New Estimates of the Effect of Marijuana and Cocaine Use on Wages
Author(s): Robert Kaestner
Source: ILR Review, Vol. 47, No. 3 (Apr., 1994), pp. 454-470
Published by: Sage Publications, Inc.
Stable URL: http://www.jstor.org/stable/2524977
Accessed: 05-03-2018 14:41 UTC
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
Sage Publications, Inc. is collaborating with JSTOR to digitize, preserve and extend access to
ILR Review
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
NEW ESTIMATES OF THE EFFECT OF
MARIJUANA AND COCAINE USE ON WAGES
ROBERT KAESTNER*
Using the 1984 and 1988 waves of the National Longitudinal Survey of Youth,
this study provides an update of several previous cross-sectional estimates of the
effect of illicit drug use on wages, as well as the first longitudinal estimates of that
effect. The cross-sectional results, which are generally consistent with the
surprising findings of previous research, suggest that illicit drug use has a large,
positive effect on wages. The longitudinal estimates, which control for unobserved
heterogeneity in the sample, are mixed: among men, the estimated wage
effects of both marijuana and cocaine use are negative, but among women, the
effect of cocaine use remains positive and large. Because the longitudinal
model is imprecisely estimated, however, those results are inconclusive.
T he adverse physical and psychological
effects of illicit drug use have been well
documented in the medical literature and
widely publicized through an extensive public
and private media campaign. Given the
public's knowledge of the negative health
consequences of illicit drug use, it is not
surprising that illicit drug use is also commonly
believed to adversely affect the social
and economic aspects of users' lives. Despite
these perceptions, a large segment of the
U.S. population engages in illicit drug use. In
a 1991 survey of 18-34-year-olds, for example,
approximately 60% of the respondents reported
having used illicit drugs at some time
*The author is Assistant Professor of Economics at
Rider College and Faculty Research Fellow of the National
Bureau of Economic Research. He thanks Andrew
Gill for comments on an earlier draft. A data
appendix with additional results, and copies of the
computer programs used to generate the results presented
in this paper, are available upon request to
Robert Kaestner, Department of Economics, Rider College,
Lawrenceville, NJ 08648.
in their lives, and approximately 23% reported
using illicit drugs within the previous
year (NIDA 1991).
One important focus of efforts to reduce
illicit drug use has been the labor market.
Concern that the ability of our workers is
being seriously impaired by illicit drug use
has spurred sizable expenditures to correct
the problem. The fundamental goal of most
drug prevention programs is to eliminate all
illicit drug use, a goal that appears unrealistic
in view of the fact that illicit drugs continue to
be widely used despite extensive dissemination
of information about their harmful effects.
Thus, the merit of a "zero tolerance"
drug policy depends critically on the effect of
illicit drug use on the individual. If illicit drug
use harms the user regardless of the particular
drug used and even when the amount of
the drug consumed is small, a zero tolerance
policy is clearly defensible; otherwise, the
wisdom of such a policy is questionable.
To date, little evidence of the type that
would satisfy most economists has been proIndustrial
and Labor Relations Review, Vol. 47, No. 3 (April 1994). ? by Cornell University.
0019-7939/94/4703 $01.00
454
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
WAGE EFFECTS OF MARIJUANA AND COCAINE USE 455
duced to justify the extent and scope of the
current drug prevention effort, particularly
in the labor market. Since earnings are one of
the best and most accessible measures of
labor market success, several recent studies,
using data from the National Longitudinal
Survey of Youth (NLSY), have examined the
impact of illicit drugs on the wages of young
adults. In general, the evidence presented in
these papers, directly contradicting the commonly
held belief that drug use has an adverse
impact on labor market outcomes, indicates
that drug use has a positive effect on
wages. The explanation most often suggested
for this finding is the probable existence of
unobserved heterogeneity in the sample; individuals
who use drugs may also, for reasons
unobserved by the researcher, be more productive
than average.
The previous studies on this subject have
all been cross-sectional in nature, and none
of them, despite their careful execution, have
been able to fully control for unobserved
characteristics that could strongly affect the
observed relationship between illicit drug
use and wages. The purpose of this paper is to
partially fill this gap in the literature by estimating,
for the first time, a longitudinal or
fixed effects model of drug use and wages. To
the extent that some of the potentially important
unobserved characteristics that have been
omitted from previous studies are individualspecific
and time-invariant, a fixed effects
wage model will provide better estimates of
the relationship between illicit drugs and
wages. In addition, using the NLSY data of
1988, I provide the most up-to-date crosssectional
estimates of the effect of illicit drug
use on the wages of young adults yet pub-
lished.
Ilicit Drug Use and Wages
There have been several studies of the
effects of illicit drug use on wages, and all of
them have used the 1984 wave of the NLSY.
Gill and Michaels (1992), using a switching
regression framework, examined the effect
of (1) use of any illicit drug and (2) use of
hard drugs (cocaine and heroin) on the wages
of a combined sample of men and women.
They found a positive effect of drug use on
wages, resulting primarily from differences
in the return to unobserved characteristics.
In a 1991 study (Kaestner 1991), I examined
the wage effect of both lifetime and recent
use of marijuana and cocaine, using a simultaneous
equations model in which drug use
and the wage werejointly determined. I found
that both marijuana use and cocaine use
positively affected wages of all four demographic
groups chosen on the basis of gender
and age. I also estimated a switching regression
model, and found positive wage effects
of cocaine and marijuana use consistent with
Gill and Michaels's (1992) results. Using a
model that was somewhat unconventional in
terms of economic theory, Kandel and Davies
(1990) estimated simple OLS wage regressions,
and found no significant effect of marijuana
or cocaine use on the wages of employed
men.' Finally, Register and Williams
(1992) examined the effect of marijuana and
cocaine use on the wages of men, including
the effect of on-thejob use and long-term
use, using an instrumental variables approach.
They found a positive wage effect of general
marijuana use, but negative effects of on-thejob
marijuana use and long-term marijuana
use. They found no significant effect of cocaine
use on the wage.
All of these previous studies started with
the premise that illicit drug use results in
deterioration of an individual's physical and
psychological well-being, and consequently
also decreases a person's productive capabilities.
Drug use and wages were hypothesized
to be negatively related.
The observed relationship between illicit
drug use and wages could very well be positive,
however, if illicit drugs are a normal
good, since wages are an important determinant
of income. Indeed, this hypothesis has
received some empirical support. Sickles and
Taubman (1991) reported findings suggesting
that individuals with higher earnings ca'Kandel
and Davies (1990) estimated a model in
which the 1985 wage was the dependent variable, and all
independent variables were measured as of 1984. Included
among the independent variables were the 1984
wage and drug use, both of which were treated as
exogenous. In addition, the authors made no attempt to
correct for selection bias due to the labor force participation
decision.
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
456 INDUSTRIAL AND LABOR RELATIONS REVIEW
pacity have a greater involvement in illicit
drug use. Kaestner (1991) and Register and
Williams (1992) explicitly recognized the
potential income effect, and estimated a structural
model of illicit drug use and wages. A
second reason for estimating a structural
model of drug use and wages is found in Gill
and Michaels (1992) and Kaestner (1991).
