PreprintPDF Available

Light cannabis and organized crime. Evidence from (unintended) liberalization in Italy

Abstract and Figures

The effect of marijuana liberalization on crime is object of a large interest by social scientists and policy-makers. However, due to the scarcity of relevant data, the displacement effect of liberalization on the supply of illegal drugs remained substantially unexplored. This paper exploits the unintended liberalization of cannabis light (C-light, i.e. with low THC) occurred in Italy in December 2016 by means of a legislative gap, to assess its effect in a quasi-experimental setting. Although the liberalization interested all the Italian territory, the intensity of liberalization in the short-run varied according to the pre-liberalization market configuration of grow-shops, i.e. shops selling industrial canapa-related products that have been able to first place the canapa flowers (C-light) on the new market. We exploit this variation in a Differences-inDifferences design using a unique dataset on monthly confiscations of drugs at province level (NUTS-3 level) over the period 2016-2018 matched with data on the geographical location of shops and socio-demographic variables. We find that the legalization of C-light led to a reduction of 11-12% of confiscation of marijuana per each pre-existing grow-shop and a significant reduction of other canapa-derived drugs (plants of cannabis and hashish). Back-to-envelope calculations suggest that forgone revenues for criminal organizations amount to at least 160-200 million Euros per year. These results support the argument that, even in a short period of time and with an imperfect substitute, the organized crime's supply of illegal drugs is displaced by the entry of official and legal retailers.
Content may be subject to copyright.
Light cannabis and
organized crime: Evidence
from (Unintended) liberalization
in Italy
European Economic Review, 2019
Vincenzo Carrieri, Leonardo Madio
and Francesco Principe
2987
2987
CORE
Voie du Roman Pays 34, L1.03.01
B-1348 Louvain-la-Neuve
Tel (32 10) 47 43 04
Email: immaq-library@uclouvain.be
https://uclouvain.be/en/research-institutes/
lidam/core/reprints..html
Light Cannabis and Organized Crime:
Evidence from (Unintended) Liberalization
in Italy
Vincenzo Carrieri
Leonardo Madio
Francesco Principe
“Magna Graecia” University
HEDG, University of York
RWI Research Network
CORE, Université Catholique de
Louvain
CESifo Research Network
HEDG, University of York
Erasmus School of Economics
HEDG, University of York
Abstract
This paper explores the unintended liberalization of light cannabis that occurred in Italy in December
2016 by means of a legislative gap in order to assess its effect on the illegal supply of marijuana.
Although liberalization interested the entire Italian territory, in the short run, the level of intensity
varied according to the pre-liberalization market configuration of grow shops, i.e., retailers that sold
industrial cannabis-related products. We exploit this variation using a differences-in-differences (DID)
design with a unique dataset on monthly confiscations of drugs at the province level during 20162018,
which is matched with data on the geographical location of shops and socio-demographic variables. We
find that the liberalization of light cannabis led to a reduction of up to 14% in marijuana confiscations
per each pre-existing grow shop and a significant decrease in both other cannabis-derived drugs and in
the number of people arrested for drug-related offences. Back-of-the-envelope calculations suggest that
forgone revenue for criminal organizations amount to at least 90170 million euros per year. These
results support the argument that the supply of illegal drugs is displaced by the entry of official and
legal retailers.
Keywords: cannabis; marijuana light; crime; illegal market; diff-in-diff.
JEL codes: K23; K42; H75; I18.
Please cite this article as: Vincenzo Carrieri, Leonardo Madio, Francesco Principe , Light Cannabis and
Organized Crime: Evidence from (Unintended) Liberalization in Italy, European Economic Review (2019), doi:
https://doi.org/10.1016/j.euroecorev.2019.01.003.
We thank the editor and two anonymous referees for their helpful comments. We are grateful to Giuseppe Bertola,
Fabio Berton, Gianmarco Daniele, Giacomo De Luca, Davide Dragone, Gloria Moroni, Paulo Santos Monteiro,
alongside all participants at the Research Student Workshop in York, 23rd Annual Conference of the Italian Health
Economics Association in Naples, and 29th Conference of the Italian Economics Association and seminar participants
at the University of Turin, University of Calabria, Erasmus school of Economics for their helpful comments.
Leonardo Madio acknowledges financial support from the Scottish Economic Society (Small Grant Scheme).
Department of Law, Economics and Sociology, “Magna Graecia” University of Catanzaro, Viale Europa, 88100
Catanzaro, Italy. Email: vincenzo.carrieri@unicz.it
CORE - Center for Operations Research and Econometrics, Universitè catholique de Louvain, Voie du Roman Pays, 34 -
L1.03.01 1348 Louvain-la-Neuve. Email: leonardo.madio@uclouvain.be
Erasmus School of Economics, Erasmus University Rotterdam, Netherlands. Email: principe@ese.eur.nl
1
1. Introduction
According to the most recent European Drug Report (EMCDDA 2016), “cannabis accounts for the largest
share in value of Europe’s illicit drug market,and it is the most consumed drug worldwide (UNODC 2016).
The illicit drug market is a long-standing problem in several countries. On the one hand, it constitutes
an enormous source of revenue for organized crime; on the other hand, it represents a cost for public
authorities, e.g., for law enforcement and public health reasons. To tackle this problem, some countries
have recently begun implementing a more liberal approach to cannabis consumption by legalizing
and/or decriminalizing its use and commercialization. In the US, recreational marijuana is liberalized in
several states (e.g., Colorado and California), and Canada legalized it in October 2018. Other countries
have, instead, legalized only its medical use, which requires a doctor’s prescription. However, the
discussion about legalization has always been accompanied by divisive arguments. On the one hand,
promoters of legalization usually argue that doing so would crowd out the illicit market, disrupt
organized crime, and reallocate police efforts toward other crimes.
1
On the other hand, opponents of
legalization contend that eliminating the social stigma associated with marijuana consumption would
induce more consumption (Jacobi and Sovinski 2016) and thus lend itself to negative impacts on social
welfare.
Several studies have looked at the effects of legalization by studying its impact on crime (Adda et al.
2014, Shephard and Blackely 2016, Brinkman and Mok-Lamme 2017, Chang and Jacobson 2017,
Gavrilova et al. 2017, Hansen et al. 2017, Chu and Townsend 2018, and Dragone et al. 2018), health-
related issues (DiNardo and Lemieux 2001, Wen et al. 2015, Sabia et al. 2017), consumption (Jacobi and
Sovinsky 2016), and the presence of spillover effects, such as school attendance and academic
achievement (Plunk et al. 2016, Marie and Zolitz 2017), housing prices (Cheng et al. 2018), traffic
fatalities (Anderson et al. 2013, Hansen et al. 2018), and in-migration (Zambiasi and Stillman 2018).
The study of the impact of legalization on violent and non-violent crimes in the US has attracted most
of the attention of economics literature. From a theoretical standpoint, the introduction of legal
marijuana retailers can have several effects on the market. Besides a potential market expansion of
marijuana users, it makes the market more competitive and more transparent and solves the problem of
moral hazard associated with the purity of drugs (see, e.g., Galenianos and Gavazza 2017). Legal
retailers can offer a controlled substitute product, potentially displacing the demand and hence the
equilibrium supply in the illegal market, whereas organized crime often operates in the regime of a
monopoly. This prediction seems to be indirectly supported by empirical evidence. For instance,
Hansen et al. (2017) used a regression discontinuity design (RDD) to study how the legalization of
marijuana impacts “drug tourism.” They explored two different instances of legalization, in Washington
and in Oregon, which showed that when Oregon legalized recreational marijuana, the quantity of
marijuana sold in Washington fell by 41%. Hence, many of Oregon’s consumers were travelling to
Washington to purchase legal marijuana, and interstate spillovers partly displaced the equilibrium
supply in Oregon’s illegal market. Using similar data, Dragone et al. (2018) found that the legalization of
recreational marijuana in Washington resulted in a reduction in thefts and rapes relative to Oregon and
the pre-legalization period. More controversial results have been found concerning the legalization of
medical marijuana. Chu and Townsend (2018) showed that violent and property crimes significantly
decreased only in California but not at the national level. Also in California, a similar pattern was
supported by Chang and Jacobson (2017), who showed that closing marijuana dispensaries generated
an increase in crime in the proximity. More generally, Gavrilova et al. (2017) studied the impact of
medical marijuana laws on drug trafficking organizations in the US. They found that the supply shock
1
In 2016, the National Anti-mafia and Anti-terrorism Directorate (DNA), expressed a positive opinion on the legalization
of marijuana. Apart from the possibility of disrupting revenues of organized crime, which is historically rooted in Italy, the
DNA claimed that it would reduce the disproportion between the monetary and non-monetary costs of law enforcement
and the small results obtained in terms of convicted criminals and drugs confiscated.
