likely to spend this much. We also reasoned that about 86
percent of expenditures on heroin could be attributed to
addicts". However, even if we make all these extreme assump-
tions, heroin expenditures cannot exceed approximately $25
billion."
Therefore, we conclude that the supply-based model for
heroin is not credible. We continue to improve this model and
expect that we will develop credible supply.-based heroin
estimates in the future.
MARIJUANA
It is difficult to develop an estimate of the size of the
U.S. retail market for marijuana from estimates of available
supply. First, the amount of marijuana that Americans cultivate
for personal use is impossible to estimate. Second, even though
a large amount of the domestic marijuana market is grown in the
For axupla. auppos* X !• Um aaount of haroln troa Waatam Haalaphara aourcaa. and auppoaa that Y
parcant of the U.S. aarkct la aat b7 thoaa Waatam Haalaphara aourcaa. Than tha worldvlda auppir of haroln aant
to tha Unltad Stataa auat a^ual X/Y.
D. Haalll and P. coolay. "national tatlaataa of Haroln Pravalanca 19S0-1987: Raaulta froa Analjala
of OAUM taargancT tooa Data.' rri Tachnleal Daport. Raaaarch Trlangla Inatltuta. 1990.
" Thla parcantajc would cartalnly ba lar^ar If «a aaauaad that 1 allllon addlcta apant S420 par iraak
on heroin.
69
Multiply 1 allllon addlcta br S420 par \rmmit by 52 ttaaka and dlelda by 0.86.
34
401
United States,'" countries in South and Central America, the
Caribbean, Asia, North Africa, and the Middle East also supply
cannabis to the domestic market.
Using available production data, conservative assumptions
lead to estimates that Americans spend S261 billion to S281
billion a year on marijuana." However, Table 13 reported
retail expenditures on marijuana as less than SIO billion.
Either the estimates of worldwide cultivation and production are
wildly Inflated, or seizures, losses, and consumption within
producer countries have been underestimated by an equal amount,
or the United States' share of the worldwide marijuana consumer
market is much smaller than suggested here. Unfortunately, the
data needed to develop better estimates of these factors are not
available, and without the Independent ability to assess the
reliability of the marijuana cultivation estimates, we cannot
develop a plausible supply-based estimate of the retail value of
the marijuana market in the United States.
70
Tha DEA •■clB«t«a that apprvalaataly 5.000 to 6.000 satrlc tona of can/)a£la vara cviltlvatad during
1««0 Iraportad In tha latlonal Rarcotlca Intalllfanca Conaisara Coaalttaa (imiCC). Tha miicg Haport mo Tha
Suaala of llMr lt Bruoa tn tha iinit.H ataiaa (Waahlngton, D . C . : national Harcotlca Intalllganca Conataara
Coaalttaa. Juna 1991). )]).
71
kmm\mm that nofia of tha doaaatlcally grvwn Marijuana la aaportad. tliat 77 parcant of all Isporta Into
tha Unltad Stataa ara Haalcan »arl]uana. and that all aartjuana producad alaaifhara la axportad to othar
countrlaa. Than If tha DIA/nicC aatlaataa ara raaaonably accurata. approilBataly 2(.100 to K.lOO aatrlc tona
vould haTa baan avallabla daring 1990. At an avaraga coat of 910 par graa. tha ratall valua of thla aaount of
•arljuana oould ha'a baan (261 to till billion dollara. If tha aaaraga prlca oara cloaar to Si par «raa. tha
ratall valua vovild ha»a baan half that lar«a i Tha mice lapnrt 1990 1.
35
402
in. SUMMARY
Because of the quality of available data, there is
considerable imprecision in estimates of the number of heavy and
casual users of drugs, the amount of drugs they consume, and the
retail sales value of those drugs. The best estimates (all for
1991) follow:
About 2.1 million Americans are heavy cocaine users;
590,000 are heavy heroin users; 5.7 million use
marijuana at least one day a week; and an unknown
number are weekly users of other illicit drugs and
legal drugs used illicitly.
• About 5.4 million Americans are casual users of
cocaine; about 380,000 are casual users of heroin;
about 13.6 million use marijuana less than once a
week; and millions more use other drugs less than
once a week.
Trends are difficult to discern. Nevertheless, it
appears that the number of heavy cocaine and heroin
users has changed little over the last four years,
except for a decrease in 1990 that could be
attributed to an Increase in the price of cocaine.
In contrast, the number of weekly marijuana users
has declined.