Those studies cast the basic consumer problem
in a household production framework,
in which the interdependence of drug use
and wages resulted from the wage being the
price of time used in producing goods.2 Given
these considerations, in this paper I use the
following two-equation model to estimate
the effect of illicit drug use on wages:
(1) Wit =f(Xit, Dit, E., Vt)
and
(2) Dit= g (Zit, Wit, E., Ut).
In equation (1), Wis the natural logarithm
of the wage, X is a vector of exogenous variables
affecting the wage, D is a measure of the
quantity of illicit drug use, E is an unobserved
person-specific effect; V is an error term, i
indexes individuals, and t indexes time. Equation
(2) has similarly defined variables, with
Z being a vector of exogenous variables that
affect drug use, and U the error term.
Empirical Model
Before empirical estimates of the model
can be obtained, two issues need to be addressed.
First, the functional form of equations
(1) and (2) needs to be specified. For
now, I assume these functions are linear.
Second, the exogenous variables for each
equation need to be chosen. For equation
(1), these will include several human capital
2The two papers do not treat the problem in exactly
the same way. In my 1991 study, I provided only a simple
model similar to that found in Stigler and Becker (1977).
In this model, the wage is simply the price of time used
to produce consumption goods, including drug consumption.
Gill and Michaels (1990) presented a much
more detailed model, in which the role played by the
wage, though similar, is more complex. In their model,
drug time is being produced, and the wage becomes
part of the full price.
variables (for example, education and experience),
demographic variables (such as age,
race, and marital status), geographic measures
(such as region), and family/personal
background variables (such as household
composition at age 14 and self-esteem scale).
The exogenous variables chosen for equation
(2) will include all of those in equation
(1), plus a measure of non-earned income,
the frequency of religious attendance in 1979,
the current number of dependent children
of various ages, and the number of illegal acts
committed in 1979. These last four variables
will identify the parameter estimates of equation
(1). I chose them based on the results of
my 1991 study, which tested this set of variables
for the over-identifying restrictions they
impose. The parameters of equation (2) are
under-identified, and therefore only the reduced
form version of this equation will be
estimated.
Both cross-sectional and panel data (fixed
effect) estimates of the model will be obtained,
and the implementation strategy will
be the same in both cases. First, reduced form
estimates of equation (2) will be obtained
using the entire sample, and from these estimates
a predicted drug use measure will be
calculated. Next, estimates of the wage equation
(1) will be obtained for a sample of
employed individuals using the predicted
drug use variable in place of its actual value,
and correcting for the sample selectivity introduced
by examining only employed individuals.3
The Heckman (1976) two-step procedure
will be used to correct for the potential
sample selectivity bias.4
3This strategy is different from that found in Kaestner
(1991). In that paper, I obtained the reduced form
estimates of equation (2) using only employed persons,
and a selectivity correction was applied to this equation
as well as the wage equation (1). The results are little
affected by which of these strategies is chosen.
4In computing the standard errors for this model, it
is important to take account of the fact that several of
the right-hand-side variables (specifically, drug use and
sample selection terms) are predicted values. Murphy
and Topel (1985) outlined the procedure for calculating
the exact standard errors for a simple version of this
type of model. The current case is non-standard, given
the inclusion of the selectivity terms. In the cross-sectional
regressions, the standard errors are those that
correct for the inclusion of the selection correction
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
WAGE EFFECTS OF MARIJUANA AND COCAINE USE 457
The choice of the fixed effects specification
of the model is based on the expectation
that the unobserved personal characteristics
that influence the wage will also be correlated
with the other variables in the model,
particularly the drug use measures. Thus, this
specification is preferred over the alternative
random effect model, which assumes that the
individual effect is uncorrelated with the other
variables. The Chamberlain (1982) "correlated
random effects" model, which allows
for the correlation between the unobserved
effect and the other explanatory variables, is
identical to the fixed effects specification
when the model is linear in the parameters,
as is the current model. The specification of
the fixed effects model used in this paper,
however, is somewhat unconventional and a
departure from what is usually found in the
literature. For simplicity, assume that equation
(1) can be written as follows for period t:
(la) Wi = b +b1 X+b2Z. + b3E. + Vi
and for period t- 1,
(lb) WVVl =a0+a1X1 +a2Z1 +b3E.+Vt1.
The X are exogenous variables that are
time-invariant (such as race), and the Z are
exogenous variables that are time-varying
(such as illicit drug use). The E in the above
equations are the unobserved person-specific
characteristics, which are assumed to be
time-invariant. The standard way to obtain
unbiased estimates of the parameters of the
model is to take the difference of equations
(la) and (lb), and run an OLS regression.
Furthermore, it is usually assumed that the a,
= bi = b, which yields the following:
(Ic) Wi-Wi = b (Z.i - Z. 1) + (Vt - Vit-).
Note that the unobserved person effect
has been eliminated by taking the difference.
term, but ignore the fact that drug use is a predicted
value. For the panel data estimates, in which there are
two correction terms and two predicted drug use measures,
the generalized correction for heteroscedasticity
suggested by White (1980) is used to estimate the variance-covariance
matrix. The latter procedure remains
incorrect, but it should represent an improvement over
OLS estimates.
An alternative specification allows the a, an
b. to differ and estimates the following:
(I d) Wi - Wil= b- a,,) + (b, - al) Xi +
b2 Zi -a2 Z4-1 + (Vit - Vz-)
Although not common, specifying a model
in which the parameter values change over
time is a theoretically sound idea. For example,
if the degree of racial discrimination
changes over time, the impact of race on the
wage will also change. Similar arguments can
be made for the remaining variables, including
the human capital variables such as education
and experience.5
There are several benefits associated with
specifying the model as in equation (Id).
First, equation (Id) imposes the fewest restrictions
on the model. The only restriction
imposed in equation (Id) is that the effect of
the unobserved person-specific characteristics
(b3) is the same in both time periods.
Second, given the way the quantity of drug
use is measured (described in detail below),
the differencing of the drug use measure
presents problems, and is not intuitively appealing.6
Thus, including the level of drug
use from each year of data in the model
5To test the restrictions implied by equation (1c), I
performed a series of Wald tests. The Wald test is
appropriate in this case because the F-test assumes
homoscedasticity. The variance-covariance matrix used
for the Wald test incorporated the White (1980) correction
for heteroscedasticity, and is only an approximation
of the true estimate.
Although not all of the restrictions could be rejected,
many were found to be invalid. In addition, the
results of the Wald tests differed by gender, with more
of the restrictions being rejected in the equations for
women than in those for men.
6The drug use measures are discrete, categorical
variables, and taking first differences of these measures
would also result in a discrete measure. The most appropriate
estimation procedure for categorical variables of
this type is an ordered probit (logit) model (Sickles and
Taubman 1991; Kaestner 1991). In the cross-sectional
model, this methodology is rejected, due to the severe
collinearity between the predicted drug use categories
that are used in the two step procedure; a person with a
high probability of being a heavy cocaine user also has
a high probability of being a moderate cocaine user. In
the panel data analysis, a fixed effects model of an
ordered categorical variable would need to be estimated,
and that task is beyond the scope of this paper
(Chamberlain 1984).