2
offered by the introduction of medical marijuana in the US resulted in a reduction in profits and thus
the incentive to settle disputes using violence.
Although these studies advance our knowledge of the impact of liberalization on crime, the interplay
between the illegal and legal markets has not yet undergone a throughout examination. In particular,
due to the scarcity of relevant data, the displacement effect of liberalization on the supply of illegal
drugs remains substantially unexplored. This paper aims to fill this gap by examining the effect of
liberalization on the confiscations of drugs sold in the illegal market and other crime-related outcomes
through a quasi-experiment that focuses on Italy. In December 2016, a legislative gap created the
opportunity to legally sell cannabis with low levels of tetrahydrocannabinol (THC), a psychoactive
chemical. As a result, some start-ups (e.g., Easyjoint, Marymoonlight) entered the market and began selling
light cannabis (C-light). Traditional and online media gave wide coverage of the phenomenon and the
rapid growth of the market. This (unintended) marijuana liberalization represents an exceptional
opportunity to test changes in the (equilibrium) supply of street marijuana in a market where illegal and
legal retailers coexist. Italy is an interesting case study to test this effect due to the presence of strong
criminal organizations that entirely control the market of illegal substances, often in partnership with
international criminal organizations. Moreover, the market of cannabis-derived drugs represents
roughly 91.4% of the illegal drugs confiscated in Italy (Dipartimento Politiche Antidroga 2017) and a
significant source of revenue for these organizations.
While liberalization occurred simultaneously in the entire territory, in the short run, the level of
intensity was not homogeneous. Specifically, from May 2017 onwards, the cannabis flowering process
was completed. Some shops already specialized in the retail of industrial cannabis (i.e., grow shops)
began selling C-light on a franchising base, exploiting large economies of scope, namely the possibility
to sell, by means of liberalization, both cannabis-related products and flowers. These shops are located
in the proximity of cannabis cultivations, which are concentrated in areas close to waterways and humid
soil (see Section 3.1 and Figure 1). In the following months, i.e., after 1 year of post-liberalization, para-
pharmacy, herbalists, and tobacco shops followed suit, exposing the Italian territory to more
homogenous market coverage. However, during the first year after liberalization, the pre-existing stock
of grow shops at the local level largely determined the local availability of C-light. In some places, the
existence of grow shops resulted in a high supply of C-light, while in other places, their absence (or the
presence of very few grow shops) resulted in a low or zero supply of C-light. Consequently, after
liberalization, areas with high numbers of grow shops were more affected by the policy change than
areas with low numbers of grow shops.
We exploit this variation in order to identify the effect of liberalization using a differences-in-
differences (DID) framework. To accomplish this, we use a unique dataset running from 2016 to
February 2018 that was built using several sources of data. Data including the quantity of illicit
substances confiscated in the Italian territory, as well as the number of people arrested for drug-related
offences, are made available by the National Police at the province level (NUTS-3), i.e., the
administrative division of the intermediate level between a municipality and a region. These data were
then matched with provincial data on the number of grow shops that were active in Italy in 2016,
which were collected from official retailers’ websites. Finally, we linked these data to provincial
demographic variables provided by the National Institute of Statistics (ISTAT). An appealing and rare
feature of this unique dataset is the availability of monthly data on confiscations and drug-related
offences. This feature, coupled with the unexpected nature of liberalization, allows us to estimate the
effect of interest within a very short window of time around the policy, when law enforcement and
police effort adjustments were extremely unlikely. In order to rule out any endogenous adjustment,
however, we also control for the provincial monthly number of police operations and find no effect of
the policy on police law enforcement in one of the model specifications. This allows us to interpret any
change in the number of drugs confiscated as changes in the equilibrium supply of illegal marijuana.
3
Under the hypothesis of a common trend in confiscations across provinces with a different number of
grow shops that existed pre-liberalization—which is largely supported in our case (see Section 6)—our
DID strategy allows us to retrieve the causal effect of liberalization on the quantity of illegal drugs
confiscated and other crime-related outcomes. This effect is likely to represent a lower bound of the
displaced supply of illegal drugs because confiscated drugs obviously represent only a share of the total
illegal market.
We find that the liberalization of C-light led to a reduction in the confiscation of illegal marijuana. Our
estimates indicate a decrease of up to 14% in monthly confiscations per each pre-existing grow shop as
a consequence of the unintended policy change. This corresponds to a decrease in elasticity of 3% in
confiscations in response to a 10% increase in the number of grow shops per province. Interestingly,
we find that liberalization also impacted the illegal supply of other cannabis-derived drugs. It led to an
8% reduction in the supply of hashish and a 32% decrease in the number of plants confiscated monthly
per each grow shop. Moreover, while we do not detect any law enforcement adjustment from police
authorities, we find a negative impact of the unintended liberalization on the total number of people
(i.e., -3%), foreigners (-3%), and minors (-15%) arrested for drug-related crimes. This result is
remarkable as these categories are often used by criminal organizations as street drug dealers, and it
provides strong support for our findings: even a mild form of liberalization, such as the one that
occurred in Italy and used an imperfect substitute product of street marijuana, can harm organized
crime. These results are robust in a number of robustness checks, placebo regressions, and alternative
approaches to statistical inference. Back-of-the-envelope calculations on the 106 provinces considered
suggest that forgone revenues for criminal organizations amount to around 90170 million euros per
year.
The rest of this paper is structured as follows: In Section 2, we discuss the (unintended) policy change.
In Section 3, a description of the dataset and main variables is provided. Identification strategy is
presented in Section 4. In Section 5, we present the main results. Section 6 provides some sensitivity
analyses and robustness checks, while Section 7 summarizes and concludes the paper.
2. Institutional setting
In December 2016, the Italian government approved a law devoted to regulating and incentivizing the
production and commercialization of industrial cannabis (also called hemp). Hemp has a variety of
commercial uses, ranging from food (e.g., cannabis flour for pizza) to clothing, from therapy to
construction, as well as biofuel. Italy has a long tradition of the cultivation of hemp. In early 1900 and
prior to World War II, it was the second largest producer in the world, just behind Russia. This is
essentially explained by the presence of several waterways, which were essential for both the production
of vapor energy and yarn processing. Most of the production served the navy, army, railway, hospital,
and tobacco industries. Today, this heritage is reflected by the higher presence of cannabis cultivations
and cannabis-related shops in areas of the country closest to waterways and with humid soils (see
Section 3.1 and Figure 1 for more details).
Industrial cannabis contains a low level of THC, the main psychoactive constituent of marijuana. While
incentivizing the cultivation of cannabis, the 2016 law did not regulate the production of flowers. As a
result of this legislative gap, after a few months, in May 2017, some start-ups saw a profitable
opportunity and began selling cannabis flowers with a low level of THC and a naturally high level of
cannabidiol (CBD). In theory, the flowers could not be consumed or smoked. According to the labels
applied to the pot, C-light could only be used for technical purposes, e.g. as collectors’ items.
Moreover, the way they are commercialized, e.g., in sealed packages that should not be opened in the
streets, differs visibly from the illegal street cannabis. Paradoxically, given its “technical use,” minors of
18 years old could buy C-light but not tobacco.
4
As a matter of fact, (unintended) liberalization of C-light took place in May 2017, when the
fluorescence process was completed, and the flowers were actually commercialized. Indeed, shops
selling cannabis-related products for industrial use (i.e., grow shops) began putting the flowers on the
market. This relies on the possibility of exploiting large economies of scope, namely the possibility of
selling both cannabis-related products and its flowers.
