The Kids Aren’t Alright: The Effects of Medical Marijuana Market Size on Adolescents

Abstract

We exploit shocks to US federal enforcement policy to assess how legal medical marijuana market size affects youth marijuana use and consequences for youth traffic-related fatalities. Using hand-collected data on state medical marijuana patient rates to develop a novel measure of market size, we find that legal market growth increases youth marijuana use. Likely mechanisms are lower prices and easier access. Youth die more frequently from alcohol-involved car accidents, suggesting complementarities for youths. The consequences of marijuana legalization for youth are not immediate, but depend on how supply-side regulations affect production and prices.

As of 2020, 67% of adolescents in the US live in a state where marijuana is medically legal, up from 30% just one decade prior (USDHHS 2019). These substantive state policy changes have occurred in the context of the US federal government’s classification of marijuana as a prohibited Schedule I substance with high potential for abuse and no currently accepted medical value. The conflict between state and federal policy has arisen in part due to disagreement about the consequences of legalization on youth use and potential externalities (ONDCP 2014). These concerns have prompted further debates about the potential welfare effects of legalizing the nonmedical use of marijuana (Anderson and Rees 2014a), a policy now enacted in 19 states and Washington D.C.

Existing work is largely reassuring in that previous studies tend to find no effect of medical marijuana laws (MMLs) on youth use (Sarvet et al. 2018, Anderson and Rees Forthcoming). However, most studies assess changes before and after legalization events and average effects across MML frameworks. While legalization may beget a sharp change in social norms and risk, other aspects of medical marijuana policy that matter more for youth consumption may take time to manifest, with differential impacts depending on the specifics of regulatory design, implementation, and enforcement. Even when marijuana is medically legal for adults aged 18 and older, age restrictions force youth to access marijuana through illicit markets or marijuana diverted from legal markets. Larger, more loosely regulated medically legal markets will likely have larger effects on the price and quantity of illicit or diverted marijuana.¹ If mechanisms related to price and access drive youth consumption decisions, this may produce policy effects that are heterogeneous over time and across states. Evaluations that fail to account for potentially heterogeneous time paths of medical marijuana market evolution across states may thus underestimate the impacts of mechanisms related to market features rather than those related to social norms or legal risk.

This paper thus uses a novel strategy to estimate how youth consumption responds to changes in the size of the legal market for medical marijuana.² We use hand-collected panel data on registered medical marijuana patient rates by state as a measure of “exposure” to legal medical marijuana markets, and we show that this measure responds to exogenous shocks to legal supply costs. Specifically, we exploit the timing of two memos that signaled changes in federal enforcement priorities regarding state medical marijuana markets. These memos had differential effects across legal states depending on how the supply-side of the market was regulated. A memo in 2009 relaxed federal enforcement policy, increasing market size and decreasing price in loosely regulated states. A memo in 2011 reversed this position, shrinking market size in loosely regulated states. Leveraging the timing of these two memos, interacted with pre-existing state laws regulating suppliers, we show that variation in federal enforcement policy drives changes in legal medical marijuana market size, with larger effects in states with generous production limits.

We find that medical market size affects youth marijuana use independent of medical marijuana legalization. Based on our results, if states reach the median medical marijuana market size of about one percent of adults registered, the prevalence of marijuana use in the past month would increase from 7.6% to 8.2% for adolescents aged 12-17 (a 8% increase), from 17.3% to 18.6% (a 7% increase) for 18-25 year-olds, and from 4.4% to 5.4% (a 23% increase) for adults over age 25. Similar to Anderson et al. (2013), who find a gradual decline in marijuana prices after legalization, we find that prices decrease with market size. Youths also report easier access to marijuana in larger legal marijuana markets. Evidence for changes in perceived risk from marijuana use are weaker, consistent with past findings that medical marijuana laws produce little change in risk perceptions among adolescents (Wen et al. 2019). Together, our results suggest that, as predicted by Grossman and Chaloupka (1997), ease of access and decreased price may be more important mechanisms for youth consumption than changes in risk perceptions or social norms.

For youth, we also find that greater marijuana access interacts with risky behaviors, such as impaired driving, which generates negative externalities. Estimating the effects of medical marijuana market size on traffic fatalities, we find that greater marijuana market size increases traffic fatalities involving drivers aged 15-20 by 6%, with large and significant effects on weekend, nighttime, and alcohol-related accidents. These results highlight that youth use patterns differ from adults, for whom vehicle fatalities are unaffected and for whom alcohol-related fatal accidents are unchanged or, in the case of older adults, significantly reduced.

Our main contribution is to highlight the role of market size as a key mediator of youth marijuana use and to demonstrate how federal and state policies interact to determine medical marijuana market size. Early pioneers in the evaluation of marijuana legalization’s effects on youth have shown that it is not legality per se that affects use (Wall et al. 2011; Harper et al. 2012; Anderson et al. 2015), and more recent work confirms this result (Anderson and Sabia 2022; Coley et al. 2019; Hollingsworth et al. 2022). We, too, confirm this result: we find no effect of the passage of MMLs on adolescent use of marijuana. However, similar to how outlet density of alcohol retailers mediates youth alcohol use (Huckle et al. 2008), we find that legal medical marijuana market size affects youth marijuana use. Likely channels from larger markets include lower prices and greater supply promoting easier access by adolescents, with the importance of social markets echoing Katzman et al.’s (2007) findings for youth tobacco use. Furthermore, our finding that state medical marijuana markets take time to develop indicates that the full effects of marijuana policy changes may only be detected by using methods that allow for dynamic policy effects (e.g., event studies used in Wen et al. (2015) and Anderson et al. (2019)); although, by showing that state legal marijuana market size depends on cost shifters that are not solely a function of time since law enactment, we demonstrate that states experience heterogeneous time paths of market evolution, suggesting that even standard event-study designs in this literature are subject to producing biased estimates (Sun and Abraham 2021).

Given marijuana liberalization policy’s potential benefits to adult populations, our work reveals potential trade offs: greater legal marijuana access for adults leads to spillovers to youth substance use and related harms. State legalization alone fails to increase youth marijuana use; however, policies with lax production limits that enable large-scale cultivation and expansion of marijuana availability, and subsequent declines in price, do seem to increase youth use. As the US government considers federal legalization, our work points to the importance of imagining, tracking, and measuring how changes in federal policy, be they formal or informal, shape pre-existing and newly emerging state-legal marijuana markets.

I. Youth Use and Legal Medical Marijuana Access

Youth marijuana use differs from adult use in several ways that interact with regulatory design. First, in the US, people under the age of 18 have very limited legal access to medical marijuana because all states have imposed legal age-limits of 18 years. While individuals under age 18 can qualify for medical marijuana, they have stricter qualifying standards and require parental consent.³ Thus, while MMLs create legal supply channels for adults to access marijuana, their effects on youth access are theoretically ambiguous depending on how legal and illicit markets interact (Amlung et al. 2019). Increased legal supply could increase access in illicit markets through legal producers diverting product, through resale or sharing of medical marijuana by legal users, or if competition between legal and illegal producers exerts downward pressure on prices in both markets. Alternatively, illicit market access could decrease if legal production crowds out illegal production. While some work suggests that the massive price declines seen in states with commercial legalization exert downward pressure on illegal marijuana prices (Meinhofer and Rubli 2021), it is unclear whether the same holds in states with legal medical marijuana markets only.

