COMMENTS 0

Informing America 's Policy on Illegal Drugs

User Sanctions and Supply Control Mark A. R. Kleiman UCLA School of Public Policy and Social Research October 1, 2001 The NRC Committee produced an impressive document, full of important recommendations. This brief commentary does not review those recommendations in detail. Rather it attempts to set out some organizing concepts and specific research topics not covered in the Committee's report. Overview

Drug consumption is a function of price, non-price factors of availability (such as search time), the formal and informal sanctions imposed for drug use, and other economic and social conditions , such as the availability of treatment for substance abuse disorders. It is also a function of habits and attitudes among current and potential users, which themselves depend in part on past drug use; this makes the problem , from a modeling perspective, a dynamic rather than a static one. One approach is to imagine a limited number of possible states of any individual with respect to any drug (e.g., never-user, new user, casual user, heavy user, former experimental user, former casual user, former heavy user) , with a matrix of transition probabilities among those states. Interventions can be thought of as attempts to change one or more transition probabilities (initiation , continuation , intensification, moderation , quit, or relapse ) or to influence the rate of damage to the user and to others resulting from a given period spent in a given state. (Moore 1990). Damage can result from drug consumption , illicit drug dealing , and drug control efforts. The social optimization problem to be solved is to minimize the sum over time of these various harms. Research should serve to elucidate the parameters of this optimization problem . User Sanctions The Multiple Purposes of User Sanctions

Any analysis of data requirements for policy should start with an analysis of goals. Setting to one side various symbolic objectives, three sets of drug control purposes might be served by sanctions (formal and informal) against drug users: (1) Reducing the extent of substance abuse by, or the damage done by substance abuse to, those sanctioned , or threatened with sanctions. Some of this functions as "prevention" (i.e., it addresses those not currently using, or not currently suffering from abuse or dependency disorder) , and some of it functions as "treatment" (i.e., it addresses those who are clinically abusing or dependent). The effect may be direct, as in the case of deterrence, or indirect, as when sanctions are used to force individuals into treatment and the treatment then provides some benefit. Reducing drug use can benefit both users and others, as when reducing the heroin purchases of a heroin-using burglar facilitates a reduction in criminal activity.

(2) Reducing the damage done by drug-abusing or drug-dependent individuals by removing them from social roles (e.g., as parents or employees in safety-sensitive jobs) where impairment due to drug-taking can damage others. (3) Shrinking the illicit drug markets by reducing effective demand , thus possibly reducing drug availability by shrinking the number of retail outlets, reducing violence and disorder incident to the markets, reducing the burden of drug law enforcement, and preventing the recruitment of new participants, especially juveniles, into drug dealing . Simply writing down , for each extant or proposed form of user sanction , which of these goals it is expected to serve , the purported mechanism of action , and the basis, if any, for making a quantitative projection as to its efficacy, would constitute a significant advance. Pareto's Law and Substance Abuse

There is considerable evidence that the amounts of any given illicit drug used by its various users tend to follow a Pareto distribution , with total consumption dominated by a small minority of heavy users. (The same is true for alcohol , though not nicotine.) Heavy, criminallyactive users of drugs that support very expensive addictions (notably cocaine and heroin) account for a very small proportion of the total population of drug users, but contribute the bulk of total illicit revenues . That suggests that programs aimed at shrinking the drug markets need to focus on this group. It also suggests that the turnover rate among these heavy-user populations is a matter of urgent research interest. Much is known about the persistence of drug abuse in some treatment cohorts ; much less is known about the larger number who never seek treatment. There seems no alternative to cohort prospective studies, as expensive and long-latency as they are , in answering these questions. Such studies could also try to answer the more ideologically sensitive questions about the correlates and consequences of drug use that never leads to diagnosable substance abuse disorder. Much discussion of the drug problem tends to assume that the risk of developing diagnosable disorder is more or less constant across potential new users of any given drug , and that therefore the only way to reduce the number of new addicts is to reduce the number of new users. There seems no good reason to make such an assumption , but neither do the data exist to support a more complicated set of assumptions. Current prevention efforts, implicitly reflecting a fixed-ratio theory, aim mostly at primary prevention (i.e ., prevention of initiation) , and are evaluated largely by their capacity to reduce initiation rates. A more nuanced model of drug use careers might support more subtle intervention strategies. Any given user sanction cou ld be examined in terms of its effects on three different rates: the rate of initiation to the use of some drug in some population , the rate of desistance among those in that population who currently use that drug , and the rate of progression to diagnosable disorder among those in that population who continue to use that drug . Tougher policies might, for example , lead to a decrease in initiation and an increase in desistance, but also to lead to an increased risk of addiction among those who continue to use. Testing and Sanctions on Probation and Parole

