*
Note
:**
Dif
fere
nce
bet
wee
nre
sponden
tw
ith
pro
ble
m-d
rinkin
gpar
ent
and
resp
onden
tw
ith
no
pro
ble
m-d
rinkin
gpar
ent
stat
isti
call
ysi
gnifi
cant
at5
per
cent;
***
dif
fer-
ence
stat
isti
call
ysi
gnifi
cant
at1
per
cent.
a.P
aren
tal
pro
ble
m-d
rinkin
g¼
1w
hen
resp
onden
tre
port
edei
ther
abio
logic
alor
nonbio
logic
alpro
ble
m-d
rinkin
gm
oth
eror
fath
er.
478 The Journal of Human Resources
Fifth, the magnitudes of the standard errors in the GMM estimations raise con-cerns about the power of the instruments to identify statistically significant results.This imprecision limits our ability to say anything definitive about causality. Omittedvariable bias could still be behind the detected associations between parental problem-drinking and children’s labor market outcomes.
The psychopathological consequences of parental drinking on adult children havereceived special attention in the area of psychology and family medicine. In econom-ics, however, none of the available estimates of the consequences of alcoholism usedfor policy purposes (to design, finance, and evaluate interventions) account for theimpact of parental drinking on children. These estimates are restricted to measuringthe direct health care costs, productivity losses, and crime costs imposed upon soci-ety by the individual with a drinking problem. This paper highlights the importanceof accounting, in addition, for the losses in productivity suffered by adult sons ofalcoholics, which are manifested through lower labor force participation and lowerwages. Furthermore, it indicates the need to research and quantify the health carecosts that children of alcoholics and society as a whole have to bare due to parentalalcoholism. Policy recommendations, and in particular the types of interventions de-signed, may change substantially if these consequences are taken into consideration.
Balsa 479
Ap
pen
dix
Ta
ble
A1
No
nre
spo
nse
an
dA
ttri
tio
n
Eli
gib
le,
but
no
tin
terv
iew
edin
19
96
–2
00
2In
terv
iew
edat
leas
to
nce
in1
99
6–
20
02
No
resp
on
seto
par
enta
lal
coh
ol
ques
tions
Res
po
nd
edp
aren
tal
alco
ho
lq
ues
tio
ns
Fin
alsa
mp
le
Mea
n(1
)(1
)v
s(3
)aM
ean
(2)
(2)
vs
(3)b
Mea
n(3
)S
tan
dar
dD
evia
tio
nM
inim
um
Max
imu
m
N¼
9,9
86
95
85
34
8,4
94
Lab
or
mar
ket
ou
tco
mes
Wee
ks
out
labor
forc
e(a
vera
ge
95,
97,
99,
01)
N/A
10.8
14***
8.6
50