• Casual use of cocaine and heroin has declined. However,
small sample sizes make the heroin data unreliable. No
decline in the casual use of marijuana was apparent.
Deriving estimates of the total expenditure on illicit drugs
and licit drugs consumed illegally is more difficult and
uncertain because those estimates require even more data about
prices paid. Nevertheless, the best estimates indicate the
following:
Americans spent about $30 billion on cocaine, $9
billion on heroin, $8 billion on marijuana, and $2
billion on other substances. Income in kind earned
by drug dealers and others probably adds about $2 to
$3 billion to the cocaine figure and another $3
billion for heroin.
• Again, estimating trends is risky, but it does not
appear that expenditures on cocaine and heroin have
changed much over the last four years. In contrast,
expenditures on marijuana and on other drugs seem to
have declined.
36
403
Estimates of the total amount of cocaine consumed are lower
than but broadly consistent with estimates of the total amount of
cocaine available for consumption:
• From the supply-side perspective, 274 to 442
metric tons of cocaine were available for
consumption in the United States.
• From the consumption perspective, Americans consumed
roughly 180 to 250 metric tons of cocaine.
Although the estimates from the supply-side perspective are
higher than those from the consumption perspective, the supply-
side estimates are surely overstated. First, they do not exclude
some losses that occur within the source countries but that
cannot be readily estimated; and second, they do not account for
domestic seizures by State and local officials. Although the
supply-side and the consumption estimates are remarkably close,
they cannot be completely reconciled.
The sizable price increase seen during 1990 is not
reflected in a comparable decrease in the supply of cocaine
available during 1990. This may have occurred because the supply
of cocaine on the street lags behind the supply of cocaine
entering the country, which lags behind the harvest of coca
leaves, so the supply-based and consumption-based estimates
should not be in lock step. Still, how the supply of cocaine
could have remained relatively constant across time while the
price of cocaine increased (and apparently the consumption of
cocaine decreased) during 1990 is a perplexing question.
Although these estimates paint a picture of drug consumption
with an extremely broad brush, and although not all estimates can
be reconciled, the approach we use provides an important
perspective on what is not known about drug production and
consumption and what needs to be known to better understand the
policy choices available to the Nation.
We make no pretense here that the model and estimates we
present in this report are fully adequate to the larger task of
informing public policy decisions. They are at best a start, but
offer important possibilities of integrating what are otherwise
often seen as disparate pieces of information about the
consumption and supply of drugs.
We expect incremental improvements to the estimates and
methods offered here, particularly as better data become
available. We also expect improvement in the model, which will
include systematic and analytic links between government policy
and drug use. Thus, it is probably best to consider this an
interim report. The estimates we present might be seen as an im-
provement over those reported in 1991 and as a prelude to
improved estimates for 1993.
37
404
Moreover, the estimates by themselves have only modest
importance - they tell us nothing more than that the drug trade
is large, a conclusion that requi^res no special study. The real
utility of these numbers is the development of a systematic
methodology for Integrating the various indicators - crops in
foreign countries, drugs seized at the borders, arrests made in
American cities, etc. - that can help policymakers to better
understand the dynamics of the drug trade and to better fashion
appropriate policy responses.
38
405
APPENDIX
ESTIMATING TYPICAL EXPENDITURES ON DRUG CONSUMPTION
This appendix discusses the methcxlology used to develop
estimates of weekly expenditures on cocaine and heroin by
arrestees who used either or both of these drugs on more than 10
days during the month before their arrest. The estimates
reported here are based on self-rep>orts by arrestees in 22
cities. These self-reports, which are for 1989 and 1990, are
from the Drug Use Forecasting (DUF) program.*
The Data
DUF respondents reported how much they spent on all drugs
combined (during a typical week) but not how much they spent on
each individual drug. They also reported the number of rtays they
used any of 22 kinds of drugs during the month before their
interview. We used regression analysis to infer expenditure
patterns for cocaine and heroin based on these data.
The greatest obstacle to accurate reporting Is a
respondents' denial of drug use.^ Therefore, drug use is
underreported. Once a respondent admits drug use, however, he or
she would seem to have less Incentive to underreport or
overreport consumption. To be included in this analysis, the
respondent had to have admitted some illicit drug use during the
last 30 days and had to have admitted some drug expenditure
during the typical week. (These different time periods were
required because of the wording of the DUF questions. ) A total
of 9,364 respondents were selected for analysis from the 1990 DUF
data and 5,151 were selected from the 1989 DUF data. We
estimated expenditure patterns for these two years separately.
