RISKS AND PRICES: THE ROLE OF USER SANCTIONS IN MARIJUANA MARKETS

I. Introduction

About 25 million Americans reported use of marijuana in the past year in 2007, despite nearly 900,000 marijuana arrests the same year (Substance Abuse and Mental Health Association, 2008; Bureau of Justice Statistics, 2009). Since 2000, there have been over 7 million arrests for marijuana possession and/or sale, and still use among the general household population remains relatively stable (SAMHSA, 2008; BJS, 2009). Given its popularity, the widespread belief that its harms are low relative to other illicit substances, and the costs related to enforcing prohibition and incarcerating drug offenders, a significant debate has continued regarding marijuana policy reform in the United States and abroad. Recent state reforms enabling the use of marijuana for medicinal purposes have brought to the forefront issues related to how patients can safely access the illegal substance, reinvigorating debates in California, Nevada, and New Mexico about broader legalization. However, to date, most states have opted for smaller steps that simply eliminate the criminal status of possession offenses involving small amounts of marijuana, making these civil crimes rather than criminal offenses. The momentum of the state reforms has encouraged debate at the Federal level, as indicated by Rep Barney Frank (D- Massachusetts) introduction of HR5843 in April 2008, which proposes the elimination of federal penalties for possession of marijuana for personal use.

An obvious question to ask is whether these small changes in legal sanctions toward users impact the market for marijuana and, if so, how. Deterrence theory would suggest that a reduction in the criminal status of possession offences should increase the demand for marijuana if people are sensitive to these legal risks. The empirical evidence supports the idea that people understand and are responsive to the risks and rewards tied to other criminal activity, even crimes committed under the influence of a substance (see e.g., Viscusi, 1986; Kenkel, 1993; Eisenberg, 2003). However, the literature examining the extent to which state changes in the legal risks associated with using marijuana, particularly those that focus on reductions in jail time and lower fines, show relatively small effect sizes or no effect at all (Thies and Register, 1993; Saffer and Chaloupka, 1999; Farrelly et al., 2001; DiNardo and Lemieux, 2001; Pacula et al., 2003; Williams et al., 2004). Several hypotheses could be offered for the conflicting results, but two interpretations are most frequently offered. First, the legal penalties for possession of marijuana may truly have little impact on demand, especially if enforcement of the policy is inconsistent or negligible. Second, people may not be aware of the sanctions or if they have changed, again suggesting that the policies will not affect demand (Pacula et al., 2005; MacCoun et al., 2009). One potential explanation that has gotten virtually no attention this far relates to the supply-side of the market. If the supply of marijuana is upward sloping as with other normal goods there could be an offsetting price increase generated by an increase in demand that would reduce the net effect of a change in these legal sanctions on use. Because data on the local price of marijuana are not widely available, most analyses examining the effect of user sanctions do not hold price constant when evaluating a change in these policies.

This paper examines the extent to which changes in user sanctions, believed to shift demand, influence the prices paid by users in local markets. Economic theory suggests that price will be affected by demand side policies if supply is upward sloping, a perfectly reasonable assumption for marijuana. To date virtually no work has been done examining the slope of the supply curve for marijuana, presumably due to lack of reliable data given the illegal nature of the good. However, it is possible to infer basic information on the slope of the supply curve by understanding how shifts in the demand for marijuana influence the equilibrium level of prices observed in the market.

In this paper, we use rich transaction-level information from marijuana purchases made by arrestees who are part of the Arrestee Drug Abuse Monitoring (ADAM) Program to estimate the impact of state marijuana policies on self-reported price per bulk gram paid. We first develop a model focused exclusively on proximal measures of transaction risk, and then use natural variation in state marijuana laws over time and within states to evaluate the impact of these demand-side policies on the equilibrium prices of marijuana observed in illegal markets. Because it is possible that demand-side policies influence sellers’ risk as well, we test the robustness of our findings with respect to these state-policies by considering the impact of other demand shifters on price, in particular alcohol consumption. The sensitivity analyses support the interpretation that reductions in user risk raise prices. There are three implications of these findings. First, the findings support the conclusion that marijuana use is sensitive to changes in the legal risks targeting users. A reduction in sanctions will increase use of marijuana. Second, the supply of marijuana is upward sloping at least in the short run, implying that temporary shortages caused by a reduction in supply or an increase in demand will impact market price. Third, shortages caused by shifts in demand (such as that generated by a reduction in user sanctions) will raise sellers’ profits, ceteris paribus.

The rest of the paper is organized as follows. In the first section we provide some background literature about state marijuana laws, the operation of marijuana markets, and marijuana prices. In Section 2 we discuss how policies targeting users of marijuana might influence marijuana prices under alternative assumptions about supply. In Section 3 we discuss the data used for this analysis and present our empirical model in Section 4. Section 5 presents the results from these models and in Section 6 we offer some discussion and conclude.

2. Marijuana Laws, Markets and Price

2.1 State Marijuana Laws

Even though the U.S. federal government retains a strict prohibition on the possession of even small amounts of marijuana, there is significant variation across the states in the legal penalties associated with this offense (Pacula et al., 2003; Chriqui et al., 2002). Given that the majority of marijuana possession offenses are tried in state courts, state law applies (Ostrom and Kauder, 1999). The proportion of these arrests leading to conviction is unknown, but those convicted of marijuana possession are often sentenced to probation and/or a fine, not jail (MacCoun & Reuter, 2001; National Research Council, 2001; Office of National Drug Control Policy (ONDCP), 2005).¹ Although twelve states are generally recognized as having “decriminalized”² possession of small amounts of marijuana, fifteen states have actually eliminated criminal sanctions associated with it and another twenty-six states have conditional discharge provisions for first time offenders (Pacula et al., 2003).

Unlike early studies examining the effect of state marijuana decriminalization policies (Single, 1989; Thies and Register, 1993; DiNardo and Lemieux, 2001), recent studies examining consumption of marijuana by youths and young adults controlling for enforcement risk and/or price find that use is sensitive to changes in the legal penalties associated with possession of marijuana (e.g., Farrelly et al., 2001; Pacula et al, 2003; Williams, 2004). Presumably, this effect is due to the fact that lower penalties reduce the legal risk if caught in possession of marijuana. However, implicit in this interpretation is the assumption that people are actually aware of these reduced sanctions and are responding to them, which recent evidence suggests is not the case among the general household population (Pacula et al., 2005; MacCoun et al., 2009). In light of the lack of an alternative interpretation for the negative correlation between severity of penalties and use may be warranted. For example, it may be the case that states that have adopted relatively lower penalties all share some unobserved characteristics, such as a strong belief in personal liberties or normative values, which are more favorable toward using marijuana. Alternatively, it may be the case that these policies targeting users also inadvertently influence the risk to sellers so that lower penalties facing users translates into a reduced risk to sellers and thus lower marijuana prices. Given that our information on marijuana prices is relatively poor, many studies have not controlled for marijuana prices when evaluating the effect of these policies on demand and thus the coefficient on the policy variable reflects the net effect (direct effect and potentially off-setting indirect effect).

Medical marijuana provisions represent another area of state policy in which there has been substantial change in recent years. As is the case with sanctions targeting users, state medical marijuana laws have the potential of influencing both the supply and the demand for marijuana within the state. Although it is believed that the medical marijuana market is relatively small (General Accounting Office, 2002; Marijuana Policy Project, 2006, Appendix F) and diversion from a relatively small medicinal market to a very large recreational market would normally not be expected to have any real impact on black market supply, current provisions for medicinal marijuana are extremely vague in terms of how patients are allowed to obtain marijuana. Nine state statutes allow for home cultivation, while several others remain silent or allow patients to obtain marijuana through “any means necessary” (Pacula et al., 2002). Large-scale diversion can easily occur in environments such as these where the government or some other oversight agency does not closely monitor the production and distribution of marijuana for medicinal purposes. More importantly, such ambiguity regarding source of supply creates legitimacy for illegal suppliers, who become the only source of marijuana for individuals unable or unwilling to grow their own. This legitimacy may translate into a reduction in the risk of supplying marijuana to the black market and a reduction in the monetary price.