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
458 INDUSTRIAL AND LABOR RELATIONS REVIEW
facilitates estimation, since standard methods
can be used to obtain the predicted
values of the level of drug use that are used in
the second stage of estimation. The predicted
values of drug use will be derived from the
reduced form estimates obtained from a
model that includes all exogenous variables
from both time periods.7 Third, taking differences
of any qualitative variable (such as
marital status) results in an arbitrary ordering
being imposed on the data, and using the
levels avoids this problem.
In addition, the panel data estimates will
be obtained from a sample of individuals
employed in both of the available time periods.
The exclusion of those who were not
employed in both of the periods introduces a
potential sample selection bias that I address
by including separate correction terms associated
with the labor force participation decision
in each year. Using an unrestricted model
also facilitates the incorporation of the sample
selection terms, since these variables enter
the model in levels. The reduced form labor
force participation equation will also be obtained
using all the exogenous variables from
both time periods.8
A potential drawback of this procedure is
that the time-varying variables are often highly
collinear, and this collinearity will affect the
precision with which the parameters are estimated.
Alternative models that use the
differenced form of the drug use variables
are estimated to test this hypothesis as it
pertains to the drug use variables, which are
the variables of interest in this paper. In
addition, several of the variables that are
technically time-varying will be treated as
time-invariant, since there is very little varia-
7The standard errors reported in the text are from an
OLS regression, and they ignore the fact that there are
several predicted values among the right-hand-side vari-
ables.
8The labor force participation model is estimated
separately for each year, but includes all the exogenous
variables in the model. For example, the respondent's
education in both of the years under consideration will
be included in the estimates of the probability of working
in period 1. This specification can be interpreted as
a reduced form version of the Chamberlain (1980)
random effects probit model, and thus accounts for
unobserved heterogeneity in the selection equation.
tion over time in these measures. These variables
include the respondent's age, since all
respondents aged about four years between
the two surveys, and the respondent's region
of residence.
Data
The data used in the analysis come from
the National Longitudinal Survey of Youth
(Center for Human Resource Research 1990).
In its starting year, 1979, the survey sample
consisted of approximately 12,000 youths aged
14-21. The survey has been updated each
year since 1979, with a broadening array of
purposes and questions. The questions elicit
detailed information on respondents' labor
market experience, family and personal background,
and illicit drug use. Central to the
purposes of this paper are the questions related
to the respondent's illicit drug use. In
1984, and again in 1988, respondents were
asked questions about their lifetime and recent
use of several illicit drugs, most notably
marijuana and cocaine.9
Several selection criteria were established
to eliminate sources of heterogeneity: the
sample included only those respondents who
were at least 18 years old in 1984, were living
independently or with their parents, but not
injail or other temporary quarters (such as a
dormitory), and who were not enrolled in
school, in the military, or self-employed. In
addition, those observations with missing data
were deleted. These restrictions resulted in
samples of approximately 7,800 individuals
in 1984 and 7,200 in 1988. A matched sample
of individuals present in both years, who
numbered approximately 5,700, was also created.
Definitions and descriptive statistics of
the variables used in the analysis can be found
in the appendix.
The illicit drug use questions are limited in
two major respects. First, as was suggested in
Mensch and Kandel (1988), there may be
some under-reporting in the NLSY1984 data,
particularly with regard to cocaine use. The
exact nature of the under-reporting is not
9The 1988 NLSY survey limited the illicit drug use
questions to include only marijuana and cocaine.
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
WAGE EFFECTS OF MARIJUANA AND COCAINE USE 459
known, but Mensch and Kandel (1988) suggested
that under-reporting is more common
among relatively light users of illicit
drugs than among heavier users, and more
pronounced among women and minorities
than among men and non-minorities. Although
the present analysis accounts for a
simple (that is, random) type of measurement
error, under-reporting remains a potential
problem.10
The second problem related to the drug
use questions is the absence of a measure of
quantity of use; only the frequency of drug
use is measured. Although frequency and
quantity have been shown to be highly correlated,
the two measures are clearly not equivalent
(Stein et al. 1988). In fact, Stein et al.
(1988) reported finding that the quantity of
drug use was a more powerful predictor of
problems associated with illicit drug use than
was frequency of use. In addition, the frequency
of use was interval-coded with relatively
large groupings (see Table 1).
Table 1 is a frequency distribution of illicit
drug use for the sample under examination,
and presents the unweighted data by gender.
One finding of note in Table 1 is the relatively
large increase between 1984 and 1988 in the
percentage of respondents reporting some
lifetime use of cocaine. In 1984, about 19.6%
of the male sample and 12.7% of the female
sample reported some prior cocaine use; by
1988 the respective figures were 32.6% and
21.2%. The observed increase in the initiation
into cocaine use over this age range is
consistent with previous studies (Kandel and
Logan 1984; Raveis and Kandel 1987). The
surprising finding is that the number of respondents
who had used cocaine in the previous
30 days decreased between 1984 and
1988, whereas the number of people who had
tried cocaine increased by approximately 66%
over that period. The relatively high levels of
lifetime cocaine use, compared to past 30 day
use, imply that there were many individuals
10The empirical strategy is to use a Two Stage Least
Squares (2SLS) estimation procedure, and thus the
drug use measures will be replaced by their respective
predicted values. This procedure is appropriate due to
both the simultaneity and measurement error prob-
lems.
who experimented with cocaine but neither
regularly used it nor became addicted to it.
Initiation into marijuana use largely ceased
over the age range observed for this sample,
as evidenced by the relatively small increase
in the prevalence of lifetime marijuana use,
and the dramatic decline in the number of
respondents who reported using marijuana
during the past 30 days, between 1984 and
1988.
The figures in Table 1 imply a general
decline in illicit drug use consistent with the
data from the recent National Institute on
Drug Abuse (NIDA) household surveys. In
general, marijuana use was much more common
than cocaine use, and the proportion of
users who reported relatively heavy marijuana
use was much greater than the proportion
reporting heavy cocaine use. It is also apparent
that men had a greater frequency of use
than women. Finally, the frequency distribution
of illicit drug use for employed individuals
(not shown) is very similar to that reported
in Table 1 (Kaestner 1994).
The levels of reported drug use in the 1988
NLSY survey are very similar to those reported
in the 1988 National Household Survey
(NHS) on DrugAbuse (National Institute
on Drug Abuse 1988). The sample of respondents
used in this study had an age range of
23-32 in 1988, and 32.6% of the men in that
sample reported having used cocaine at some
time in their lives, slightly higher than the
32.3% figure reported in the NHS survey for
a similarly aged (26-34) group of men."
70.6% of the men in this sample reported
having used marijuana at some time, compared
to 68.1% of men in the NHS. Also
consistent with the NHS results are the responses
of women in the present sample. Of
the women in this sample-who, like the
men, were aged 23-32-21.2% reported having
used cocaine at some time and 58.9%
I The NHS numberswould be expected to be higher,
since the NHS sample was somewhat older than the
sample examined here, and therefore had a greater
chance of having initiated use. In addition, the NLSY
oversamples blacks and respondents from the South,
two groups that have reported levels of illicit drug use
below that of the entire population (Kozel and Adams
1985).
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
460 INDUSTRIAL AND LABOR RELATIONS REVIEW
Table 1. Distribution of the Total Sample by Gender and Frequency of Drug Use.