14.9
08
0.0
00
52.0
00
Wee
ks
unem
plo
yed
(ave
rage
95,
97,
99,
01)
N/A
1.7
86
1.8
43
5.0
02
0.0
00
52.0
00
Ln
ho
url
yw
age
(ave
rag
e9
5,
97
,9
9,
01
)N
/A2
.59
12
.60
20
.72
00
.00
95
.68
7P
aren
tal
alco
ho
lp
rob
lem
sP
roble
m-d
rinkin
gm
oth
er0.0
44
N/A
0.0
43
0.2
028
0.0
00
1.0
00
Pro
ble
m-d
rinkin
gfa
ther
0.1
61**
N/A
0.2
04
0.4
027
0.0
00
1.0
00
Dem
ogra
phic
sA
ge
19
79¼
15
0.1
15
0.1
25
0.1
40
0.3
47
0.0
00
1.0
00
Ag
e1
97
9¼
16
0.1
48
0.1
25
0.1
38
0.3
45
0.0
00
1.0
00
Ag
e1
97
9¼
17
0.1
36
0.1
46
0.1
36
0.3
43
0.0
00
1.0
00
Ag
e1
97
9¼
18
0.1
13
0.1
54
0.1
39
0.3
46
0.0
00
1.0
00
Ag
e1
97
9¼
19
0.1
46
0.1
10
0.1
20
0.3
25
0.0
00
1.0
00
Ag
e1
97
9¼
20
0.1
36
0.1
27
0.1
06
0.3
07
0.0
00
1.0
00
Ag
e1
97
9¼
21
0.1
10
0.1
03
0.1
10
0.3
13
0.0
00
1.0
00
(co
nti
nu
ed)
480 The Journal of Human Resources
Ap
pen
dix
Ta
ble
A1
(Co
nti
nu
ed)
Eli
gib
le,
but
no
tin
terv
iew
edin
19
96
–2
00
2In
terv
iew
edat
leas
to
nce
in1
99
6–
20
02
No
resp
on
seto
par
enta
lal
coh
ol
qu
esti
on
s
Res
po
nd
edp
aren
tal
alco
ho
lq
ues
tio
ns
Fin
alsa
mple
Mea
n(1
)(1
)v
s(3
)aM
ean
(2)
(2)
vs
(3)b
Mea
n(3
)S
tan
dar
dD
evia
tion
Min
imum
Max
imum
Ag
e1
97
9¼
22
0.0
31
0.0
34
0.0
25
0.1
56
0.0
00
1.0
00
Wh
ites
0.5
15
0.4
10
**
*0
.49
30
.50
00
.00
01
.00
0H
isp
anic
0.2
13
**
0.2
92
**
*0
.18
80
.39
10
.00
01
.00
0B
lack
0.2
59
**
*0
.27
2*
*0
.30
70
.46
10
.00
01
.00
0O
ther
race
0.0
13
0.0
26
**
*0
.01
10
.10
60
.00
01
.00
0F
amil
yb
ack
gro
un
d1
97
9/
age
14
AF
QT
43
.09
7*
**3
3.8
35
***
39
.03
42
8.6
63
1.0
00
99
.00
0A
FQ
Tm
issi
ng
0.1
80
**
*0
.17
2*
**
0.0
40
0.1
97
0.0
00
1.0
00
Mar
ried
in1
97
90
.08
80
.10
10
.08
60
.28
10
.00
01
.00
0F
ore
ign
lan
gu
age
0.2
58
**
0.3
46
**
*0
.22
80
.41
90
.00
01
.00
0N
ot
reli
gio
us
19
79
0.1
22
**
0.0
98
0.1
04
0.3
05
0.0
00
1.0
00
Cat
ho
lic
0.3
54
**
0.3
96
**
*0
.31
90
.46
60
.00
01
.00
0B
apti
st0
.19
9*
**
0.2
01
**
*0
.27
40
.44
60
.00
01
.00
0A
tten
ded
reli
gio
us
serv
ices
19
79
0.4
35
**
0.4
49
0.4
66
0.4
99
0.0
00
1.0
00
Sin
gle
-par
ent
fam
ily
0.1
84
0.2
05
0.1
77
0.3
82
0.0
00
1.0
00
Ste
ppar
ent
fam
ily
0.0
70
0.0
73
0.0
81
0.2
72
0.0
00
1.0
00
Oth
ern
on
inta
ctfa
mil
yst
ruct
ure
0.0
60
0.0
62
0.0
63
0.2
43
0.0
00
1.0
00
Balsa 481
Nu
mb
ero
fsi
bli
ng
s1
97
93
.75
93
.97
23
.87
52
.67
20
.00
02
2.0
00
Ln
aver
age
fam
ily
inco
me
19
79
/19
801
0.5
38
10
.40
0*
*1
0.5
07