In addition, the social contexts in which youths and adults use marijuana differ in ways that may differentially influence polysubstance use or other risky behavior. For example, youth may be more likely to consume marijuana at parties while also consuming alcohol (Lipperman-Kreda et al. 2017). Youths also tend to take greater risks than adults (O’Donoghue and Rabin 2001). Together, this could differentially affect the likelihood that adults and youth combine marijuana and alcohol. Indeed, descriptively, adolescent and young adult drinkers are far more likely to consume marijuana with alcohol; in 2019, among past-month drinkers, 18% of youth aged 12-25 report having used marijuana while drinking, whereas 7% of adults over age 25 report this behavior (SAMHSA 2020). These simultaneous use behaviors have particular implications for binge drinking and driving while intoxicated.

A final difference between youth and adult use concerns behavioral responses to market features. When consuming addictive goods, youth are more price elastic than adults (Grossman and Chaloupka 1997), and evidence from cigarette markets supports that youth are largely reliant on social markets for accessing age-restricted goods (Katzman et al. 2007). Given that legal users’ costs in the formal market affect supply in social markets (Hansen et al. 2013), medical marijuana market size may have an effect on youth use separate from the effect of legalization alone. Consistent with this, when Salomonsen-Sautel et al. (2012) surveyed adolescents in substance abuse treatment in Denver, Colorado, they found 74% reported using diverted medical marijuana. In 2013, a national survey found that, among high-school seniors, almost 18% of past-year users report access through another person’s medical marijuana “prescription” (Boyd et al. 2015); among high-school seniors in states with MMLs, the share was 34% (Wadley and Barnes 2013). This suggests substantial diversion from legal to illegal markets that supply adolescents and young adults.

Based on the literature to date, the average effect of creating a legal market for medical marijuana through state legislation fails to increase, and may even decrease, adolescent use (Sarvet et al. 2018, Anderson and Rees Forthcoming, Coley et al. 2019).⁴ Early studies of recreational marijuana laws find mixed evidence on whether legalizing adult nonmedical marijuana use generates unintended spillovers to adolescent markets, with some studies finding an increase (Hollingsworth et al. 2022; Cerdá et al. 2017) and others finding a negative or null effect (Anderson and Sabia 2022).

Studies evaluating the effects of legalizing medical marijuana dispensaries have found mixed evidence for youth. While some find no effect of medical dispensaries on adolescent use (Johnson et al. 2017; Shi et al. 2018), others that use empirical designs more robust to time-varying effects show some evidence that legalizing dispensaries increases marijuana use by youth (Pacula et al. 2015; Wen et al. 2015). However, findings that dispensary laws have significant effects even without the existence of operational dispensaries raises questions on how to interpret these results (Anderson and Rees 2014b). While some subsequent research indicates that the presence of legally protected and operational dispensaries may be a more important driver of changes in substance use (Hollingsworth et al. 2022; Powell et al. 2018), this literature has primarily focused on adults.

II. State Markets and Federal Enforcement Over Time

As of December 2020, 36 states and Washington D.C. had enacted laws providing protections for the use of medical marijuana. In most states, medical marijuana patients must register in order to access medical marijuana and be protected from state arrest. While states vary slightly in the specific information required of applicants, typically individuals must obtain a physician’s certification that they have a qualifying medical condition, provide verification of residency, and submit an application and registration fee to the state authority (Houser and Hiller 2020). Without registration, marijuana possession and purchase remain illegal. Registration typically must be renewed every one or two years (depending on the state), and most states explicitly authorize patient permit revocation for failing to update registration information, fraud, marijuana distribution to a non-patient, or other violations of MML policy (Klieger et al. 2017).

Conceptually, the number of registered marijuana patients is a function of legal marijuana demand, which depends on whether a patient’s benefits from registration exceed the costs. Benefits from registration include access to greater product variety, quality, and safety, lower search costs, and lower prices; these benefits vary with state production limits. After legalization, impacts on product availability and costs may also spill over to illicit markets if there is leakage from the legal to illegal market or if dealers compete with legal retailers. Patient costs include registration fees, finding a doctor to provide a recommendation, and perceived risk from state and federal enforcement.

Increases in federal enforcement, actual or perceived, affect both patients’ willingness to register and state-legal producers’ willingness to supply, shrinking the legal market. Decreases in enforcement have the opposite effect. However, supply-side responses to changes in federal enforcement will likely be greater, i.e. more elastic, in states with looser restrictions on producers. For example, in loosely regulated states there is more room to increase supply in response to relaxed federal enforcement while staying within the boundaries of state-level policy. This also means that when federal enforcement tightens, loosely regulated states have larger markets that may be ripe targets for federal enforcement, causing them to shrink more. Thus, variation in federal enforcement policy over time interacts with variation in state-level marijuana regulation intensity to change medical marijuana market size in a way that varies across states.

Federal enforcement policy has varied over time. Before 2009, the federal government made direct threats to MML states, stating that even users and suppliers in compliance with state policy would remain subject to federal prosecution (French 2005). Two federal memos altered this punitive stance: the Ogden memo in 2009 and the Cole memo in 2011.

The Ogden Memo, announced on October 19, 2009, maintained the government’s commitment to prosecuting large-scale traffickers of marijuana, but emphasized that “prosecution of individuals with cancer or other serious illnesses who use marijuana as part of a recommended treatment regimen consistent with applicable state law, or those caregivers in clear and unambiguous compliance with existing state law who provide such individuals with marijuana, is unlikely to be an efficient use of limited federal resources” (Ogden 2009). In sum, the Ogden Memo de-prioritized prosecuting medical marijuana users and suppliers in states with legal medical markets.

On June 29, 2011, the US government reversed this stance, citing concerns with the “increase in the scope of commercial cultivation, sale, and distribution and use of marijuana for purported medical purposes” (Cole 2011). The Cole Memo stated that individuals involved in medical marijuana sales and distribution would be subject to federal enforcement action; in the months leading up to and following the memo, the Drug Enforcement Administration stepped up raids on medical marijuana producers (Houser and Rosacker 2014). Together, the Ogden and Cole Memos create three federal enforcement regimes: strict enforcement before 2009, lax enforcement between October 2009 and June 2011, and strict enforcement after June 2011.

Our analysis exploits the differential effects of these changes in federal enforcement policy on medical marijuana market size, with variability driven by pre-existing state regulations constraining the state-legal behaviors of medical marijuana suppliers. We distinguish between two types of pre-existing state medical marijuana regimes for which we expect suppliers to respond differently to changes in the federal enforcement stance: (1) states with loosely regulated supply, whereby suppliers faced non-existent or minimally binding production caps and were not subject to a state licensing regime, and (2) states with strictly regulated supply, whereby suppliers could only produce for one patient or had to conform to the state licensing regulations. We anticipate that marijuana producers in states with more loosely regulated supply will have a more elastic response to changes in federal enforcement policy compared with producers in MML states with stricter supply regulations or without MMLs.

III. Data

Our paper offers a novel measure of medical marijuana market size, the number of registered medical marijuana patients per capita. In this section, we discuss this measure in detail and summarize sources for the policy and outcome data.