Lifetime Lifetime
Frequency of Men Women Frequency of Men Women
Year Cocaine Use N % N % Marijuana Use N % N %
1984 0 2930 80.4 3655 87.3 0 1107 30.4 1804 43.1
1-9 360 9.9 286 6.8 1-9 916 25.1 1161 27.7
10-39 165 4.5 135 3.2 10-39 385 10.6 481 11.5
40-99 93 2.6 60 1.4 40-99 356 9.8 274 6.5
100-999 73 2.0 37 0.9 100-999 444 12.2 296 7.1
1000+ 23 0.6 15 0.4 1000+ 436 12.0 172 4.1
1988 0 2204 67.4 3084 78.8 0 962 29.4 1608 41.1
1-2 366 11.2 292 7.5 1-2 457 14.0 725 18.5
3-9 249 7.6 222 5.7 3-9 433 13.2 451 11.5
10-39 236 7.2 191 4.9 10-39 455 13.9 514 13.1
40-99 106 3.2 72 1.8 40-99 271 8.3 227 5.8
100+ 109 3.3 51 1.3 100+ 692 21.2 387 9.9
Past 30 Day Past 30 Day
Frequency of Men Women Frequency of Men Women
Year Cocaine Use N % N % Marijuana Use N % N %
1984 0 3458 94.9 4057 96.9 0 2622 72.0 3613 86.3
1-2 89 2.4 71 1.7 1-2 243 6.7 200 4.8
3-5 41 1.1 24 0.6 3-5 196 5.4 113 2.7
6-9 30 0.8 17 0.4 6-9 158 4.3 78 1.9
10-19 17 0.5 15 0.4 10-19 194 5.3 82 2.0
20-39 7 0.2 2 0.0 20-39 130 3.6 61 1.5
40+ 2 0.1 2 0.0 40+ 101 2.8 41 1.0
1988 0 3122 95.5 3818 97.6 0 2770 95.5 3590 91.8
1-2 73 2.2 60 1.5 1-2 113 2.2 123 3.1
3-5 33 1.0 22 0.6 3-5 104 1.0 64 1.6
6-9 20 0.6 1 0.0 6-9 77 0.6 34 0.9
10-19 12 0.4 7 0.2 10-19 102 0.4 41 1.0
20-39 4 0.1 2 0.1 20-39 62 0.1 27 0.7
40+ 6 0.2 2 0.1 40+ 42 0.2 33 0.8
Source: Figures are derived from the National Longitudinal Surveys of Labor Market Experience-Youth Cohort
(Center for Human Resource Research 1990).
reported having used marijuana at some time,
rates that closely match the corresponding
figures of 21.0% and 56.2% for the 26-34year-old
women in the NHS. This finding
raises questions about the extent of underreporting
in the NLSY, and particularly about
whether there was in fact substantial underreporting
in 1984, as suggested by Mensch
and Kandel (1988). Furthermore, Sickles and
Taubman (1991) reported findings from an
unpublished NLS study that suggest, contrary
to Mensch and Kandel's criticism, that
the self-reports in the NLSY are reliable.
For both marijuana and cocaine, five separate
measures of drug use were used in the
cross-sectional analyses: a linear measure of
lifetime use that takes on values of 0-5, corresponding
to the categories in Table 1; a linear
measure of lifetime use that uses the midpoints
of the categories observed in Table 1;
two similar measures for past 30 day use; and
a dummy variable indicating a relatively heavy
amount of lifetime use and non-zero past 30
day use. 12 It should be noted that the intervals
used to code the illicit drug use responses
12Heavy use of cocaine is defined as lifetime use of 40
or more times, and heavy use of marijuana is defined as
lifetime use of 100 or more times.
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
WAGE EFFECTS OF MARIJUANA AND COCAINE USE 461
changed between the two surveys. The estimates
for the drug use equations were obtained
by OLS methods for the linear measures,
and by a probit regression model for
the dummy variable indicating heavy use.
For the longitudinal models, the coding
intervals used to group the data according to
lifetime drug use were standardized between
the two years, and only the simple (nonmidpoint)
linear measures were used in the
analysis.'3 These variables took on values from
0 to 4, indicating the following frequencies of
use: 0, 1-9, 10-39, 40-99, and 100 or more
times. In addition, a differenced version of
this linear lifetime measure was created, and
this variable took on values ranging from 0,
for no change in use, to 4, for changes in
reported lifetime use exceeding 99 (that is, in
effect, reports of 100 or more instances of use
in the past four years). Since lifetime drug
use cannot decline, all analyses were estimated
twice-once with consistency of response
imposed on the data by using the 1984
value if the value reported in 1988 was below
that in 1984, and a second time with no such
adjustment of contradictory values. The results
do not differ according to which method
is used, and the results reported in the text
are for data in which consistency of the response
has been imposed.
As noted above, the grouped nature of the
data limits the value of this differenced variable,
since much of the variation in use over
time is unobservable. For example, individuals
who were in the highest category of use in
1984 will never be observed to have an increase
in use. To better differentiate between
types of users, two additional variables were
created to be used in conjunction with this
differenced measure of lifetime use. The first
is a dummy variable indicating no reported
use in either survey, and the second is a
dummyvariable indicating initiation into use
"3The large intervals used in the NLSY make it difficult
to use midpoints. There is little information on the
nature of the true distribution of drug users within
intervals, and any estimate of the mean within the
interval would be ad hoc. In addition, estimating the
mean of the open-ended interval would also be errorridden.
Because of these concerns, I used the
untransformed data.
between the two surveys.'4 Measures created
for past 30 day drug use were similar, except
that the two additional dummyvariables were
not used in the analysis of the effect of recent
drug use.
Cross-Sectional Results
One purpose of this paper is to use the
1988 NLSYsurveyyear information to update
the previous cross-sectional estimates of the
effect of illicit drug use on wages. The model
used to generate these estimates closely follows
that in my 1991 study (Kaestner 1991),
with only a few minor modifications.'5
Table 2 lists the parameter estimates of the
effect of illicit drug use on the wage, and a full
set of estimates is provided in the appendix.
Five separate models were estimated for each
gender group and both years of data. For
example, lifetime cocaine use and past 30 day
cocaine use were not entered into the same
model, due to the high degree of collinearity
between the two predicted measures of illicit
drug use. The same reasoning would also
apply to the other models in which separate
drug use measures were entered.
The parameter estimates associated with
the 1988 survey year data are generally similar
to those for 1984. In fact, the estimates of
the effect of illicit drug use on the wage are
large, positive, and frequently significant in
both years, across both gender groups, and
for both types of drugs. The only negative
effect observed among the estimates is associated
with heavy cocaine use (as defined in
this paper) among the 1988 female sample.
This coefficient has a very low level of significance,
and its importance should be dis-
14The predicted values of these variables are obtained
by maximum likelihood probit methods. The
probits are estimated independently of each other.
15The differences are related to aspects of the sample,
exogenous variables, and estimation strategy. First, in
this study, individuals living in temporary quarters have
been deleted, as have individuals with missing information
related to the measure of self-esteem. Next, age and
AFQT test score enter the model as quadratics. Finally,
as mentioned in note 3, the reduced form drug estimates
were obtained using the entire sample, as opposed
to only the employed.