1.0
92
0.0
00
12
.91
4F
amil
yin
com
em
issi
ng
0.0
86
**
0.1
16
**
*0
.07
00
.25
60
.00
01
.00
0In
pov
erty
19
79
0.2
33
0.3
02
**
*0
.25
00
.43
30
.00
01
.00
0P
over
tyst
atus
mis
sing
0.0
72**
0.0
75**
0.0
58
0.2
33
0.0
00
1.0
00
Rec
eive
dp
ub
lic
assi
stan
ce1
97
90
.13
70
.17
8*
*0
.15
00
.35
70
.00
01
.00
0P
ub
lic
assi
stan
cem
issi
ng
0.0
30
0.0
54
**
0.0
38
0.1
91
0.0
00
1.0
00
Wh
ite
coll
arfa
ther
0.2
69
**
0.2
06
**
0.2
39
0.4
26
0.0
00
1.0
00
Wh
ite
coll
arfa
ther
mis
sin
g0
.05
00
.05
40
.06
30
.24
40
.00
01
.00
0A
dult
fem
ale
inhouse
hold
work
ed0.4
77***
0.4
94
0.5
30
0.4
99
0.0
00
1.0
00
Work
ing
stat
us
of
fem
ale
adult
:m
issi
ng
0.0
08
0.0
11
0.0
09
0.0
94
0.0
00
1.0
00
Hig
hes
ted
uca
tio
nfa
mil
y:
no
hig
hsc
ho
ol
0.3
17
0.3
95
**
*0
.33
80
.47
30
.00
01
.00
0H
igh
est
edu
cati
on
fam
ily
:h
igh
sch
oo
l,n
oco
lleg
e0
.50
10
.47
70
.50
90
.50
00
.00
01
.00
0H
igh
est
edu
cati
on
fam
ily
:co
lleg
eo
r+
0.1
82
**
*0
.12
90
.15
20
.35
90
.00
01
.00
0F
amil
yed
uca
tio
nm
issi
ng
0.0
39
0.0
41
0.0
37
0.1
90
0.0
00
1.0
00
Fam
ily
had
ali
bra
ryca
rd0
.72
9*
*0
.69
80
.69
90
.45
90
.00
01
.00
0R
esid
ency
1979
or
age
14
Liv
edin
sam
ere
sid
ence
sin
ceb
irth
(by
19
79
)0
.43
60
.41
2*
**
0.4
61
0.4
99
0.0
00
1.0
00
Urb
anre
sid
ency
atag
e1
40
.79
90
.82
50
.79
50
.40
40
.00
01
.00
0C
ou
nty
of
resi
den
cycr
ime
rate
in1
97
56
0.7
95
**
*6
0.2
28
**
*5
3.9
31
35
.57
30
.11
04
06
.87
0C
ou
nty
crim
era
tem
issi
ng
0.0
73
*0
.09
4*
**
0.0
57
0.2
32
0.0
00
1.0
00
Co
un
tyu
nem
plo
ym
ent
rate
in1
97
96
.30
36
.41
8*
6.2
52
1.9
51
2.4
00
13
.20
0U
nem
plo
ym
ent
rate
mis
sin
g0
.03
10
.05
2*
**
0.0
33
0.1
78
0.0
00
1.0
00
Reg
ion
of
resi
den
cy1979:
Nort
h0.2
46
0.2
18*
0.2
52
0.4
34
0.0
00
1.0
00
Reg
ion
of
resi
den
cy1979:
South
0.2
94***
0.3
32*
0.3
74
0.4
84
0.0
00
1.0
00
Reg
ion
of
resi
den
cy1979:
Wes
t0.2
05
0.2
16
0.1
89
0.3
92
0.0
00
1.0
00
Educa
tion
No
hig
hsc
ho
ol
or
GE
D0
.16
1*
**
0.1
69
**
*0
.11
10
.31
40
.00
01
.00
0H
igh
sch
oo
lg
rad
uat
e0
.36
6*
**
0.4
23
0.4
37
0.4
96
0.0
00
1.0
00
(co
nti
nu
ed)
482 The Journal of Human Resources
Ap
pen
dix
Ta
ble
A1
(Co
nti
nu
ed)
Eli
gib
le,
but
no
tin
terv
iew
edin
19
96
–2
00
2In
terv
iew
edat
leas
to
nce
in1
99
6–
20
02
No
resp
on
seto
par
enta
lal
coh
ol
qu
esti
on
s
Res
po
nd
edp
aren
tal
alco
ho
lq
ues
tio
ns
Fin
alsa
mp
le
Mea
n(1
)(1
)v
s(3
)aM
ean