A. Registration Rates as a Measure of Market Size

The number of registered medical marijuana patients provides an observable measure of legal market size. Though the number of registered medical marijuana patients is a good measure of legal medical marijuana market size, until now it was unavailable because records are not centralized.⁵ We collected monthly data for 1999 to 2013 from a number of sources, including contact with state officials, state department websites, news articles, and academic papers. Our period of inquiry ends in 2013 to focus attention on medical markets, since 2014 marks the first year of commercial recreational marijuana availability within the US.⁶

Because our outcome variables are annual state-level prevalence measures, we use an annual measure for market size: the number of registered patients in December of the year. For states that lack data for December of a given year, we linearly interpolate missing end-of-year registration rates using the nearest available months of registered patient counts.⁷ The final measure of market size is the annual registration rate, calculated as the percent of adults registered as medical marijuana patients at the end of a given year (i.e., the number of registered patients per 100 adults aged 18 and older in that state in December). A heat map of registration rate variation across states and over time is shown in Figure A.1. For our main analyses, we exclude Washington, Maine, and California. Washington is the only state with no registration system in place, while California and Maine have voluntary patient registries. While we obtained the voluntary registration data available from California after 2005, these represent a notable undercount of the true number of medical marijuana patients in the state (ProCon.org 2014). However, for completeness, we report results of all primary analyses including California’s voluntary registration rate data in Appendix C.

B. Other Data Sources

We combine our novel measure of medical marijuana market size with measures of regulation intensity, marijuana use, mechanisms, and externalities. Table 1 provides summary statistics for the marijuana use measures and covariates.

Table 1:

Summary Statistics (2002-2013), Comparing States with Registered Medical Marijuana Patients to States Without

	All	Registration Rate=0	Registration Rate>0	Mean Difference [Standard Error]
Legal Market Size
Registration Rate	0.06 (0.32)	0.00 (0.00)	0.68 (0.83)	0.68*** [0.08]
Registered Patients	2846.59 (15833.94)	0.00 (0.00)	30063.68 (42915.77)	30063.68*** [4091.86]
Marijuana Consumption
Past-month use, age 12-17	7.10 (1.27)	6.93 (1.12)	8.85 (1.38)	1.92*** [0.14]
Past-year initiation, age 12-17	5.81 (0.87)	5.69 (0.77)	7.03 (0.94)	1.34*** [0.10]
Past-month use, age 18-25	17.29 (3.65)	16.84 (3.30)	21.65 (4.01)	4.81*** [0.41]
Past-year initiation, age 18-25	6.93 (1.30)	6.78 (1.20)	8.36 (1.34)	1.58*** [0.14]
Past-month use, age 26+	4.38 (1.31)	4.13 (0.94)	6.78 (1.84)	2.65*** [0.18]
Past-year initiation, age 26+	0.16 (0.05)	0.15 (0.04)	0.22 (0.07)	0.07*** [0.01]
State Covariates
% Under 30	0.41 (0.02)	0.41 (0.03)	0.40 (0.02)	−0.01*** [0.00]
% Male	0.49 (0.01)	0.49 (0.00)	0.50 (0.01)	0.01*** [0.00]
Ln(Population)	15.88 (0.80)	15.96 (0.76)	15.13 (0.78)	−0.82*** [0.08]
Unemployment rate	6.57 (2.03)	6.45 (1.91)	7.80 (2.64)	1.35*** [0.27]
Ln(Real state per capita income)	10.35 (0.14)	10.35 (0.14)	10.35 (0.13)	0.00 [0.01]
Decriminalization law	0.25 (0.44)	0.23 (0.42)	0.51 (0.50)	0.29*** [0.05]
BAC 0.08 law	0.94 (0.23)	0.94 (0.23)	0.95 (0.21)	0.01 [0.02]
Ln(Real beer tax)	−1.81 (0.73)	−1.78 (0.72)	−2.08 (0.71)	−0.30*** [0.08]
Ln(Real cigarette tax)	−0.38 (0.89)	−0.43 (0.90)	0.06 (0.58)	0.49*** [0.07]
N	576	466	110

Notes and sources: CA, WA, and ME are excluded. Data on substance use comes from NSDUH ((2002-2003)-(2013-2014)). Decriminalization data is from Alford (2013) and MPP.org. Beer tax data is from Brewer’s Almanac and BAC laws are from the National Highway Traffic Safety Administration. Cigarette excise tax data is from Impacteen.org and TaxFoundation.org. Remaining state-level covariates are from BLS and Census State Statistical Abstracts. Means are weighted by state-year population, with standard deviations in parentheses. Standard errors from t-test of difference in means with unequal variance in square brackets.

^***p<0.01

^**p<0.05

^*p<0.1

State Regulation Type:

Pacula et al. (2015) highlighted how variation in specific dimensions of MML policy can have meaningful policy impacts. Production allowances are a key feature of state-level policy that conceptually should drive price, product composition, and ease of access in the legal market (Sevigny et al. 2014; Smith 2020). We thus define “Loose” supply restrictions as those states with MMLs prior to the Ogden Memo where (1) caregivers were allowed to serve an unlimited number of patients, regardless of dispensary legality; or (2) where caregivers were allowed to serve multiple patients and state-licensed dispensaries were not legal.⁸ These states represent those where producers or caregivers faced minimal costs from state regulatory oversight and were subject to less restrictive (or no) constraints on the amount of marijuana they could legally supply.

Marijuana Consumption:

Measures of marijuana consumption come from the National Survey on Drug Use and Health (NSDUH) small area estimates (SAMHSA 2014a). Funded by the Substance Abuse and Mental Health Services Administration (SAMHSA), the NSDUH is an annual survey of the US population twelve years of age or older. Based on the survey microdata, SAMHSA publishes aggregate state-level estimates of the prevalence of past-month use and past-year initiation of marijuana, available separately for age groups of 12-17, 18-25, and over 25 years of age in overlapping two-year averages starting with 2002-2003. These published estimates, which have been used in prior studies of MMLs (Hollingsworth et al. 2022; Choi et al. 2019), have two key limitations. First, data are self-reported and thus subject to reporting bias; while we would expect changes in marijuana’s legal status to be a larger determinant of truthful reporting, it is possible that changes in market size may induce changes in willingness to report. Second, data are only available in two-year averages, which means our registration rate data fail to correspond exactly with the use data, reducing precision. For this paper, primary analyses of two-year averaged outcomes are done using the last month of the first year of the two-year average for linkage with registration rates (e.g., 2013-2014 consumption data are linked to registration rate data as of December 2013).⁹ Despite these limitations, the NSDUH is the only publicly available dataset that provides representative state-level estimates of marijuana consumption for all individuals 12 years of age and older.¹⁰

Youth Access and Risk Perceptions:

The NSDUH Restricted-use Data Analysis System (R-DAS) uses the same underlying NSDUH microdata to provide state-level estimates for a wider range of measures. Unlike the published NSDUH small area estimates, the R-DAS imposes suppression constraints for cells with small sample sizes and only provides state-level estimates in non-overlapping two-year averages (e.g., 2002-2003, 2004-2005). We use the R-DAS to obtain information on youth perceptions about marijuana to better understand the potential mechanisms by which changes in legal marijuana markets affect youth marijuana use. Our outcome variables are the share of youths aged 12-17 who agree with statements related to marijuana access and risk. For access, they include: (1) marijuana is “somewhat” or “very” easy to obtain, (2) most students their age use marijuana, and (3) they purchased marijuana in the past year. For risk, they include: (1) monthly marijuana use poses a “great risk,” (2) their friends “somewhat” or “strongly” disapprove of trying marijuana, and (3) their state’s maximum penalty for possessing one ounce of marijuana is prison time.