The dependent variable (EXPEND) was the weekly expenditures
on all drugs. This variable was skewed (a few individuals
reported very high amounts). Consequently, weekly expenditure
was converted to a logarithm before estimating the regression.
We then converted the predictions back to the original dollar
scale.
The number of days that a respondent consumed each of four
categories of drugs were the Independent variables. We collapsed
drugs into four general categories: COCAINE (powdered and crack),
HEROIN (black tar and other), MARIJ (marijuana and hashish —
'No qucsuoa wu asked about the amouni of czpendiiura on dnip u the 1988 DUF diu, lo do uulyss wu perfonned (a
ihai ycu ->
' Drug uaen also havt dilliculiy recalling bow oftcs ihcy used a dru^ bow mucfa Ibey used, and bow much ibey paid fa
IL However. ihi> inaccuiacy. uolikc uuesuoiul denial ot deocpuoD. probaMv avaagea out wbcs (he dau are agfregaied.
Al
406
combined in the DUF interview), and OTHER. Cocaine, heroin, and
marijuana were the only drugs consumed by a large percentage of
the arrestee population. OTHER comprised a large number of
infrequently consumed substances. Except for MARIJ, each
variable comprised at least two drugs.
The category variiable represents the maximum number of days
any one of those drugs was consumed. For example, if powdered
cocaine had been consumed on 15 days and crack cocaine had been
consumed on 20 days, then COCAINE was coded as "20 days."
We expected the relationship between expenditures and days
of consumption to be nonlinear, but the logarithmic translation
may not have been adequate to capture that nonlinearity .
Consequently, each of the above category variables was raised to
the second power, thus creating additional independent variables:
C0CAINE2, HEROIN2, MARIJ2, and 0THER2.
Cocaine, heroin, and other drugs are frequently consumed in
combination. For example, heroin users often use cocaine, a
stimulant, to moderate the effect of heroin, a depressant.
However, someone who uses a combination of heroin and cocaine on
a daily basis is unlikely to consume the same amount of heroin
and cocaine that is consumed by two people who are daily users
and exclusive In their drug use.
Consequently, two interaction terms were added to the
regression. COKEHER equals COCAINE x HEROIN. C0KEHER2 -
COKEHERV300. The division by 300 facilitates the computing
algorithm, but otherwise has no substantive importance for the
analysis. The consumption of other drugs wSs relatively
infrequent, so we did not add an interaction term to the
regression for this variable.
During 1990, DUF respondents were asked whether they had
consumed any drugs in addition to those listed in the interview.
A variable OTHERDRG denotes that some other drug had been
consumed (1-yes, 0-no). This question was not asked during 1989.
A2
407
Estimation
We used ordinary least squares to estimate the regressions.
Results are presented In Table Al .
Table Al
Statistical Results for Regression Azvalysls
of Drug Expenditures, 1989-1990
Descriptive
Analysis
1989
1990
Standard
Standard
Variable
Mean
Deviation
Mean
Deviation
EXPEND
4.5
1.6
4.4
1.6
COCAINE
11.0
12.1
9.3
11.5
C0CAINE2
267.7
369.5
218.4
344.5
HEROIN
3.9
9.5
4.0
9.7
HER0IN2
105.8
279.3
109.6
283.6
h4ARIJ
6.5
9.4
5.8
9.0
MARIJ2
131.8
272.4
115.1
257.8
OTHER
2.4
6.9
2.5
7.0
0THER2
52.5
192.7
55.8
196.8
COKEHER
63.2
209.8
55.1
195.2
C0KEHER2
160.0
605.6
137.1
556.6
OTHERDRG
Regression
0.8
Results
1989
1990
Variable
Parameter
T- Score
Parameter T-Score
CONSTANT
3.1325
87.64
3.1438
90.41
COCAINE
0.1516
22.58
0.1503
30.27
C0CAINE2
-0.0023
-10.42
-0.0022
-13.36
HEROIN
0.1462
10.23
0.145
14.38
HER0IN2
-0.0021
-4.36
-0.0021
-6.27
MARIJ
-0.0207
-3.14
-0.0035
-6.96
MARIJ2
0.0012
5.17
0.0017
9.87
OTHER
0.0565
5.25
0.0788
10.62
OTHER2
-0.001
-2.8
-0.0018
-6.78
COKEHER
-0.0061
-10.07
-0.0053
-12.33
C0KEHER2
0.0013
6.77
0.001
7.42
OTHERDRG
0.0483
1.61
R-Square
0.46
0.48
Number of
Cases
5,151
9,364
Source: DUF,
1989 and 1990
A3
408
The model's explanatory power (R^-0.46 for 1989 and 0.48 for
1990) appears remarkable given the presumed measurement error in
these data. Residuals were plotted against the number of days
that the respondent reported using cocaine, heroin, marijuana,
other drugs, and the interaction term. These plots indicate that
the logarithmic transformation does a sufficient job of inducing
normality among the residuals and that the model specification
does not systematically distort the relationship between days of
use and amount of money spent.