Medicinal marijuana laws could also influence the demand for recreational marijuana through one of three alternative mechanisms: (1) a change in the legal risk to recreational users of marijuana, (2) a change in the perceived harm associated with using marijuana, and (3) an increase in the social availability. States with medical necessity defenses allow individuals who have received permission from a physician to use marijuana for medicinal purposes without risk of prosecution. Thus, these policies directly influence the legal risk of users who obtain medicinal allowances. It is unclear how big a population this is relative to recreational users. Second, attitudes and perceptions of marijuana as a harmful drug, which numerous studies have identified as an important correlate with drug use (Bachman et al., 1998; Pacula et al., 2001), are likely to be diminished by policies recognizing its medicinal value because these allowances formally recognize that marijuana can have a beneficial impact on one’s health. Finally, medicinal allowances might also influence the social availability of marijuana by increasing the number of social encounters in which youths are around individuals who have used or are using marijuana. Social availability can be further increased if teenagers live in households or near homes where patients and/or their caregivers are growing marijuana, as it is possible for them to share or steal the drug.

2.2 Marijuana Markets

Unlike the markets for cocaine and heroin, marijuana markets are relatively understudied. Although ethnographic information provides insights about how particular markets operate (ONDCP, 2004), very little work has been done modeling marijuana markets in general (Wilkins and Sweetsur, 2006) and no work has been done examining the supply curve in the United States. This is perhaps due to the fact that the primary source of information used to evaluate other drug markets, the Drug Enforcement Agency’s System To Retrieve Information from Drug Evidence (STRIDE), is severely limited in two important ways when it comes to marijuana markets. First, unlike observations for cocaine, heroin and other drugs included in the STRIDE database, marijuana is not chemically assayed when purchased or seized by agents. Thus, the STRIDE database does not contain any information on marijuana potency or quality. Second, marijuana purchase observations in STRIDE are highly concentrated in a single city, Washington, DC. These two factors make the STRIDE data relatively less useful for examining the dynamics of these markets.

Data sources providing systematic information on markets geographically dispersed across the nation have begun to emerge, although none contain information on potency and all of the information is obtained from a buyer’s perspective. The National Survey on Drug Use and Health (NSDUH) has been collecting information from marijuana users on their experiences in marijuana markets since 2001 and the Arrestee Drug Abuse Monitoring (ADAM) Program has been asking arrestees about their last purchase since 2000. Both surveys ask respondents about their drug use behavior, and among recent marijuana users, their methods and circumstances for procuring the drug. In an analysis of the 2001 NHSDA data, Caulkins and Pacula (2006) show that most respondents who report use of marijuana in the past year obtain marijuana indoors (87 percent), from a friend or relative (89 percent), and for free (58 percent). Among those who do buy their marijuana, these purchases generally involve small quantities (less than 10 grams), are acquired from a friend (79 percent), and are exchanged indoors (62 percent). Similar transaction characteristics are reported by youth (SAMHSA, 2003) and arrestees (Taylor et al., 2001) and are consistent with evidence obtained from ethnographic studies (ONDCP, 2004). Thus it seems clear that marijuana markets function differently than that of cocaine and heroin markets (Caulkins et al., 1999; Caulkins and Reuter, 1998; Caulkins, 1997). Most notably, open air markets and purchases from strangers are generally rare as compared to other drugs. However, Caulkins and Pacula identify at least two similarities between marijuana markets and other illicit drug markets. First, they find evidence of quantity discounts, so that people who buy in larger quantities pay less per gram than those who buy in smaller quantities. Second, they find that a minority of users account for the vast majority of purchases.

2.3 Marijuana Prices

There is substantial ethnographic evidence of large geographic differences in the black market price of marijuana (ONDCP, 2002; 2004). For example, information reported in a summary of ethnographic studies conducted by ONDCP shows that an ounce of commercial-grade marijuana in Phoenix, Arizona and San Diego, California sold for between $60–$100, while an ounce of commercial-grade marijuana in New York City and Philadelphia, Pennsylvania sold for between $100–$200 in Fall 2002 (ONDCP, 2004). A number of factors could generate this sort of variation including differences in the cost of production (particularly labor, given that this is a labor-intensive commodity), regional differences in quality, transportation costs, imperfect information, search costs, and differential risks to buyers and sellers participating in these geographic market.

Only two studies empirically examine the role of specific factors on the prices paid for marijuana. In an early study, Caulkins (1995) examines data from nine states reported in the Middle Atlantic-Great Lakes Organized Crime Law Enforcement Network (MAGLOCLEN) and tests two conjectures about how prices vary from location to location within the United States. The first conjecture is that prices increase as one moves away from the source, which in the case of marijuana is largely domestic. The second conjecture is that prices are lower in larger markets, as proxied by population. He examines these hypotheses by evaluating prices for cocaine, LSD, and marijuana. In the case of marijuana, Caulkins only finds support for the first hypothesis. The mean price per bulk gram paid for marijuana got statistically higher as the cities got more distant from the mid-west/Appalachian growing region. These results, however, are based on a very limited number of price observations (fewer than 60 in total) that were not representative of all transactions within these states. Thus, although they are suggestive that transportation costs might explain some of the geographic variation in marijuana prices, the evidence cannot be viewed as definitive.

Caulkins and Pacula (2006) evaluate how differences in transaction-level characteristics influencing individual buyers’ and sellers’ risk correlate with the self-reported price paid per bulk gram of marijuana using data from the 2001 NSDUH. Transaction-specific characteristics evaluated in the models include from whom they purchased (a friend, relative or stranger), where they purchased (inside a private dwelling, inside a public building, outside, or some other location), and proximity to home (near where they live). They find that, contrary to intuition, the quantity-adjusted self-reported price per bulk gram is positively associated with purchases made in private dwellings, where risk of detection is presumed to be lower. Purchases made outside or inside public buildings were associated with lower prices on average. They also find no association between whom they purchased from and price paid. Given that the public use data contain no geographic information and no information on the potency of marijuana traded, their analyses could not account for location-specific factors or differences in potency. Thus, the authors interpret their counter-intuitive findings as potentially reflecting omitted variable biases.

Although it is possible that imperfect information and variation in search costs, transportation costs, production costs and quality generate some of the geographic variability in price, it seems implausible that variation in any of these factors could generate the substantial differences in marijuana street prices that are currently observed across cities.

3. Theoretical Framework

3.1 Supply and Demand Analysis

The impact of changing user sanctions on price depends on how it influences user and seller risk as well as one’s assumptions about supply. Assumptions about supply are necessary as no empirical analyses examining the elasticity of supply have been conducted, so theories develop based on alternative plausible assumptions (e.g. Becker, Murphy, and Grossman, 2006).

If the supply of marijuana is perfectly competitive or one presumes a relatively short run adjustment period, then it is appropriate to think of the supply function as infinitely elastic. If state depenalization or medicalization policies truly only affect the user’s risk of using marijuana and not the enforcement or risks to sellers, then adoption of these policies should not influence the monetary price of marijuana.

