HEALTH ECONOMICS, VOL.

5: 77-92 (1996)
HEALTH PROMOTION

THE IMPACT OF ALCOHOL CONSUMPTION AND

MARIJUANA USE ON HIGH SCHOOL GRADUATION
TETSUn YAMADA’, MICHAEL KENDIX’ AND TADASHI YAMADA’
‘ Department of Economics, Rutgers University, USA;
‘I Health Care Financing Administration, Ojice of
Research & Demonstrations. USA; Institute of Socio-Economic Planning, University of Tsukuba,Japan.

SUMMARY
In this study we use data from the National Longitudinal Survey of Youth (NLSY) to estimate the relationship
between high school graduation, and alcohol and marijuana use among high school students. We also estimate the
demand determinants for each of these substances.
Our results show that there are significmt adverse effects of alcohol and marijuana use on high school
graduation. In particular, increases in the incidence of frequent drinking, liquor and wine consumption, and
frequent marijuana use, significantly reduce the probability of high school graduation. Our results also show that
beer taxes, liquor prices and marijuana decriminalization have a significant impact on the demand for these
substances.
These findings have important policy implications. A ten percent increase in beer taxes, reduces alcohol
consumption among high school students, which in turn raises the probability of high school graduation by about
three percent. A 1 percent increase in liquor prices raises the probability of high school graduation by over 1
percent. Raising the minimum drinking age for liquor also reduces liquor and wine consumption, and thus,
improves the probability of high school graduation. Although the relationship between marijuana decriminalization
and marijuana use is not significant, decriminalization is found to reduce the probability of becoming a frequent
drinker. This result suggests that marijuana use and frequent drinking are substitute activities.
Illicit substance abuse reduces the rate of high school completion, reduces expected future earnings and creates
potential health problems. Thus, high-school-based preventive programs which discourage alcohol consumption and
marijuana use are highly recommended, in order to alleviate these problems.

KEY WORDS-alcohol; marijuana; education

During the 198Os, the Federal Government placed
a high priority on helping young people who were
at risk of alcohol and drug abuse. In 1984,
Congress passed the Federal Uniform Drinking
Age Act (Public Law 98-363), which had a
significant effect on young persons’ consumption
of alcoholic beverages. The Anti-Drug Abuse
Act of 1986 (Public Law 99-570), allowed a
number of Federal agencies to address alcohol and
drug abuse problems. The role of government at
the Federal, State and Local level was broadened
by the Anti-Drug Abuse Act of 1988 (Public Law
100-690). A program entitled Healthy People
2000, has been co-sponsored by the Department
of Labor and the Public Health Service. This
program focuses on the objectives set for the year
2000. It is aimed at young people at risk of subs-
tance abuse and is designed to reduce their alcohol
and drug use. In addition to these efforts, other
programs have been directed toward the general
population; for example, in recent years there has
been a significant increase in the frequency of

Address for correspondence: Tetsuji Yamada, Department of Economics, Rutgers University, The State University of NJ,
Camden, New Jersey 08102, U.S.A