Interpretation
We converted predictions based on the regression reported in
Table Al from logarithms to natural units using two approaches.
When Ln(S) is the predicted value of the original regression,
then the median value in the original units is Median($) -
Exponential ( Ln($) ), and the mean value in the original units is
Mean(S) - Exponential ( LN( S ) +0^/2 ) .
When cocaine is the only drug consumed, estimating expendi-
tures on cocaine is straightforward. First, substitute zeros for
all independent variables other than COCAINE and C0CAINE2.
Second, use the regression results to make predictions when
COCAINE - 1, COCAINE - 2, COCAINE - 30. Similar calculations
yield estimates for expenditures on heroin when heroin is the
only drug consumed. See Figures Al and A2, which also report
expenditures for other drugs consumed alone.
For example. Figure Al Indicates that when cocaine is
consumed 10 days a month, the median weekly expenditure is
somewhat more than $80. It is about $200 a week when cocaine is
consumed on 20 days a month, and it is about $300 a week when
cocaine is consiuned on 30 days a month.
Note from Figure Al that, when broken down by daily
expenditure, spending on heroin and cocaine is about the same.
However, this does not mean that when cocaine and heroin are
consumed in combination, expenditures on each are equally
divided. More likely, one of the drugs is the drug of
preference, and the other is used frequently but at a lower
dosage.
When cocaine and heroin were consumed in combination, we
attributed greater expenditure to what appeared to be the
dominant drug. Let $ represent the predicted dollar expenditure
on drugs by individuals who consume cocaine and heroin but no
other drugs.
A4
409
Let N, represent the number of days a month that an Individual
consumed cocaine, and let N^ represent the number of days a month
that Individual consumed heroin. Expenditures on cocaine and
heroin are estimated as:
$. = $
Nc
N^ * ADJ-Nf,
$,=$-$,
where ADJ - 0.5 when N^ > N^ and ADJ - 2.0 otherwise. According
to this formulation, when cocaine is consumed on more days than
heroin, at least two-thirds of the drug expenditure is attributed
to the purchase of cocaine. When heroin is consumed on more days
(or the same number of days) as cocaine, then at least two- thirds
of the drug expenditure is attributed to heroin. As a practical
matter, this rule dictates that respondents who say thcat they use
both heroin and cocaine daily spend two-thirds of the money on
heroin and one- third on cocaine. This division seems appropriate
given evidence that such individuals typically are long-
established heroin users who add a small amount of cocaine to
their consumption.'
Typical Expenditures
Using results from the above regression, coupled with
assumptions about how joint expenditures on heroin and cocaine
should be apportioned, we estimated the median and mean
expenditures for cocaine and heroin for every respondent who used
either drug heavily. We averaged those estimates over all
respondents who admitted using cocaine or heroin on more than 10
days during the month before their arrest. Results are reported
in Table A2.
Typical expenditures appear to have remained stable during
1989 and 1990. The chief problem in interpreting these numbers
is that the medians are so different from the means. Which
should be used as "typical" expenditures? Evidence presented
later seems to Indicate that the median is preferable. Before
turning to this evidence, the matter of earnings from income in
kind must be considered.
' D Hum. Ab< Assoduej Inc . "Tucking ihc Piev»Jcnc« of Heioio Uie.* pipet prepired foi ONDCP. July 1992.
AS
410
TABLE A2
Mean and Hedlan Expenditures on Cocaine and Heroin, 1989-1990
1989
1990
Expenditures on cocaine by those
who use cocaine heavily
( Mean )
(Median)
$434
$211
$430
$217
Expenditures on heroin by those
who use heroin heavily
( Mean )
(Median)
$513
$250
$519
$261
Source: DUF, 1989 and 1990.
Accounting for Income In Kind
Heavy drug users support their drug use through legitimate
sources and through crime, especially drug dealing. Returns from
dealing are often in the form of drugs as a payment for services
rendered, or "income in kind." How does income in kind affect
our estimates?