2
As a consequence, in May 2017, the local
availability of C-light was essentially determined by the presence of these shops. As explained above,
the geographical concentration of grow shops is largely historically rooted in relation to the existence of
cannabis cultivations, which tend to be concentrated in areas with humid soil and close to large
waterways. The pre-liberalization market configuration of grow shops, indeed, represents a useful
source of exogenous variation in the local availability of C-light.
Our strategy allows us to retrieve the causal effect of the policy in the short run (i.e., up to 1 year after
liberalization). Indeed, in the long run, the selling of C-light was not only circumscribed to grow shops.
C-light became so popular that herbalist and tobacco shops, along with para-pharmacists, also began
selling it, covering most of the Italian provinces with different intensities and timelines. In February
2018, 87 out of 106 provinces covered in our study were served by at least one in-store retailer.
Italy is an interesting case study for the analysis of the displacement effect of C-light liberalization on
the illegal drug market. Indeed, Italy has historically been pervaded by the presence of organized crime,
with four main criminal organizations born in the southern regions (Camorra in Campania, Sacra
Corona Unita in Apulia, ‘Ndrangheta in Calabria, and Stidda and Mafia in Sicily) but operating in the
entire Italian territory. Drug trafficking is the most significant activity pursued by these organizations
and is often jointly run with other international criminal organizations. Illegal revenue from the
consumption of drugs accounts for 14.2 billion euros in Italy alone. The market of cannabis-derived
drugs represents roughly 91.4% of the entire market of illegal drugs, corresponding to around 3.5
billion euros (DNA 2017).
3. Data
Our empirical analysis is based on a unique dataset recording longitudinal information on all 106 Italian
provinces. We built the dataset by merging information from several sources. Monthly data on the
drugs seized in each province by police forces were made available by the Direzione Centrale per i Servizi
Antidroga (Central Direction for Anti-drugs Services), which plays a role in coordination of the Italian
police forces with respect to anti-drug operations. Our dataset contains information about kilograms of
marijuana, hashish, and the number of cannabis plants seized monthly in each Italian province. For all
provinces, we collected information about the monthly number of anti-drug operations conducted by
police forces and the number of people arrested for drugs-related crimes, including most sensitive sub-
categories, such as foreigners and minors.
Moreover, we collected data about the pre-policy (October 2016) territorial diffusion of grow shops.
These are retailers of cannabis-related products that are used as treatment intensity variables in our
empirical analysis. These data were collected using web scraping techniques, along with data on the
official dealers of C-light after liberalization.
3
This information will not be used in our empirical strategy
2
As most of the early producers were already in the grow shops network, a local grow shop was chosen as the first point for
the commercialization of C-light.
3 Data on the post-liberalization market came from the websites of the main producers (i.e., Easyjoint, Marymoonlight,
RealHemp, XXXJoint), and we accessed archived copies of their early pages using the Internet Archive Wayback Machine
https://archive.org/web/. This is a website that memorizes, at different points in time, the content of a given webpage.
Data were collected on a monthly basis after the policy and using the last accessible page for each month. The data on the
pre-policy number of grow shops comes from http://www.growshopmaps.com/, which maps the grow shops available in
the Italian territory. The last archived copy of the map before the policy is October 2016. Data were also collected for
March 2016 to control for the number of grow shops per province before the (fake) policy used in the placebo analysis.
5
as this might be due to an endogenous entry of these shops on the market. However, we will use these
data for a descriptive analysis in Section 3.1. Data were aggregated at a provincial basis. This led to a
balanced panel dataset composed of roughly 2,700 observations, from January 2016 to February 2018.
Concerning demographic characteristics, we use data from the Italian National Institute of Statistics
(ISTAT) on population size, density, the territorial extensions of the provinces, and the presence of
freight ports.
3.1 Descriptive statistics
The full list of variables included in our dataset is presented in Table 1, along with mean values and
standard deviations. Concerning our outcomes of interest, monthly confiscations by Italian police
forces at the province level amounted to an average of 33 kilograms of marijuana, 12 kilograms of
hashish, and 103 plants of cannabis.
[Table 1 around here]
Compared to the entire illegal drug market and traffic, cannabis-derived substances (i.e., both herbal
and resin) account for more than 90% of the total amount of confiscated drugs, according to our data.
However, large heterogeneity exists among provinces with respect to monthly confiscations. This
ranges from no confiscations, to confiscations of a few grams, to maxi-confiscations (tons). The mean
values mask a number of important features of our data regarding drug confiscations. First, as shown in
the non-parametric distribution reported in Table 2, the distribution is highly right skewed. Second, we
observe several zeros in the confiscation variables that must be taken into account in the empirical
strategy.
[Table 2 around here]
Concerning grow shops, we observe an average of 2.7 shops per province in the period before the
policy. Additionally, in this case, the mean value masks a high spatial heterogeneity. Figure 1 shows the
spatial distribution of both grow shops (left) in the pre-policy period (October 2016) and C-light shops
(right) in February 2018 over the Italian territory. Each province is colored according to number of the
shops existing in the territory.
[Figure 1 around here]
Panel (A) of Figure 1 shows that grow shops were located mostly in provinces on the seaside, in the Po
valley (Pianura Padana), and in Veneto. These locations are all close to waterways and are characterized
by the presence of very humid soil, which make the cultivation of cannabis, and thus its
commercialization, more favorable. For instance, the same soil characteristics also make favorable the
cultivation of rice, which is indeed located mostly in the same parts of the country. As discussed in
Section 1, grow shops were the first to put flowers on the market by exploiting large economies of
scope provided by the possibility of selling both cannabis and its flowers. As a matter of fact, these
shops faced very small marginal costs when adding the new product to their catalogue.
Panel (B) of Figure 1 shows that spatial heterogeneity reduced substantially in February 2018, i.e., 10
months after liberalization. A comparison between Panels (A) and (B) reveals two interesting features.
First, C-light retailers are more concentrated in areas characterized by a higher pre-policy concentration
of grow shops. This is not surprising because grow shops were the first C-light retailers after
liberalization. Second, it shows that liberalization caused a massive entry in the market, especially in
6
provinces not previously covered by grow shops. This phenomenon essentially interested herbalists and
tobacco shops. The geographical distribution of the grow shops and of C-light retailers reinforces the
idea that although the policy was national, its treatment effect was rather heterogeneous in the short
run as the pre-policy market coverage of grow shops was not spread uniformly over the Italian
territory. However, it became more homogenous in the long run, i.e., in February 2018. This pattern is
one of the main rationales for our decision to focus on the short-run effect of liberalization (see Section
4 for more details).
Finally, Table 1 offers some additional insights regarding other crime measures included in our dataset.
It is important to highlight the fact that the monthly average of provincial anti-drug operations is
approximately 16. This represents a massive effort in terms of both human resources and budget for
Italian law enforcement agencies. On average, 16 people per province were arrested on a monthly basis,
with foreigners accounting for almost more than 50% of total arrests. Interestingly, we observe a
monthly average of one minor arrested per province.
4. Identification strategy
To identify the effect of unintended liberalization of C-light on the illegal drug supply, we exploit the
local availability of the product in order to set up a DID study design. Although the legislative gap was
national, the treatment effect was rather heterogeneous over the Italian territories in the short run due
to the differentiated extent of the physical availability of the product.
Following the early approach by Card (1992), we employ a DID with continuous treatment, which uses
the number of grow shops already in existence in each province before liberalization as the intensity
treatment variable. As explained in Section 2, these shops were the first to sell C-light as a result of the
opportunity of exploiting large economies of scope given by the possibility of selling both cannabis-
related products and its flowers. The distribution of these shops across provinces is driven by province-
specific geographical and morphological factors, which make the cultivation and commercialization of
cannabis for industrial use more favorable in some areas, as discussed in Section 3. Importantly, this is
likely to be essentially unrelated to local demand for illegal drugs, which might, instead, determine an
endogenous entry of shops to sell C- light.