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
462 INDUSTRIAL AND LABOR RELATIONS REVIEW
Table 2. Cross-Sectional Estimates of the Effect of Illicit Drug Use
on the Natural Logarithm of the Wage.
(Standard Errors in Parentheses)
1984 1988
Drug Type/Model Men Women Men Women
Marijuana Use
1 Lifetime Use .042** .022 .031 .044*
(0,1,2,3,4,5) (.022) (.024) (.028) (.028)
2 Lifetime Use .0002** .0002 .001 .002
(midpoints) (.0001) (.0002) (.001) (.001)
3 Heavy Use .161 .166 .424** .246
(0,1) (.123) (.173) (.192) (.286)
4 Past 30 Day Use .065* .052 .034 .153
(0,1,2,3,4,5,6) (.003) (.047) (.060) (.122)
5 Past 30 Day Use .011+ .010 .006 .012
(midpoints) (.006) (.008) (.012) (.017)
Cocaine Use
1 Lifetime Use .096* .124* .046 .153***
(0,1,2,3,4,5) (.052) (.066) (.044) (.051)
2 Lifetime Use .001* .002** .003 .010**
(midpoints) (.001) (.001) (.002) (.004)
3 Heavy Use .012 .449 1.033** -.361
(0,1) (.276) (.364) (.418) (.509)
4 Past 30 Day Use .313** .515** .315 .493
(0,1,2,3,4,5,6) (.142) (.227) (.221) (.573)
5 Past 30 Day Use .098** .169** .077* .055
(midpoints) (.046) (.079) (.048) (.144)
Observations 2852 2619 2907 2724
Notes: In all models the actual value of drug use is replaced by its predicted value. In models 1, 2, 4, and 5, the
prediction method was an OLS regression. In model 3, a probit procedure was utilized, and the dummy variable was
replaced by the predicted probability. All models include the inverse mills ratio associated with the Heckman two step
sample selection correction.
*Statistically significant at the .10 level; **at the .05 level; ***at the .01 level.
counted accordingly. The coefficient on heavy
cocaine use for men in 1988, however, has a
somewhat unbelievably large, positive effect,
which is statistically significant. In general,
the results listed in Table 2 contradict prior
expectations regarding the effect of illicit
drug use on the wage, and are consistent with
results that have been previously reported.
The results for the 1984 survey data are virtually
the same as those I reported three years
ago (Kaestner 1991), in the study that provides
the framework for the analysis used
here.
Respondents in the cross-sectional sample
from 1988 were, on average, four years older
than those in the 1984 sample. Therefore, if
the detrimental effects of drug use are cumulative,
the potential for observing a negative
relationship between illicit drug use and wages
should be greater in 1988 than in 1984. The
results reported above, however, do not support
this hypothesis. For men, the magnitude
and significance of the results tend to diminish
somewhat across years, but for women the
opposite is true. Moreover, the changes in
magnitude may be explained by slight differences
between the units of measurement of
the drug use variables in the two survey years.
On the other hand, drug use may have adverse
effects on wages that become noticeable
only over a period longer than four
years.
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
WAGE EFFECTS OF MARIJUANA AND COCAINE USE 463
Fixed Effect Estimates
The cross-sectional results are surprising,
and most observers would probably consider
them inadequate as a description of the true
relationship between drug use and wages.
The cross-sectional estimates are questionable
primarily due to the possibility that there
are unobserved characteristics that have been
omitted from the analysis that are correlated
with both drug use and the wage. To address
this problem, I created a limited panel of data
consisting of a matched sample of respondents
present in both years and implemented
a fixed effects estimator. The model used to
generate these estimates is specified above,
and is a relatively unrestrictive version of a
model of first differences.
Before examining the fixed effects estimates,
note that the cross-sectional models of
Table 2 were re-estimated using the matched
sample. If attrition in the sample is an important
source of potential bias, the cross-sectional
results using the matched sample
should differ from those reported in Table 2.
On the other hand, if attrition is not a problem,
the estimates from a cross-sectional analysis
should be similar. In fact, when the crosssectional
models were re-estimated on the
matched sample, the results (not shown)
differed very little. The signs associated with
the drug use coefficients were identical, and
the magnitudes of the drug effects were very
close to those reported in Table 2. The standard
errors associated with the estimates,
however, were larger, and the effects were
therefore reduced in significance.
There were two exceptions to these generalizations,
and both involved the effect of
past 30 day marijuana use on the female
wage. In 1984 and 1988 the estimates of this
effect were substantially smaller for the
matched sample than for the full sample.
Thus, to the extent that the cross-sectional
and panel data estimates differ, that difference
does not stem from changes in the
sample, except possibly for the effect of past
30 day marijuana use on the wages of women.
Table 3 lists the parameter estimates associated
with the illicit drug use measures; a
complete set of results is contained in the
appendix. For simplicity, the discussion of
the results will be in terms of the effect of
illicit drug use on the level of the wage, even
though the underlying model is of wage
changes.
Drug use by men. All of the estimates of the
effect of illicit drug use on wages of men are
negative. They are not statistically significant,
however, and the large size of the standard
errors associated with the parameter estimates
underscores the need to approach these
estimates cautiously. The magnitudes of these
estimates are substantial, but they are difficult
to interpret given the way illicit drug use
is measured. For example, a one-unit increase
in cocaine use over the preceding
four-year period would be expected to reduce
an individual's wage anywhere from 2
(model 2) to 22% (model 3 for a new user),
and a one-unit increase in marijuana use
would be expected to reduce the wage by
between 9% (model 2) and 52% (model 3 for
a new user).
The problem lies in the interpretion of
what is meant by a one-unit increase, since
the groupings into drug use categories were
somewhat irregular. The specification of the
illicit drug use variables in model 3 helps
clarify the interpretation. For both marijuana
and cocaine, an increase in use that is also
associated with an individual's initiation into
use has a more adverse impact on an
individual's wage than an increase in use for
a previous user.16 In the case of cocaine,
approximately 80% of all observed increases
in use-531 male respondents, or 21.3% of
the total male sample-were for people initiating
use, with over half of these cases having
a total reported use of only 1 to 9 times over
the four-year period. For marijuana, approximately40%
ofthe observed increases in use -
589 respondents, or 23.6% of the total male
sample-were for individuals who initiated
use during this period, and about a third of
these cases were individuals who reported
use of only 1 to 9 times during the periodY7
16The coefficients on the change in use and initiation
into use variables should be considered in an
additive fashion when deriving the total wage effect of
initiation into use. The total effect is not, however, the
simple sum of the two estimates.
17The figures on initiation into use of marijuana that
are reported by the matched sample indicate that more
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
464 INDUSTRIAL AND LABOR RELATIONS REVIEW
Table 3. Fixed Effects Estimates of the Effect of Illicit Drug Use
on the Natural Logarithm of the Wage.