If we assume supply is upward sloping, there are two mechanisms by which lowering the user sanctions could increase the retail price. First, increase use due to lower penalties could shift the demand curve to the right and subsequently increase the price at least in the short run. Second, if relaxing user sanctions causes police officers to spend more time targeting marijuana sellers, then the subsequent increase in seller risk could cause the supply curve to shift in, resulting in higher prices. However, there is also reason to believe that relaxing user sanctions will actually lower the legal risk to sellers. This could happen if, for example, marijuana depenalization policies reduce the likelihood that police pursue low-end marijuana transactions or if sellers are able to pass themselves off as users rather than sellers (by carrying less with them and selling smaller consumption bundles). There is some evidence suggesting that retail dealers are aware of the differential penalties and willing to use them to reduce their own legal consequences. Sevigny and Caulkins (2004) show that 20 percent of state inmates in prison on charges for drug possession or possession with intent to sale actually self-report that they were selling drugs.³ Thus, the net impact of lower sanctions on equilibrium prices is theoretically ambiguous.

If we assume supply is downward sloping,⁴ lower user sanctions can only lead to higher retail prices if it leads to a net increase in the legal risk to sellers. Therefore, if we hold the legal risk to sellers constant we should anticipate that a reduction in penalties should lead to a lower equilibrium price.

Findings of the impact of relaxing user penalties on equilibrium prices can therefore provide insight into the slope of the marijuana supply curve. If relaxing penalties leads to no change in prices, then the supply curve is either perfectly elastic or upward sloping (assuming seller risks are also affected by this change). If relaxing penalties lowers equilibrium prices, supply is either downward sloping or upward sloping (which assumes the decline in sellers’ risks is greater than the decline in buyers’ risks). If relaxing penalties leads to a higher equilibrium price, then the supply curve for marijuana must be upward sloping.

3.2 Modeling the Price of Marijuana

The basic model for prices in marijuana markets is based on a search theoretic framework, where price dispersion emerges because of the need to search for sellers, the fact that there is asymmetric information between buyers and sellers about the product, and that non-pecuniary aspects of the exchange (e.g. location, relationship between buyer and seller) exist that influence the legal risk of getting caught in any given exchange. Galenionos et al (2009) provide a theoretical formalization of this model for the case of heroin markets, but it was first described generally by Reuter and Kleiman (1986). Given our focus on the impact of legal risk to the buyer, we develop a simple reduced form model of the average price for marijuana (P̄_jt) within local market j at time t that shows the mechanisms through which buyer’s risk and seller’s risk operate.

The legal risk of selling marijuana is typically modeled as a function of the seller’s risk of getting arrested (A^S_jt) and the penalties imposed on sellers if arrested (PEN^S_jt). These risks are specific to the markets (j) and time period (t) in which sellers operate. Previous research shows that there are also transaction-specific factors that can influence the risk of getting caught in a given transaction (Caulkins and Pacula, 2006; Ramchand et al., 2006). If we let T_ijt represent a vector of variables capturing where and how the transaction takes place, including information regarding the relationship between the buyer and seller (e.g. strangers, friends, acquaintances), then one can capture the legal risk of bringing a particular bundle to market (i.e. the seller’s risk) as:

$${L^S}_{\mathit{ijt}}=L^S(T_{\mathit{ijt}},{A^S}_{jt},{{\mathit{PEN}}^S}_{jt})$$ (1)

If we let $\overline{T_{jt}}$ represent the average transaction-specific risk the typical seller and buyer are willing to take on, then we can write the average expected risk the seller faces bringing his supply to market as:

$$\overline{L_{jt}^S}=L^S(\overline{T_{jt}},{A^S}_{jt},{{\mathit{PEN}}^S}_{jt})$$ (2)

Assuming homogenous sellers, one can then write the supply of marijuana (S_jt) in market j at time t as a function of this average legal risk, the other costs of production and transporting (Z_jt) and the average price sellers get for marijuana in that market ( $\overline{P_{jt}}$), or

$$S_{jt}=S(\overline{L_{jt}^S},Z_{jt},\overline{P_{jt}}).$$ (3)

Buyers in marijuana markets face their own legal risks when they choose to obtain marijuana and use it. In addition to their own risk of arrest (A^D_jt), which may or may not be the same arrest risk as sellers, users face a different set of penalties in most states if they are caught in possession of marijuana (PEN^D_jt). As with sellers, there are non-pecuniary transaction characteristics that influence the risk buyers face in any given exchange as well. Thus, the legal risk to a buyer engaged in transaction i in market j at time t (L^D_ijt) can be written as follows:

$${L^D}_{\mathit{ijt}}=L^D(T_{\mathit{ijt}},{A^D}_{jt},{{\mathit{PEN}}^D}_{jt}).$$ (4)

If we again let $\overline{T_{it}}$ represent the average transaction-specific risk the typical buyer and seller are willing to take on, then the average expected risk to buyers can be written as:

$$\overline{L_{jt}^D}=L^D(\overline{T_{jt}},{A^D}_{jt},{{\mathit{PEN}}^D}_{jt}).$$ (5)

Once these legal risks for buyers have been defined, one can write the demand for marijuana in market j at time t as follows:

$$D_{jt}=D(\overline{L_{jt}^D},X_{jt},\overline{S{C}_{jt}},\overline{P_{jt}})$$ (6)

where X_jt represents population demand factors for market j, such as income and age demographics, $\overline{{SC}_{jt}}$ represents the average cost to the buyer involved in searching for a supplier, and $\overline{P_{it}}$ represents, as before, the average price people expect to pay in market j at time t.

We know that in equilibrium the average price paid for a given quantity of marijuana will be a function of supply and demand factors, as given by equations (3) and (6) above. Thus, we can write the following reduced form equation for the equilibrium average price observed in the market:

$${\overline{P_{jt}}}^{\ast }=P(\overline{T_{jt}},{A^S}_{jt},{A^D}_{jt},{{\mathit{PEN}}^S}_{jt},{{\mathit{PEN}}^D}_{jt},Z_{jt},\overline{{SC}_{jt}},X_{jt}).$$ (7)

Fundamentally, we are interested in determining whether the vector of penalties that are intended to influence users (PEN^D_jt) are found to be important attributes of the monetary price of marijuana after controlling for all of the other factors that can influence the price for a specific bundle. In this analysis, we have several different measures of the penalties that we wish to consider. First, we will examine the fines and allowances for conditional discharge associated with possession of small amounts of marijuana. Then we will examine the penalties associated with the use and provision of medical marijuana, as these are provisions that are aimed primarily at reducing the risk of the medical user.

4. Data and Empirical Model

Information about marijuana transactions is obtained from the Arrestee Drug Abuse Monitoring (ADAM) Program (formerly Drug Use Forecasting System). From 1987 to 2003, the U.S. Department of Justice interviewed arrestees in urban booking facilities about their drug use patterns as well as tested them for drug use.⁵ The program was terminated after 2003, but a smaller version got started again in 2007. The purpose of the study is to provide local law enforcement and other local officials with reliable estimates of the prevalence of drug abuse and related problems in the population of arrestees in their local market. ADAM sites were originally selected through applications from those local markets interested in participating, but the number of sites increased substantially from 24 in 1996 to 43 in 2003. ADAM data collection takes place four times a year at each site, usually for one or two weeks per calendar quarter. Arrestees are approached within 48 hours of their arrest and asked to participate in the study. Although participation in ADAM is strictly voluntary, response rates hover around 80 to 90 percent (National Institute of Justice, 2000).