CCC 1057 -9230/%/010077- 16 Received 8 August 1994

Q 19% by John Wiley & Sons, Ltd. Accepted 7 November 1995
78 T.YAMADA, M. KEh'DlX A N D 1:YAMADA
television warning messages against drunk driving
and drug abuse.'
W e the aforementioned programs and legisla-
tive measures have been partially successful in
reducing youth drug and alcohol abuse, these
problems persist and remain a major social issue.
According to the General Report on Youth and
Alcohol, published by the U.S. Department of
Health and Human Services in 1992, 89.5 percent
of high school seniors in the class of 1990 had
consumed alcohol at least once during the two
weeks before the survey date. Further, 32.2 percent
experienced a binge of five or more drinks in a
row. Junior and senior high school students con-
sume an average of 1.1 billion bottles and cans of
beer each year.2
The General Report also discusses issues related
to drug use. According to the Report, the perceived
availability of illicit drugs among 8th graders (13
to 14 years old) is 11.4% for heroin, 12.4% for
LSD,14.3% for crack cocaine, 14.5% for cocaine
powder and 25.9% for marijuana. These numbers
increase for 10th grade students to 17.4% for
heroin, 23.6% for LSD,25.9% for crack cocaine,
26.7% cocaine powder and 53.9% for marijuana.
The numbers are even higher for seniors; for
example, 30.6 % for heroin and 83.3% for mari-
juana. Policies to combat illicit drug use have
varied.'*3 For example, some states and localities
have decriminalized marijuana use, while others
have substantial penalties for even its causal use.
In total, according to the Report, about three
million persons aged between 10 and 17 experi-
ence various problems resulting from alcohol and
drug abuse.
There is a large literature on the efficacy of
decriminalization of drug use and its relationship
to alcohol and illicit drug use. Goldstein and
Kalant4 report that policies which simultaneously
reduce demand and supply are the key to alleviat-
ing the drug problem. Some studies have suggested
that young ople regard alcohol and marijuana as
substitutes.' In addition, there is an apparent
relationship between the age of initial drug use,
and the likelihood of moving from soft to hard
drug use, among senior high school students.
Kandel and Yamaguchi' report that the progression
to various drug classes runs from the use of
alcohol or cigarettes, to the use of marijuana, and
finally to hard drugs. They find that, on average,
each step takes about two years.
Studies of the demand for alcoholic beverages
have reported various findings. Excise taxes on
beer are highly effective in reducing beer con-
sumption among persons aged 16 to 21, since beer
is generally their drink of The estimated
own price elasticities of alcohol use for youths are
elastic, i.e., between - 1.53 and - 1.54 for beer and
between -3.29 and -4.08 for liquor." This sug-
gests that there are substantial negative price
effects on youth alcohol use. Youths who are
frequent beer drinkers are more price sensitive
than those who drink less." Further, among col-
lege students, those who consume alcohol less
frequently, have a higher probability of a success-
ful college career."
enn nett'^ investigates the impact of prevention
policies on drug abuse among youths, and finds
these programs to be highly effective. In an analy-
sis of the demand for psychoactive drugs,
Garrison, et 01.'~ report that suicidal behaviour is
associated with illicit substance use among high
school students.
The primary objective of our research is to
investigate the sensitivity of high school gradua-
tion to alcohol and marijuana consumption, using
a behavioural structural model, and to estimate the
demand determinants of alcohol and marijuana
use by high school students, using a reduced form
model. We hypothesize that high school students'
accomplishments are sensitive to substance abuse,
and this behaviour is in turn determined by the
pecuniary and other related costs of alcohol and
drug use. In particular, early alcohol abuse has a
large influence on human capital formation,
directly impacting health status and lifetime earn-
ings. Youth alcohol abuse has been analyzed using
a life-cycle framework, explicitly modelling the
impact on educational attainment, occupational
choice and earning^.'^"^ Indeed, unless one
considers indirect effects, such as educational
accomplishment, marriage and divorce (or separa-
tion), the full economic costs of alcoholism will
not be taken into account, leading to an underesti-
mation of the potential gains of early interven-
tion." Applying an instrumental variable approach
to data from the 1988 Alcohol Survey of the
National Health Interview Survey, Mullahy and
Sindelar" find a large negative impact of problem
drinking on employment. Adolescents who
progress to using cocaine, begin by smoking
cigarettes, drinking alcohol, or smoking mari-
juana.' The progression to more serious forms of
drug use leads to a deterioration in health, which
can be viewed as a deterioration of human capi-
tal.'3.'9.20Although the results are not always
 ALCOHOL, DRUGS AND HIGH SCHOOL GRADUATION
conclusive, many studies show that both marijuana
and cocaine use have negative effects on wages
among men, and long-term use or on-the-job use
of marijuana negatively affects wages in the youth
population.21vt2
Despite repeated reports relating high school
dropout rates to drug and alcohol use and the
implication for social and health policies, rela-
tively few studies have documented the effects of
alcohol consumption and marijuana use on high
school completion. Specifically, our paper
addresses the impact of high school students’
alcohol consumption and marijuana use on high
school graduation, and the policy implications
from our estimates of these effects.
The organization of the paper is as follows. The
next section describes our empirical framework
and the data. The empirical results are then pre-
sented and are followed by the policy implications
and conclusions.
EMPIRICAL FRAMEWORK AND DATA
The analysis of substance abuse in the traditional
human capital framework specifies health status as a
determinant of human capital.16-18-23 Human capital,
through health status, generally declines with
ageing, illness, accidents and substance abuse
(alcohol and illegal drug use). 19.20 Substance abuse,
especially early abuse, is viewed as reducing pro-
ductive human capital, which reduces educational
achievement and lowers tential lifetime income.
Mullahy and Sindelarls*Rodiscussthe effects of
drinking on earnings and educational achievement,
in a life cycle context. Cook and Moore” use a
simple recursive, two period model, where past
drinking determines whether the student stays in
school. Early onset of alcohol and drug abuse can
establish addictive behaviour, which may lead to the
use of harder drugs; for example, heroin and
cocaine. In sum, viewed from a life cycle perspec-
tive, the decline in human capital may lead to
reduced eamings, higher absenteeism, reduced
employment and higher unemployment. Another
perspective on substance abuse is the so-called
‘RationalAddiction’ model, which has been devel-
oped and tested by Becker and Murphy,24Becker,
Grossman and Murphy,2sand Grossman.%Among
other things, this paradigm predicts that individuals
with higher inter-temporal discount rates are more
likely to become addicted, and that in the long-run
demand for addictive goods tend to be more elastic
79
than the demand for non-addictive goods. The
rational addiction model is applied to alcohol con-
sumption by Chaloupka er a1.,2’ using pooled time
series and cross section m ual data at the U.S.state
level, and Grossman, Qlaloupka and Sirtalan,=
using panel data from the High School Senior
Swvey, conducted between 1976 and 1985. They
fmd that the model of rational addictive behaviour
explains frequent and heavy alcohol consumption.
ChaloupkaBapplies the model of rational addiction
to cigarette consumption. He linds that more heavily
addicted (myopic) individuals are more responsive
to changes in price than less heavily addicted (less
myopic) individuals. Becker, Grossman and
Murphyx estimate elasticities, which show that
long-run responses to price changes exceed short
run responses.
Some of the aforementioned studies are set in a
long run or life cycle context, where a loss of
human capital arises from heavy illicit substance
use, addictive or otherwise. Our analysis is
simplified in that we examine the contem-
poraneous effects of substance abuse on
adolescents who are in an early stage of their
development. Nevertheless, our analysis has the
usual implications for the development of human
capital and its impact on lifetime experience.
A typical high school student is assumed to be a
rational individual (i), who maximizes his or her
utility. An individual’s utility is assumed to be a
function of high school graduation (G) and a
composite of other consumption goods (Z):
U i = U(Gi,Zi) U’ > O and U”<O. (1)
High school graduation, G,is a function of illicit
substance use (S), for example, alcohol consump
tion or marijuana use:
G ;= Gi(S;,Xi), S E X, (2)
where the vector X i contains the observable
characteristics of the individual; for example, his
or her family structure, ethnic background, poten-
tial academic performance, and other factors that
may influence the graduation outcome. The student
chooses the levels of S;and Zi to maximize utility,
subject to a budget constraint. We assume that
utility increases when high school graduation
becomes more likely. When Si is positive, there
are two effects on utility. First, utility increases
due to the direct impact of Si. Secondly, there is
an indirect effect. If Si exceeds some threshold
80 T. YAMADA, M. KENDIX AND T. YAMADA
level, which may vary across individuals, the
probability of high school graduation is lower, and
utility declines. In essence, we assume that a
rational student trades off a higher probability of
high school graduation, for alcohol and marijuana
use. Thus, the model contains no explicit inter-
temporal effects. Implicitly, however, these enter
through Gi, since the utility gained from high
school graduation is partially derived from the
expected future benefit of increased human capi-
tal. Ideally, the price of a composite good should
be included directly in the model. In our structural
behavioural model, however, our treatment of the
price of composite good follows that of Cook and
Moore,’’ and Chaloupka and Laixuthai.’ That is,
we use variables that measure the implicit relative
costs of consuming alcohol or marijuana; for
example, alcohol taxes, the minimum drinking age
and marijuana decriminalization.
The empirical specification for high school
graduation (G) and substance use (S) can be
expressed as follows:
Gj=Sja+ 0;s+ S€,y+ Qjd + e;, (3)
and
Sj = P;K+ Lip-+ D j q + SEiC + Q;E + E; (4)
where eiand E ; represent the unobservable deter-
minants of high school graduation and illicit
substance use, respectively. D, is a vector of
regressors capturing demographic factors such as
sex, race and family structure. SEi is a vector of
socioeconomic factors, which includes poverty,
parents’ educational attainment and religion. Qi is
a vector of exogenous measures of student
quality, namely, the ASVAB score and class rank.
In the equation modelling substance use (4), we
include additional variables. Pi is a vector of
regressors containing the beer tax rate and liquor
price. L j is a vector of exogenous factors measur-
ing legal factors, such as marijuana decriminaliz-
ation and the minimum drinking age for beer and
liquor. We use two different approaches to
measure the impact of illicit substance use
(alcohol and drug use in general) on high school
graduation. First, we use a probit model to
estimate the direct effect of substance use on
graduation. Secondly, we model alcohol and
marijuana demand. In the demand estimations we
use a probit to model frequent drinking and mari-
juana use (dichotomous dependent variables) and
least squares to estimate liquor and wine con-
sumption (a continuous dependent variable). An
alternative approach would have been to use a
reduced form model, replacing the alcohol and
marijuana use variables in the probability of high
school graduation equation with other determi-
nants. We leave this approach for future study, and
confine our analysis to a structural behavioural
model.
The data for our study is from a panel survey;
the National Longitudinal Survey of Youth
(NLSY). The NLSY began collecting data on
12,686 persons between the ages of 14 and 21 in
1979 and has continued each year. We use a
sample of high school students who were in the
twelfth grade (mostly 17 and 18 year olds) during
the 1981-82 academic year; 1035 students in the
NLSY sample were attending the twelfth grade
in 1982. The NLSY deliberately over-samples
minorities, enabling more precise estimates of the
impact of race variables. The set of 1035
individuals is reduced to a sample of 672 persons
in our analysis, since not all the variables were
available for all respondents, in particular, some
of the key variables such as frequent marijuana
use, frequent drinking, poverty indicator, aca-
demic test scores and high school class rank. In
order to maintain as large a sample size as poss-
ible, we used a dummy variable to indicate when
the class rank is missing. The summary statistics
of variables used are reported in Table 1. Calcula-
tions of summary statistics for some of the main
variables, using the full 1035 person sample are
provided for comparison in Table 2. There is no
significant difference between the summary statis-
tics calculated for the 1035 and 672 person
samples. There is a possibility that our results
could change, if all the variables for all the stud-
ents were available, however, there is no a priori
reason to expect such a change based on the above
comparison.
We focus on students in the twelfth grade during
the 1981-82 academic year because this was the
first year in which questions were asked about
concurrent drinking activity. Most of the cited
literature that is relevant to our analysis is from a
more recent period than our data. Consequently,
there is a small but finite possibility of a funda-
mental change in behaviour between the early
1980s and more recent times. Unfortunately, there
are insufficient observations beyond the early
1980’s in the NLSY for the purpose of this study.
Other recent studies have used older data; for
 ALCOHOL, DRUGS AND HIGH SCHOOL GRADUATION 81
Table 1. Variable definitions and statistics
~~ ~~~