It has been proposed (Kleiman 2001) that drug testing and sanctions administered by probation and parole agencies could be a highly effective and cost-effective means of reducing drug use and crime. That proposal has been questioned (Brownsberger 2001) . A version of that proposal is now being put into practice in Maryland. Our understanding of whether this idea is sound would benefit from both clinical trials and field research on active programs. The NRC report proposes studying testing-and-sanctions as an adjunct to treatment. An alternative study would randomize a group of probationers into four cells: testing-and-sanctions with mandated treatment, testing-and-sanctions alone, mandated treatment alone, and no

intervention. (Nothing would prevent those assigned to the testing-and-sanctions-alone or the nointervention group from seeking out formal treatment or participation in group self-help; all they would lack is the mandate to attend treatment.) Under a more sophisticated design that would better replicate real-world conditions, those assigned origina lly to testing-and-sanctions alone might have a treatment mandate added after some number of positive test results or missed tests. That sort of clinical trial can help answer the question whether testing and sanctions can reduce drug consumption , improve health outcomes, and reduce reoffending rates among probationers and parolees. Knowing whether doing so would have a significant effect on the overall drug market requires community-level trials, which could also help answer the even more difficult questions about the administrative practicability of such programs. (Kleiman 1997)

The Importance of Heterogeneity

Since drugs and users are highly heterogeneous , generalizations need to be made with caution , and policies need to be focused where they are most appropriate and do the most good. Aspects of the heterogeneity of drug users largely ignored by current policies constitute additional important research topics. In particular, a majority of illicit drug users use only cannabis. So making policy about "drug users" generally while having crack addicts in mind can easily lead to unintended results. Similarly, only a minority of drug users meet clinical criteria for abuse or dependency, and of those in turn only a minority suffer from the "chronic, relapsing" form of the disorder. (Heyman 2001) , So basing po licy on the assumption that the typical drug user suffers from a chronic, re lapsing brain disease will likely lead to mistakes. (That's obvious in the case of legal nonmedical drugs: no one today would propose making policy toward anyone who drinks any alcohol at all based on what would be appropriate for a skid-row alcoholic.) That suggests a research agenda : For each kind of user sanction , what is the distribution of drug use patterns among those subject to that sanction? (E.g ., of those arrested for possession of cocaine, how many meet clinical criteria for cocaine abuse or cocaine dependency? Of those fired or refused employment due to positive drug tests, how many actually engage in drug use patterns that would reduce their productivity or increase their accident rates on the jobs they hold , or for wh ich they are applying?) Spontaneous Desistance

Much is known about the drug use careers of those who seek treatment. Much less is known about the larger number of users whose drug use ceases , or substantially decreases, without any formal intervention. One way to think about user sanctions is as a means of increasing the rate of "spontaneous" desistance or de-escalation of drug use. But any study of such effects would require , as a precondition , a data-supported model of the spontaneous desistance process itself. Understandably, this is of little interest to agencies who see their business as promoting drug treatment, and whose prevention efforts rely on the myth of addiction as both a likely result of any illicit drug use and a chronic, relapsing condition . But someone has to fund it, or our ignorance wil l remain. Workplace Drug Testing

As the report notes, the widely-believed proposition that employee drug testing helps companies by weeding out unproductive and unsafe employees rests on a very slender evidentiary base. While workplace drug testing can serve goals wider than the enterprise, the questions about whether such testing helps enterprises serve their own goals remain important. Since the test results are reported on a yes-or-no basis are quantitative at their base, a simple

study suggests itself, one that could be run on either the military data or on the data from any large company. Identify employees whose drug tests came in just below whatever arbitrary cutoff is called a "positive test. " Compare their measured workplace performance and disciplinary records to demographically similar employees whose drug tests were at zero. (The study should be disaggregated by drug class; there is no reason a priori to assume that the results for, e.g. , cocaine are similar to those for cannabis.) Since these results were reported as negative, neither the supervisor nor the employee will be aware of the actual result, so that knowledge cannot contaminate the findings . Measurable differences between "true zeros" and "borderlines" would suggest that use of the drug in question is either causally linked to workplace problems or is a marker for other characteristics so linked . The absence of measurable differences would cast substantial doubt on the beliefs underlying current practice . Making Sense of the Arrest Statistics