From 2000 to 2003, questions about recent drug purchases (i.e., how the dealer was contacted, relationship with dealer, location of transaction) were included to capture information about local drug markets. For those randomly-selected arrestees who agreed to be interviewed, over 42% reported obtaining marijuana in the 30 days prior to the interview. To isolate the effect of penalties on the price paid for marijuana, we limit the analysis sample in three important ways. First, we exclude transactions where the marijuana was obtained with means other than cash, such as property or sex. Second, because the penalties we are examining are only for possession of small amounts of marijuana, we only include transactions involving an ounce or less. In this way, we are able to isolate the effect of penalties on the retail transactions they are meant to influence. Third, we only consider transactions where the amount purchased was reported in grams and ounces since we have no way of converting “joints”, “bags”, and “packets” to grams in a meaningful way across disperse geographic locations.⁶

State-level data on marijuana penalties and medicalization laws came primarily from original data collected by lawyers and policy analysts at The MayaTech Corporation (See Pacula, et al. (2003) for a detailed description of the data collection procedures).⁷ Detailed information on the statutory implementation of the depenalization policies is available for the first three quantity amounts specified in each law, and was supplemented with information on medical marijuana policies from the Marijuana Policy Project (2006). In this analysis we use information on state statutory fines and conditional discharge provisions for possession offences involving 10 grams or less of marijuana.⁸ We limit our analysis to those arrestees who live in states where the maximum fine for possessing and selling marijuana is statutorily defined.

The type of medicalization provision is important because it indicates the number of patients likely to be affected by the provision as well as the legal rights. Therapeutic research program provisions and rescheduling laws are generally viewed as more restrictive as they are passive laws and do not provide active defenses to patients except if use is federally sanctioned. The physician recommendation laws and medical necessity defenses provide greater protection to patients and potentially a larger number of patients might be affected.

These statutory laws represent one dimension of the legal risk of buying and selling drugs. Another dimension is the degree of law enforcement in the area in which the drugs are sold. To proxy this enforcement risk, we divide the number of marijuana arrests (for possession and sales separately) by the total number of arrests at the county-level. These arrest data are pulled from the FBI’s Uniform Crime Reports (UCR), which collects and reports information on the number of arrests and crimes reported to the police for every city in the country. While the shortcomings of these data are well documented (e.g. O’Brien, 1985; Maltz and Targonski, 2002), they remain the only national source of geographically disaggregated crime and arrest data in the United States.⁹

These ratios capture the relative importance of marijuana enforcement to police officers in the county and have been used in other studies (e.g., Farrelly et al., 2001; Pacula et al., 2003). While the ideal measure to capture enforcement would be possession arrests per unit of consumption, annualized consumption information is not available at the county level. DeSimone and Farrelly (2003) demonstrate with data from the 1990–1997 National Household Surveys on Drug Abuse that the effect of the probability of arrest on self-reported use rates for adults is negative regardless of whether they denominated by number of users or number of Index I arrests. Thus, we believe marijuana arrests as a fraction of all arrests in a county is a reasonable proxy for the probability of being arrested for possession.

Transaction-specific risk and costs associated with search come from specific information obtained in the ADAM survey pertaining to buyer’s last purchase of marijuana in the past 30 days. Specifically, respondents identify whether they made their last purchase from a regular source, occasional source, or new source and whether the exchange occurred in the buyer’s neighborhood or not. While capturing the nature of the relationship between buyer and seller, this information also provides useful insights regarding the amount of effort the buyer had to exert to obtain marijuana (i.e. search costs). Information on how the dealer was contacted (by phone or page, approached in public, went to a house or apartment, and other), whether the purchase location was in a public or private, and whether the marijuana was purchased directly from a dealer or a go-between was also included.

To capture differences across counties in the demand for marijuana and size of the market (Jacobson, 2005), we merge data about county demographic and socio-economic characteristics. Information on the demographics of the county population, such as age, race/ethnic categories, percent male are obtained from the U.S. Census. Information pertaining to per capita income and local unemployment rates are obtained from the Bureau of Economic Analysis and Bureau of Labor Statistics, respectively.

Descriptive statistics about these enforcement variables, as well as information about marijuana transactions and state penalties for the analytic sample employed here are reported in Table 1. Because of the geographic dispersion of the ADAM sites, we see that the maximum fine for possessing 10 grams of marijuana ranges from $100 to $150,000 in our analytic sample.¹⁰ For this sample, the average nominal maximum fine is $16,401 and the median is $500.¹¹ Almost 88 percent of arrestees in the analysis sample live in states where possession of 10 grams can be conditionally discharged. We focus on fines and allowances for conditional discharges because they have been shown to be correlated with demand in previous studies (Farrelly et al., 2001; DeSimone and Farrelly, 2003; Pacula et al., 2003).

Table 1.

Summary Statistics for ADAM Arrestees Who Purchased up to 1 Oz of Marijuana with Cash Only (N = 13,370 unless otherwise noted)

Variable	Mean	Std Dev.	Min.	Max.
Conditional discharge for possessing 10 grams	0.879	0.326	0	1
Nominal maximum fine for possessing 10 grams in dollars (N=9708)	16401.29	45297.62	100	150000
Classified as a “decriminalization” state	0.506	0.499	0	1
Nominal maximum fine for selling 10 grams in dollars (N=9,984)	49410.8	106000.7	100	500000
Physician recommendation or medical necessity	0.356	0.479	0	1
Current Therapeutic Research Program	0.423	0.494	0	1
Alcohol outlets per 10,000 population in county	2.909	1.255	0.747	6.163
MJ possession arrests/All arrests (FBI-CI>=70) (N=8,579)	0.046	0.024	0.002	0.123
MJ sales arrests/All arrests (FBI-CI>=70) (N=8,579)	0.005	0.004	0.0003	0.027
Nominal transaction cost in dollars	46.14	78.59	1	4500
Total grams purchased	10.236	9.380	0.050	28.350
Real price per bulk gram in dollars (1982–84=100)	5.39	15.58	0.06	361.38
Purchased in buyer’s neighborhood	0.401	0.490	0	1
Purchased from regular source	0.487	0.500	0	1
Purchased from occasional source	0.348	0.476	0	1
Purchased from a new source	0.165	0.371	0	1
Contacted dealer: By phone or page	0.432	0.495	0	1
Contacted dealer: Go to house or apt.	0.237	0.425	0	1
Contacted dealer: Approached in public	0.201	0.400	0	1
Contacted dealer: Other method	0.131	0.337	0	1
Purchase location: In a house or apartment	0.608	0.488	0	1
Purchase location: Public--Indoors	0.122	0.327	0	1
Purchase location: Public--Outdoors	0.267	0.444	0	1

Approximately 36 percent of the arrestees in the analysis sample live in states where there is either a provision for physicians to recommend marijuana or allowance for a medical necessity defense for those who use marijuana for medicinal purposes. There also appears to be tremendous variation in the enforcement of marijuana laws across the ADAM counties. For example, in 2000 over 10.5 percent of all arrests made in Douglas County, NE were for marijuana possession; the equivalent figure for Bernalillo County, NM was less than 0.4 percent.¹²

With respect to the last marijuana purchase, roughly 49 percent of the sample bought from their regular source, 35 percent bought from an occasional source, and 16 percent made the purchase from a new source. Most buyers contacted the dealer by making a phone call or going to a house or apartment building. Approximately 60 percent of the transactions occurred in a house, apartment, or public building and this is consistent with the claim that most marijuana transactions do not occur in bustling street markets (Kleiman, 1992). The price paid per bulk gram ranges from approximately $0.10 to $680, with the median and mean values being $4.44 and $10.19, respectively ($2004). This is strikingly similar to the price per bulk gram derived from the Drug Enforcement Administration’s System to Retrieve Information from Drug Evidence (2000-$9.21, 2001-$10.06, 2002-$11.65, 2003-$11.94; Caulkins et al., 2004).¹³ Table 2 displays the mean prices paid per bulk gram by marijuana policy. Using data from the entire sample, Panel A suggests that in states with more liberal marijuana laws (conditional discharge, physician recommendation, decriminalization), arrestees pay more per bulk gram for marijuana. Panel B examines the arrestees from three states that changed medical marijuana policies during the sample period (Colorado, Hawaii, and Nevada). In almost every case, the mean price paid was higher after the physician recommendation or medical necessity law was enacted. But since only one of the three pair-wise comparisons is statistically significant at the 10 percent level, we cannot rule out that these differences are simply attributable to chance. We also cannot rule out whether these differences are attributable to some other confounding factor. This motivates the multivariate analyses.