(2)
(2)
vs
(3)b
Mea
n(3
)S
tan
dar
dD
evia
tio
nM
inim
um
Max
imu
m
So
me
coll
ege
0.2
25
0.2
47
0.2
41
0.4
28
0.0
00
1.0
00
Co
lleg
eg
rad
uat
e0
.15
5*
0.0
88
0.1
19
0.3
24
0.0
00
1.0
00
Gra
du
ate
stu
die
s0
.09
30
.07
30
.09
20
.28
90
.00
01
.00
0H
igh
est
gra
de
com
ple
ted
13
.10
71
2.7
68
**
*1
3.1
82
2.4
94
0.0
00
20
.00
0R
esid
ency
19
96
–2
00
2U
rban
resi
den
cy1
99
6–
20
02
N/A
0.7
56
0.7
48
0.3
58
0.0
00
1.0
00
Liv
edin
the
No
rth
19
96
–2
00
2N
/A0
.17
1*
**
0.2
24
0.4
17
0.0
00
1.0
00
Liv
edin
the
So
uth
19
96
–2
00
2N
/A0
.37
10
.38
90
.48
80
.00
01
.00
0L
ived
inth
eW
est
19
96
–2
00
2N
/A0
.22
9*
**
0.1
88
0.3
90
0.0
00
1.0
00
Ch
ang
edre
gio
ns
19
96
–2
00
2N
/A0
.04
70
.04
90
.21
50
.00
01
.00
0R
egio
no
fre
sid
ency
mis
sin
g1
99
6–
20
02
N/A
0.5
60
**
*0
.20
50
.40
30
.00
01
.00
0C
ounty
of
resi
den
cyunem
plo
ym
ent
rate
1996–2002
N/A
6.4
10***
6.0
87
2.3
07
2.0
40
18.1
25
Fam
ily
char
acte
rist
ics
19
96
–2
00
2M
arri
ed1
99
6–
20
02
N/A
0.4
68
0.4
74
0.4
99
0.0
00
1.0
00
Sep
arat
ed1
99
6–
20
02
N/A
0.1
39
**
*0
.19
10
.39
30
.00
01
.00
0N
um
ber
of
kid
sN
/A1
.31
41
.40
11
.23
00
.00
08
.75
0
Balsa 483
Hea
lth
Dep
ress
ive
sym
pto
ms
11
.64
01
2.0
12
11
.85
18
.39
50
.00
05
6.0
00
Dep
ress
ive
sym
pto
ms
mis
sin
g0
.67
8*
**
0.2
02
**
*0
.02
90
.16
70
.00
01
.00
0M
ajo
rd
epre
ssio
n0
.14
00
.15
00
.12
50
.33
10
.00
01
.00
0H
ealt
hli
mit
atio
ns
N/A
0.1
72
0.1
67
0.3
73
0.0
00
1.0
00
Hea
lth
lim
itat
ions
mis
sing
N/A
0.0
11**
0.0
04
0.0
67
0.0
00
1.0
00
Alc
oh
ol
use
traj
ecto
ryA
ge
star
ted
dri
nk
ing
16
.64
61
6.5
64
16
.59
32
.21
82
.00
02
5.0
00
Sta
rted
dri
nkin
gbef
ore
age
15
0.2
06
0.2
20
0.2
09
0.4
07
0.0
00
1.0
00
Alc
oh
ol
dep
end
ent
in1
98
9o
r1
99
40
.08
0*
*0
.05
9*
**
0.1
04
0.3
05
0.0
00
1.0
00
Alc
oh
ol
dep
end
ent
mis
sin
g0
.50
20
.10
7*
**
0.0
28
0.0
53
0.0
00
1.0
00
Lab
or
mar
ket
exp
erie
nce
(by
19
94
)L
abor
mar
ket
exper
ience
(yea
rs)
N/A
10.6
76
11.3
09
4.3
77
0.0
00
16.9
62
Lab
or
mar
ket
exper
ience
mis
sing
N/A
0.9
74***
0.3
63
0.4
81
0.0
00
1.0
00
N9
58
53
48
,49
4
a.*
Dif
fere
nce
bet
wee
nm
ean
(1)
and
mea
n(3
)is
stat
isti
call
ysi
gnifi
cant
at5
per
cent;
***
dif
fere
nce
stat
isti
call
ysi
gnifi
cant
at1
per
cent
b.*
Dif
fere
nce
bet
wee
nm
ean
(2)
and
mea
n(3
)is
stat
isti
call
ysi
gnifi
cant
at5
per
cent;
***
dif
fere
nce
stat
isti
call
ysi
gnifi
cant
at1
per
cent.