High school graduation

Dummy variable equal to 1 if the student graduated at the end of the 1982 academic year, and 0 otherwise.
(Mean = 0.939,S.D. = 0.240)
Frequent drinker
Dummy variable equal to 1 if the student reported at least 2 days drank during the past week, when alcohol was
consumed. (Same as in Cook and Moore, 1993),and 0 otherwise.
(Mean = 0.142,S.D. = 0.349)
Liquor and wine consumption
Number of liquor and wine drinks the student reported consuming during the past week.
(Mean = 0.625,S.D. = 2.094)
Frequent marijuana user
Dummy variable equal to 1 if the student reported using marijuana during every calendar month of the school year
between September 1981 and June 1982,and 0 otherwise.
(Mean = 0.111, S.D. = 0.314)
Male
Dummy variable equal to 1 for males and 0 for females.
(Mean = 0.510,S.D. = 0.500)
Black
Dummy variable equal to 1 for blacks and 0 otherwise.
(Mean = 0.235,S.D. = 0.424)
Hispanic
Dummy variable equal to 1 for hispanics and 0 otherwise.
(Mean = 0.134,S.D.= 0.341)
Intact Family
Dummy variable equal to 1 if the student lived with own parents at the age of 14:either biological, adoptive or
step-parents, and 0 otherwise.
(Mean = 0.817,S.D. = 0.387)
More than 4 siblings
Dummy variable equal to 1 if the student has more than 4 siblings and 0 otherwise.
(Mean = 0.194,S.D. = 0.395)
Parents’ high school education
Dummy variable equal to 1 if both parents completed 12 years of education (high school education), and 0
otherwise.
(Mean = 0.542,S.D. = 0.499)
Poverty line
Dummy variable equal to 1 if the student’s family is below the poverty line in 1982,and 0 otherwise.
(Mean = 0.225,S.D. = 0.418)
Class rank percentile
Percentile ranking of student in their class in 1981:the numerical value equals 1 if the student is ranked in the top 1
percent.
(Mean = 30.695,S.D.= 32.533)
Missing class rank
Dummy variable equal to 1 if the class rank of the student is missing, and 0 otherwise.
(Mean = 0.352,S.D. = 0.478)
ASVAB Score
Armed Services Vocational Aptitude Tests administered in 1980. Score is the sum of the average of the three
scores on the Mathematics tests and the average of two scores on the English tests.
(Mean = 94.755,S.D. = 16.932)
(continued)
82 T.YAMADA, M.KENDIX AND T.YAh4ADA
Table 1. Continued

Religion
Dummy variable equal to 1 if the student reported attending religious services at least once a week, and 0 otherwise.
(Mean = 0.366,S.D. = 0.482)
Alcoholic parent
Dummy variable equal to 1 if the student reported that either parent had a drinking problem, and 0 otherwise.
(Mean = 0.202,S.D. = 0.401)
Beer tax
Tax on per case of beer in the student’s state of residence.
(Mean = 0.436,S.D. = 0.421)
Liquor price
Liquor price in the student’s state: average price of I1 different brands of liquor, reported in the Annual Statistical
Review 1982,published by the Distilled Spirits Council of the United States.
(Mean = 8.169,S.D. = 0.601)
Marijuana decriminalization
Dummy variable equal to 1 if using marijuana was decriminalized in the student’s state of residence in 1982,and 0
otherwise.
(Mean = 0.279,S.D.= 0.449)
Age 18 for beer
Dummy variable equal to 1 if the minimum age for drinking beer was 18 in the student’s state of residence in 1982,
and 0 otherwise.
(Mean = 0.273,S.D. = 0.446)
Age 18 for liquor
Dummy variable equal to 1 if the minimum age for drinking liquor was 18 in the student’s state of residence in
1982,and 0 otherwise.
(Mean = 0.148,S.D. = 0.355)
Note: Mean and S.D.are the sample mean and standard deviation, respectively. 672 observations.