Table 2.

Price Paid per Bulk Gram of Marijuana by Arrestees Subject to Different Policies ($2004)

		No policy	Policy in Effect
Panel A: Price paid by arrestees under different state laws
Conditional discharge	Mean	$6.55	$10.69	***
	N	1252	9118
Physician recommendation/Medical necessity	Mean	$6.91	$16.13	***
	N	6682	3688
Decriminalization	Mean	$9.54	$10.81	**
	N	5118	5252
Panel B: Prices paid by arrestees in states that changed physician recommendation/medical necessity laws since 2000
Colorado	Mean	$3.59	$4.87	*
	N	94	297
Hawaii	Mean	$18.11	$23.44
	N	13	143
Nevada	Mean	$7.78	$9.17
	N	233	239

Note: All figures are adjusted for inflation and reported in $2004. A two-sided sample t test is used to compare means. Statistically meaningful differences are indicated as follows:

^*significant at 10%;

^**significant at 5%;

^***significant at 1%.

Preliminary examination of the distribution of the observed price per pure bulk gram paid for marijuana revealed that the dependent variable is significantly skewed to the right. To ensure the proper specification of our models with a skewed dependent variable, we adopt a strategy that involves the testing of both the conditional mean and mean-variance relationship of our dependent variable (Deb, et al., 2003). A Box-Cox test is first used to determine the appropriate transformation of the dependent variables dependent on our controls, and findings from these tests are confirmed using Pregibon’s link test (Pregibon, 1980).¹⁴ Following these, we use a general linear model with a log link function:

$$\text{E}({\mathit{RPPG}}_{\mathit{ijt}})=\mathit{exp}({\delta }_0+T_{\mathit{ijt}}{\delta }_I+I_{\mathit{ijt}}{\delta }_2+A_{jt}{\delta }_3+{{\mathit{PEN}}^S}_{jt}{\delta }_4+{{\mathit{PEN}}^D}_{jt}{\delta }_5+X_{jt}{\delta }_6+{\mathit{Year}}_t{\delta }_7+{\mathit{Quarter}}_t{\delta }_8+{\mathit{State}}_j{\delta }_9)$$ (5)

where RPPG is the real price per bulk gram paid at the last transaction, Year, Quarter, and State represent a vector of dummy variables noting the year and quarter of the transaction and state where it occurred, and μ is the residual. All other variables are as specified in Section 3.2.

Once we identify the proper transformation of our dependent variables, we then determine the relationship between the transformed dependent variable’s mean and its variance using a modified Park test (1966).¹⁵ Based on the size of gamma (which is reported in the tables), we specify the mean-variance relationship with the Gamma distribution:

$$\text{V}({\mathit{PPG}}_{\mathit{ijt}})=\text{E}({{\mathit{PPG}}_{\mathit{ijt}}}^2).$$ (6)

Since we are primarily interested in the significance of state policy variables, we cluster the standard errors at the state-level to account for correlation among the residuals for arrestees in the same state.

5. Results

We begin our analysis by examining the relationship between retail prices and transaction risk. Table 3 models the real price per bulk gram of marijuana as a function of transaction-level variables that are reported within the ADAM survey while controlling for time-invariant state-level characteristics. One main finding is that arrestees who contact dealers by going to a private dwelling (house or apartment) generally pay lower prices than those who contact a dealer in public (the omitted category). This may represent differences in search costs in that people going to private dwellings may have a particular dealer in mind, while those buying in public may be looking for an available dealer (73% of those who contacted a dealer in public make the exchange outdoors). The coefficient on contacting the dealer by other means is also negative and statistically different from zero. It might also represent lower transaction risk, as indoor purchases are harder to detect by law enforcement.

Table 3.

Examination of the Effects of Transaction Specifics on the Log Real Price per Bulk Gram of Marijuana

	(1)	(2)	(3)	(4)
Purchased from occasional source	0.042	0.047	0.028	0.030
	[0.053]	[0.055]	[0.051]	[0.050]
Purchased from a new source	0.027	0.038	−0.014	−0.009
	[0.071]	[0.068]	[0.072]	[0.072]
Purchased in buyer’s neighborhood	−0.027	−0.027	−0.023	−0.023
	[0.063]	[0.063]	[0.059]	[0.059]
Purchased in Public--Indoor		−0.022		−0.095
		[0.081]		[0.086]
Purchased in Public--Outdoor		−0.049		−0.124
		[0.061]		[0.079]
Contact dealer: Phone or pager			0.046	−0.014
			[0.051]	[0.073]
Contact dealer: Go to house or apt.			−0.094*	−0.179**
			[0.051]	[0.081]
Contact dealer: Other method			0.287***	0.249**
			[0.085]	[0.097]
Bought directly from dealer				0.076
				[0.090]
Constant	−2.842	−2.715	−5.003	−5.085
	[6.339]	[6.386]	[5.821]	[5.706]
State fixed effects	Yes	Yes	Yes	Yes
County-level variables	Yes	Yes	Yes	Yes
Individual-level variables	Yes	Yes	Yes	Yes
Quarter and year fixed effects	Yes	Yes	Yes	Yes
Observations	10370	10370	10370	10370
Gamma from Modified Park test	1.97	1.97	2.07	2.09
BIC	−85422	−85228	−85339	−85334

Notes: All models estimated using GLM with link(log) and family(Gamma). Individual-level variables also included as controls in the model are gender, age categories (21 – 25, 26 – 35, 36 and older), race/ethnicity, marital status, employment status, and high school completion (graduate or GED). County-level variables also included in the model are unemployment rate, real per capita personal income, gender and ethnic distributions, and age distribution of the county population. Reference groups for variables listed above are as follows: source (regular source), contact (approached in public), and purchase location (in a house or apartment). Standard errors clustered at the state-level in all models. Statistical significance using a two-tailed test is indicated as follows:

^*significant at 10%;

^**significant at 5%;

^***significant at 1%.

We find no association between the price paid per bulk gram and whether the sale took place in the buyers’ neighborhood, contrary to results presented in Caulkins and Pacula (2006). The differential results may be due to a difference in the populations examined (arrestees versus a household sample) or may be due to our ability to account for regional market differences in this analysis which went unaccounted for in the previous analysis.

Table 4 presents the results from models that include penalties for possessing small amounts of marijuana. Because we have virtually no variation in these penalties within states during the time period being examined, state fixed effects are not included in the model and results should not be interpreted as causal. Further, the first two columns exclude the vector of county-level variables which 1) do not change much from 2000–2003 and 2) effectively identify many of the states in our analysis sample since most states only include one ADAM site. Column 1 begins by specifying the model without factors that may be correlated with the risks buyers and sellers face, such as arrest rates and sales penalties. We find that higher maximum fines (which indicate higher legal costs to buyers and are associated with less use) are indeed associated with lower prices. Furthermore, states that allow for dismissal of charges for first time offenders (conditional discharge provisions that have been shown to be positively associated with consumption among youth and young adults) are associated with higher prices. These preliminary findings suggest that the legal risks to buyers may be important correlates of price.

Table 4.