Arrest is an important component of the total user-sanctions picture , but remarkably little is known about it other than the bare numbers. "Arrest" is not a transparent category, since it can include both "custodial arrests" (reflecting the common meaning of the term) and "citations" or "desk appearance tickets" (categorized as "arrests" in the Uniform Crime report). Fishbein (2001) has found that, at least in New York and California , about half of all marijuana "arrests" are noncustodial. The nationwide pattern remains unknown , as does the extent (if any) to which the soaring marijuana arrest figures of the 1990s are attributable to non-custodial events. Nor is much known about the consequences of simple cannabis possession arrests, which constitute more than half of all drug possession arrests. Thomas (1999) estimates that at anyone time more than 15,000 prison or jail inmates are behind bars for simple possession of marijuana , but it would be of interest to draw a representative sample of cannabis arrests nationally and follow them out to their conclusions. The Thomas study raises an additional point: the connections between charge at arrest, charge at conviction , and actual conduct are poorly understood. Some substantial fraction of those serving time for simple possession were in fact dealing , with the charge either "broken down" in the course of plea negotiation or used as a substitute because prosecutors could not prove all of the elements of drug dealing. A solid analysis of the effects of "user sanctions" would require clarification of this point, lest user sanctions and dealer sanctions be confused . Again , a study starting with arrest events would be valuable , but it would have to include interviews with prosecutors, police , and arrestees . Supply Control

Demand Responsiveness to Price

As Boyum (1992) pOints out, the change in drug consumption as a function of drug price is not best thought of as a demand elasticity in the classical sense, which holds income constant. A key question is whether, and how much, drug users can and do vary their incomes - from licit work, from illicit activity including theft, prostitution , and drug dealing , and from gifts from family members - as the cost of drug-taking varies. But this is not the only reason that point estimates of aggregate elasticities are unlikely to be helpful. We need to understand how changes in price influence the dynamics of drug consumption , which means estimating the change in the probabilities of various transitions resulting from changes in price . Unfortunately, the heroin market seems to be providing a natural experiment. Heroin prices in some areas of California are reportedly at astonishingly low levels: $30-60/gm . at greater than 50% purity, suggesting that a new user, without an established tolerance, could have a full heroin experience for about the price of a candy bar. As always, there are anecdotes about the spread of use in "new" populations , and as always the established data sets do not allow us to conclusively confirm or disconfirm the anecdotes. This research opportunity should not be

passed up. Cohort prospective studies , employing ethnographic as well as statistical approaches, ought to be mounted in areas where heroin prices have fallen sharply. That won 't tell us anything about the effects of a hypothetical cocaine price increase on cocaine desistance, but it will te ll us something important about a drug likely to be a major problem . Price Responsiveness to Enforcement