The possibility of endogenous entry is essentially ruled out in our study for two main reasons. First, due
to the particular nature of the liberalization process, liberalization was unintended, being a consequence
of a legislative gap, and was thus unannounced. This ruled out any possibility of an anticipation effect.
Second, our monthly data allow us to compare variations in very small windows after the time during
which the policy took place (less than a year).
Focusing on a short time period after liberalization is also useful in ruling out long-term trends in
cannabis confiscation, which (if negative) might lead to upward biased estimates. This may be caused
both by contractions in demand and changes in the efficiency of law enforcement agencies toward the
drug war. The latter is often a sensitive point in the empirical analysis on drug confiscations. However,
this is unlikely to represent a threat to our identification strategy for a number of reasons. First, due to
the unexpected nature of liberalization, making adjustments in police efforts was very unlikely,
especially in the very short time windows that we consider (i.e., 10 months after the policy). Second, in
the period to which our analysis refers, we did not find evidence of any new public hiring of police
and/or measure to displace police forces in specific geographical areas of the country, e.g., the only
public displacement of police forces occurred in August 2017 in Foggia (Apulia) to repress the so-called
Mafia Garganica. When excluding this province from our analysis, our results are unaffected. In
addition, the last round of police hiring was launched in May 2017, and the ranking of admitted people
was released only in May 2018. Presumably, these (new) policemen will be hired through 2018. More
importantly, any change in law enforcement should be systematically different across provinces
experiencing different intensities of liberalization to represent a threat to our strategy. This appears
even more unlikely in our context for the reasons discussed above (the unexpected nature of
7
liberalization and the short time window) but also because law enforcement is administered at national
level in Italy, i.e., local police have little responsibility for the concerns of anti-drug operations. In any
case, to rule out any residual concern on this point, we also include the monthly number of anti-drug
operations conducted in each province as a measure of police effort in our estimates, and we further
narrow the time windows and control for province-specific trends in the DID model. Finally, we test
whether the policy change had any effect on police efforts, showing that no law enforcement
adjustment occurred. These checks are reported in detail in Section 5 of the paper.
Importantly, while the basic DID compares two groups (treatment and control) over pre- and post-
policy periods, in our framework, the treatment variable is continuous, and every observational unit is
identified by the intensity of the exposure to the policy. A similar strategy has been used by other
empirical papers, i.e., Gaynor et al. (2013), to test the impact of hospital competition on healthcare
quality. In this framework, the impact of the policy change is identified by the interaction between the
pre-existing number of grow shops and the dummy that identifies the post-policy period.
Thus, our model takes the following form:
!
"# $ %&'()* + %,'()*
- ./(0)"1,2&3 + %4
5
6"# + 7"+8"#
(1)
where
!
it is the quantity of drug (i.e., marijuana, hashish, plants of cannabis) confiscated at the time
*
(month-year period) in the province
9
,
'()*
is the indicator of the post-liberalization period and takes
the value of 1 from May 2017 onwards and 0 otherwise, while Shops is our treatment intensity variable,
namely the number of grow shops pre-policy (October 2016). Xit is a vector of time-variant covariates
that includes the number of anti-drug police operations and time-variant province-specific
characteristics (i.e., total population and population density),
7"
is unobserved province fixed effects
(which includes the pre-policy configuration of grow shops), and
8
is the error term. To deal with the
high presence of zeros in the outcome variable, as shown in Section 3.1, we use a log transformation of
the dependent variable.
4
This allows us to have a more straightforward interpretation of the impact of
liberalization, i.e., on the share of illegal marijuana displaced.
The main coefficient of interest in equation (1) is
%
2, which measures the change in the illegal market
supply post-liberalization per each pre-policy grow shop. As known, this identification strategy relies on
the common trend assumption, which requires that in the absence of (unintended) liberalization,
provinces would have experience parallel trends in confiscation independently from the presence of
grow shops. In Section 6, we test this key assumption in different ways encompassing both graphical
regression techniques and alternative approaches to statistical inference. All of these checks provide
strong support for the common trend assumption in our setting.
5. Results
[Table 3 around here]
Table 3 reports our results from the DID regression, as in equation (1), according to several
specifications. For all specifications, we report estimates that include standard errors clustered at the
province level that are robust to correlated province-level shocks in drug confiscations. The number of
clusters (106 provinces) should rule out concerns about the validity of inference in our estimates.
However, in Section 6, we demonstrate that our results are robust also for different approaches to
statistical inference (i.e., randomization tests based on simulated liberalizations).
5
4
We use zero-skewness log transformation in the spirit of the Box and Cox (1964) transformation. This actually adds a
value k to the zeros before operating the log transformation so that the skewness of the dependent variable is reduced to
zero.
5
We find qualitatively similar results when considering non-linear models, such as the Tobit and Poisson (results are
available upon request to the authors). However, their interpretation is problematic in a DID setting (see, e.g., Puhani (2012)
and Blundell and Dias (2009)).
8
We find a negative and statistically significant DID coefficient in all specifications: this supports the
thesis that the liberalization of C-light displaced the market of illegal marijuana. In Columns (1–2), we
report the results of operating a log-transformation of the dependent variable. According to this
specification, we find that the liberalization of C-light resulted in a reduction of 11.5% in confiscations
of illegal marijuana for each pre-policy grow shop. In other words, while the policy impacted all Italian
provinces, those provinces served by grow shops before the policy experienced a more intense
reduction in the amount of seized marijuana. This occurred mainly in provinces where grow shops
were located: the greater the number of grow shops in the market, the greater the displacement effect.
A startling result is that such a displacement arises also with an imperfect substitute of street marijuana
with a low level of THC. Moreover, such a displacement occurs despite a general but not statistically
significant increasing trend for the amount of confiscated marijuana, as indicated by the Post coefficient.
Finally, the number of police operations positively impact the amount of the illicit substance seized:
one more operation leads, on average, to an increase of 7% in marijuana confiscated per pre-policy
grow shop.
As seasonality is a concern related to crime as well as marijuana consumption, we follow Draca et al.
(2011) to seasonally difference the data and then wash out any province-specific seasonality. Results are
reported in Column (3) and show that our effect of interest still holds and is also reinforced in its
magnitude (-0.14). Moreover, in order to have a more intuitive interpretation of the policy impact, we
also present a specification including the log of grow shops. This allows us to estimate the elasticity of
the policy effect. According to this specification reported in Column (4), we find that a 10% increase in
the number of grow shops led to a 3.3% decrease in monthly marijuana confiscations.
Finally, to account for an eventual dynamic of law enforcement at different territorial levels, we run the
DID model that includes province-specific time trends. As shown in Column (5), our results are
consistent with the baseline estimates: a negative DID effect of -10.6% is found. This result provides
further evidence in support of the parallel trend assumption and allays any concerns regarding
province-specific police effort adjustment after the policy.
5.1 Other measures of crime
[Table 4 around here]
In Table 4, we report estimates of the DID model on other crime-related outcomes. This helps to shed
further light on the effect of the policy change. In Column (1), we report estimates on the total number
of anti-drug police operations, which represents a proxy for the police effort. We do not find any
significant impact on the number of operations. This result indicates no significant adjustment in police
efforts in reply to the liberalization and supports our arguments on the unexpected nature of
liberalization. In Columns (2) and (3), we investigate the effect on two other cannabis-derived drugs. In
Column (2), we focus on the plants of cannabis. Our results indicate that for any pre-existing grow
shop, C-light liberalization caused a reduction of 32% in the cannabis plants confiscated. In Column
(3), we focus on hashish. This is the resin of the cannabis plants and is a processed product that is
usually stronger and more concentrated than marijuana. Our results suggest that the liberalization of C-
light led to a reduction of approximately 8% in hashish confiscated by police forces. All in all, these
findings indicate that liberalization generated a spillover on the entire cannabis drug market.
In the last three columns of Table 4, we show the effect of the policy change on arrests for drug-related
crimes. Column (4) shows a 3% reduction in the total number of arrests. Among these, we document a
significant decline in the number of arrests of foreigners ( -3% as reported in Column (5)) and of
underage individuals (-15% but significant only at 10% as reported in Column (6)). Overall, these
results suggest that liberalization had a negative and significant effect on organized crime, especially
among categories more often used by criminal organizations as drug dealers in the streets.