Examination of User Penalties on the Log Real Price per Bulk Gram of Marijuana

	(1)	(2)	(3)	(4)	(5)	(6)
Conditional discharge allowed for possessing 10g	0.847***	0.776***	0.370***	0.516***	0.605***	0.390***
	[0.189]	[0.228]	[0.094]	[0.196]	[0.201]	[0.134]
Ln(Real maximum fine for possessing 10g)	−0.156***	−0.125**	−0.008	−0.056	0.004	0.065
	[0.029]	[0.054]	[0.016]	[0.069]	[0.054]	[0.047]
Ln(Real maximum fine for selling 10g)		−0.048	−0.162***
		[0.051]	[0.012]
Classified as a decriminalization state					0.257**	0.540***
					[0.131]	[0.149]
MJ possession arrests/All arrests						−8.721***
						[3.038]
MJ sales arrests/All arrests						6.202
						[12.314]
Purchased from occasional source	0.059	0.029	0.002	0.073	0.061	0.120**
	[0.055]	[0.053]	[0.053]	[0.060]	[0.061]	[0.060]
Purchased from a new source	−0.002	−0.012	−0.043	−0.045	−0.046	−0.022
	[0.085]	[0.090]	[0.082]	[0.088]	[0.090]	[0.103]
Purchased in buyer’s neighborhood	0.055	0.071	0.025	0.039	0.035	0.010
	[0.065]	[0.068]	[0.062]	[0.059]	[0.062]	[0.058]
Contact dealer: Phone or pager	−0.056	−0.097	−0.102	−0.108	−0.086	−0.034
	[0.088]	[0.088]	[0.073]	[0.076]	[0.076]	[0.077]
Contact dealer: Go to house or apt.	−0.269***	−0.301***	−0.286***	−0.272***	−0.258***	−0.191**
	[0.086]	[0.086]	[0.074]	[0.081]	[0.082]	[0.083]
Contact dealer: Other method	0.141*	0.130	0.122	0.128	0.155	0.190
	[0.084]	[0.090]	[0.089]	[0.098]	[0.104]	[0.124]
Purchased in Public-- Indoor	−0.161**	−0.217***	−0.214***	−0.158*	−0.154*	−0.089
	[0.079]	[0.052]	[0.057]	[0.088]	[0.088]	[0.101]
Purchased in Public-- Outdoor	−0.105	−0.108	−0.149*	−0.173**	−0.163*	−0.092
	[0.094]	[0.093]	[0.081]	[0.085]	[0.087]	[0.097]
Bought directly from dealer	0.077	0.054	0.103	0.151*	0.152*	0.218***
	[0.103]	[0.102]	[0.091]	[0.083]	[0.082]	[0.073]
County-level variables	No	No	Yes	Yes	Yes	Yes
Observations	9708	9322	9322	8708	8708	6917

Notes: All specifications include quarter and year fixed effects. State fixed effects are excluded from these models due to insufficient variation in state penalty variables. Individual and county-level characteristics also included in the models are listed under Table 3. The same reference categories apply for included transaction variables and statistical significance is as indicated in the notes to Table 3. Columns 4–6 do not include 1,000 observations from Arizona. The sample size decreases by an additional 1,791 observations in Column 6 because enforcement data are limited to local markets with UCR Coverage Indicator scores >=70.

In Column 2 we include a measure of the real maximum fine associated with selling small amounts of marijuana. It may be the case that states with tougher penalties on users are also tougher on sellers, and hence associations in the previous two regressions merely reflect unaccounted variation in sellers’ risks. However, the results in Column 2 suggest that this is not the case as the inclusion of penalties for the sale of marijuana does not change the principal findings for our possession penalties and the transaction-level variables. Columns 1 and 2 still indicate that those who contacted the dealer at an apartment or home paid less while those who made the purchase in a residence paid more than those who made the exchange inside a public location.

Column 3 includes the vector of county-level covariates along with the user penalties and the maximum fine for selling. While the coefficient on conditional discharge remains positive and highly significant, it becomes roughly 50 percent smaller. It is also important to note how the coefficients on the maximum fines essentially change from Column 2 to Column 3. In Column 2, the coefficient on the log of the real maximum fine for possession is −0.125 and statistically significant while the coefficient on the log of the real maximum fine for selling is statistically indistinguishable from 0. In Column 3, which includes the vector of county-level variables, the possession fine increases to almost 0 (−0.008) and the selling fine is now −0.16 and statistically significant at the one-percent level. The instability in the coefficients on these fine measures is likely attributable to the fact that they do not vary and are correlated with the county factors that also do not really change during our sample period.

The ideal measure of the monetary fine associated with marijuana possession would be a statistic generated from assigned fines instead of statutory maximums. However, geographically-disaggregated data about conviction rates and assigned fines for marijuana possession are simply not available. We use the statutory maximum as a proxy for the expected fine associated with using and buying marijuana, but realize the limits of this approach when using data from Arizona—which has a statutory maximum of $150,000 and accounts for almost 10% of our ADAM sample. Column 4 of Table 4 drops the 1,000 observations from Arizona and not surprisingly, the maximum fine is no longer significant. Despite the smaller sample, the coefficient on conditional discharge is still positive and significant. The remaining columns in Table 4 exclude observations from Arizona.

Next, we introduce an indicator variable for the 11 “decriminalization” states that reduced criminal sanctions for marijuana possession in the 1970s.¹⁶ While Pacula et al (2003) suggest that the traditional interpretation of this variable as a measure of the legal risk is inappropriate, decriminalization status remains an important correlate of demand even when measures of the legal risk are included in the analysis. One interpretation of the robustness of the decriminalization variable to the inclusion of other measures of legal risk is that this policy more accurately reflects unobserved differences in the enforcement of the policy or differences in unobserved state sentiment toward marijuana. Another interpretation is that it is an information effect in that users/potential users may see decriminalization as a clear signal of weaker penalties, even if the penalties are not that different from states that have not decriminalized. The coefficient on decriminalization in Column 5 is positive and significant and does not minimize the practical and statistical significance of the conditional discharge variable. In fact, the inclusion of decriminalization actually increases the coefficient on conditional discharge. The fact that both conditional discharge and decriminalization are positively associated with price lends further support to the conclusion that shifts in demand will influence price. However, one cannot rule out the possibility that these policies might also influence risk to sellers because of corresponding shifts in enforcement that accompany the policy itself.

Column 6 includes county-level enforcement measures from the Uniform Crime Reports. These enforcement measures should provide a better approximation of the general enforcement risk, and hence expected penalty, associated with any transaction. In addition, by including measures of enforcement risk in the model, we can evaluate the impact of the demand-side policies holding enforcement (at least arrest rates) constant. After including these variables, the coefficients on conditional discharge and dealer contact remain statistically significant with smaller absolute values. The negative sign on the ratio of marijuana possession arrests to all arrests is consistent with the hypothesis that as expected sanction for possession increases, price decreases via a shift in demand. We find no significant association between sales arrests on price per gram.

In general, Tables 3 and 4 support the notion that the market prices for marijuana are sensitive to transaction risks. Moreover, the results in Table 4 are consistent with the hypothesis that users’ risks, which are at least partially determined by demand-side policies regarding possession penalties for marijuana, are also important correlates of price. However, there are at least two limitations of the analyses presented thus far. First, the results presented in Table 4 are based entirely on cross-sectional variation in policies. Thus, unobserved state factors that are correlated with penalties for possession of marijuana may be driving the associations in Table 4 rather than a real penalty effect. Second, all of the state laws evaluated thus far could also influence the risk faced by sellers if dealers can effectively hide themselves as users or if enforcement of low-end marijuana transactions changes as a result of these policies.