As the report points out, prices have moved perversely over the past twenty years , going down as enforcement effort, even measured as a ratio to the size of the market, has gone up sharply. But the report's call for data collection in support of systems modeling seems premature. We lack anything like an adequate theory of price and quantity determination in the illicit drug markets. Enforcement agents' impressions of what goes on do not correspond closely to models of economic rationality. This does not mean that enforcement personnel are right and economic theory inapplicable, but it does suggest caution before diving in with the comparativestatics assumptions of Reuter and Kleiman (1986). [For an exposition of some of the theoretical puzzles, see Kleiman 1989, Chapters 4 and 7, Kleiman 1992, Chapter 5, and Boyum 1992.) The report is correct that "Economic theory is unequivocal about the direction of the effect of drug law enforcement on total supply." But being unequivocal is not the same as being correct. The wages of retail crack dealers appear to have plunged over the past fifteen years , from the $30/hour plus drugs estimated by Reuter et al (1990) to below the minimum wage. This helps explain the fall of cocaine prices despite enforcement, and in turn seems to be explained , at least in part, by the development of a large group of prison releasees , previously convicted of felony crack dealing and therefore without licit-market alternatives to drug dealing . It is more than conceivab le that greater cocaine enforcement in 1990 led to lower cocaine prices in 1995 and 2000 . This would not have been the case had the dealers in question been fully informed, fully rational , and fully self-controlled . But the very rationality assumption that makes the predictions of comparative-statics analysis unequivocal are at least questionable as applied to the drug markets. That drug consumers are imperfectly rational no one familiar with the phenomena would doubt. (Kleiman 1999. But see Becker, Murphy, and Grossman 1994). But imperfect rationality also seems to be an important characteristic of many drug sellers (many of whom are, after all, users as well.) Thus any sound economic analysis of the drug markets will have to incorporate concepts of imperfect rationality drawn from behavioral economics and the psychology of heuristics and biases: hyperbolic discounting , loss aversion , self-confidence bias, the availability heuristic. This is not good news for the economics of illicit markets. Just as offenders would be easier to control operationally were they perfectly rational , they would be easier to model scientifically. At a minimum , imperfect rationality introduces more free parameters into a field already suffering from a surplus of degrees of freedom and a deficit of data points. But this is not a case in which the econom ist's natural preference for elegance can safely be allowed to outweigh the demands of descriptive accuracy, no matter how conceptually and computationally ugly the description may have to be. The problem first raised by Caulkins (1990) , about the extent to which drug price changes work through the distribution chain multiplicatively as opposed to additively, remains a central one. (The multiplicative hypotheSiS, which seemed hard to reconcile with cost minimization by rational actors, may be much easier to make sense of in the context of imperfect rationality.) That problem also remains largely unexplored since Caulkins's original paper. We should not assume that the right model has the same parameter values, or even the same structure, from drug to drug or from time to time .

A related question deals with the shape of the aggregate supply curves for the various drugs. The diagrams in the introductory textbooks show supply curves sloping upward, since expanded production requires tapping into ever less productive resources , as in the classical case of wheat production and soil productivity. But the largest single cost in the illicit markets is the cost of suffering , or trying to avoid, punishment. If enforcement capacity directed at a given drug grows less slowly in the short run than does the volume of that drug sold , then it follows that the enforcement attention per unit of drugs tends to fall as volumes rise. (Kleiman 1993) This suggests an industry-wide economy of scale. Another potential contributor to such an effect would be increased efficiency in retail trade; the greater the density of buyers, the smaller the expected waiting time between customers for a dealer. Whether such effects are sufficient to make some drug supply curves slope downward is an open question , and one that deserves empirical attention . Much more also needs to be known about what determ ines geog raph ic patterns of drug distribution , at scales ranging from city blocks to regions of the country. Why is it a stable fact that heroin is available in East New York but not in central Maine? (After the OxyContin experience, it is no longer a supportable hypothesis that there was no demand for high-purity opiates in rural areas .) What was the process, at the nuts and bolts level, by which cocaine availability spread nationally, and why is methamphetamine still available only in some places? In the face of these basic questions, launching an expensive new data-collection effort deSigned to produce a time series of drug prices in various geographic markets seems premature. Data collection needs to be theory-driven . There is a strong case for carrying through in a few areas, as a proof of concept, an effort to greatly improve retail-level price data. In this connection , it may be important that novice users are likely to pay higher prices than do established heavy users, and that estimating, e.g., the newuser price of heroin in a given market may require a different approach than estimating the established-user price. Recruiting and managing shoppers who look like novices may be easier than recruiting and managing shoppers who look like drug addicts. On the other hand, the human-subjects protection issues might be greater. If we could assume that the two prices moved up and down together in some predictable pattern , measuring one would be tantamount to measuring the other; however, that assumption cou ld easily prove false . Given a finite data-collection budget, there is a strong case for putting a substantial share of it into one-time efforts that look more like research than they do like routine data collection in support of an ongoing set of time series. An example would be intensive interviewing of convicted drug dealers now in prison, with questions designed to elicit information about how prices and volumes are determined. Measuring Drug A vailabiJity

Now that survey techniques have been shown to be capable of measuring search time (Rocheleau et al. 1993), there is a strong case for beginning to gather city-by-city search time data for some of the major drugs. The accumulation of those data will allow much more sensitive modeling of price effects. In addition , we might begin to be able to model both the search-time elasticities of demand and the impact on availability of changes in enforcement intensity and techniques. The Multiple Purposes of Drug Law Enforcement

Enforcement efforts aimed at drug dealers are often thought of as "supply control ," but enforcement can also decrease (or increase) the damage created by the conduct of the illicit markets. Learning how to do drug law enforcement to shape the illicit markets into less destructive forms may offer social gains as great, or greater, than learning how to use enforcement to raise price or shrink availability. Purely observational stUdies are not the only, and

may not be the best, way to understand the capacity of enforcement to change behavior by market participants; this is a good target for "intervention research ," where research workers participate in designing and implementing interventions.