9
6. Robustness checks
In this section, we deal with a number of robustness checks. First, we check whether our results are
robust in the presence of maxi-confiscation activities. As marijuana mostly arrives via the maritime
route (e.g., the Balkan route), the presence of freight ports in some provinces may artificially increase
the quantity of drugs confiscated in these provinces. We thus conduct a robustness check that controls
for the presence of large freight ports at the province level. Table 5 (Columns (1) and (2)) presents the
main results with and without freight ports. We show that the estimates of the DID approach are
identical, regardless of the subsample considered. The marginal effect of pre-existing grow shops on
marijuana confiscated is approximately 9–10%, thus very close to the one shown in Section 5.
Second, given the skewness of our dependent variable, we perform a subsample analysis of provinces
falling below and above the mean of pre-policy confiscations (average annual confiscations in 2016).
Results are reported in Table 5 (Columns (3) and (4)) and show a 10% reduction among provinces
below the mean and a reduction of approximately 6-7% in confiscations among provinces above the
mean. These results are qualitatively similar to the ones obtained by considering provinces with and
without ports, and they show a rather homogeneous treatment effect.
6
[Table 5 around here]
Third, we perform additional robustness checks for concerns regarding the time window analyzed in
our quasi-experiment. We first reduce the time before the policy to make the pre- and post-policy
periods more symmetric. Hence, we consider the period between May 2016 and February 2018. In this
case, the estimated DID coefficient is -0.11 and hence is consistent with the main results. Then, we
study a symmetric period around the policy (May 2017) by considering six months before and after the
unintended implementation of the policy. In this case, the DID coefficient is -0.09 yet is qualitatively
similar to the main results. It is important to note here that this result is relevant to further allay
concerns about law enforcement adjustments. By further narrowing the time window and focusing on
the short run after the policy, the probability of any systematic law enforcement adjustment across
periods became even more unlikely.
As an additional set of robustness checks, we deal with the validity of the common trend assumption in
our setting. One usual concern when using a DID model specification is that the results can be driven
by pre-policy trends and by the presence of confounding factors. While the presence of confounding
factors is rather limited in our case because the liberalization of C-light was unintended, and the
industry was not regulated in the past, we conduct several tests to ensure that the common trend
hypothesis was satisfied. First, we make a graphical inspection of the common trend assumption. The
graphical analysis of the common trend assumption in the basic DID framework requires both groups
to follow a parallel path. In Figure 2, we present trends in marijuana confiscations according to terciles
of pre-policy grow shops at the province level. This allows us to verify the robustness of the
assumption for different levels of the treatment variable. Interestingly, the different lines follow a very
parallel pre-policy path that leads us to be confident about the credibility of the common trend
hypothesis. A small difference in confiscations across terciles is observed in the months of November
and December. This is not surprising as confiscation rates exhibit a reduction in these months due to a
reductions of police operations (e.g., because of Christmas holidays). Indeed, the average number of
police operations per province is around 12 in December vs around 16 in the other months. When
6
It is important to notice that when doing subsample analyses, the number of clusters available is reduced (i.e., to around 20
for “above” the mean of confiscation estimates) and approximates to a level that might be problematic for statistical
inference (Cameron and Miller (2015), for instance, suggest a “safe” threshold of 50 clusters in state-year panel data). For
this reason, in Table 5, we use bootstrapped clustered standard errors for these estimates and, more generally, we suggest a
more cautious interpretation of these results.
10
accounting for seasonality, the common trend in the pre-policy period becomes even more evident, as
shown by Figure 3.
[Figure 2 around here]
[Figure 3 around here]
However, in our framework, the presence of a continuous treatment variable makes the graphical
solution less straightforward. Thus, we complement our graphical analysis with a placebo regression.
We essentially test the effect of a fake policy for May 2016, and we control for its effect up to the
period covered by the real policy. In other words, we shorten our sample period and consider only the
period running from January 2016 to April 2017 by anticipating the policy from May 2017 to May 2016.
Indeed, due to the sample cut, the total number of observations considered is reduced to 1,676. As the
number of grow shops used in our main analysis refer to October 2016, we used the Internet Archive
Wayback Machine to collect data on grow shops for March 2016—two months before the fake policy.
[Table 6 around here]
Table 6 shows that a fake policy has no significant effect on illegal marijuana confiscated at the
province level. All other controls, such as the time trend, the number of operations, and the number of
grow shops in March 2016, are instead positive and highly significant.
Lastly, to reduce any residual concern about possible violations of common trend assumptions, we also
perform a permutation test based on a Monte Carlo simulation. The permutation test also allows us to
explore the robustness of the results to assumptions about the structure of the error distribution. This
is a strategy that is increasingly used in many empirical applications (i.e., Wing and Marier, 2014;
Carrieri and Principe, 2018). Indeed, although we rely on a sufficient number of clusters (106) in our
empirical analysis, inference in the DID setting might be sensitive to the choice of the cluster unit and
the approach to the statistical difference (Bertrand et al., 2004; Donald and Lang, 2007).
Formally, we randomly select a set of different time periods and treatment intensities (Month x Year x
Number of Shops) in order to simulate the effect of a “fake liberalization” and estimate the average
treatment effect in our DID framework by using the fake policy in place of the real one. Then, we
simulated the model 5,000 times and stored the estimated coefficient in order to plot the non-
parametric distribution of placebo estimates. The key assumption of this randomization test based on
placebo laws is that the fake liberalization should not generate any effect on the marijuana confiscation
since the timing of the policy change is randomly assigned. Thus, on average, the estimated effect
should be zero.
Figure 4 presents the non-parametric distribution of placebo estimates of the (unintended)
liberalization of C-light on marijuana confiscation. As the mean of the distribution is virtually zero, the
estimator is unbiased. Moreover, the average treatment effect we estimate (about -11%) falls in the very
extreme left tail of the distribution. As a result, this increases the confidence that the liberalization-
driven reduction in illegal marijuana supply was not generated by chance.
[Figure 4 around here]
11
7. Conclusions
Marijuana is the most popular drug in Europe. According to the EMCCDA, approximately 7% of the
European population smoked cannabis (23.5 million people) in 2016, with a peak of approximately
13.9% among young adults (17.1 million). In Italy, cannabis is even more popular in the young
population, with 19% of young adults using it in 2016. However, as marijuana remains illegal, users face
no alternative to purchasing it in the streets, which generates revenue for those active in the black
market. Proponents of legalization of cannabis identify the possibility of diverting revenue from the
illegal and untaxed market to the legal one as their main argument. However, a potential for the
legalization of cannabis has always brought about a polarized discussion, and due to the scarcity of
relevant data, the displacement effect of liberalization on the supply of illegal drugs remained
substantially unexplored. The main purpose of this paper was to fill this gap.
We looked at the effect of liberalization of light pot in Italy through a quasi-experiment that occurred
in December 2016 by means of a legislative gap that created the opportunity to legally sell cannabis
with a low level of tetrahydrocannabinol (C-light). To identify the effect of interest, we exploited the
fact that the intensity of liberalization in the short run varied according to the pre-liberalization market
configuration of grow shops, i.e., shops selling industrial cannabis-related products that have been able
to put cannabis flowers (light cannabis) on the new market by exploiting large economies of scope (i.e.,
the opportunity to sell both cannabis-related products and its flowers). Pre-policy localization of these
shops essentially depends on the proximity to cannabis cultivations, which are concentrated in areas
close to waterways and humid soil.