Table 2. Statistics of some basic variables by 1035
observations
High school graduation [Mean = 0.930,S.D. = 0.255J
Male Wean = 0.515,S.D. = 0.5001
Black Wean = 0.284,S.D.= 0.451 J
Hispanic Wean = 0.157,S.D. = 0.3641
Intact Family [Mean = 0.745,S.D.= 0.4361
More than 4 siblings Wean = 0.231, S.D. = 0.4221
example the study by Grossman, Chaloupka and
Sirtalan” cited above, and Cook and Moore’’ who
use the NLSY data.
Data on illicit substance use other than alcohol,
uses retrospective responses to survey questions.
Thus, the data on marijuana use is retrospective
and was not obtained until the 1984 interview.
Finally, it should be emphasized that we do not
consider the educational outcome or experience of
our sample of students beyond 1982. It is entirely
possible that some students who did not graduate
on time, were able to graduate later. In a sense,
therefore, we are concentrating on the academic
progress of twelfth graders, rather than graduation
per se.
During the course of our investigation, we
experimented with a number of different variables
to measure alcohol consumption and marijuana use
among high school students. In their study of
college students’ alcohol consumption, Cook and
Moore’’ define a ‘frequent drinker’ as a respon-
dent who drinks alcohol two or more days in the
past week. They also use a ‘frequent drunk’ vari-
able, defined as a respondent who reports at least
four occasions of consuming six or more drinks
during the past month. In our study, we use two
definitions of alcohol use; first, Cook and Moore’s
‘frequent drinker’ and second, the number of
liquor and wine drinks the student reported con-
suming in the past week It should be noted that a
12 ounce can of beer, a shot of liquor and a 4
ounce glass of wine (the standard definition of
one drink in the data) contain the same amount
 ALCOHOL,DRUGS AND HIGH SCHOOL GRADUATION
of alcohol (see the relevant studies by Grossman
et 0 1 . ~ ~ ’ Coate and Grossman,” Grossman,*
Chaloupka et aL2’).However, despite the fact that
the alcohol content of one drink (wine, beer or
liquor) is the same, we found more significant
results for students using more concentrated forms
of alcoholic drinks, namely, wine and liquor.
Among the respondents in our data set, 25 percent
of students admitted consuming at least one beer
in the past week, compared with only 17 percent
reporting at least one drink of either liquor or
wine.
As mentioned above, questions about illicit drug
use were not asked in the NLSY before 1984. The
respondents were asked to list the months in which
they used marijuana from 1979 onward. We use
the responses to this set of questions regarding use
during the 1981-82 academic year (September
1981 to June 1982). If the student responded ‘yes’
to using marijuana in each of the ten months
during the academic year, they were classified as a
‘frequent marijuana user.’ In this way, we focus
on chronic and sustained use of marijuana, and its
impact on high school students’ graduation. We
recognize that alternative measures of marijuana
use could be derived from the data. For example,
we could have used those students who admitted
use in at least half the months during the academic
year, or the number of months during the aca-
demic year when marijuana was used. Given the
method of data collection for this question, how-
ever, we thought it best to concentrate on high
levels of use. For example, since the questions
were not asked until 1984, it is unlikely that an
occasional user would remember the exact months
in which he used marijuana. A regular user, how-
ever, would be able to answer the questions
accurately, since they would simply respond ‘yes’
to all the months. We recognize the unreliable
nature of data reported retrospectively, however,
this is the only data available on marijuana use for
this cohort of high school seniors. Aside from
errors in the variables due to genuine inability to
recall past events, there is the potential for respon-
dents to lie about their illicit drinking and
marijuana use. The designers of the NLSY are
well aware of these difficulties and take extensive
measures to ensure anonymity. Nevertheless, there
may be systematic under-reporting of these activi-
ties, as discussed by Cook and Moore.’* In
addition, we recognize that using other drugs, such
as cocaine or LSD,may have more serious effects
on academic achievement. Unfortunately, the data
83
on other drug use are not as detailed as for mari-
juana use. For example, students are asked only
whether they have ever used a certain drug and
their age at first use, and there is no data on use
during this cohort’s high school year. Finally, we
note that some students may have dropped out
prior to their senior year due to substance abuse,
thereby lowering the observed incidence of subs-
tance abuse.
The work of Heien and Pompelli” and Manski
et aZ.32has shown the importance of demographic
and family structure. We, therefore, include a
number of demographic and family-related vari-
ables in our model of high school graduation.
These include gender, race, family structure (an
intact family measure and the number of siblings
in the family), parents’ education, the degree of
religious activity and a poverty indicator. Table 1
contains the variables’ definitions. ‘Intact family’
is a dummy variable equal to one for those respon-
dents who lived with both parents at age 14. This
accounts for the impact of family support and
stability of home life, which may effect graduation
probabilities. Since a typical student spends only
six or seven hours at school, for 180 days per year,
a significant part of the ‘education process’ takes
place at home. One would, therefore, expect that
parents with more education will provide a super-
ior home learning environment. Furthermore,
previous studies of family structure and high
school graduation have shown that the probability
of high school graduation increases with both
parents’ educational attainn~ent.’~.~~ We use a
dummy variable, indicating whether both parents
have completed high school, to estimate the effect
of parents’ educational attainment on the student’s
completion of high school. Based on previous
studies by Register and William~,~’ and Laixuthai
and Chaloupka,’ we hypothesize a negative causal
relationship between practicing religion and subs-
tance abuse rates. We, therefore, include a
‘Religion’ variable to account for this effect in our
reduced form demand equation of substance use.
A study of the effect of youthful drinking on
the likelihood of college graduation by Cook and
Moore” uses income as an exogenous variable. A
study by DiNardo and Lemieux‘ uses unemploy-
ment and income to explain young persons’
alcohol and marijuana use. A study of the relation-
ship between the earliest symptoms of alcoholism
and years of schooling by Mullahy and Sindelar”,
includes the head of family’s job type; for
example, blue or white collar. Our analysis uses a
84 T. YAMADA, M. KENDIX AND T. YAMADA
poverty level dummy variable, since a family
living below the poverty line has fewer resources
to devote to their child’s education, which lowers
the probability of high school graduation. We also
estimated our probit model, replacing the poverty
indicator with family income. The overall empiri-
cal results were not significantly affected. These
additional results are not reported here but are
available from the authors on request. The vari-
ables indicating whether the adolescent’s family
was intact and whether they had more than four
siblings, are used to measure the quality of avail-
able parental support other than that due to
income. Weak parental and peer support may have
a negative impact on educational attainment and
may increase the likelihood of alcohol and illegal
drug use.13 Furthermore, given the theoretical
and empirically observed negative associations
between the number of children and income, large
family size may be proxying for lower socioecon-
omic status.
We include two variables to represent the
student’s academic potential. First, the combined
English and Mathematics scores from the
ASVAB tests; and second, the respondent’s
percentile class ranking. We hypothesize that
higher ASVAB scores and higher class ranking
are associated with a higher probability of
graduation. It is possible that the ASVAB test
score and class ranking variables may be
influenced by substance abuse, and it may be
inappropriate to use them as control variables.
However, not controlling for student quality in a
model of graduation would cause the estimated
coefficients on the substance abuse variable to be
overestimated. A number of observations on the
student’s class ranking are missing, therefore, we
use a dummy variable labelled ‘missing class
rank’ to account for observations where the
individual class ranking is unavailable.
In addition to our probit estimates of high
school graduation, we also estimate reduced form
demand equations for alcohol consumption and
marijuana use. This allows us to evaluate the
impact of policy variables on students’ illicit
substance use. Our reduced form demand equa-
tions include the following policy variables; the
student’s home state’s beer tax, the price of liquor
beverages in the state, marijuana decriminalization
and minimum drinking age indicators.
We tested for heteroskedasticity and general
misspecification in the probit estimations by
considering an error structure where Var( e ;) =
[exp(a‘wi)I2, where ej is the error term for the
i-th student, a’ is a vector of parameters and wi is
a vector of observations on variables for the i-th
student. In order to test for this particular form of
heteroskedasticity, we collected data on the unem-
ployment rate of the area in which the student
lives, a dummy variable indicating whether the
student has ever been charged with a crime, a
dummy variable equal to one for students residing
in an urban area, and a dummy variable equal to
one for students who attend a state run school as
compared to a private school. We ran four separate
regressions for each of our probit models, letting
w , equal each of the four variables in turn
described above. The coefficient, a , was not
significant at the 5 percent level in any of these
regressions. We also included the four variables
described above as regressors in each of the probit
equations. Using a likelihood ratio test, these
variables were never jointly significant at the 5
percent level in any of the probit models. We
recognize the limitations of these tests, neverthe-
less, these results support the null hypothesis of
homoskedastic errors and the absence of
misspecification in general (see Davidson and
Ma~Kinnon,~’ Kmenta,34 Kennedy,35).
EMPIRICAL RESULTS
The summary statistics and definitions of the
variables used in this study are reported in Table
1. Table 3 reports the probit estimates of high
school graduation from the behavioural structural
model. The High School Graduation (HSG)-
Frequent drinker model estimates the effects of a
student being a ‘frequent drinker’ defined as a
student who drinks alcohol twice or more days in
the past week, on high school graduation. The
HSG-Liquor and wine model estimates the effect
of ‘liquor and wine consumption’ on high school
graduation. The HSG-Marijuana user model
estimates the effect of being a ‘frequent mari-
juana user’ on high school graduation. Table 4
reports the illicit substance demand equation
estimates, using a reduced form model. Our
calculations of the marginal effects and elastici-
ties associated with changes in alcohol and
marijuana use are reported in Table 5. These
calculations are based on the estimated
coefficients reported in Tables 3 and 4.
In this section we begin by discussing the results
for the high school graduation probit estimations,
 ALCOHOL, DRUGS AND HIGH SCHOOL GRADUATION 85
Table 3. Estimated probit coefficients of high school graduation