Directly Communicated Threats

General deterrence relies on the capacity of potential offenders to estimate enforcement risks. The dynamics of drug enforcement, where the risk faced by anyone dealer is a decreasing function of the number of dealers in the same area (Kleiman 1993) means that slow learning on the part of offenders can make focused crackdowns extremely expensive from an enforcement viewpoint. (Press 1987) David Kennedy has proposed the use of explicit and precise threats to shift dealer behavior without making large numbers of arrests; police would announce in advance that a given market was to be shut down as of some future date. That proposal is an excellent candidate for intervention research . Cross-Cutting Issues: The Epidemic Cycle

Caulkins and collaborators, in a series of papers (Caulkins , 2001 , Reuter 2001 , Baveja, Caulkins, et al. 1997, Tragler, Caulkins and Feichtinger 2001 , Behrens, Caulkins, Tragler, et a/. 1999 and Behrens Caulkins, Tragler and Feichtinger 2000), have shown how the effects of various kinds of drug control efforts are likely to vary across the epidemic cycle . (A different set of questions are raised by endemic drugs such as cannabis, and by the possibility of transition from epidemic to endemic patterns.) Such modeling can quickly outstrip the available data, but that does not make the effects it delineates any less important. Understanding the epidemic cycle , and how the drug control effort should be optimized across different drugs at different points in their cycles, is an essential research frontier, not less important for being difficult. Cross-Drug Effects

Drugs, like other goods, may be substitutes or complements for one another in a static context: a heroin user who fortuitously acquires some cocaine is likely to use more heroin as a resu lt, while one who comes on a supply of illicit methadone or OxyContin is likely to use less heroin as a result. Again , these effects are embedded in a dynamic process, and it is possible for two drugs to be substitutes contemporaneously but complements over time: the use of heroin today tends to increase the demand for any other opiate tomorrow, especially if heroin becomes hard to get due to a price increase or diminished earnings capacity. These cross-drug effects have crucial implications for policy. If cannabis and alcohol are short-term and long-term substitutes, and if the substitution is anything like one-for-one in hours of intoxication, then attempts to reduce cannabis use look far less attractive than they seem at first blush . If, on the other hand , they are instead complements, the case for discouraging juvenile cannabis use (and juvenile alcohol use, for that matter) is becomes much stronger. There are reports of juvenile users switching from cannabis to heroin as the price of heroin falls while the price of cannabis stays high : an interesting alternative to the famous "gateway." This issue applies to user sanctions as well as price: one effect of urine testing may be to shift demand from cannabis, with its long detection window, to drugs with shorter detection widows such as heroin or to drugs not tested for such as LSD . (It has been asserted that the rise of heroin use in the US Army in Vietnam followed immediately on the institution of urine testing for this reason.)

The current state of the question is about as unsatisfactory as it could be: two able researchers , using somewhat different methods and data sets, reached opposite conclusions about the cross-relationships between cannabis and alcohol (Model 1991 , Pacula 1998). That's hardly surprising , given the difficulty of the problem . But the result has been , instead of a burst of well-funded activity in the area, stagnation , due to the ind ifference of the federal research funding agencies to the question . Th is needs to be changed . It is so obvious as hardly to need saying that price and ava ilability are central to the etiology of substance abuse disorder. There is no excuse , then , for the National Institute on Drug Abuse to ignore econom ic phenomena and economic modeling . Arguably, the drug markets ought to be the province of the National Institute of Justice, but its persistent and seemingly irreversible poverty (the NIJ budget for investigator-in itiated proposals is approximately one percent of the NIDA budget) means that assigning this problem to NIJ (without assigning NIJ a chunk of the NIDA budget) means assigning it to the circular file .

References

Becker, Gary S.; Grossman , Michael ; Murphy, Kevin M. "An Empirical Analysis of Cigarette Addiction" American Economic Review 8,4 nO.3 (June 1994): 396- 418. Boyum , David . 1992. "Reflections on Economic Theory and Drug Enforcement. " dissertation, Harvard University.