We employed a unique dataset on monthly confiscations of drugs and other crime-related measures at
the province level (NUTS-3) over the period from 20162018, which was matched with data on the
geographical location of grow shops as well as socio-demographic variables. Several features of our
analysis allowed us to make a number of contributions to existing literature on the interplay between
the legal and illegal market of drugs. First, the availability of monthly data, coupled with the unexpected
nature of liberalization, allowed us to estimate our effect of interest in a very short time window around
the policy. This makes changes in law enforcement and police efforts extremely unlikely (however,
controlled in several ways) and allowed us to interpret any change in the number of drugs confiscated
as changes in the equilibrium supply of illegal marijuana. Second, unlike previous evidence coming quite
exclusively from the US, we estimated the displacement effect in a European country and, in particular,
a country where criminal organizations are historically rooted and have very strong monopolistic power
and entire control over the smuggling of the drug, often jointly with international criminals. Last but
not least, the availability of data on confiscations allowed us to quantify the amount of illegal drugs
displaced by a mild form of liberalization and then to assess, albeit roughly, the foregone revenues for
criminal organizations. This is useful in order to evaluate the main expected beneficial effect from a
liberalization process.
According to our differences-in-differences (DID) estimates, for any grow shop serving a local market
before the policy, liberalization led to a contraction of up to 14% of the monthly confiscation of illegal
marijuana. In terms of elasticity, we documented that a 10% increase in the number of grow shops per
policy caused a 3.3% reduction in the confiscation of illegal marijuana. These results are robust to a set
of different checks and model specifications. We also found significant spillover effects on other
cannabis-related drugs, i.e., a reduction of 33% in the number of cannabis plants illegally cultivated and
8% for hashish. Moreover, we showed that the unintended policy also caused other indirect effects on
organized crime, such as a reduction of 3% in the number of people arrested for drug-related crimes.
Interestingly, we also found that the number of foreigners arrested for drug-related offences fell by 3%
and roughly 15% for the number of minors arrested. Overall, the policy had the beneficial effect of
reducing the number of people incarcerated for drug-related offences.
12
These estimates allow for a back-of-the-envelope calculation of the forgone revenues for the criminal
organizations. Considering that the average number of grow shops at the province level is around 2.76
and that the marijuana price is estimated to be 7–11 euros per gram (ECCMDA 2016), our estimates
over the 106 provinces imply that forgone revenue due to C-light liberalization ranges from 90–170
million euros per year, on average.
7
This refers to the only market of marijuana and thus excludes other
cannabis-related drugs, such as plants of cannabis and hashish.
The implied forgone revenue, despite being statistically different from zeros, is not very high if
evaluated as a share of revenue for the entire market of illegal cannabis-related drugs, estimated to be
around 3.5 billion euros in Italy. In particular, our estimates suggest that the liberalization of C-light led
to a reduction in revenue from street marijuana of around 3–5% of the entire cannabis-related market.
This may suggest underplaying the role of liberalization as a way to fight criminal organizations in the
short run. However, it is important to highlight that we are able to estimate only a (very) lower bound
of the real displacement effect of liberalization. This is for a number of reasons. First, it is because we
used data on confiscations of illegal drugs, which obviously represent only a lower bound of the stock
of marijuana available in the illegal market. Second, it is because we dealt with liberalization of a rather
imperfect substitute of the marijuana available in the illegal market. Indeed, C-light contains a low level
of THC and a naturally high level of CBD, the cannabidiol. Consequently, it has much less of a
recreational effect. Last but not least, liberalization was unintended and received very little attention by
the media, at least in the short term to which our analysis refers. This implies that product
advertisement was very low, at least in the short run.
Despite this, our estimates indicate that even a mild form of liberalization, such as the one that
occurred involuntarily in Italy, can accomplish the purpose of reducing the quantity of marijuana sold
in the illegal market and the related revenues for organized crime, and this is likely to also encompass a
variety of cannabis-related drugs. This result is ultimately supported by the fact that while police forces
did not adjust their effort, anti-drug operations resulted in fewer arrests.
Our results also have a number of other interesting implications. Besides the positive effect on crime,
our results show that a substitution effect on the demand side exists between high- and low-THC
products. According to the EMCDDA (2016), the potency of the substance has been increasing in
recent years, reaching an average percentage of THC in street marijuana of 10.8, with peaks of 22%,
relative to 0.2–0.6% of THC permitted by the current Italian regulation. Given the substitution pattern,
we may speculate that all potential consequences of the direct liberalization of recreational marijuana
(e.g., negative effects on school achievement as shown by Plunk et al. 2016) are not likely to arise with
this light substance. Evidence in this direction may inform policymakers about a mixed approach to
legalization which, on the one hand, diverts illegal consumption toward legal consumption, disrupting
the black market, and on the other hand, it also reduces the negative externalities associated with the
abuse or misuse of these substances.
This paper also sets the ground for future research. This may also be devoted to investigating, in the
Italian context, the effect of this mild form of liberalization on other violent and non-violent crimes.
This might be particularly relevant in the long run, with a more efficient reallocation of police resources
toward the repression and prevention of other crimes. Lastly, it might be beneficial for an assessment
of potential forgone tax revenue resulting from C-light liberalization. This may represent another
argument in favor of liberalization, especially in times of strict public budget constraints.
7
The estimates were made on the basis of the DID treatment effect of -11.5 % and -14% as shown in Table 3 and using the
average confiscation mean of 32.89 kg. By considering a price of 7 euro/gr, forgone revenues range between 92.95 113
million euros, according to the DID treatment effect parameter. Similarly, by considering a price of 11 euro/gr, forgone
revenues account for 146177 million euros.
13
References
Adda, J., McConnell, B., & Rasul, I. (2014). Crime and the depenalization of cannabis possession:
Evidence from a policing experiment. Journal of Political Economy, 122(5), 1130- 1202.
Anderson, M. D., Hansen, B., & Rees, D. I. (2013). Medical marijuana laws, traffic fatalities, and
alcohol consumption. The Journal of Law and Economics, 56(2), 333-369.
Bertrand, M., Duflo, E., & Mullainathan, S. (2004). How much should we trust differences-in-
differences estimates?. The Quarterly Journal of Economics, 119(1), 249-275.
Blundell, R., & C. Dias, M. (2009). Alternative Approaches to Evaluation in Empirical
Microeconomics, Journal of Human Resources, 44(3), 565-640.
Box, G. E., & Cox, D. R. (1964). An analysis of transformations. Journal of the Royal Statistical Society.
Series B (Methodological), 211-252.
Brinkman, J., & Mok-Lamme, D. (2017). Not in my backyard? Not so fast. The effect of marijuana
legalization on neighbourhood crime. FRB of Philadelphia Working Paper No. 17-19.
Cameron, C. & Miller, D. (2015). A Practitioner's Guide to Cluster-Robust Inference, Journal of Human
Resources, 50(2), 317-373.
Card, D. (1992). Using regional variation in wages to measure the effects of the federal minimum
wage. ILR Review, 46(1), 22-37.
Carrieri, V., & Principe, F. (2018). WHO and for how long? An empirical analysis of the consumers’
response to red meat warning (No. 18/08). HEDG, c/o Department of Economics, University of
York.
Chang, T. Y., & Jacobson, M. (2017). Going to pot? The impact of dispensary closures on crime. Journal
of Urban Economics, 100, 120-136.
Cheng, C., Mayer, W. J., & Mayer, Y. (2018). The effect of legalizing retail marijuana on housing values:
Evidence from Colorado. Economic Inquiry, 56(3), 1585-1601.
Chu, Y. W. L., & Townsend, W. (2018). Joint culpability: The effects of medical marijuana laws on
crime. Journal of Economic Behavior & Organization.
DiNardo, J., & Lemieux, T. (2001). Alcohol, marijuana, and American youth: the unintended
consequences of government regulation. Journal of Health Economics, 20(6), 991-1010.
Dipartimento Politiche Antidroga (2017). Relazione annuale al Parlamento sullo stato delle
tossicodipendenze in Italia. June 2017.
Donald, S. G., & Lang, K. (2007). Inference with difference-in-differences and other panel data. The
review of Economics and Statistics, 89(2), 221-233.
Draca, M., Machin, S., & Witt, R. (2011). Panic on the streets of London: Police, crime, and the July
2005 terror attacks. American Economic Review, 101(5), 2157-81.
14
Dragone, D., Prarolo, G., Vanin, P., & Zanella, G. (2018). Crime and the legalization of recreational
marijuana. Journal of Economic Behavior & Organization. Forthcoming:
https://doi.org/10.1016/j.jebo.2018.02.005
EMCDDA (2016). European Drug Report 2016: Trends and developments.