Marijuana Prices:

Data on marijuana prices were collected from two sources, High Times magazine (1997-2011) and Priceofweed.com (2010-2013).¹¹ High Times is an online magazine where users can submit the price they paid for their last marijuana purchase. Following prior work (Jacobson 2004; Anderson et al. 2013), we classified marijuana as “high” or “low” quality using strain name. Priceofweed.com is a website that collects user-submitted data in real-time on the price of marijuana purchases and classifies them into “high”, “medium”, or “low” quality. Outlying price values (below the 1st and above the 99th percentile) were dropped and then data were aggregated to the state-quarter level. Because medicinal marijuana supply tends to be of higher-potency products and high-quality strains (Anderson et al. 2013; Sevigny et al. 2014), and both sources have far more records on “high” versus “low” quality items, we focus our analyses on price per gram for “high” quality marijuana.¹²

Traffic Fatalities:

We use data from 1999 to 2013 from the Fatality Analysis Reporting System (FARS), a census of motor vehicle fatalities in U.S. states and Washington, D.C. that has been collected annually by the National Highway Traffic Safety Administration (NHTSA) since 1975. Drawing on information obtained from police reports and other administrative records, the FARS provides detailed information on the circumstances of the accident, in addition to information on the characteristics of involved vehicles, vehicle occupants, and non-occupants. We use this information to analyze traffic fatalities separately by age of the driver for all accidents, focusing on age of the driver to better reflect which individuals are changing their behavior in response to increased medical marijuana availability. While FARS provides information on both alcohol and drug involvement in fatal accidents, we only present results specific to alcohol-involved accidents; testing for cannabinoids among traffic accident decedents varies widely across jurisdictions and over time (Slater et al. 2016), these decisions are likely endogenous to changes in marijuana policy and prevalence of use, and NHTSA explicitly recommends against comparing drug testing result data in FARS across jurisdictions or across years (Berning et al. 2022).

IV. Empirical Framework

We use three main frameworks to assess how use and negative health externalities respond to medical marijuana market size. We first establish that our measure of market size responds to federal enforcement efforts. Next, we exploit this natural experiment to test how changes in market size affect youth use. Finally, we assess changes in fatal car accidents in response to changes in market size.

A. Medical Marijuana Market Size and Federal Enforcement

To isolate supply-side drivers of growth in legal medical marijuana markets, the ideal policy variation would exogenously shift production costs in some medically legal states but not others. We predict that the Ogden and Cole Memos approximated this ideal by differentially shifting production costs based on pre-existing state supply regulations. In states with loose medical marijuana production limits, the 2009 Ogden Memo induced producer entry and increased supply. The 2011 Cole Memo reversed this trend, with a larger chilling effect expected to occur in loosely regulated states.

To provide descriptive evidence in support of these hypotheses, we estimate a dynamic event study following the recommendations of Borusyak et al. (2021) to avoid identification issues (although we note that in our context we do not have staggered adoption). Here, the outcome is the monthly medical marijuana patient registration rate, and the policy change of interest is the Ogden Memo release in October 2009. Our interest with this analysis is to show whether loosely regulated MML states experienced a differential time path in registration rate trends following the Ogden and Cole Memos.

To better align with our marijuana consumption data, which are measured at a coarser time period unit and over a shorter time frame, we then approximate this event study regression model by estimating a more parsimonious (but more restrictively parameterized) linear trend-break model:

$$\begin{gathered} R{R}_{jt}={\gamma }_0+{\gamma }_1[m-Ogden{]}_t\ast Loos{e}_j+{\gamma }_2[m-Cole{]}_t\ast Loos{e}_j \\ +\delta X_{jt}+u_j+v_t+{\mu }_{jt} \end{gathered}$$ (1)

where RR_jt is patient registration rate in state j at time t, [m – Ogden]_t is the number of months following the Ogden Memo, [m – Cole]_t is the number of months following the Cole Memo, and Loose_j is an indicator variable for states with loose supply restrictions as of 2009. The model controls for state and year fixed effects, as well as time-varying state covariates X_jt described below.

B. How does Market Size Affect Youth Use?

Our main empirical framework uses the natural experiment of changes in market size, as proxied by registration rates, in response to changes in federal enforcement, which varies across states based on their production limits. Specifically, the following model is estimated separately by age group:

$$Y_{jt}=\alpha +\beta R{R}_{jt}+\gamma X_{jt}+u_j+v_t+{\epsilon }_{jt}$$ (2)

where Y_jt is our outcome in state j in year t and RR_jt is the per capita registration, where our outcomes include marijuana use, prices, and attitudes.¹³ The coefficient β identifies the percentage point change in the outcome due to a one percentage point increase in registration rates. State u_j and time v_t fixed effects control for time-invariant state characteristics and national trends. For all specifications, to account for heteroskedasticity and serial correlation, robust standard errors are clustered at the state level (Bertrand et al. 2004).

The main threat to identification is that registration rates and illegal youth use may be jointly determined by time-varying unobservables. Our identification strategy is to model youth use with a selection on observables framework. In addition to state u_j and time v_t fixed effects, we include time-varying state covariates X_jt that potentially affect marijuana use. These covariates can be categorized as: (1) demographics influencing marijuana use, (2) economic characteristics, and (3) substance-related policies influencing marijuana consumption. Comparing mean differences in Table 1, we see that states with registered patients have higher levels of past-month marijuana use prevalence for all age groups, are more likely to have decriminalized marijuana, have higher cigarette taxes, and have younger and more male populations. To address this, all regressions control for the state covariates listed in Table 1.

Even after controlling for state and year fixed effects and state-year covariates, we may remain concerned about endogeneity between youth recreational marijuana use and registration rates.¹⁴ For instance, β will be biased upward if changes in perceived health risks from marijuana use led to changes in both medical marijuana uptake by adults and youth use. We offer two alternative estimation strategies to address endogeneity concerns.

The first is an instrumental variables approach, where we instrument for registration rates by way of two-stage-least-squares using Equation (1) as the first-stage specification. The instruments are the timing of the federal memos and differences in initial MML supply restrictions. The instrumental variable estimates are valid as long as the exclusion restriction is satisfied. In the context of Equations (1) and (2), this occurs if $\mathbb{E}[{ϵ}_{jt}{Z}_{jt}\mid {\mu }_j,v_t,X_{jt}]=0$, where Z_jt are the instruments of the interaction between (1) a time-invariant indicator for whether a state eventually enacted an MML with supply restrictions and (2) trend breaks based on the exogenous timing of the Ogden and Cole Memos. As state and year fixed effects are included, the identification is not threatened by level differences between states or by national trends in marijuana consumption (e.g., state geographic proximity to marijuana exporting countries, or national trends in support for marijuana legalization). However, the exclusion restriction will be violated if changes in federal enforcement following the Ogden and Cole Memos had differential effects on demand in states with loose compared to strict production restrictions through channels other than exposure to legal marijuana supply.

Our second specification is similar to our main specification, described in Equation (2), except that we also control for perceived risk of using marijuana. The measure we use is the percent of the relevant-aged population perceiving marijuana use as posing “great risk.” Given it is possible that perceived risk is also affected by market size (i.e., perceived risk is endogenous to patient registration rates), including perceived risk as a covariate is intended to partial out the component of market size reflecting risk perceptions in order to generate an estimate of β that more accurately isolates the role of market access or supply exposure in influencing adolescent marijuana use. Including risk perception as a control should give a sense of how use changed in response to market size, holding risk perceptions constant.