(Standard Errors in Parentheses)
Cocaine Use Marijuana Use
Drug Type/Model Men Women Men Women
1 Lifetime Use 1984 -.224 .828* -.079 -.277
(0,1,2,3,4) (.428) (.428) (.284) (.277)
Lifetime Use 1988 -.137 .607* -.086 -.254
(0,1,2,3,4) (.275) (.316) (.219) (.271)
2 Change in Lifetime -.024 .276 -.093 -.250
Use, 1988-84 (.155) (.223) (.208) (.270)
3 Change in Lifetime -.183 .538 -.032 -.129
Use, 1988-84 (.317) (.380) (.219) (.399)
New User -.225 .748 -.523 -.812
(.586) (.687) (.530) (.928)
Never Used -.274 .521** .130 -.159
(.338) (.255) (.238) (.206)
4 Past 30 Day Use, 1984 -.106 .159 -.096 .024
(0,1,2,3,4,5,6) (.467) (.343) (.105) (.130)
Past 30 Day Use, 1988 -.094 .525 -.186 .086
(0,1,2,3,4,5,6) (.405) (.700) (.152) (.252)
5 Change in Past 30 -.101 .166 -.051 -.002
DayUse, 1988-84 (.409) (.343) (.106) (.112)
Observations 1858 1623 1858 1623
Notes: In all models the actual value of drug use is replaced by its predicted value, and the prediction method was
an OLS regression. Note that the signs associated with the 1984 drug variable coefficients have been reversed, since
the regression package assumes that the model is additive. All models include the inverse mills ratio for each year.
*Statistically significant at the .10 level; **at the .05 level; ***at the .01 level.
It is not known whether the negative effect
associated with the initiation measure of drug
use is due primarily to the phenomenon of
initiation itself, or to initiation into heavy use
over a relatively short period of time. If the
latter reason accounts for the negative effect,
we would expect this effect to be greater in
the case of marijuana, since a larger proportion
of the individuals who started using
marijuana than of those who started using
cocaine became relatively heavy users. This
expectation is in fact supported, as illustrated
in Table 3.
There is, however, one anomalous result
regarding the male sample. The wages of
those men who never used cocaine is expected
to be 27% lower than the wages of
similar individuals who previously used cocaine,
but did not increase their use during
initiation into marijuana use took place than is implied
in Table 1.
the four-year period between 1984 and 1988.
Since most individuals reported a low level of
lifetime use, this result implies that individuals
who experimented with cocaine fared
better than those who did not. Initiation into
cocaine use over this period, however, does
have a negative effect on the wage, as does an
increase in use. Thus, those individuals who
tried cocaine while relatively young, but did
not increase their use afterward, are expected
to have the highest wage, even compared to
non-users.
Drug use by women. The findings for the
female sample are qualitatively differentfrom
those for the male sample. Cocaine use appears
to have had a large positive effect on
women's wages. A one unit increase in cocaine
use increased the wage by between 28%
(model 2) and 75% (model 3, new user).
These effects are extremely large, and in
some cases reach commonly accepted levels
of significance. The results obtain both for
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
WAGE EFFECTS OF MARIJUANA AND COCAINE USE 465
lifetime cocaine use and for past 30 day cocaine
use. The model 3 estimates for women
differ sharply from those for the male sample.
These estimates suggest that the women who
tried cocaine at a relatively young age tended
to have a lower wage than women who either
never used cocaine or initiated use between
1984 and 1988. Over 83% of all observed
increases in use-447 respondents, or 14.1 %
of the total female sample-were individuals
who first tried cocaine during this period,
and over 60% of these individuals reported
lifetime use of 1-9 times by 1988. Thus, very
few of the observed increases in use are among
relatively heavy users. Initiation into cocaine
use during this period did not adversely affect
women's wages.
The effect of marijuana use was negative
for the female sample, although past 30 day
marijuana use does have a positive coefficient.
As was the case for the other estimates
reported in Table 3, the standard errors associated
with these estimates are large, and
result in relatively low levels of significance.
Conclusion
This study has had two purposes: first, to
update previous cross-sectional estimates of
the effect of illicit drug use on the wage, using
data from the 1988 wave of the NLSY; and
second, using both the 1984 and 1988 NLSY,
to provide a set of longitudinal estimates of
the effect of illicit drug use on the wage. The
longitudinal estimates are preferred, since in
theory this methodology controls for potentially
important unobserved individual characteristics
that cause the cross-sectional estimates
to be biased. I performed both analyses
separately by gender.
In regard to the first objective, my estimates
of the effect of illicit drug use on the
wage using the 1988 NLSYdata are consistent
with those I obtained in 1991 using the 1984
NLSY data. Both sets of estimates indicate
large, positive, statistically significant effects
of illicit drug use on the wage, for both
gender groups and for both marijuana and
cocaine use.
These findings raise several disturbing
questions related to drug prevention policy.
Much time and money has been and continues
to be invested in deterring and preventing
drug use, and this investment is based on
the proposition that illicit drug use adversely
affects users. Few economists would be willing
to argue that illicit drug use is somehow
beneficial for young people, and is a characteristic
that is rewarded in the labor market.
Thus, an explanation of the cross-sectional
results that would confirm our commonly
held beliefs would be comforting. The most
commonly invoked explanation of that kind
has been that important unobserved characteristics
that are positively correlated with
both wages and illicit drug use underlie the
empirical relationship found in cross-sectional
models.
The longitudinal estimates presented in
this paper, however, provide only partial support
for that explanation. To some extent,
the large standard errors associated with the
fixed effects estimates can be interpreted to
mean that there are really no effects of illicit
drug use on the wage that are significantly
different from zero. Unfortunately, the standard
that must be imposed to draw that conclusion
may be too stringent for this type of
empirical analysis. If the sign and magnitude
of the effects of illicit drug use on the wage
are examined, the evidence is mixed. Among
the male sample, illicit drug use tended to be
negatively related to wages-a finding in accord
with most people's expectations, and
one that would support continued vigilance
against drug use. Similarly, among the female
sample, lifetime marijuana use appears
to have been negatively associated with the
wage; but recent marijuana use (use within
the previous 30 days) had a positive effect,
and, more important, cocaine use (both recent
and lifetime) was positively related to
the wage, and had quite a large impact.
There are two possible explanations for
the results reported in this paper. The mixed
nature of the preferred estimates those
obtained using the longitudinal data-implies
that there is a wide range of wage effects
among people who consume the same amount
of drugs. Thus, illicit drug use may be a highly
idiosyncratic phenomenon that has a variety
of consequences that depend on the individual,
the type of drug, or some combination
of the two. The adverse physical and
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
466 INDUSTRIAL AND LABOR RELATIONS REVIEW
psychological consequences of drug use are
known to be quite individual-specific, and a
given level of use may therefore lead to a wide
variety of labor market outcomes. Another
possibility is that some drug users choosejobs
in which their drug use has the least impact
on their productivity, but others do not. In
addition, the drug use measures incorporated
in the analysis were extremely crude,
and exactly what they measure may differ
drastically from person to person. These considerations
may explain why the observed
wage effects of illicit drug use differ for men
and women and are influenced by the type of
drug and timing of its use.
The second possible explanation for the
results obtained in this paper is that the
current analysis, although a significant improvement
on past work, is still inadequate.
The fixed effects methodology used in this
paper controls for characteristics that do not
change over time, but among a sample of
young adults in their twenties, there may be
few characteristics that remain constant.