Models
~

HSG -Frequent HSG-Liquor HSG -Marijuana

Independent variable drinker and wine user

Constant -0.2688 -0.3384 -0.245 8

(0.26) (0.32) (0.23)
Frequent drinker -0.4144'
(1.64)
Liquor and wine consumption -0.0608b
(1.96)
Frequent marijuana user -0.5013b
(1.98)
Male 0.1132 0.0977 0.1280
(0.56) (0.49) (0.63)
Black 1.0623a 1.0626" 1.1051'
(3.57) (3.60) (3.75)
Hispanic 0.0415 0.024 0.03 18
(0.16) (0.10) (0.13)
Intact family 0.5387b 0.5 1Ub 0.5241
(2.3 1) (2.21) (2.22)
More than 4 siblings -0.3684' -0.3460 -0.3899'
(1.67) (1.56) (1.77)
Both parents graduated 0.2519 0.2093 0.2507
high school (1.09) (0.92) (1.09)
Poverty -0.0799 -0.0920 -0.6507
(0.36) (0.41) (0.29)
ASVAB Score 0.0262' 0.026ga 0.0265'
(3.67) (3.73) (3.69)
Class rank percentile -0.0065 -0.0067 -0.0068
(0.73) (0.74) (0.74)
Dummy missing class rank - 1.5628' - 1S562 -1.5633b
(2.35) (2.31) (2.29)
Religion 0.3494 0.3920' 0.2704
(1.51) (1.68) (1.14)
Log likelihood - 106.5 - 105.9 -105.9
Restricted log likelihood - 154.4 - 154.4 - 154.4
Likelihood ratio test 95.8 97.0 97.0
McFadden's R-squared 0.3 1 0.32 0.3 1
Note: Absolute value of t-statistics are in parentheses.
', and are statistically significant at the I%, 5% and 10%level, respectively.

reported in Table 3. We then describe how these High school graduation

results are used to calculate the marginal and
elasticity effects reported in Table 5. Secondly, we The results reported in Table 3 show that high
discuss the results from our demand estimations, school graduation is sensitive to, and is negatively
reported in Table 4. We then discuss how this set associated with, alcohol consumption and mari-
of results is used to calculate the values reported juana use. The alcohol use and marijuana use
in Table 5 . variables are collinear and their effects are,
Finally, we provide a brief review of the social therefore, modelled separately in different
and demographic factors' impact on our equation structural behavioural equations. 'Frequent drink-
estimates. er' is significant at the 10 percent level in the
86 T. YAMADA. M. KENDM AND T. YAMADA

HSG-Frequent drinker model. ‘Liquor and wine
consumption’ and ‘Frequent marijuana user’ are
statistically significant at the 5 percent level in the
HSG-Liquor and wine and HSG-Marijuana user
models, respectively.
All the marginal values in Table 5 are expressed
in percentage points, except those from the ‘liquor
and wine’ equation estimation, in which case the
changes in liquor and wine drinks are in quantity
units. A detailed description of the method of

Table 4. Estimated coefficients of high school student’s demand for alcohol and marijuana
Models
PROBIT OLS PROBIT
Frequent Liquor Marijuana
Independent variable drinker and wine user

Constant - 1.8817 -1.1824 -0.0248

(1.64) (0.92) (0.02)
Beer tax -0.5079b -0.1778 -0.2446
(2.32) (1.51) (1.15)
Liquor price 0.031 1 0.0893 -0.2285
(0.28) (0.73) (1.83)
Marijuana decriminalization -0.3788b 0.0205 -0.0430
(2.25) (0.11) (0.24)
Age 18 for beer 0.0938 0.2047
(0.59) (0.89)
Age 18 for liquor 0.7678b -0.1676
(2.27) (0.59)
Alcoholic parent 0.1550 0.3547 0.2084
(0.98) (1.38) (1 -26)
Male 0.1355 -0.0957 0.3425b
(1.01) (0.56) (2.31)
Black -0.8401” -0.4460b 0.0024
(3.w (2.43) (0.01)
Hispanic -0.2952 -0.2495 -0.3666
(1.38) (1.06) (1.46)
Intact family -0.2038 -0.2282 -0.0612
(1.13) (1.07) (0.34)
More than 4 siblings 0.2034 0.4238 0.1263
(1.17) (1.31) (0.71)
Poverty -0.3777b -0.3847b -0.2180
(2.02) (2.02) (1.15)
ASVAB Score 0.0072 0.0092‘ 0.0020
(1.36) ( 1.95) (0.37)
Class rank 0.0097a 0.0094’ 0.0089”
(2.96) (2.67) (2.46)
Dummy missing class rank 0.4287‘ 0.4971 0.6687’
(1.89) (1.99) (2.62)
Religion -0.3129b 0.0312 -0.8070”
(2.12) (0.16) (4.25)
Log likelihood -238.02 -202.65
Restricted log likelihood -273.50 -232.79
Likelihood ratio test 70.96 60.28
McFadden’s R-squared or R ’ 0.13 0.06 0.13
Note: Absolute values of t-statistics are in parentheses. White t-statistics are reported for the ‘Liquor &
Wine’equation.
*. band are statistically significant at the 1%, 5% and 10%level, respectively.
 ALCOHOL,DRUGS AND HIGH SCHOOL GRADUATION 87
Table 5. Marginal effects and elasticities
Impact of alcohol consumption and marijuana use on high school graduation and high
school student’s demand for alcohol and marijuana

Models
Dependent variable

(Demand for alcohol and marijuana)

Independent High school Frequent (Liquor (Marijuana
variable graduation drinker) and wine) user)
Frequent drinker -4.3
(-0.650)
Liquor and wine -0.3
consumption (-0.201)
Frequent marijuana -5.6
user (-0.662)
Beer tax (0.300*) -9.2 -0.18 -3.5
(-0.282) (-0.124) (-0.138)
Liquor price (1.152*) 0.6 0.09 -3.3
(0.324) (1.167) (-2.429)
Marijuana (0.104) -7.2 0.02 -0.7
decriminalization (-0.141) (0.Oool) (-0.018)
Age 18 for beer 1.9 3.5
Age 18 for liquor 0.77 -2.9
Note: The marginal effects are expressed in percentage points except for those in the liquor and wine
equation, where the changes in liquor and wine drinks are in quantity units. m e method of calculation
of the marginal effect is described in the appendix.
The values in parentheses are calculated elasticities. In the case of the high school graduation
equations, the ‘Frequent drinker’ and ‘Frequent marijuana user’ variables are not continuous,
Feerefore, readers should interpret these calculations with caution.
denotes the implied reduced form calculation of the elasticity of high school graduation. For
example, the total effect of beer taxes (B) on high school graduation (G) is a In g/a In b = a,
(a In frequent drinker/a In b) + a2 (a In liquor and wine/a In b) + a, (a In marijuana user/a In b),
where the a’s represent the estimated elasticities from the high school graduation equations.