C. Externalities from Youth Use

To study whether increased marijuana use generates health externalities, we examine how growth in the legal market affects traffic fatalities. Since fatalities are count data (non-negative integer values) with over-dispersion, we employ a negative binomial model, which combines the Poisson distribution of event counts with a gamma distribution of the unexplained variation in the true mean event counts.¹⁵ Table D.2 presents estimates from Poisson regressions.

In this model, for each cause of death, the mean number of deaths μ_jt in state j at time t is given by the following equation:

$$E({\mu }_{jt})=exp(\beta R{R}_{jt}+\gamma X_{jt}+u_j+v_t+ln(po{p}_{jt}))$$ (3)

where ln(pop_jt) is the natural log of the population of state j in year t and its coefficient is constrained to 1, which allows estimated effects (β) to be expressed in terms of mean proportionate changes in the death rate associated with a one percentage-point change in the registration rate. To control for average state characteristics and national trends, state and year fixed effects are included rather than modifying the likelihood function, as recommended by Allison and Waterman (2002). The other explanatory variables are as in Equation (2), although we include additional state-level traffic-related covariates (see Anderson et al. (2013)).

V. Results

We begin by presenting results from the first stage, where we relate changes in federal enforcement to medical marijuana patient registration rates. Next, we present results for marijuana consumption outcomes, as well as exploration of potential mechanisms that may drive these results. After exploring potential mechanisms, we assess mortality from motor vehicle accidents.

A. Market Size, Federal Memos, and Use

Figure 1 explores registration rate trends over time among loosely regulated MML states relative to other states using an event study framework. Panel A presents estimates for loosely regulated states compared to all other states (strictly regulated or non-MML), and Panel B presents estimates for loosely regulated states compared to strictly regulated states only; both figures show the timing of the Ogden Memo (green line) and Cole Memo (red line). Small but statistically significant evidence of trends over the pre-period in both Panels (p<0.001) suggests that loosely regulated MMLs had lower registration rates than strictly regulated MMLs (averages of the lead coefficients are −0.13 and −0.12 for Panels A and B, respectively) that were increasing prior to the Ogden Memo. However, this pre-trend appears negligible compared to the large, significant differential registration rate trends experienced in states with loose production limits following the Ogden Memo (p<0.001; averages of the lag coefficients are 1.52 and 1.38 for Panels A and B, respectively). There does seem to be evidence of anticipatory behavior prior to the Ogden Memo, although the timing of this seems to coincide with a March 2009 statement by Attorney General Eric Holder that federal agents would only target marijuana distributors who violated both federal and state law. For the purposes of our main estimation strategy, which relies on annualized data, this within-year anticipatory behavior should have little effect on our results.

Figure 1: Trends in Medical Marijuana Patient Registration Rates.

Notes: Green line marks the month of the Ogden Memo reducing federal enforcement (Oct 2009), and red line marks the month of the Cole Memo reinstating federal enforcement (June 2011). Coefficients and 95% confidence intervals are from a model predicting state registration rates (number of registered patients per 100 adults) as a linear function of leads and lags relative to the Ogden Memo (excluding two leads), interacted with an indicator for whether a state had loose regulations on medical marijuana producers prior to the memo; models control for state and year-month fixed effects. Panel (b) excludes non-MML states.

For loosely regulated states, the 2011 Cole Memo had similar effects in the opposite direction.¹⁶ This likely reflects both direct actions taken by the federal government to shut down large-scale suppliers as well as perceptions that suppliers were now subject to federal enforcement risk. However, there were also some state-level policy changes taken in light of the shifting federal stance (e.g., Montana Senate Bill 423) that may have impacted both medical marijuana supply and demand; and in sensitivity analyses, we control for state-level policy changes over the period to address this concern. Overall, however, given our focus on how medical marijuana market size affects adolescent marijuana use, it is important to highlight that changes in medical marijuana market size do not reflect changes in youth demand since restrictions on the conditions by which youth under age 18 can access medical marijuana legally mean minors represent a negligible fraction of registered patients.

Table 2 presents results from the linear trend break model specified in Equation 1, reporting both first-stage and reduced-form estimates of how the 2009 Ogden and 2011 Cole Memos changed registration rates and marijuana use across states with and without loose production limits. We find that between the Ogden and Cole Memos, loosely regulated states saw an average increase of 2% of the adult population register relative to other states. In contrast, after the 2011 Cole Memo these states saw reductions in registration rates, declining on average by 0.03% monthly.¹⁷ In Table A.6, we further show that our categorization of loose regulations based on theoretical production limits fits the registration rate data better than other potential policy classifications (e.g., based on qualifying conditions).

Table 2:

Trend Break at Federal Memos

	Registration	Past-Month Use			Past-Year Initiation
	Rate	12-17	18-25	26+	12-17	18-25	26+
[m-Ogden]*LooseReg	0.110*** (0.023)	0.027** (0.012)	0.146*** (0.032)	0.077*** (0.024)	0.008 (0.029)	0.046* (0.026)	0.001* (0.001)
[m-Cole]*LooseReg	−0.137*** (0.044)	−0.013 (0.024)	−0.182** (0.089)	−0.053 (0.047)	−0.015 (0.066)	−0.096 (0.064)	0.001 (0.001)

Notes and sources: For comparison with the annualized marijuana consumption analyses, the analyses of registration rates (i.e., the number of registered patients per 100 adults) have been restricted to observations in December of each year spanning 2002 to 2013. Thus, N=576 (47 states + DC, 12 years from 2002-2013) for each regression. For registration rate results with interactions by strict regulatory structure or using monthly data, see Table A.3 and Table A.4, respectively. Regressions include state and year fixed-effects and state-level covariates (listed in Table 1). [m – Ogden] is the number of months following the Ogden Memo, which relaxed federal enforcement policy, and [m – Cole] is the number of months following the Cole memo, which increased federal enforcement policy. LooseReg_j is an indicator variable for states with MMLs prior to the Ogden Memo where caregivers were allowed to serve an unlimited number of patients, or where caregivers were allowed to serve multiple patients and state-licensed dispensaries were not legal. Robust standard errors (in parentheses) are clustered at the state-level.

^***p<0.01

^**p<0.05

^*p<0.1

Federal memos substantially altered the size of the legal market, with larger effects in states with looser supply regulations. Columns two through seven explore the relationship between the Ogden and Cole memos and past-month use and past-year initiation by age group. The direction of the response —increase after Ogden, decrease after Cole —is consistent across age groups, though the magnitude of the response and its precision varies. In general, the adult groups are most sensitive to changes in market size, and estimates for past-month use are more precise. The small and statistically insignificant estimates for adolescent initiation suggest that changes in medical marijuana market size may have greater impact on the intensive versus extensive margin of adolescent use. Additionally, differential effects of the Ogden Memo are larger and more precise than those of the Cole Memo, with the exception of effects on past-month use among young adults. This could reflect, for example, a larger supply response to the Ogden versus Cole memo or asymmetric consumption elasticities to increases versus decreases in the cost of marijuana (e.g., due to habit formation or changes in tolerance). Raw data on use and initiation relative to the federal memos are shown in Figure E.1.