In light of these limitations, we also consider the influence of state medical marijuana laws on equilibrium prices in Table 5. Unlike the penalties for possession, there has been some variation in state medical marijuana laws within the states even within our short time period being examined; thus, it is possible to include state fixed effects in our models that account for other unobserved state factors that might be associated with the policy and price. Moreover, state medical marijuana laws are in theory supposed to create protection for medical users but not provide protection to drug dealers and organized marijuana sellers.

Table 5.

GLM Estimates of Log Real Price Per Bulk Gram Including Medical Marijuana

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Physician recommendation or medical necessity	0.872***	0.729***	0.669***	0.574***	0.771***	0.165*	0.160*
	[0.133]	[0.160]	[0.172]	[0.136]	[0.118]	[0.086]	[0.086]
Current Therapeutic Research Program			0.139
			[0.144]
Conditional discharge for possessing 10g				0.432***
				[0.138]
Ln(Real maximum fine for possessing 10g)				0.057
				[0.043]
Classified as a decriminalization state				0.155*
				[0.089]
MJ possession arrests/All arrests					−7.706***	−1.638	−1.777
					[1.960]	[4.042]	[4.167]
MJ sales arrests/All arrests					14.778	25.282	30.306
					[9.057]	[18.247]	[18.646]
Ln(County population)							−0.119**
							[0.054]
Purchased from occasional source	0.038	0.020	0.018	0.041	0.073	0.072	0.071
	[0.053]	[0.051]	[0.051]	[0.058]	[0.052]	[0.054]	[0.054]
Purchased from a new source	−0.041	−0.054	−0.059	−0.058	−0.033	−0.018	−0.019
	[0.083]	[0.082]	[0.081]	[0.092]	[0.087]	[0.084]	[0.084]
Purchased in buyer’s neighborhood	0.000	−0.029	−0.029	0.004	−0.053	−0.050	−0.050
	[0.064]	[0.060]	[0.060]	[0.066]	[0.052]	[0.053]	[0.053]
Contact dealer: Phone or pager	−0.072	−0.051	−0.053	−0.082	−0.002	0.007	0.008
	[0.080]	[0.080]	[0.078]	[0.078]	[0.079]	[0.077]	[0.077]
Contact dealer: Go to house or apt.	−0.263***	−0.233***	−0.234***	−0.245***	−0.162**	−0.133	−0.131
	[0.091]	[0.086]	[0.084]	[0.085]	[0.081]	[0.084]	[0.084]
Contact dealer: Other method	0.166*	0.225**	0.220**	0.168*	0.241**	0.262**	0.265**
	[0.091]	[0.106]	[0.105]	[0.101]	[0.117]	[0.109]	[0.109]
Purchased in Public--Indoor	−0.119	−0.127	−0.130	−0.150*	−0.069	−0.067	−0.067
	[0.090]	[0.087]	[0.086]	[0.085]	[0.099]	[0.101]	[0.101]
Purchased in Public--Outdoor	−0.120	−0.130	−0.128	−0.157*	−0.065	−0.079	−0.077
	[0.098]	[0.087]	[0.086]	[0.089]	[0.089]	[0.088]	[0.088]
Bought directly from dealer	0.119	0.092	0.097	0.146*	0.134	0.126	0.130
	[0.096]	[0.088]	[0.088]	[0.084]	[0.096]	[0.093]	[0.093]
State fixed effects	No	No	No	No	No	Yes	Yes
County-level variables	No	Yes	Yes	Yes	Yes	Yes	Yes
Observations	10370	10370	10370	8708	8579	8579	8579

Notes: All models include quarter and year fixed effects. Individual and county level variables are also included (see notes under Table 3) and the same reference categories apply as in Table 3. Column 4 does not include 1,000 observations from Arizona and 662 observations from states that do not have a maximum fine for possessing 10 grams. The sample size decreases by 1,791 observations in Columns 5–7 because enforcement data are limited to jurisdictions with UCR Coverage Indicator scores >=70. Standard errors clustered at the state-level in all models. Statistical significance using a two-tailed test is indicated as follows:

^*significant at 10%;

^**significant at 5%;

^***significant at 1%.

Columns 1 through 7 of Table 5 include an indicator variable that equals 1 if the state actively allows physician recommendation and/or medical necessity defense.¹⁷ This measure captures policies that protect physicians and/or their patients from state prosecution for discussing, recommending, or using marijuana as medicine. The models are first evaluated excluding state fixed effects so that comparison with results in Table 4 can be directly made. In Columns 3–4 other state penalty measures are included and in Columns 5–7 the enforcement measures are incorporated. Across the board, the coefficient on physician recommendation/medical necessity defense is positive and statistically significant. In the specification with state fixed effects, enforcement controls, county-, transaction-, and individual-level characteristics (Column 6), the results suggest that arrestees in states with active medical marijuana programs pay $1.40 more for a gram of marijuana compared to arrestees in non-medical states.¹⁸

In Column 7, the natural log of the county population is included as a regressor. This provides another proxy for market demand and allows us to test Caulkins’s (1995) conjecture and Jacobson’s (2005) finding that the price is lower in larger markets. The coefficient on county population is negative and statistically significant, thus confirming the hypothesis that price per gram in lower in larger markets. The coefficient and standard error on physician recommendation/medical necessity defense remains virtually unchanged. Since laws that allow physicians to recommend marijuana or patients to use a medical necessity defense should primarily influence consumers (rather than dealers),¹⁹ the positive association with price suggests that the market supply curve for marijuana is upward sloping because a change in demand leads to an increase in equilibrium prices. It is important to note, however, that all states with recommendation/medical necessity laws also have provisions for home cultivation and some allow dispensaries or shops to sell marijuana for medicinal purposes. If allowances for home cultivation and the sale of marijuana reduced the risk to sellers of producing marijuana, then prices should fall holding other factors constant. But because we see an overall positive effect of these policies on price, it suggests that the shift in demand exceeds the shift in supply that could occur with a reduction in risk to sellers.

Finally, to address any remaining concerns that sellers’ risk—instead of buyers’ risk—is driving our results, Table 6 reports how per capita alcohol outlets relate to the price paid for marijuana by arrestees.²⁰ Several studies that include estimates for the monetary price of alcohol and marijuana find that alcohol and marijuana are economic complements (Farrelly et al., 1999; Pacula, 1998; Saffer & Chaloupka, 1999; Williams et al., 2004). Various studies have also demonstrated a statistically significant positive association between alcohol outlet density (on-premise and off) and alcohol consumption among young adults and adults (Nelson, 2008), drunk driving (Gruenewald, et al., 2002; Treno et al., 2003; Wechsler et al., 2002), and homicides and assaults (Gorman et al., 2001; Lipton & Gruenewald, 2002). Given these bodies of literature, outlet density should be positively related to marijuana use, and hence marijuana prices if higher demand increases price.²¹

Table 6.