Ph .D

Baveja , Alok, Jonathan P. Caulkins, Wensheng Liu , Rajan Batta , and Mark H. Karwan . 1997. 'When Haste Makes Sense: Cracking Down on Street Markets for Illicit Drugs." Socio-Economic Planning Sciences, Vol. 31 , No. 4, pp.293-306. Behrens, Doris A. , Jonathan P. Caulkins, Gernot Tragler, Josef Haunschmied , and Gustav Feichtinger. 1999. "A Dynamic Model of Drug Initiation: Implications for Treatment and Drug Control." Mathematical Biosciences. Vol. 159, pp.1-20. Behrens, Doris A. , Jonathan P. Caulkins, Gernot Tragler, and Gustav Feichtinger. 2000. "Optimal Control of Drug Epidemics: Prevent and Treat - But Not at the Same Time." Management Science , Vol. 46 , No.3, pp.333-347 . Brownsberger, W. N. 2001 . "Lim its on the Role of Testing and Sanctions," in Philip B. Heymann and William N. Brownsberger, eds., Drug Addiction and Drug Policy, Harvard Un iversity Press. Cau lkins, Jonathan P. 2001 "The Dynamic Character of Drug Problems." UN Bulletin on Narcotics. Forthcoming 2001 . Caulkins, Jonathan P. 1990. "The Distribution and Consumption of Illicit Drugs: Some Mathematical Models and Their Policy Implications." MIT Operations Research Center. Fishbein , P. 2001 "Disaggregating Cannabis Arrest Statistics ," unpublished manuscript, UCLA School of Public Policy and Social Research . Heyman, G.M. 2001 . "Is Addiction a Chronic, Relapsing Disease?" in Philip B. Heymann and William N. Brownsberger, eds., Drug Addiction and Drug Policy, Harvard Un iversity Press. Kleiman , MAR. 2001. "Reducing Drug Use and Crime with Sanctions , Testing , and Treatment," in Philip B. Heymann and William N. Brownsberger, eds., Drug Addiction and Drug Policy, Harvard University Press.

Kleiman , MAR. 1999 "Addiction , Rationality, Behavior and Measures: Some Comments on the Problems of Integrating Econometric and Behavioral Economic Research ," in Frank J. Chaloupka, Warren K. Bickel, Michael Grossman , and Henry Saffer, eds., The Economic Analysis of Substance Use and Abuse: An Integration of Econometric and Behavioral Economic Research , The University of Chicago Press, Chicago, National Bureau of Economic Research Kleiman , MAR. 1997 "Coerced Abstinence: A Neo-Paternalistic Drug Policy Initiative," in Lawrence A. Mead, ed. , The New Paternalism, Brookings Institution Press Kleiman , MAR. 1992 Against Excess: Drug Policy for Results; New York: Basic Books Kleiman , MAR. 1989. Marijuana : Costs of Abuse, Costs of Control; Greenwich , Conn .: Greenwood Press Model, Karyn 1991 . "The Effect of Marijuana Decriminalization on Hospital Emergency Room Drug Episodes: 1975-1978" Department of Economics, Harvard University Cambridge, MA Moore, Mark H. 1990. "An Analytic View of Drug Control Policies," Working Paper, Program in Criminal Justice Policy and Management, Harvard University. Pacula, Rosalie L. 1998."Does Increasing the Beer Tax Reduce Marijuana Consumption?" Journal of Health Economics 17(5):557-586 . Press, Aric 1987 "Piecing Together the System : The Response to Crack" (New York: New York Bar Association) Reuter, Peter, Robert MacCoun , and Patrick Murphy 1990. Money from Crime . The RAND Corporation . Reuter, Peter and MAR. Kleiman "Risks and Prices : An Economic Analysis of Drug Enforcement," (with Peter Reuter), in Michael Tonry and Norval Morris (eds.), Crime and Justice: An Annual Review of Research , Vol. 7, 1986. Rocheleau , Ann Marie et al. 1993 "Measuring Heroin Availability: A Demonstration ," Washington , D.C .: Office of National Drug Control Policy. Thomas, C. 1999 "Marijuana Arrests and Incarceration in the United States" Drug Policy Analysis Bulletin #7. Tragler, Gernot, Jonathan P. Caulkins, and Gustav Feichtinger. (2001) "Optimal Dynamic Allocation of Treatment and Enforcement in Illicit Drug Control." Forthcoming in Operations Research, Vol. 49 NO. 3.