B. Youth Marijuana Use and Market Size

Table 3 reports the estimated effects of growth in legal medical marijuana market size on past-month marijuana use, by age group. We find that growth in registration rates significantly increases past-month marijuana use for all age groups, with the largest effects for older adults. An additional one percentage point of the adult population registering as medical marijuana patients predicts a significant increase in reported past-month use of 8% among 12-17 year olds, 7% for 18-25 year olds, and 23% for adults over age 25. Estimates remain significant, although magnitudes are reduced, in specifications that control for state-specific linear trends (Panel B).

Table 3:

Change in Use with Change in Medical Patient Registration Rate

	Ages 12-17			Ages 18-25			Ages 26+
Panel A: No State-Specific Trends
Registration Rate	0.534*** (0.190)	0.494** (0.204)	0.465** (0.195)	1.258*** (0.307)	1.207*** (0.354)	1.126*** (0.292)	1.010*** (0.214)	0.931*** (0.232)	0.866*** (0.233)
MML=1		0.355 (0.273)	0.327 (0.287)		0.452 (0.547)	0.374 (0.576)		0.698*** (0.192)	0.636*** (0.207)
Legal disp open=1			0.267 (0.396)			0.738 (0.641)			0.591 (0.357)
Panel B: With State-Specific Trends
Registration Rate	0.364*** (0.105)	0.363*** (0.105)	0.355*** (0.104)	0.769*** (0.252)	0.778*** (0.242)	0.739*** (0.244)	0.420*** (0.121)	0.422*** (0.121)	0.400*** (0.109)
MML=1		−0.072 (0.270)	−0.074 (0.271)		1.155* (0.602)	1.144* (0.600)		0.255 (0.156)	0.249 (0.162)
Legal disp open=1			0.169 (0.110)			0.769 (1.067)			0.439 (0.405)
Mean Outcome	7.1	7.1	7.1	17.3	17.3	17.3	4.4	4.4	4.4

Notes and sources: N=576 (47 states + DC, 12 years from 2002-2013). Registration rate is the number of registered medical marijuana patients per 100 adults in state j at time t. Regressions include state and year fixed effects and the time-varying state covariates listed in Table 1. Robust standard errors (in parentheses) are clustered at the state-level.

^***p<0.01

^**p<0.05

^*p<0.1

Table 3 also compares the effects of medical marijuana market growth to the effects of MML enactment and the opening of legally protected dispensaries. We fail to find that MMLs or the opening of legally protected dispensaries have a statistically significant effect on youth use, with or without state trends. Including state-specific linear trends causes the estimates for MML passage to become small and insignificant and, for adolescents, the coefficient switches sign. In contrast, with or without state trends, estimates of the effects of registration rates remain positive and significant for all age groups. The registration rate estimates are less sensitive to trend inclusion than those of the binary policy variables, which could be due to the non-monotonicity of trends in registered patient counts or the binary measures confounding preexisting trends with the dynamic effects of the policy (Wolfers 2006). Additionally, the magnitude of the coefficient on registration rate fails to change much with the inclusion of the binary policy variables. Similar stability of estimates on registration rates is shown in models that control for other time-varying features of MML policy, such as non-specific pain provisions (see Table B.3).

Table 4 instead examines how changes in legal medical marijuana market size affect the extensive margin of use. The effects on past-year initiation are similar to those for past-month use, but smaller in magnitude: an increase of 6% in the share of 12-17 year-olds, 5% for 18-25 year-olds, and 22% for adults over age 25. Again, results are relatively consistent regardless of whether we include controls for MML enactment, the opening of legally protected dispensaries, or the other MML provisions included in Table B.3. However, results for the relationship of legal market growth with marijuana use initiation are somewhat more sensitive to the inclusion of state-specific trends; in specifications that control for state-specific linear trends (Panel B), the magnitudes of our point estimates are similar, but the standard errors are greatly increased such that the associations of registration rates with initiation are no longer significant for adolescents or young adults.

Table 4:

Change in Initiation with Change in Medical Patient Registration Rate

	Ages 12-17			Ages 18-25			Ages 26+
Panel A: No State-Specific Trends
Registration Rate	0.345*** (0.079)	0.304*** (0.089)	0.299*** (0.095)	0.356** (0.141)	0.318** (0.145)	0.349** (0.150)	0.037*** (0.011)	0.032*** (0.011)	0.030** (0.012)
MML=1		0.362 (0.263)	0.357 (0.267)		0.339 (0.235)	0.368 (0.239)		0.041* (0.021)	0.039* (0.021)
Legal disp open=1			0.051 (0.239)			−0.280 (0.204)			0.021 (0.021)
Panel B: With State-Specific Trends
Registration Rate	0.469 (0.332)	0.471 (0.333)	0.490 (0.323)	0.393 (0.398)	0.394 (0.402)	0.416 (0.397)	0.018*** (0.004)	0.018*** (0.004)	0.017*** (0.004)
MML=1		0.176 (0.249)	0.182 (0.247)		0.165 (0.405)	0.172 (0.405)		0.011 (0.014)	0.011 (0.014)
Legal disp open=1			−0.393 (0.459)			−0.428 (0.449)			0.017 (0.014)
Mean Outcome	5.8	5.8	5.8	6.9	6.9	6.9	0.2	0.2	0.2

Notes and sources: N=576 (47 states + DC, 12 years from 2002-2013). Registration rate is the number of registered medical marijuana patients per 100 adults in state j at time t. Regressions include state and year fixed effects and the time-varying state covariates listed in Table 1. Robust standard errors (in parentheses) are clustered at the state-level.

^***p<0.01

^**p<0.05

^*p<0.1

Table 5 reports sensitivity analyses comparing our main results (columns 1) to results using federal memo trend breaks as instruments (columns 2) and to results from regression models that control for risk perceptions (columns 3). While our first-stage F-statistic (27.8) exceeds common standards for “strong” instruments, recent guidance suggests the IV results may still have issues of power and inference, and thus should be taken with caution (Lee et al. 2022; Keane and Neal 2021). For past-month use, the IV and risk-adjusted results tend to be somewhat smaller than the main results, although they are not qualitatively or statistically different than those from the main specification. This suggests that, for the intensive margin of use, changes in registration rates and use are not being driven by some unobserved factor correlated with both recreational and medical demand for marijuana.

Table 5:

Robustness Checks Accounting for Potentially Endogenous Drivers of Registration Rates

	Age 12-17			Age 18-25			Age 26+
	Main (1)	IV (2)	-Risk (3)	Main (1)	IV (2)	-Risk (3)	Main (1)	IV (2)	-Risk (3)
Panel A: Past-Month Use
Registration Rate	0.534*** (0.190)	0.393* (0.229)	0.494** (0.202)	1.258*** (0.307)	1.320*** (0.276)	1.258*** (0.321)	1.010*** (0.214)	0.994*** (0.236)	0.986*** (0.217)
Hansen J p-value		0.35			0.99			0.14
Endogeneity p-value		0.18			0.58			0.96
Panel B: Past-Year Initiation
Registration Rate	0.345*** (0.079)	0.043 (0.104)	0.312*** (0.074)	0.356** (0.141)	0.146 (0.139)	0.356** (0.152)	0.037*** (0.011)	0.023*** (0.007)	0.037*** (0.011)
Hansen J p-value		0.87			0.28			0.10
Endogeneity p-value		0.05			0.88			0.03

Notes and sources: N=576 (47 states + DC, 12 years). CA, ME and WA are excluded due to lack of reliable registration data. Registration rate is measured as the number of registered patients per 100 adults in the state. Regressions include state and year fixed effects, and the time-varying state covariates listed in Table 1. For all regressions, robust standard errors (in parentheses) are clustered at the state level. Column (1) replicates the primary results from Tables 3 and 4. Column (2) presents second-stage results from IV estimation using the trend break analysis of Table 2 as the first stage (first-stage F-statistic equal to 27.8). Column (3) replicates the analyses of Column (1) but includes an additional covariate representing marijuana risk perception of the relevant age group.