Examining the Impact of Alcohol Outlet Density on the Log Real Price per Bulk Gram of Marijuana

	(1)	(2)
Ln(Per capita alcohol outlets in county)	1.161**	1.460***
	[0.486]	[0.512]
MJ possession arrests/All arrests		1.305
		[3.099]
MJ sales arrests/All arrests		46.717
		[30.335]
Purchased from occasional source	0.029	0.069
	[0.048]	[0.053]
Purchased from a new source	−0.015	−0.024
	[0.079]	[0.092]
Purchased in buyer’s neighborhood	−0.030	−0.056
	[0.050]	[0.054]
Contact dealer: Phone or pager	−0.012	0.007
	[0.074]	[0.079]
Contact dealer: Go to house or apt.	−0.180**	−0.134*
	[0.078]	[0.075]
Contact dealer: Other method	0.248**	0.260**
	[0.103]	[0.111]
Purchased in Public--Indoor	−0.100	−0.069
	[0.071]	[0.083]
Purchased in Public--Outdoor	−0.128*	−0.073
	[0.073]	[0.075]
Bought directly from dealer	0.072	0.123
	[0.089]	[0.095]
County-fixed effects	Yes	Yes
County-level variables	Yes	Yes
Individual-level variables	Yes	Yes
Quarter and year fixed effects	Yes	Yes
Observations	10,730	8,579

Notes: All models estimated using GLM with link(log) and family(Gamma). The sample size decreases by 1,791 observations in Column 2 because enforcement data are limited to jurisdictions with UCR Coverage Indicator Scores >=70. Individual and county level variables, described in the note under Table 3, are also included. Standard errors clustered at the state-level in all models. Statistical significance using a two-tailed test is indicated as follows:

^*significant at 10%;

^**significant at 5%;

^***significant at 1%.

In Column 1 we identify a positive and statistically significant relationship between ln(per capita alcohol outlets) in a county and the real price paid per bulk gram of marijuana by arrestees in that county, suggesting that higher demand for alcohol, and subsequently marijuana, increases the equilibrium price for marijuana. These models include county-, quarter-, and year-level fixed effects as well as individual-, county-, and transaction-level controls. The standard errors are clustered at the county-level. This model does not include county-level marijuana enforcement controls and these variables are added in Column 2. As in the previous tables, adding these variables decreases the sample size by almost 1,800 observations as some counties have low-quality arrest information and are excluded from the analyses. The coefficient on our alcohol measure remains positive and highly significant, providing additional support for our contention that buyer’s risk is driving our results.

6. Discussion

This paper finds that variations in policies associated with user risk are associated with variations in self-reported marijuana prices per gram. Results exploiting cross-sectional variation in possession penalties suggest that states with lower penalties have higher prices due to increased demand that more than offsets the effect on price of any increase in supply. This provides new evidence that user sanctions can influence demand. Given the variation examined is purely cross-sectional, the direction of causality is not entirely certain. Thus, we also examine state medical marijuana laws and using within state temporal variation in the adoption of active medical marijuana policies, we find evidence that policies aimed at users do in fact lead to changes in prices in a fashion that is consistent with anticipated shifts in demand. Robustness checks that examine changes in alcohol consumption proxied through measures of alcohol outlet density support the conclusion that supply is upward sloping in these markets.

There are a number of plausible explanations for an upward sloping supply curve for marijuana even in the long run. As has been published elsewhere (Kleiman, 1989; Reuter and Kleiman, 1986), a substantial fraction of the monetary price of marijuana represents the economic value of risk associated with engaging in an illicit market. Although it is frequently presumed in other areas of economics that producers are risk-neutral, such a notion is inconsistent with empirical evidence showing that marijuana sellers receive large compensation for incurring risk in bringing the drug to market. A more plausible assumption is that producers of marijuana are not risk-neutral.²² If we further presume that the population of potential suppliers to the market has heterogeneous risk preferences, it is easy to identify a potential barrier to entry into this market - the risk premium. If the current monetary compensation for incurring risk is relatively low, then some potential suppliers may choose to stay out of the market. Only by increasing the risk premium (and hence monetary price of supplying the good) will new suppliers decide to enter the market.

An upward sloping supply curve could also exist if input prices rise with increases in output. Although in legal markets rising input prices are relatively less common for agricultural products, the illicit nature of the marijuana market means that suppliers need to compensate field workers for the risk they incur participating in the production of an illegal good. If individuals who make up the potential pool of laborers have heterogeneous preferences in risk, then it is possible that input costs could rise with the level of production as those with the lower risk preference will have to be paid a higher premium to be enticed to work in this market versus a legal market. The extent to which heterogeneous risk preferences among potential laborers could influence the cost of production depends on the distribution of high and low risk preferences in the population of workers and the relative tax imposed on legitimate wages.

There are a number of important caveats that need to be considered when interpreting these results. First, the data that are employed in this study are drawn from a convenience sample of heavy and/or regular users who are frequently engaged in the marijuana market. We do not know to what extent purchases described here are representative of the full population of purchases made in the market. Furthermore, we only observe purchases over a small window of time (16 quarters over the course of 4 years). However, these ADAM data do represent a very good source of information on marijuana prices in the U.S., as it is clear that arrestees are frequent users who engage regularly in the market (Caulkins and Pacula, 2006).

A second and perhaps more important limitation of the study is the lack of information on the quality of marijuana. The omission of information on potency is clearly important because it is possible that several transaction-specific characteristics that we interpret as indicative of risk are also highly correlated with quality. To the extent that certain transaction patterns are correlated with the quality of marijuana, then there may be some bias in interpreting the transaction coefficients as measuring differences in risk. However, unless the quality of marijuana is systematically correlated with state-level policies, it is not clear how the omission of potency from this analysis could potentially bias the findings regarding the effect of state policies.

While the finding of an upward sloping supply curve in marijuana markets may not be surprising, it does have important implications for thinking about the current state and federal debates about decriminalization. First, it suggests that price increases may offset some of the consumption effects of reducing the legal risk to users on demand. This helps us better understand the variable results obtained in the literature examining the effects of this policy in the United States. It also clearly demonstrates that consumption does in fact respond to a change in the legal status, so much so that price rises because of it. Second, it suggests that lowering legal risks for users are associated with higher marijuana prices, which ceteris parabis, implies higher profits for drug dealers in the short-run (and possibly the long run if the long run supply curve is also upward sloping). Evaluating the welfare and moral implications of increasing profits for dealers is beyond the scope of this paper, but it is important to note that the increase in profits could lead to a rise in the number of dealers in the long run if barriers to entry, such as heterogeneous risk preferences across sellers, do not exist.

Appendix Table A1.

Marijuana Penalties and Medicalization Policies as of Jan 1, 2003 for States in our Analysis Sample

State	Obs.	Conditional Discharge for 10g	Max Fine for Possess 10g	Classified “Decrim”	Physician Rec/Med. Necessity	Current TRP	Home Cultivation
AL	135	Yes	$2,000			Yes
AK	367	Yes	$1,000	Yes	Yes		Yes
AZ	1,000	Yes	$150,000	Yes
CA	1,384	Yes	$100	Yes	Yes	Yes	Yes
CO	391		$100	Yes	Yes*		Yes*
DC	17	Yes	$1,000
FL	172	Yes	$1,000
GA	168	Yes	$1,000			Yes
HI	156	Yes	$1,000		Yes*		Yes*
IL	137	Yes	$1,500			Yes
IN	461	Yes	$5,000
IA	303	Yes	$1,000
LA	107	Yes	$500
MA	15	Yes	$500			Yes
MI	114	Yes	$2,000
MN	386	Yes	$300*	Yes		Yes
MO	30		$1,000
NE	329	Yes	$100	Yes
NV	472	Yes	$5,000		Yes*		Yes*
NM	303	Yes	$100			Yes
NY	397		$100	Yes		Yes
NC	188	Yes	$200	Yes
OH	434		$100	Yes
OK	662	Yes	NA
OR	376	Yes	$1,000	Yes	Yes		Yes
PA	94	Yes	$500
TX	577	Yes	$2,000			Yes
UT	313	Yes	$1,000
WA	882	Yes	$250		Yes	Yes	Yes

Notes:

^*Denotes change during sample period. Arizona and Louisiana have physician prescription laws on the books, they are purely symbolic and do not afford medical users legal protection (Marijuana Policy Project, 2006). In Arizona, a law was signed in 1997 which required all drugs dispensed as medicine must be approved by the federal FDA. This, in essence, repealed the medical marijuana law (Annas, 1997).