These shortcomings of the analysis point to
refinements that should be made in future
studies to allow the identification of important
personality traits and patterns of drug
use that affect the wage.
I believe this study has taken an important
new step in the analysis of the effect of illicit
drug use. It is the first analysis to exploit
longitudinal data, and therefore controls for
unobserved person-specific effects that are
important determinants of the wage. It has
provided some important insight into the
effect that these unobserved factors have on
the relationship between illicit drug use and
wages, and it has provided a foundation for
future work on that subject.
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
WAGE EFFECTS OF MARIJUANA AND COCAINE USE 467
APPENDIX TABLE 1
DESCRIPTIVE STATISTICS OF VARIABLES USED IN THE ANALYSIS
Men, 1988 Women, 1988 Men, 1984 Women, 1988
Variable Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.
Age 27.420 2.240 27.580 2.268 23.270 2.257 23.33 2.219
Black 0.260 0.439 0.271 0.445 0.250 0.433 0.252 0.434
Hispanic 0.164 0.370 0.154 0.361 0.158 0.365 0.157 0.364
AFQT 64.300 23.330 65.100 20.950 63.670 23.220 65.190 21.110
Experience 7.379 2.804 6.131 3.228 3.860 2.010 3.265 2.143
Education 12.680 2.394 12.780 2.266 12.170 2.058 12.370 2.036
No. Life Employers 7.430 4.174 6.269 3.945 4.889 3.000 4.099 2.782
No. Children 0-2 0.162 0.401 0.203 0.426 0.115 0.342 0.227 0.460
No. Children 3-5 0.185 0.450 0.331 0.559 0.087 0.328 0.245 0.495
No. Children 6-18 0.187 0.537 0.513 0.853 0.039 0.244 0.144 0.452
Health 0.040 0.196 0.051 0.220 0.033 0.180 0.050 0.217
Never Married 0.455 0.498 0.328 0.470 0.670 0.470 0.505 0.500
Sep-Divorce 0.105 0.306 0.163 0.370 0.053 0.225 0.101 0.301
Religious Attendance 3.111 1.679 3.419 1.696 2.983 1.655 3.358 1.687
Self-Esteem 32.360 4.095 31.980 4.124 32.370 4.039 32.060 4.156
Rotter Index 8.684 2.371 8.820 2.396 8.654 2.389 8.827 2.407
No. IllegalActs 11.800 31.250 5.051 18.260 12.750 36.400 4.949 17.870
Two Parents at Age 14 0.775 0.418 0.748 0.434 0.773 0.419 0.756 0.430
Live at Home 0.213 0.409 0.139 0.346 0.455 0.498 0.324 0.468
Non-Earned Income 7645 13033 13500 16335 10057 14360 11613 13599
Missing Income 0.184 0.388 0.156 0.363 0.185 0.389 0.160 0.367
Urban 0.792 0.406 0.781 0.413 0.784 0.411 0.796 0.403
Northeast 0.184 0.388 0.178 0.382 0.185 0.389 0.176 0.381
North-Central 0.243 0.429 0.236 0.425 0.250 0.433 0.234 0.423
South 0.378 0.485 0.403 0.490 0.364 0.481 0.397 0.489
Observations 3270 3912 3644 4188
Definitions: Age: age in years. Black: indicates respondent is black. Hisp: indicates respondent is hispanic. AFQT:
combined score on armed forces qualifications test. Experience: actual years of labor market experience. Education:
highest grade completed. No. Life Employers: number of previous employers. No. Children 0-2: number of dependent
children less than 3 years of age. No. Children 3-5: number of dependent children 3-5 years of age. No. Children 6-18:
number of dependent children 6-17 years of age. Health: indicates respondent is limited in activity. Never Married:
indicates respondent was never married. Sep-Divorce: indicates respondent is separated or divorced. Religious
Attendance: frequency of religious attendance as of 1980. Self-Esteem: psychological scale measuring feelings of selfworth.
Rotter Index: psychological scale measuring a person's locus of control. No. Illegal Acts: number of illegal acts
respondent reported in 1980. Two Parents at Age 14: respondent had two-parent household at age 14. Live at Home:
respondent resides in family home. Non-Earned Income: all non-earned income of respondent's household. Missing
Income: indicates income variable is missing. Urban: indicates respondent lives in urban area. Northeast: indicates
respondent lives in Northeast region. North-Central: indicates respondent lives in North-Central region. South:
indicates respondent lives in the South.
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
468 INDUSTRIAL AND LABOR RELATIONS REVIEW
APPENDIX TABLE 2
CROSS-SECTIONAL PARAMETER ESTIMATES FROM THE WAGE MODEL
Men, 1988 Women, 1988
Variable Coefficient Std. Error Coefficient Std. Error
Constant 1.721 1.594 0.628 0.505
Age -0.033 0.117 0.085 0.032
Age Squared 0.000 0.002 -0.002 0.001
Black -0.024 0.027 0.061 0.029
Hispanic 0.023 0.028 0.090 0.031
AFQT 0.004 0.004 -0.003 0.004
AFQT Squared -0.000 0.000 0.000 0.000
Experience 0.106 0.046 -0.005 0.035
Experience Squared -0.004 0.002 0.001 0.001
Education -0.019 0.036 -0.050 0.035
Education Squared 0.003 0.002 0.002 0.002
Exper x Education -0.000 0.002 0.003 0.002
AFQT x Education -0.000 0.000 0.001 0.000
No. Life Employers -0.016 0.003 -0.021 0.004
Health -0.067 0.050 -0.037 0.043
Never Married -0.111 0.025 -0.033 0.028
Sep-Divorce -0.103 0.037 0.004 0.032
Self-Esteem 0.006 0.003 0.007 0.002
Rotter Index -0.007 0.004 -0.006 0.004
Two Parents at Age 14 0.001 0.023 0.013 0.021
Live at Home -0.136 0.031 -0.121 0.027
Urban 0.122 0.078 0.159 0.084
Northeast 0.033 0.096 0.192 0.101
North-Central -0.117 0.089 0.038 0.092
South -0.025 0.083 0.005 0.087
Urban x Northeast 0.073 0.100 -0.090 0.105
Urban x South -0.084 0.084 -0.058 0.089
Urban x North-Central 0.064 0.089 -0.103 0.094
Lifetime Cocaine 0.046 0.044 0.153 0.051
Inverse Mills Ratio 0.273 0.161 -0.118 0.079
Adjusted R-Square .239 .293
Observations 2907 2724
Notes: Estimates from other models are available from the author upon request. The reported standard errors are
only approximate, since they do not take into account the predicted nature of illicit drug use.