^***p<0.01

^**p<0.05

^*p<0.1

In contrast, our results for initiation among adolescents and young adults become small and statistically insignificant when instrumenting for registration rates. This suggests that changes in youth marijuana use initiation associated with changes in the size of the medical marijuana market may be driven less by changes in marijuana supply than by changes in other factors correlated with medical marijuana patient registration rates through non-supply-specific channels, such as social approval. Additionally, these findings suggest that increased past-month use of marijuana among youth may be driven primarily by increased frequency of use among those who have previously tried marijuana rather than by inducing new consumers into the market.

C. Mechanisms: Prices, Access, and Attitudes

Although age restrictions are a powerful tool to reduce youth use of tobacco, alcohol, and other substances, social and illicit markets persist and remain difficult to govern (DiFranza 2012; Gendall et al. 2014). As medical marijuana markets grow, they may lower search costs, increase product variety, or decrease the quality-adjusted price of illegal marijuana (Alford 2013; Anderson et al. 2013). Table 6 reports estimates for changes in marijuana price in response to changes in medical market size. Across two data sources, we find that prices decrease with medical marijuana market size, consistent with other studies that assess price changes (Anderson and Rees Forthcoming).

Table 6:

Legal Market Growth and Price of High-Quality Marijuana

	Priceofweed.com (2010-2013)	High Times (1996-2011)
Registration Rate	−0.024** (0.011)	−0.071*** (0.026)
Average price per oz	327.52	381.84
Average price per oz (Loose)	297.84	336.84
Average price per oz (Strict)	322.37	375.44
N	584	1193

Notes and sources: Outcome is the natural log of the price per ounce in nominal US dollars. Registration rate is measured as the number of registered patients per 100 adults in the state. Excludes CA, ME, and WA. Data are at the state-quarter level. Regressions include state, year, and quarter fixed effects and are weighted by the number of user submissions in the state-quarter. Standard errors in parentheses are clustered by state.

^***p<0.01

^**p<0.05

^*p<0.1.

Beside price, medical marijuana supply may spillover to the illegal market serving adolescents through resale or sharing of medical marijuana by legal users. The leftmost columns of Table 7 explore how changes in medical marijuana market size affect the share of youths aged 12-17 that agree with various statements related to marijuana availability. A one percentage-point rise in registration rates significantly increases the share of 12-17 year-olds who believe marijuana is easy to obtain (3%), who believe most students their age use marijuana (5%), and who report buying marijuana in the past year (6%); these estimates are jointly significant with a p-value of 0.02.

Table 7:

Effects of Legal Market Growth on Adolescent Perceptions of Risk and Access

	Adolescent Access			Adolescent Risk Perceptions
	Easy to Obtain	Most Students Use	Bought in Past Year	Great Risk Monthly Use	Friends Disapprove	Max Penalty is Prison
Registration Rate	1.370*** (0.292) [2.7]	1.287** (0.600) [5.2]	0.428** (0.186) [6.0]	−0.824** (0.326) [−2.6]	−0.303 (0.551) [−0.0]	−0.461 (0.378) [−1.3]
Joint P-value		0.02**			0.12
Mean Outcome	50.0	24.9	7.1	31.9	65.2	34.2

Notes and sources: N=288 (47 states + DC, 6 sets of non-overlapping two-year averages, e.g. 2002-2003, 2004-2005, etc.). Registration rate is measured as the number of registered patients per 100 adults in the state. All specifications include state and year fixed effects and time-varying state covariates. The dependent variable is the prevalence measure for 12-17 year olds from the NSDUH Restricted-use Data Analysis System (2014b). Standard errors in parentheses are clustered by state, and implied percent changes are in square brackets.

^***p<0.01

^**p<0.05

^*p<0.1.

Columns on the right report effects on risk perceptions. If registered adult users are more visible to adolescents, youths may increase consumption due to lower perceived risks associated with social disapproval, formal sanctions, or health consequences. While growth in the legal market affects adolescent perceptions of both risk and availability, the results of Table 7 suggest changes in access may be more important. Estimates for statements related to access are larger and more precise than those for risk perceptions. These results support ease of access to marijuana as the primary channel for increased use rather than decreased risks of legal penalties or social disapproval.¹⁸ Given challenges with isolating changes in perceived risk from perceived access (e.g., perceptions of greater access may lead youth to think that marijuana is less risky), these results are intended as suggestive evidence only. However, when instrumenting for registration rates using the federal memos, results are similar for perceived access outcomes (Table E.2), further supporting that supply-side shocks to medical marijuana markets influence youth access to marijuana.¹⁹

D. Potential Health Externalities

To provide some evidence for how medical marijuana market size affects health externalities, Table 8 presents initial results of the effects of registration rates on traffic fatalities by age of driver. Estimates are shown separately for total, weekend, weekday, daytime, nighttime, alcohol-involved, and BAC>0.08 crashes.²⁰

Table 8:

Registration Rates and Traffic Fatalities, by Age of Driver Involved

	Age 15-20	Age 21-24	Age 25-44	Age 45-64
Panel A: All Accidents
Registration Rate	0.067** (0.033)	0.052 (0.065)	0.018 (0.013)	−0.013 (0.032)
Mean outcome	240.6	207.0	633.5	441.7
Panel B: Weekday Accidents
Registration Rate	0.020 (0.032)	−0.015 (0.065)	0.031* (0.018)	−0.041 (0.032)
Mean outcome	129.3	107.5	360.7	276.9
Panel C: Weekend Accidents
Registration Rate	0.126*** (0.048)	0.134* (0.073)	0.005 (0.025)	0.042 (0.046)
Mean outcome	111.2	99.24	272.1	164.4
Panel D: Daytime Accidents
Registration Rate	0.013 (0.029)	−0.001 (0.092)	0.045* (0.025)	0.007 (0.037)
Mean outcome	106.4	79.17	298.6	257.4
Panel E: Nighttime Accidents
Registration Rate	0.115** (0.055)	0.099 (0.078)	0.000 (0.027)	−0.033 (0.035)
Mean outcome	133.3	127.0	332.4	183.0
Panel F: Alcohol-Involved Accidents
Registration Rate	0.054* (0.028)	0.054 (0.093)	0.013 (0.020)	−0.121** (0.048)
Mean outcome	48.80	71.66	189.8	78.95
Panel G: BAC>0.08 Accidents
Registration Rate	0.103*** (0.036)	0.046 (0.101)	0.013 (0.052)	−0.128** (0.063)
Mean outcome	28.47	48.81	136.0	54.40

Notes and sources: N=720 (47 states + DC, 15 years) (NHTSA 2014). Results from negative binomial specification. Includes all fatal accidents. Registration rate is measured as the number of registered patients per 100 adults in the state. All specifications include state and year fixed effects, time-varying state covariates, and state-specific linear trends. Robust standard errors (in parentheses) are clustered at the state level.

^***p<0.01

^**p<0.05

^*p<0.1.