484 The Journal of Human Resources
References
Aizer, Anna. 2004. ‘‘Home Alone: Supervision after School and Child Behavior.’’ Journal ofPublic Economics 88(9–10):1835–48.
Amuedo-Dorantes, Catalina, and Traci Mach. 2002. ‘‘The Impact of Families on JuvenileSubstance Use.’’ Journal of Bioeconomics 4(3):269–82.
Anda, Robert, Charles Whitfield, Vincent Felitti, Daniel Chapman, Valerie Edwards, ShantaDube, and David Williamson. 2002. ‘‘Adverse Childhood Experiences, Alcoholic Parents,and Later Risk of Alcoholism and Depression.’’ Psychiatric Services 53(8):1001–09.
Antecol, Heather, and Kelly Bedard. 2007. ‘‘Does Single Parenthood Increase the Probabilityof Teenage Promiscuity, Substance Use and Crime?’’ Journal of Population Economics20(1):55–71.
Balsa, Ana. 2006. ‘‘The Impact of Parental Drinking on Children’s Utilization of Health Care
Services.’’ Coral Gables: University of Miami. Unpublished.Berger, Lawrence. 2005. ‘‘Income, Family Characteristics, and Physical Violence toward
Children.’’ Child Abuse and Neglect 29(2):107–33.The Book of States. 1947-64. Lexington, KY.: Council of the State Governments.Booth, Brenda, and Weiwei Feng. 2002. ‘‘The Impact of Drinking and Drinking
Consequences on Short-term Employment Outcomes in at-Risk Drinkers in Six SouthernStates.’’ The Journal of Behavioral Health Services and Research 29(2):157–66.
Cameron, Lisa, and Jenny Williams. 2001. ‘‘Cannabis, Alcohol and Cigarettes: Substitutes orComplements?’’ Economic Record 77(236):19–34
Chatterji, Pinka, and Sara Markowitz. 2001. ‘‘The Impact of Maternal Alcohol and IllicitDrug Use on Children’s Behavior Problems: Evidence from the Children of the NationalLongitudinal Survey of Youth.’’ Journal of Health Economics 20(5):703–31.
Chatterji, Pinka, and Jeff DeSimone. 2005. ‘‘Adolescent Drinking and High School Dropout.’’NBER Working Paper No. W11337. Available at SSRN: http://ssrn.com/abstract¼723306
Connell, Arin, and Sherryl Goodman. 2002. ‘‘The Association between Psychopathology inFathers Versus Mothers and Children’s Internalizing and Externalizing Behavior Problems:A Meta-Analysis.’’ Psychological Bulletin 128(5):746–73.
Dee, Thomas, and William Evans. 2003. ‘‘Teen Drinking and Educational Attainment:Evidence from Two-Sample Instrumental Variable Estimates.’’ Journal of LaborEconomics 21(1):178–209.
Dobkin, Patricia, Richard Tremblay, Lyse Desmarais-Gervais, and Louise Depelteau. 1994.‘‘Is Having an Alcoholic Father Hazardous for Children’s Physical Health?’’ Addiction89(12):1619–27.
Ellis, Deborah, Robert Zucker, and Hiram Fitzgerald. 1997. ‘‘The Role of Family Influencesin Development and Risk.’’ Alcohol Health and Research World 21(3):218–26.
Feng, Weiwei, Wei Zhou, J.S. Butler, Brenda Booth, and Michael French. 2001. ‘‘The Impactof Problem-drinking on Employment,’’ Health Economics 10(6):509–21.
French, Michael, and Johanna Maclean. 2006. ‘‘Underage Alcohol Use, Delinquency, andCriminal Activity.’’ Health Economics 15(12):1261–81.
Garis, Dalton. 1998. ‘‘Poverty, Single-Parent Households, and Youth at-Risk Behavior: AnEmpirical Study.’’ Journal of Economic Issues 32(4):1079–06.