calculation is provided in the appendix. It should
be noted that in the probit model of high school
graduation, the ‘frequent drinker’and ‘frequent
marijuana user’ regressors are not continuous,
therefore, readers should exercise caution when
interpreting the elasticity calculations for these
variables.
In Table 5, the column entitled ‘high school
graduation’, reports that students who have con-
sumed alcohol at least two days during the past
week, that is, ‘frequent drinkers’, have a 4.3
percent lower probability of graduation. Second,
ten drinks of liquor and wine per week reduces the
probability of high school graduation by 0.3
percent. Third, frequent marijuana use lowers the
probability of high school graduation by 5.6
percent. The estimated elasticities of high school
graduation with respect to ‘frequent drinker’,
‘liquor and wine consumption’ and ‘frequent
marijuana user’, are -0.65, -0.201 and -0.662,
respectively. As discussed above, the alcohol data
are constructed such that one reported beer drink
has the same pure alcohol content as either one
drink of wine or a drink of liquor. Our finding that
liquor and wine consumption has a significant
negative impact on high school graduation is
somewhat surprising, since we could not find a
significant relationship between total alcoholic
drinks consumed and graduation. We can only
speculate as to the underlying reason for this
finding. One possibility is that wine and liquor
produce intoxication more quickly and, therefore,
their consumption is indicative of more serious
substance abuse problems.
In Table 5 , applying our beer tax elasticity to a
calculation of the implied total effect of beer
taxes on high school graduation, we find that a 10
percent increase in beer taxes raises the probability
88 T.YAMADA. M.KENDIX AND T.YAMADA
of high school graduation by about 3 percent.‘marijuana user’, are reported in Table 4. The least
The effect of the beer tax, liquor prices andsquares estimates of ‘liquor and wine consump-
marijuana decriminalization are calculated using
tion’ use White t-statistics. The White t-statistics
are of similar order to the OLS t-statistics. We
the weighted sum of the estimated elasticities
with respect to ‘frequent drinker’, ‘liquor and
conclude, therefore, that our ‘liquor and wine’
wine’ and ‘marijuana user’ in the demand demand equation does not have serious problems
equations and high school graduation equations.
arising from heteroskedastic errors or general
For example, the total implied effect (elasticity)
misspe~ification.”*~~ Liquor prices and beer taxes
are included in all the specifications, in order to
of a beer tax on high school graduation from the
reduced form is a(ln g/a In b ) = a1(a In Frequent
examine whether alcohol and marijuana are substi-
drinker/a In b ) + a,@ In Liquor and wine/a In b )+
tutes. The state beer excise tax is used as a proxy
a,@ In Marijuana user/d In b), where the a’s are
for beer prices and, of course, is also an important
the estimated elasticities from the high school
policy variable.
graduation equations. The estimated coefficient for the beer tax
Using the NLSY data, Cook and Moore” find variable is statistically significant in the frequent
that an increase in the state beer tax from $.lo to
drinker model (see Table 4). The marginal and
$1.00 per case, raises the probability of college
elasticity calculations for beer tax with respect to
graduation by 6.3 percent. Our results are, there-
the demand for alcohol and marijuana are
fore, consistent with other investigators who have
reported in Table 5. The elasticity of the beer
shown that increases in the price of alcoholic
tax with respect to the probability of being a
beverages are an effective means of improvingfrequent drinker is -0.282 (=a In frequent
drinker/d In tax), which is much smaller than for
educational attainment. Dinardo and Lemieux6find
that alcoholic beverages are normal goods. Given
the youth population as a whole. For example, the
normality and the relatively elastic demand for
beer price elasticity for youths aged between 16
and 21 ranges from - 1.53 to -1.54 (‘beer partici-
alcoholic beverages among high school students,
pation’) as reported by Grossman, et d.,”and
there is a potentially significant role for alcoholic
-1.18 (frequent drinking) and -0.59 (fairly
beverage taxes in affecting youths’ educational
frequent drinking) as reported by Coate and
attainment and general welfare. There may also be
other benefits from higher alcohol taxes, including
Grossman.” There are a number of possible
larger tax revenues and improved educational reasons for the discrepancy between our results
achievement, the latter of which not only benefits
and those cited above. First, there is a difference
students, but has positive external societal benefits.
in the age of the respective cohorts; our study
Alcohol abuse imposes external costs, such as loss
only uses data on high school seniors, mostly
of life, injury and property damage caused byaged between 17 and 18. Secondly, our definition
drunken driving, crime, an increase in health care
of a ‘frequent drinker’ is different from Gross-
costs, productivity loss and lower educational
man, er al.’s ‘beer participation’ variable. Their
achievement. If taxes on alcohol are lower than
data sources are also different. They use data from
appropriate in the United States, the external costs
the National Health and Nutrition Examination
associated with alcohol abuse may exceed the Survey (NHANES) I, and Coate and Grossman”
use data from NHANES 11. Thirdly, the price of
deadwei ht welfare losses associated with taxes in
general.’*37 Our calculation of the total implied
beer is not fully captured by the beer tax. Due
effect of liquor prices on high school graduation in
to the elastic demand for beer among high
Table 5 , shows that a 1 percent increase in liquor
school students, the beer-price elasticity with
prices is associated with a 1.152 percent increase
respect to beer tax (=aIn Price/a In tax) is about
in the probability of graduation. These calcula-
0.18. In other words, a 1 percent increase beer
tions suggest that higher liquor prices have ataxes raises the price of beer by only 0.18
substantial positive impact on high school comple-
percent.
tion; at least as effective as increases in beer taxes.
Our estimates of the effects of the price of
liquor are not statistically significant in the
Demand for alcohol and marijuana ‘frequent drinker’ or the ‘liquor and wine’ con-
sumption equations (see Table 4); although the
The ‘demand’ equations estimating ‘frequent liquor price is significant in the marijuana user
drinker’, ‘liquor and wine’ consumption, and equation. Previous analyses of the relationship
 ALCOHOL, DRUGS AND HIGH SCHOOL GRADUATION
between alcohol and marijuana use have shown
different results. Some studies of US youth,
examining the relationship between marijuana
decriminalization and alcohol use, suggest that
alcohol and marijuana are substitute^.^ Other
investigators examining the relationship between
alcohol prices and marijuana use, suggest they are
complements.6 Our estimate of the impact of
liquor prices in the marijuana user equation (see
Table 4) is negative and statistically significant at
the 10 percent level, implying that they are com-
plements. The elasticities reported in Table 5 show
that a 1 percent increase in liquor prices, reduces
the probability of frequent marijuana use by 2.4
percent.
In contrast to the complementary relationship
described above, marijuana decriminalization has
a significant, negative impact on the probability
of being a ‘frequent drinker’. The presence of
marijuana decriminalization in a state, reduces the
probability of a student being a frequent drinker
by 7.2 percent (see Table 5). This result implies
a substitute relationship between marijuana use
and frequent drinking, which is consistent with
the work of Chaloupka and Laixuthai.’ In sum-