Growth in medical marijuana market size increases fatal motor vehicle accidents for drivers aged 15-20. Effects on all, weekend, nighttime, and alcohol-related accidents are significant. A one percentage-point increase in registration rates increases fatalities involving young drivers by 15% on weekends and 13% at nighttime respectively. For alcohol-involved accidents and accidents where the driver had a BAC above 0.08, effects are 6% and 12%, respectively. Our result suggesting a potential complementarity between alcohol and marijuana for youth is in contrast to Crost and Guerrero (2012), who find that marijuana use decreases after young adults reach the minimum drinking age. However, we note that we are not directly testing for complementarities. Specifically, while the effects of marijuana on driver impairment are modest when compared to the effects of alcohol (Sewell et al. 2009; Downey et al. 2013), using marijuana and alcohol together may have additive or multiplicative effects on driver impairment (Sewell et al. 2009; Downey et al. 2013). Therefore, if youth marijuana use is increasing in contexts where alcohol consumption also occurs (e.g., unsupervised parties; Lipperman-Kreda et al. (2017)), this could increase alcohol-related accidents, as well as those on the weekend or at night, regardless of whether alcohol use is also changing.

In contrast to our results for adolescents and young adults, medical market size has much less of an impact on drivers 21 years or older. Though our cohorts are different, a lack of an increase in traffic fatalities among adults is consistent with Anderson et al. (2013), who find that MML passage significantly reduces traffic fatalities for adults aged 21-39, due in large part due to decreased alcohol use, and Fink et al. (2020), who find no effect of MML on driving under the influence of alcohol for adults.

E. Additional Robustness Checks

Results for past-month marijuana use are similar across specifications that drop time-varying covariates, weight by state population, or impose various sample restrictions, such as excluding states that required substantial imputation of the registration rate data (Tables B.6-B.8). Although estimates become less precise and often insignificant with the inclusion of state-specific trends, results for past-year initiation are also generally robust, with the exception that estimates for younger adults switch sign across specifications (Tables B.9-B.11). We also show that, adjusting for fixed effects, changes in medical marijuana market size are largely unrelated to changes in time-varying characteristics of states that are not included as covariates in our main specifications (Table E.3). One exception is that registration rate trends are significantly associated with the percentage of the state population aged 65 or older. However, this association becomes small and insignificant once state-specific trends are adjusted for, and it is unlikely that this aspect of the state’s age composition represents a substantial confound for youth marijuana use.

We run a series of additional robustness checks. For our primary outcomes, we re-estimate our main models, dropping each MML state one at a time to assess whether any single state is driving our results. Point estimates and 95% confidence intervals are shown in Figures B.1 (marijuana consumption) and D.1 (traffic accidents). Results are all well-within the confidence intervals of our main estimates. Finally, we use randomization inference procedures to assess the validity of our estimated inferential statistics. Specifically, we re-estimate our primary analyses, permuting registration rate data to be randomly reassigned to random states, repeating this 1000 times per analysis, and using the resultant empirical distribution of estimates to conduct hypothesis testing for the main estimates. Table E.4 shows that p-values from randomization inference largely confirm the regression-based inference, with a key exception: based on randomization-inference p-values, harmful effects on youth alcohol-involved and high-BAC fatal traffic accidents are no longer significant. Given the relatively rare nature of these outcomes for youth, we explored whether the inflated Type I error rates from the cluster-based standard error adjustment were due to potential model overfitting by re-estimating a more parsimonious model (retaining state and year fixed effects and state-specific linear trends) and repeating the permutation exercise. Randomization-based and cluster-based p-value correspond better in this more parsimonious specification, with both supporting a significant positive association of medical marijuana market size with alcohol-involved and high-BAC fatal traffic accidents involving youth drivers (Table E.5).

VI. Conclusion

Using newly collected data and a novel identification strategy, we study how adolescent marijuana use and associated public health externalities change in response to medical marijuana market size. Contrary to prior research that focused on passage of MMLs and found no effect, we find youth marijuana use responds to medical marijuana market size. Our estimates suggest that if a state where marijuana is illegal were to reach the median per capita medical marijuana registration rate, past-month marijuana use would increase by about 8% among 12-17 year-olds. Likely mechanisms are increases in youth access to marijuana through social markets and decreases in price, which our results suggest matter more for the intensive margin of use rather than inducing marijuana use initiation.

Growth in the legal market increases traffic fatalities by young drivers overall, on weekends, at night, and involving alcohol. This is a novel result given that most studies consider how marijuana legalization affects adults (Anderson and Rees Forthcoming). Like these studies, we find that medical marijuana market size is associated with fewer fatal car accidents for adults. We suspect the difference in effects for youth and adults may be due to heterogeneity in risk-taking and differences in the frequency and setting of co-use of marijuana and alcohol. As policies like Colorado’s HB 1230 relax restrictions on public consumption, our results point toward the need for more research on how these policies would affect marijuana and alcohol co-use and driving, particularly for youth and young adults.

Our paper contributes a novel approach for estimating how legal market size affects marijuana use, with some potential limitations. First, to interpret our estimates as strictly representing the effects of supply-side legal marijuana market exposure on adolescent behaviors, our identification strategy assumes that changes in federal enforcement were uncorrelated with state-level demand. Second, potential cross-border spillovers of marijuana supply would tend to bias our estimates downward. Finally, marijuana use is self-reported and crudely measured (i.e., changes in prevalence). Though commonly used in the literature, these measures are crude and make it difficult to identify problematic or harmful use. They are also subject to reporting bias. Future work should consider richer data on use, including frequency, context, potency, acquisition patterns, and co-use with alcohol.

With these limitations in mind, our work has implications for optimal marijuana policy. While marijuana may help older adults stay longer in the workforce (Nicholas and Maclean 2019) and may offer therapeutic benefits (Bradford and Bradford 2016), our paper suggests optimal marijuana policy should explicitly be designed to render illicit access more costly for youth. One intriguing idea, put forth by Anderson et al. (2019), is that recreational legalization may help if recreational dispensaries crowd out illicit marijuana dealers and make it more difficult for teens to buy marijuana. Indeed, unlike the loosely regulated medical marijuana markets evaluated in our study, all states that have legalized commercial sales of marijuana for nonmedical purposes have established state-licensing schemes and marijuana product tracking systems, which may help prevent diversion to youth markets. That said, given recreational legalization is relatively recent and occurred first in states that already had lax medical marijuana regimes, early results on adolescent outcomes may evolve over time as commercial markets expand and prices fall.

Our findings also have implications for interpreting the existing literature evaluating the effects of marijuana liberalization policies and for designing future evaluations. Marijuana policy research, and policy research more generally, often uses a binary treatment variable and the canonical difference-in-differences framework to estimate causal effects. Studies using this framework have largely found null effects of MMLs on youth outcomes, but our findings suggest that past results may not adequately capture effects due to changes in the supply of marijuana in a state, which may be a particularly important mechanism for influencing youth substance use behavior. Particularly as states continue to consider legalizing commercial markets for marijuana, future work should use approaches that can account for the heterogeneous dynamics of evolving marijuana markets to investigate policy implications for marijuana-specific and related health outcomes.

Finally, our results highlight the vital role that changes in federal policy, even if not formalized in law, play in shaping marijuana markets and use at the state level. As policymakers consider federal legalization, the implications for adolescent substance use outcomes may hinge critically on how suppliers are regulated, and the mechanisms in place to monitor and enforce such regulations. We expect our understanding of state marijuana legalization may change dramatically under a more liberal federal environment.