Galenianos, M., & Gavazza, A. (2017). A structural model of the retail market for illicit drugs. American
Economic Review, 107(3), 858-96.
Gavrilova, E., Kamada, T., & Zoutman, F. (2017). Is legal pot crippling Mexican drug trafficking
organisations? The effect of medical marijuana laws on US crime. The Economic Journal. Forthcoming:
https://doi.org/10.1111/ecoj.12521.
Gaynor, M., Moreno-Serra, R., & Propper, C. (2013). Death by market power: reform, competition,
and patient outcomes in the National Health Service. American Economic Journal: Economic Policy, 5(4),
134-66.
Hansen, B., Miller, K. S., & Weber, C. (2017). How extensive is inter-state diversion of recreational
marijuana? (No. w23762). National Bureau of Economic Research.
Hansen, B., Miller, K. S., & Weber, C. (2018). Early evidence on recreational marijuana legalization and
traffic fatalities (No. w24417). National Bureau of Economic Research.
Jacobi, L., & Sovinsky, M. (2016). Marijuana on main street? Estimating demand in markets with
limited access. American Economic Review, 106(8), 2009-45.
Marie, O., & Zölitz, U. (2017). “High” achievers? Cannabis access and academic performance. The
Review of Economic Studies, 84(3), 1210-1237.
Plunk, A. D., Agrawal, A., Harrell, P. T., Tate, W. F., Will, K. E., Mellor, J. M., & Grucza, R. A. (2016).
The impact of adolescent exposure to medical marijuana laws on high school completion, college
enrollment and college degree completion. Drug & Alcohol Dependence, 168, 320-327.
Puhani, P. A. (2012). The treatment effect, the cross difference, and the interaction term in nonlinear
“difference-in-differences” models. Economics Letters, 115(1), 85-87.
Sabia, J. J., Swigert, J. & Young, T. (2017). The effect of medical marijuana laws on body weight. Health
Economics, 26(1), 6-34.
Shepard, E. M., & Blackley, P. R. (2016). Medical marijuana and crime: Further evidence from the
western states. Journal of Drug Issues, 46(2), 122-134.
UNODC (2016). United Nations Office on Drugs and Crime. World drug report 2016.
Wen, H., Hockenberry, J.M., & Cummings, J.R. (2015). The effect of medical marijuana laws on
adolescence and adult use of marijuana, alcohol, and other substances. Journal of Health Economics, 42,
64-80.
Wing, C., & Marier, A. (2014). Effects of occupational regulations on the cost of dental services:
evidence from dental insurance claims. Journal of Health Economics, 34, 131-143.
15
Zambiasi, D., & Stillman, S. (2018). The Pot Rush: Is Legalized Marijuana a Positive Local Amenity?.
IZA Discussion Paper No. 11392.
Figures and Tables
Figure 1. Distribution of Pre-policy Grow Shops and Post-policy C-Light Dealers
Panel (A) shows the number of grow shops selling cannabis-related products before the introduction of the policy. The
number of grow shops refer to October 2016 and the data have been retrieved using the Internet Archive Wayback Machine
on the website http://www.growshopmaps.com/. Panel (B) shows the number of post-policy dealers in February 2018.
Data have been collected using the Internet Archive Wayback Machine on the websites of the four major retailers.
16
Figure 2. Common Trend – Monthly confiscation of marijuana per terciles of pre-policy grow
shops
Normalized illegal marijuana confiscation rates within terciles of the treatment intensity variable
17
Figure 3. Common Trend – Seasonally Adjusted Pre-Liberalization Trend
Seasonally-adjusted pre-liberalization illegal marijuana confiscation rates. Normalized within terciles of the
treatment intensity variable.
18
Figure 4. Kernel density distribution of placebo liberalization
Kernel density distribution of 5000 placebo estimates of the effects of the liberalization on
illegal marijuana confiscations.
0100000 200000 300000 400000
Density
-4.000e-06 -2.000e-06 0 2.000e-06 4.000e-06
Placebo Treatment Effect
kernel = epanechnikov, bandwidth = 1.628e-07
19
Table 1. Descriptive Statistics
Variable
Description
Mean
Std. Dev.
Marijuana
Monthly amount of Marijuana confiscated per province (in kilos)
32.89
244.00
Hashish
Monthly amount of Hashish confiscated per province (in kilos)
12.71
65.17
Plants
Monthly number of Plants of cannabis confiscated per province
102.64
864.49
Grow Shops
Number of grow shops pre-policy per province (October 2016)
2.76
4.24
C-light Shops
Monthly number of retailers per province selling C-light post-policy
0.71
2.14
Operations
Monthly number of police anti-drug operations per province
16.20
28.26
Arrests
Monthly number of arrested people per province
16.00
31.93
Foreigners
Monthly number of foreigners arrested per province
8.31
17.29
Minors
Monthly number of minors (<18 years old) arrested per province
0.86
1.66
Territorial controls
Density
Population density of the province
272.96
380.71
Population
Number of inhabitants per province
571,929.90
616,651.80
Square km
Land area covered by the province
2,849.74
1,739.51
Nr.
Observations
106 provinces X 26 months
2,756
20
Table 2. Distribution of Marijuana Confiscations: Summary Statistics and Percentiles
Mean
32.89
Standard Deviation
244.00
Skewness
18.37
Kurtosis
500.31
Minimum
0
P10
0
P25
.02
P50
.35
P75
3.23
P90
25.65
P95
70.00
P99
878.78
Maximum
8193.02
Summary statistics and relevant percentiles of the monthly confiscation of
marijuana. All values expressed in kilograms
21
Table 3. Differences-in-Differences Regression
(1)
(2)
(3)
(4)
(5)
Marijuana
Δ Marijuana
Marijuana
Δ Marijuana
DID
-0.115***
-0.116***
-0.140***
-0.327***
-0.106*
0.033
0.029
0.042
0.123
0.059
Post
0.217
0.218
0.243
0.104
-0.472
0.238
0.236
0.248
0.240
0.482
Police operations
0.069***
0.069***
0.064***
0.019
0.019
0.018
Δ Police operations
0.099***
0.109***
0.022
0.027
Other controls
Yes
No
Yes
Yes
Yes
Province FE
Yes
Yes
Yes
Yes
Yes
Year FE
Yes
Yes
Yes
Yes
Yes
Month FE
Yes
Yes
No
Yes
No
Province x Time
No
No
No
No
Yes
N
2,756
2,756
1,484
2,756
1,484
Log transformation for all outcomes. (1) with (2) without controls. (3) Seasonally differenced data (4) Log-Log specification
(grow shops in log). (5) Seasonally differenced data and province-specific trend included. S.E. clustered at province-level in
italics. ***, **, *, indicate significance at 1%,5% and 10%, respectively.
22
Table 4. Differences in Differences Regression – Other Measures of Crime
(1)
(2)
(3)
(4)
(5)
(6)
Operations
Plants
Hashish
Arrests
Foreigners
Minors
DID
-0.001
-0.320***
-0.077**
-0.030***
-0.026**
-0.148*
0.005
0.082
0.034
0.010
0.011
0.085
Post
0.038
2.056***
0.856***
0.057
0.024
-0.306
0.053
0.770
0.273
0.058
0.062
0.678
Police operations
0.149***
0.050***
0.029***
0.028***
0.114***
0.052
0.019
0.009
0.008
0.043
Other controls
Yes
Yes
Yes
Yes
Yes
Yes
Province FE
Yes
Yes
Yes
Yes
Yes
Yes
Year FE
Yes
Yes
Yes
Yes
Yes
Yes
Month FE
Yes
Yes
Yes
Yes
Yes
Yes
N
2,756
2,756
2,756
2,756
2,756
2,756
Log transformation for all outcomes. S.E. clustered at the province-level in italics. ***, **, *, indicate significance at 1%, 5%
and 10%, respectively.