mary, our study finds only ambiguous evidence
regarding the substitutecomplement relationship
between alcohol and marijuana. Finally, our
estimates of the relationship between marijuana
decriminalization and marijuana use are not
significant, a similar result to that of DNardo and
hmieux.6
Aside from their policy relevance, the ‘age 18
for beer’ and ‘age 18 for liquor’ variables are used
as alcohol availability indicators for high school
students. Whereas Coate and Grossman” use a
continuous measure for the minimum drinking
age, we use simple dummy variables. We experi-
mented with various minimum age specifications
and the most significant were the ones reported.
We can only speculate as to this finding, but one
possibility is that a minimum drinking age of 18
(the lowest in any jurisdiction) indicates that the
locality has a relaxed attitude towards youth
alcohol use. Hence, this variable may be measur-
ing unobservable social attitudes to alcohol
consumption, in addition to the minimum drinking
age. While the ‘age 18 for beer’indicator is never
significant, the analogous variable for liquor is
statistically significant at the 1 percent level, in
the ‘liquor and wine’ equation (see Table 4).
Currently, the minimum drinking age in all juris-
dictions is twenty-one years, however, in 1982 the
89
age requirement varied across States. Our findings
suggest that an 18 years minimum drinking age for
liquor (‘age 18 for liquor’), raises liquor and wine
consumption among high school students. Hence,
the policy of imposing higher minimum drinking
ages may have reduced the consumption of liquor
and wine, and indirectly improved the educational
attainment of high school students.
Socioeconomic and demographicfactors
The results for all the high school graduation
models (HSG-Frequent drinker, HSG-Wine and
liquor, and HSG-Marijuana user, see Table 3),
show a significant and positive coefficient estimate
for blacks. These results are consistent with other
studies by Cook and MooreI2 and Manski, et a1.32
The impact of an intact family structure is positive
and significant at the 5 percent level in all three
equations. Students living with both parents have a
higher probability of graduating from high school.
Students living with more than four siblings have a
lower probability of graduation. The ASVAB test
score variable, used to proxy a student’s academic
potential, is statistically significant at the 1 percent
level; higher scores are associated with an
increased probability of high school graduation.
We find only modest evidence that students attend-
ing religious services more than once a week, are
more likely to graduate from high school (see the
results for HSG-Liquor and wine model in Table
3). We do find, however, that the religion variable
is negative and statistically significant in the
demand analysis; namely, in the frequent drinker
and frequent marijuana user equations (see Table
4). This implies an indirect effect, whereby stud-
ents engaged in religious activities are less likely
to be substance abusers and hence more likely to
graduate high school.
CONCLUSION
In this study we have used data from the NLSY on
672 individuals who were in the twelfth grade in
1982, in order to analyze the impact of alcohol
and marijuana use on high school graduation, and
the demand for alcohol and marijuana.
Our analysis shows that there are substantial
negative effects on high school graduation prob-
abilities associated with alcohol and marijuana
use, using our measures of use; frequent drinker,
liquor and wine consumption, and frequent
90 T.YAMADA. M.KENDIX AND T.YAhfADA
marijuana user. Our elasticity calculations show
that a 10 percent increase in these activities lowers
the probability of high school graduation by 6.5
percent (frequent drinker), 2.01 percent (liquor
and wine consumption) and 6.62 percent (frequent
marijuana user). These results indicate that high
school students who use alcohol and marijuana
are exposing themselves to significant risk of
academic failure.
Our findings have a number of important policy
implications. A 1 percent increase in liquor prices
raises the probability of high school graduation by
1.152 percent. The estimated elasticity of the
probability of being a frequent beer drinker with
respect to the beer tax is approximately -0.28,
thus a 10 percent increase in the beer tax reduces
frequent drinking among high school students, and
in turn raises the probability of high school
graduation by about 3.0 percent. Raising the
minimum drinking age for liquor has a significant
impact on reducing the consumption of liquor and
wine, and, consequently, improves the probability
of high school graduation.
Our results show no significant relationship
between marijuana decriminalization and mari-
juana use among high school students, however,
marijuana decriminalization is found to reduce the
probability of becoming a frequent drinker. This
result indicates a substitute relationship between
marijuana use and frequent drinking. If raising rhe
cost of using marijuana, reduces marijuana con-
sumption, a number of policy implications follow.
Since raising taxes on illegal drugs, e.g. marijuana,
is not a viable policy option, imposing harsher
penalties for using and selling marijuana might
reduce the demand and supply of marijuana.
These types of policies have been used extensively
during the past ten years, but there is no consensus
as to their effectiveness. On the other hand, if
alcohol and marijuana are substitutes, policy
makers need to consider the full ramifications for
youth alcohol use of increasing the non-pecuniary
costs of marijuana consumption.
The educational success of students is import-
ant, being directly connected to occupational
choice, employment opportunity and earnings.
Loss of societal human capital due to academic
failure through alcohol and marijuana use, may be
considerable. Lower earnings due to lack of
education, lower personal esteem, lack of decision
making skills and weak social skills, leads to anti-
social activities, which further raise the social
costs of substance abuse. Appropriate alcohol
policy, for example tax policy, and preventive
programs should be expanded in order to improve
educational performance for high school students.
This is especially true for young adolescents, for
whom early intervention is likely to have the best
payoff in terms of affecting lifetime alcohol and
drug use. Early onset of alcohol and drug use is a
crucial risk factor for progressing to more serious
forms of substance abuse, and may in turn be
linked to severe alcohol and drug addiction in later
life. High school-based preventive programs,
which discourage alcohol and marijuana use, are
highly recommended as a means of reducing both
the direct and indirect societal costs of substance
abuse.
AF’PENDIX
Calculating the marginal effect of substance use
on the probability of graduation
There are two methods used to calculate the margi-
nal effects, depending upon whether the variable is
a continuous variable, e.g. liquor and wine con-
sumption; or a qualitative dummy variable (i.e., a
discrete variable), e.g. frequent drinker and fre-
quent marijuana user.
Case 1-Continuous variable (liquor and wine
consumption). We calculate the model’s mean
predicted probability of graduation for the whole
sample:
where Xi is the vector of observations for the i-th
student, B is a vector of estimated coefficients, @
is the normal cumulative density function and N is
the sample size. Next we calculate
where Xli is the same as X i except that one is added
to the value of X,, where Xij (thej-th element inxi)
is the variable of merest; for example, the amount
of wine and liquor consumed. The marginal effect
of increasing all students consumption by one unit,
on the probability of graduation is A, minus A,,.
 ALCOHOL, DRUGS AND HIGH SCHOOL GRADUATION
Case II-Qualitative dummy variable (frequent
drinker, frequent marijuana user). Again, let X,