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
WAGE EFFECTS OF MARIJUANA AND COCAINE USE 469
APPENDIX TABLE 3
PARAMETER ESTIMATES FROM THE FIXED EFFECT MODEL OF WAGES
Men Women
Variable Coefficient Std. Error Coefficient Std. Error
Constant 0.301 1.734 2.301 1.529
Black -0.084 0.051 0.038 0.043
Hispanic 0.042 0.066 0.144 0.062
AFQT -0.000 0.001 0.001 0.001
Self-Esteem -0.004 0.004 0.005 0.003
Rotter Index -0.007 0.006 0.008 0.006
Two Parents at Age 14 -0.095 0.038 0.017 0.029
Age -0.037 0.161 -0.231 0.128
Age Squared 0.000 0.003 0.004 0.003
Education 1984 -0.143 0.197 -0.017 0.108
Education Sq 1984 0.007 0.007 0.002 0.004
Education 1988 -0.137 0.172 -0.019 0.109
Education Sq 1988 0.008 0.006 0.003 0.004
Experience 1984 0.003 0.001 0.004 0.001
Experience Sq 1984 0.000 0.000 -0.000 0.000
Experience 1988 0.003 0.002 0.001 0.002
Experience Sq 1988 0.000 0.000 0.000 0.000
No. Life Employers 1984 0.017 0.013 -0.013 0.013
No. Life Employers 1988 0.014 0.010 -0.007 0.010
Urban 0.111 0.101 0.183 0.096
Northeast 0.130 0.124 0.495 0.132
North-Central 0.156 0.111 -0.278 0.116
South 0.100 0.118 0.246 0.100
Urban x Northeast -0.061 0.123 -0.379 0.130
Urban x North-Central -0.158 0.106 -0.253 0.113
Urban x South -0.138 0.120 -0.238 0.099
Live at Home 1984 -0.036 0.032 -0.027 0.032
Live at Home 1988 -0.080 0.039 -0.132 0.038
Never Married 1984 -0.028 0.042 -0.062 0.051
Sep-Divorce 1984 0.115 0.067 0.039 0.048
Never Married 1988 -0.020 0.042 -0.072 0.053
Sep-Divorce 1988 -0.044 0.083 0.125 0.061
Health 1984 0.059 0.070 -0.084 0.064
Health 1988 -0.041 0.050 -0.053 0.064
Lifetime Cocaine 1984 -0.224 0.428 0.828 0.428
Lifetime Cocaine 1988 -0.137 0.275 0.607 0.316
Inverse Mills Ratio 1988 0.489 0.270 0.162 0.161
Inverse Mills Ratio 1984 -0.008 0.171 -0.095 0.105
Adjusted R-Square .078 .079
Observations 1858 1623
Notes: Estimates from other models are available upon request from the author. The standard errors reported are
based on the method proposed by White (1980).
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms
470 INDUSTRIAL AND LABOR RELATIONS REVIEW
REFERENCES
Center for Human Resource Research. 1990. NLSHandbook
1990. Columbus: Ohio State University.
Chamberlain, Gary. 1984. "Panel Data." In Zvi Griliches
and Michael Intrilligator, eds., Handbook ofEconometrics,
Vol. 2. Amsterdam: North Holland, pp. 1248-1313.
. 1982. "Multivariate Regression Models for Panel
Data." Journal ofEconometrics, Vol. 18, No. 1, pp. 5-46.
. 1980. "Analysis of Covariance with Qualitative
Data." Review of Economic Studies, Vol. 47, No. 2, pp.
225-38.
Gill, Andrew M., and RobertJ. Michaels. 1990. "Drug
Use and Earnings: Accounting for Self Selection of
Drug Users." Working Paper 11-90, California State
University-Fullerton.
._ 1992. "Does Drug Use Lower Wages?" Industrial
and Labor Relations Review, Vol. 45, No. 3, pp. 93-105.
Goldstein, Avram, and Harold Kalant. 1990. "Drug Policy:
Striking the Right Balance." Science, Vol. 249 (September
28), pp. 1513-21.
Gust, Steven W., J. Michael Walsh, Linda B. Thomas,
and Dennis J. Crouch. 1991. Drugs in the Workplace:
Research and Evaluation Data, Volume 2. Washington,
D.C.: U.S. Department of Health and Human Services.
Hayghe, Howard V. 1991. "Survey of Employer AntiDrug
Programs." In Steven W. Gust,J. Michael Walsh,
Linda B. Thomas, and DennisJ. Crouch, eds., Drugs in
the Workplace: Research and Evaluation Data, Volume 2.
Washington, D.C.: U.S. Department of Health and
Human Services.
Heckman, James J. 1976. "The Common Structure of
Statistical Models of Truncation, Sample Selection
and Limited Dependent Variables. " Annals ofEconomic
and Social Measurement, Vol. 5, No. 4, pp. 475-92.
Kaestner, Robert. 1991. "The Effect of Illicit Drug Use
on the Wages of Young Adults." Journal of Labor Economics,
Vol. 9, No. 4, pp. 381-412.
._ 1994. "The Effect of Illicit Drug Use on the Labor
Supply of Young Adults." Journal of Human Resources,
Vol. 29, No. 1, pp. 126-55.
Kandel, Denise B., and Mark Davies. 1990. "Labor Force
Experiences of a National Sample of Young Adult
Men." Youth and Society, Vol. 21, No. 4, pp. 411-45.
Kandel, Denise B., andJohn Logan. 1984. "Patterns of
Drug Use from Adolescence to Early Adulthood."
American Journal of Public Health, Vol. 74, No. 7, pp.
660-66.
Kozel, NicholasJ., and Edgar H. Adams. 1985. Cocaine
Use In America: Epidemiological and Clinical Perspectives.
Washington, D.C.: U.S. Department of Health and
Human Services.
Mensch, Barbara S., and Denise B. Kandel. 1988.
"Underreporting of Substance Use in a National Longitudinal
Youth Cohort." Public Opinion Quarterly, Vol.
52, No. 1, pp. 100-124.
Murphy, Kevin M., and Robert H. Topel. 1985. "Estimation
and Inference in Two-Step Econometric Models."
Journal of Business and Economic Statistics, Vol. 3, No. 4,
pp. 370-79.
National Institute on Drug Abuse. 1986. Drug Abuse in
the Workplace. Washington, D.C.: GPO.
__. 1988. National Household Survey on Drug Abuse.
Rockville, Md.: U.S. Department of Health and Human
Services.
__. 1991. National Household Survey on Drug Abuse.
Rockville, Md.: U.S. Department of Health and Human
Services.
Raveis, V. H., and Denise B. Kandel. 1987. "Changes in
Drug Behavior from the Middle to Late Twenties:
Initiation, Persistence, and the Cessation of Use."
American Journal of Public Health, Vol. 77, pp. 607-11.
Register, Charles A., and Donald R. Williams. 1992.
"Labor Market Effects of Marijuana and Cocaine Use
Among Young Men." Industrial and Labor Relations
Review, Vol. 45, No. 3, pp. 106-23.
Sickles, Robin, and Paul Taubman. 1991. "Who Uses
Illegal Drugs." American Economic Review, Vol. 81, No.
2, pp. 248-51.
SteinJudith A., Michael D. Newcomb, and P. M. Bentler.
1988. "Structure of Drug Use Behaviors and Consequences
Among Young Adults: Multitrait Multimethod
Assessment of Frequency, Quantity, Work Site, and
Problem Substance Use." Journal ofApplied Psychology,
Vol. 73, pp. 595-605.
Stigler, George, and Gary Becker. 1977. "De Gustibus
Non Est Disputandum. " AmericanEconomicReview, Vol.
67, No. 2, pp. 76-90.
This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:41:05 UTC
All use subject to http://about.jstor.org/terms