Hansen, Lars. 1982. ‘‘Large Sample Properties of Generalized Method of MomentsEstimators.’’ Econometrica 50(3):1029–54.
Hayashi, Fumio. 2000. Econometrics. Princeton, N.J.: Princeton University Press.Jennison, Karen, and Kenneth Johnson. 1998. ‘‘Alcohol Dependence in Adult Children
of Alcoholics: Longitudinal Evidence of Early Risk.’’ Journal of Drug Education 28(1):19–37.
Balsa 485
———. 2001. ‘‘Parental Alcoholism as a Risk Factor for DSM-IV-Defined Alcohol Abuseand Dependence in American Women: The Protective Benefits of Dyadic Cohesion inMarital Communication.’’ American Journal of Drug and Alcohol Abuse 27(2):349–74.
Jones, Alison, Deborah Miller, and David Salkever. 1999. ‘‘Parental Use of Alcohol andChildren’s Behavioural Health: A Household Production Analysis.’’ Health Economics8(8):661–83.
Koch, Steven, and Kerry McGeary. 2005. ‘‘The Effect of Youth Alcohol Initiation on HighSchool Completion.’’ Economic Inquiry 43(4):750–65.
Koch, Steven, and David Ribar. 2001. ‘‘A Siblings Analysis of the Effects of AlcoholConsumption Onset on Educational Attainment.’’ Contemporary Economic Policy19(2):162–74.
Markowitz, Sara, and Michael Grossman. 1998. ‘‘Alcohol Regulation and Domestic Violencetowards Children.’’ Contemporary Economic Policy 16(3):309–20.
Markowitz, Sara, and John Tauras. 2006. ‘‘Even for Teenagers, Money Does Not Grow onTrees: Teenage Substance Use and Budget Constraints.’’ Cambridge: NBER Working Paper12300.
Mattson, Sarah, and Edward Riley. 1998. ‘‘A Review of the Neurobehavioral Deficits inChildren with Fetal Alcohol Syndrome or Prenatal Exposure to Alcohol.’’ Alcoholism:Clinical & Experimental Research 22(2):1–4.
McDonald, Ziggy, and Michael Shields. 2001. ‘‘The Impact of Alcohol Consumption onOccupational Attainment in England.’’ Economica 68(271):427–53.
McGue, Matt. 1997. ‘‘A Behavioral-Genetic Perspective on Children of Alcoholics.’’ AlcoholHealth and Research World 21(3):210–17.
Mullahy, John, and Jody Sindelar. 1996. ‘‘Employment, Unemployment, and Problem-Drinking.’’ Journal of Health Economics 15(4):409–34.
National Institute on Alcohol Abuse and Alcoholism (NIAAA). 2000. Tenth Special Report tothe U.S. Congress on Alcohol and Health. Rockville, Md.: National Institutes of Health.
Picone, Gabriel, Frank Sloan, Justin Trogdon. 2004. ‘‘The Effect of the Tobacco Settlementand Smoking Bans on Alcohol Consumption.’’ Health Economics 13(10):1063–80.
Schuckit, Marc. 1999. ‘‘New Findings in the Genetics of Alcoholism.’’ Journal of theAmerican Medical Association 281(20):1875–76.
Seo, G. 1998. ‘‘The Impact of Maternal Problem-Drinking on Children’s DevelopmentalOutcomes: Focus on Parenting As Mediator.’’ Ph.D. Dissertation. The Ohio StateUniversity. Unpublished
Williams, Jenny, Lisa Powell, and Henry Wechsler. 2003. ‘‘Does Alcohol ConsumptionReduce Human Capital Accumulation? Evidence from the College Alcohol Study.’’Applied Economics 35(10):1227–39.
Windle, Michael. 1996. ‘‘On the Discriminative Validity of a Family History of Problem-drinking Index with a National Sample of Young Adults.’’ Journal of Studies on Alcohol57(4):378–86.
Yamada, Tetsuji, Michael Kendrix, and Tadashi Yamada. 1996. ‘‘The Impact of AlcoholConsumption and Marijuana Use on High School Graduation.’’ Health Economics 5(1):77–92.
486 The Journal of Human Resources