It is sometimes asserted that most heayy users pay for their
drug use by dealing or assisting others who deal in illicit
substances. For example, Johnson and colleagues* report that in
their sample of New York City heroin users, dally users spent an
average of $7,601 a year on heroin but consumed about $13,189
worth of heroin a year. Regular users (defined in this study as
those who use heroin between three and six times a week ) spent
$4,019 a year on heroin but consumed about $6,431 worth of the
drug a year. The difference between expenditures and consumption
represents in-kind earnings in the form of drugs. If this were a
typical pattern, then the expenditures on drugs computed from DUF
should be inflated by ( $13, 189/$7, 601 ) or 1.73 for daily users,
and (S6,431/$4,019) or 1,6 for regular users. For reasons
reported here, such multipliers seem much too large.
First, consider a hypothetical illustration of a drug
market. Suppose that mid-level dealers have 100,000 units of
drug X to sell and, at SI a unit, demand $100,000 for their
drugs. Suppose initially that there are 100 heavy users but no
' B Johjuon a »L, Taking Care of Business The Economics of Crime by Heroin Abusers (Lenngtai. Mi»»»chus«U:
Lexington Books, 198S).
A6
411
other users. Then each heavy user must generate SI, 000 of Income
for the mid-level dealers.
The 100 heavy users could support their use partly from
dealing, but clearly they cannot support their use entirely from
dealing. Selling exclusively to each other would not raise the
5100,000 expected by mid-level dealers. The S100,000 must come
from some other source.
Now these 100 users could support one-third of their
consumption by selling 66,666 units for 31.50 a unit to each
other. This is a Pyrrhic marketing success, of course, but it
would be effective if alternative income sources sometimes
provided sufficient funds to make purchases and sometimes did
not. The total revenue generated is S100,000, enough to satisfy
the middlemen. This division of drugs would approximate what was
observed by Johnson and colleagues in New York.
This solution requires that the street price of drug X be 50
percent higher than the price to mid-level dealers. In reality,
retail prices for cocaine and heroin are about one-third higher
than wholesale prices,' so income in kind would seem more likely
to add about 33 percent to our estimates of drug consumption
based on transactions in which money was exchanged.
Moreover, many sellers do not themselves consume drugs (at
least not at heavy use rates). To extend the above illustration,
suppose that one-half the 100,000 units of drug X are sold by
entrepreneurs who do not themselves consume drugs . Then the 100
heavy users could support only about 17 percent of their own drug
use by selling drug X at SI. 5 a unit to other users. The market
opportunities for heavy users to support their own consumption
with income in kind is limited by sales by nonusers. Sales by
nondrug users may be sizable.
Reuter and colleagues report that only 11 percent of the
dealers they interviewed retained one-half or more of the drugs
for personal consumption; 30 percent retained less than one-half,
"usually only 'a little of it.'"* If "a little of it" means 15
percent, then these figures suggest that about 10 percent of the
drugs that were available to these dealers were retained for
personal consumption. If Reuter 's dealers are typical of those
who sell drugs, then the expenditure figures based on dollar
transactions should be increased by 0.1/0.9, or about 11 percent
to account for in-kind income.
' W Rhoda ud R. Hym. AM AHoaoei Idc. Tkt root of niiai Dru(i. 1981-1991.' piper prepvtd far ONDCT. Uty
IS. 1991
* P Rtuta ei «1., Monry from Crumt A Satdf of *t Uonomtci of Dng Dealing in Waihinfton, DC (Saat* Mobjc*.
C»lilomii Rind Corpottuon. 1990) RA>fD publicauoa R-J89«-Rf. p 61.
A7
412
In his field study conducted in 1982 and 1983 of 15 street-
level heroin dealers in Detroit, Mieczkowski' reports that
dealers are typically not heavy users: "...although runners
appear by and large to be recreational drug users, they are not
addicted to heroin." Mieczkowski ' s findings suggest that income
in kind represents a smaller percentage of drug consumption than
was reported by Johnson.
Altschuler and Brounstein* interviewed 387 ninth and tenth
grade, minority, inner-city boys from Washington, D.C. during
1988. Of the 387, 7 percent used drugs but did not sell them, 9
percent sold drugs but did not use them, and 4 percent both sold
and used drugs. These findings suggest that many drug sales are
made by dealers who are not heavy users.
Williams tracked the drug (cocaine and crack) dealing of
eight New York Juveniles who belonged to a teenage drug ring
called the Cocaine Kids, or the Kids. Williams reports that
"...virtually all cocaine suppliers expect retail dealers to
return with cash amounting to about 60 to 75 percent of potential
retail sales of their consignment."' If this profit margin is