represent observations on the variable of interest;
for example, frequent marijuana user. First, con- dsinker or a frequent marijuana user is calculated
sider the sub-sample of (size No) of students with as the average of G,minus Goand Hominus HI.
X, = 0 and calculate the model’s mean predicted
probability of high school graduation for these No
persons;
Next, using this same No person sub-sample,
calculate the mean predicted probability of high
school graduation if the dummy variable of inter-
est for these students were switched to equal 1:
where X l i is the same as X i except the dummy
variable of interest is changed to equal 0. GI
minus Gois the marginal impact on the probability
of graduation if, for example, a non-frequent
drinker or a non-frequent marijuana user were to
become a frequent drinker or a marijuana user.
Next consider the sub-sample of students with
X,= 1 and calculate the model’s mean predicted
probability of graduation for this sub-sample of
N, = N - No persons:
where X iis now the vector of observations for
students whose dummy variable of interest equals
1. Next, using the same N, person sub-sample,
calculate the mean predicted probability of high
school graduation with the value of X, switched to
equal 0:
where X I j is the same as X i except the dummy
variable of interest is changed to equal 0. Ho
minus H,is the marginal impact on the probability
of high school graduation if, for example, a
frequent drinker or a frequent marijuana user
91
were to become a non-frequent drinker or a non-
frequent marijuana user.
The overall marginal impact of being a frequent
ACKNOWLEDGEMENTS
We are indebted to Professors Michael Grossman and
Bernard Okun. We gratefully acknowledge financial
support from the Rutgers University Research Council
(grant #2-02158) and the Kikawada Foundation in
Japan. The views presented here are those of the
authors and do not necessarily represent those of their
affiliated organizations.
REFERENCES
1. Saffer, H. Alcohol advertising bans and alcohol
abuse: an international perspective. Journal of
Health Economics 1991; lO(1): 65-79.
2. U.S. Department of Health and Human Services,
Preventing adolescent drug use: from theory to
practice, OSAP Prevention Monograph-8, Office
for Substance Abuse Prevention, 1991.
3. U.S. Department of Health and Human Services,
General reports on youth and alcohol, Office of
the Inspector General, 1992.
4. Goldstein, A. and Kalant, H. Drug policy: striking
the right balance. Science 1990; 249: 1513-1521.
5. Chaloupka, F. and Laixuthai, A. Do youths substi-
tute alcohol and marijuana? Some econometric
evidence, National Bureau of Economic Research,
Working Paper No.4662, 1994.
6. DiNardo, J. and Lemieux, T. Alcohol, marijuana,
and American youth: the unintended effects of
government regulation, National Bureau of Econ-
omic Research, Working Paper No.4212,1992.
7. Kandel, D. and Yamaguchi, K. From beer to crack:
developmental patterns of drug involvement.
American Journal of Public Health 1993; %3(6):
851-855.
8. Grossman, M. Health benefits of increases in
alcohol and cigarette taxes. British Journal of
Addiction 1989; 84: 1193-1204.
9. Laixuthai, A. and Chaloupka, F. Youth alcohol use
and public policy, National Bureau of Economic
Research, Working Paper No.4278,1993.
10. Grossman, M., Coate, D. and Arluck, G. M. Price
sensitivity of alcohol beverages in the United
States: youth alcohol consumption. In: Holder H D,
(ed) Control issues in alcohol abuse prevention:
strategies for Srates and communities, Greenwich,
Connecticut: JAI Press Inc. 1987: 169-198.
92 T.YAIUADA, M.KENJXX AND T.YAMADA
11. Coate, D. and Grossman, M. Effects of alcohol
beverage prices and legal drinking ages on youth
alcohol use. Journal of Law & Economics 1988;
31: 145-171.
12. Cook, P. J. and Moore, M. J. Drinking and school-
ing. presented at the NBER Higher Education
Working Group and the Third Annual Health
Economics Workshop, June 1992; Center for the
Study of Business, Regulation, and Economic
Policy, Duke University, Working Paper No.93-6,
1993; Journal of Health Economics 1993; 12:
41 1-429.
13. Bennett, G. Substance abuse among the young. In:
Bennett G. B., Woolf, D. (eds) Substance abuse.
New York: Delmar Publisher, 1991: 142-156.
14. Garrison, C. Z., McKeown, R. E., Valois, R. F. and
Vincent, M. L. Aggression, substance use, and
suicidal behaviours in high school students. Ameri-
can Journal of Public Health 1993; 83(2):
179- 184.
15. Mullahy, J. and Sindelar, J. Life-cycle effects of
alcoholism on education, earnings, and occupation.
Inquiry 1989; 26: 272-282.
16. Mullahy, J. and Sindelar, J. Alcoholism, work, and
income. Journal of Labor Economics 1993; ll(3):
494-520.
17. Mullahy, J. and Sindelar, J. Alcoholism and
income: the role of indirect effects. Milbank Quar-
terly 1994; 72: 359-375.
18. Mullahy, J. and Sindelar, J. Employment, unem-
ployment, and problem drinking. National Bureau
of Economic Research, Working Paper No.5123,
1995.
19. Grossman, M. On the concept of health capital and
the demand for health. Journal of Political Econ-
omy 1972; 80(2): 223-255.
20. Grossman, M. The demand for health: a theoretical
and empirical investigation. New York: Columbia
University Press, 1972.
21. Register, C.A. and Williams, D.R. Labor market
effects of marijuana and cocaine use among young
men. Industrial and Labor Relations Review 1992;
45(3): 435-448.
22. Kaestner, R. New estimates of the effect of mari-
juana and cocaine use on wages. Industrial and
Labor Relations Review 1994; 47(3): 454-470.
23. Mullahy, J. and Sindelar, J. Health, income, and risk
aversion: assessing some welfare costs of alcohol-
ism and poor health. National Bureau of Economic
Research, Working Paper No.4649,1994.
24. Becker, G. S. and Murphy, K. M. The theory of
rational addiction. Journal of Political Economy
1988; 96(4): 675-700.
25. Becker, G. S., Grossman, M. and Murphy, K. M.
Rational addiction and the effect of price on con-
sumption. American Economic Review 1991; 81 (2):
237 -24 1.
26. Grossman, M. The economic approach to addictive
behaviour. In: Bloss, G., Hilton, M. (eds) Econ-
omic and socioeconomic issues in the prevention of
alcohol related problems. Rockville, MD: National
Institute on Alcohol Abuse and Alcoholism, 1993.
27. Chaloupka, F., Grossman, M., Becker, G.S. and
Murphy, K. M. Alcohol addiction: an econometric
analysis. presented at the AIlied Social Science
Association Conference, Anaheim, CA: 1993.
28. Grossman, M., Chaloupka, F. and Sirtalan, I. An
empirical analysis of alcohol addiction: results
from the monitoring the future panels. Mimeo,
1994.
29. Chaloupka, F. Rational addictive behaviour and
cigarette smoking. Journal of Political Economy
1991; 99(4): 722-742.
30. Becker, G. S., Grossman, M. and Murphy, K. M.
An empirical analysis of cigarette addiction.
American Economic Review 1994; 84(3): 396-418.
31. Heien, D. and Pompelli, G. The demand for
alcoholic beverages: economic and demographic
effects. Southern Economic Journal 1989; 55(3):
759-770.
32. Manski, C. F., Sandefur, G. D., McLanahan, S. and
Powers, D. Alternative estimates of the effect of
family structure during adolescence on high school
graduation. Journal of the American Statistical
Association 1992; 87(417): 25-37.
33. Davidson, R. and MacKinnon, J. G. Convenient
specification tests for logit and probit models.
Journal of Econometrics 1984; 25: 241-262.
34. Kmenta, J. Elements of econometrics, New York:
Macmillan, 1986.
35. Kennedy, P. A guide to econometrics, Cambridge,
Massachusetts: MIT Press, 1993.
36. Grossman, M., Chaloupka, F., SaiTer, H. and
Laixuthai, A. Effects of alcohol price policy on
youth. National Bureau of Economic Research,
Working Paper No.4385, 1993.
37. Grossman, M., Sindelar, J. L.. Mullahy, J. and
Anderson, R. Alcohol and cigarette taxes. Journal
of Economic Perspectives 1993; 7(4): 21 1-222.
38. Sickles, R. and Taubman, P. Who uses illegal
drugs? American Economic Review 1991; 81 (2):
248- 251.