23
Table 5. Differences in Differences Regression – Robustness
(1)
(2)
(3)
(4)
(5)
(6)
Without
ports
With
ports
< Mean
> Mean
>May16
+\- 6
months
DiD
-0.093**
-0.096**
-0.096*
-0.065**
-0.115***
-0.094***
0.038
0.039
0.054
0.032
0.031
0.033
Post
0.093
0.445
0.150
0.230
1.121***
0.950**
0.279
0.423
0.266
0.683
0.170
0.373
Operations
0.106**
0.042***
0.126***
0.032***
0.066***
0.057***
0.041
0.012
0.016
0.009
0.018
0.016
Other
controls
Yes
Yes
Yes
Yes
Yes
Yes
Ports
No
Yes
-
-
-
-
Year FE
Yes
Yes
Yes
Yes
Yes
Yes
Month FE
Yes
Yes
Yes
Yes
Yes
Yes
N
2,080
676
2,288
468
2,332
1,272
Log transformation for all outcomes. Subsample analyses with provinces: (1) with (2) without ports (3) below (4) above
the mean of confiscations in 2016 (i.e., approximately 11 kg). Time windows: (5) May 2016-Feb 2018 (6) Nov 2016-Nov
2017. S.E. clustered at the province-level in italics., bootstrapped S.E. clustered at the province-level for model
specifications (3) and (4), 1000 replications. ***, **, *, indicate significance at 1%, 5% and 10%, respectively.
24
Table 6. Placebo Policy: May (2016)
(1)
Marijuana
DID (Mar2016)
0.039
0.049
Post May (2016)
1.568***
0.215
Police operations
0.084***
0.021
Other controls
No
Province FE
Yes
Year FE
Yes
Month FE
Yes
N
1,696
Log transformation of the dependent variable. S.E. clustered at the province-level in
italics. ***, **, *, indicate statistical significance at 1%, 5% and 10%, respectively.
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
First-pass evidence is provided that the legalization of the cannabis market across US states is inducing a crime drop. We exploit the staggered legalization of recreational marijuana enacted by the adjacent states of Washington (end of 2012) and Oregon (end of 2014). Combining county-level difference-in-differences and spatial regression discontinuity designs, we find that the policy caused a significant reduction in rapes and property crimes on the Washington side of the border in 2013-2014 relative to the Oregon side and relative to the pre-legalization years 2010-2012. The legalization also increased consumption of marijuana and reduced consumption of other drugs and both ordinary and binge alcohol. Four possible mechanisms are discussed: the direct psychotropic effects of cannabis; substitution away from violence-inducing substances; reallocation of police effort; reduced role of criminals in the marijuana business.
Article
This paper studies the effects of marijuana legalization on neighborhood crime and documents the patterns in retail dispensary locations over time using detailed micro-level data from Denver, Colorado. To account for endogenous retail dispensary locations, we use a novel identification strategy that exploits exogenous changes in demand across different locations arising from the increased importance of external markets after the legalization of recreational marijuana sales. The results imply that an additional dispensary in a neighborhood leads to a reduction of 17 crimes per month per 10,000 residents, which corresponds to roughly a 19 percent decline relative to the average crime rate over the sample period. Reductions in crime are highly localized, with no evidence of spillover benefits to adjacent neighborhoods. Analysis of detailed crime categories provides insights into the mechanisms underlying the reductions.
Article
This paper examines the amenity value of legalized marijuana by analyzing the impact of marijuana legalization on migration to Colorado. Colorado is the pioneering state in this area having legalized medical marijuana in 2000 and recreational marijuana in 2012. We test whether potential migrants to Colorado view legalized marijuana as a positive or negative local amenity. We use the synthetic control methodology to examine in‐ and out‐migration to/from Colorado versus migration to/from counterfactual versions of Colorado that have not legalized marijuana. We find strong evidence that potential migrants view legalized marijuana as a positive amenity with in‐migration significantly higher in Colorado compared with synthetic‐Colorado after the writing of the Ogden memo in 2009 that effectively allowed state laws already in place to be activated, and additionally after marijuana was legalized in 2013 for recreational use. When we employ permutation methods to assess the statistical likelihood of our results given our sample, we find that Colorado is a clear and significant outlier. We find no evidence for changes in out‐migration from Colorado suggesting that marijuana legalization did not change the equilibrium for individuals already living in the state. (JEL I18, R23, K42, C22)
Article
Over the last few years, marijuana has become legally available for recreational use to roughly a quarter of Americans. Policy makers have long expressed concerns about the substantial external costs of alcohol, and similar costs could come with the liberalization of marijuana policy. Indeed, the fraction of fatal accidents in which at least one driver tested positive for tetrahydrocannabinol has increased nationwide by an average of 10% from 2013 to 2016. For Colorado and Washington, both of which legalized marijuana in 2014, these increases were 92% and 28%, respectively. However, identifying a causal effect is difficult due to the presence of significant confounding factors. We test for a causal effect of marijuana legalization on traffic fatalities in Colorado and Washington with a synthetic control approach using records on fatal traffic accidents from 2000 to 2016. We find the synthetic control groups saw similar changes in marijuana‐related, alcohol‐related, and overall traffic fatality rates despite not legalizing recreational marijuana. (JEL K42, I12, I18)
Article
Most U.S. states have passed medical marijuana laws. In this paper, we study the effects of these laws on violent and property crime. We first estimate models that control for city fixed effects and flexible city-specific time trends. To supplement this regression analysis, we use the synthetic control method which can relax the parallel trend assumption and better account for heterogeneous policy effects. Both the regression analysis and the synthetic control method suggest no causal effects of medical marijuana laws on violent or property crime at the national level. We also find no strong effects within individual states, except for in California where the medical marijuana law reduced both violent and property crime by 20%.
Article
Does legalizing retail marijuana generate more benefits than costs? This paper provides a first step toward addressing that question by measuring the benefits and costs that are capitalized into housing values. We exploit the time-series and cross-sectional variations in the adoption of Colorado's municipality retail marijuana laws (RMLs) and examine the effect on housing values with a difference-in-differences strategy. Our estimates show that the legalization leads to an average 6% increase in housing values, indicating that the capitalized benefits outweigh the costs. In addition, we find suggestive evidence that this relatively large housing value appreciation is likely due to RMLs inducing strong housing demand while having no discernible effect on housing supply. Finally, we show that the effect of RMLs is heterogeneous across locations and property types. (JEL K20, R28)
Article
This paper investigates how legal cannabis access affects student performance. Identification comes from an exceptional policy introduced in the city of Maastricht in the Netherlands that discriminated access via licensed cannabis shops based on an individual's nationality. We apply a difference-in-difference approach using administrative panel data on course grades of local students enrolled at Maastricht University before and during the partial cannabis prohibition. We find that the academic performance of students who are no longer legally permitted to buy cannabis substantially increases. Grade improvements are driven by younger students and the effects are stronger for women and low performers. In line with how cannabis consumption affects cognitive functioning, we find that performance gains are larger for courses that require more numerical/mathematical skills. Our investigation of underlying channels using course evaluations suggests that performance gains are driven by an improved understanding of the material rather than changes in students' study effort. © The Author 2016. Published by Oxford University Press on behalf of The Review of Economic Studies Limited.
Article
We show that the introduction of medical marijuana laws (MMLs) leads to a decrease in violent crime in states that border Mexico. The reduction in crime is strongest for counties close to the border (less than 350km), and for crimes that relate to drug trafficking. In addition, we find that MMLs in inland states lead to a reduction in crime in the nearest border state. Our results are consistent with the theory that decriminalization of the production and distribution of marijuana leads to a reduction in violent crime in markets that are traditionally controlled by Mexican drug trafficking organisations. This article is protected by copyright. All rights reserved.
Article
Jurisdictions that sanction medical or, more recently, recreational marijuana use often allow retail sales at dispensaries. Dispensaries are controversial as many believe they contribute to local crime. To assess this claim, we analyze the short-term mass closing of hundreds of medical marijuana dispensaries in Los Angeles. Contrary to popular wisdom, we find an immediate increase in crime around dispensaries ordered to close relative to those allowed to remain open. The increase is specific to the type of crime most plausibly deterred by bystanders, and is correlated with neighborhood walkability. We find a similar pattern of results for temporary restaurant closures due to health code violations. A likely common mechanism is that“eyes upon the street” deter some types of crime.