Crime and the legalization of recreational marijuana

Full text

Crime and the legalization of recreational marijuana
Davide Dragone∗Giovanni Prarolo∗Paolo Vanin∗Giulio Zanella∗†
January 31, 2017
Abstract
We provide first-pass evidence that the legalization of the cannabis market across US
states may be inducing a crime drop. Exploiting the recent staggered legalization en-
acted by the adjacent states of Washington (end of 2012) and Oregon (end of 2014) we
find, combining county-level difference-in-differences and spatial regression discontinu-
ity designs, that the legalization of recreational marijuana caused a significant reduction
of rapes and thefts on the Washington side of the border in 2013-2014 relative to the
Oregon side and relative to the pre-legalization years 2010-2012. We also find evidence
that the legalization increased consumption of marijuana and reduced consumption of
other drugs and both ordinary and binge alcohol.
Keywords: cannabis, recreational marijuana, crime
JEL codes: K23, K42
∗University of Bologna, Department of Economics. Piazza Scaravilli 2, 40126 Bologna BO, Italy
†Corresponding author. E-mail: giulio.zanella@unibo.it.

1 Introduction
Gary Becker was a strong advocate of the legalization of drugs (Becker and Murphy, 2013),
particularly — in the wake of the first wave of legalization of recreational cannabis in the
US — of marijuana (Becker, 2014). Becker and Murphy (2013) claimed that the largest
costs of a prohibitionist approach to buying and selling drugs in the US “are the costs of
the crime associated with drug trafficking”, predicting that legalizing this market would
“reduce the role of criminals in producing and selling drugs [and] improve many inner-city
neighborhoods”: “Just as gangsters were largely driven out of the alcohol market after the
end of prohibition, violent drug gangs would be driven out of a decriminalized drug market”.
That is, letting the drug market emerge from illegality would make illegal activities in this
market not pay, thus greatly reducing fertile ground for crime, a central theme in Becker’s
economic approach to crime (Becker, 1968).
The present paper provides evidence in favor of these conjectures exploiting the full
legalization of the cannabis market recently enacted by some states in the US. Although
possessing, using, selling and cultivating marijuana is illegal under US federal law,1between
2012 and 2016 eight states have legalized recreational marijuana: Colorado and Washington
in 2012, Alaska and Oregon in 2014, California, Nevada, Maine and Massachusetts in 2016.2
The comparison between Washington (WA) and Oregon (OR) offers an experimental oppor-
tunity to study the effect of such legalization on crime because these are neighboring (hence
similar, in many respects) states that legalized cannabis for recreational use at about the
same time, but with a 2-year time lag that induces a quasi-experiment, and sufficiently early
to allow the observation of crime rates for at least two years from official sources. Combin-
ing difference-in-differences (DID) and spatial regression discontinuity (SRD) designs at the
county level to identify the causal impact of the legalization of cannabis for recreational use
on crime rates we find that the legalization reduced rapes by about 4 per 100,000 inhabitants
1Except for restricted uses, cannabis has been illegal under US federal law since the Marihuana Tax Act
of 1937. The Controlled Substance Act of 1970 (Title II of the Comprehensive Drug Abuse Prevention and
Control Act, Public Law 91-513) classified marijuana and tetrahydrocannabinols among the drugs listed in
Schedule I, which have high potential for abuse and no accepted medical value.
2Many more states have passed medical marijuana laws. These, however, do not legalize the supply side
of the market. Making marijuana legal for recreational purposes is the strongest form of legalization of the
cannabis market.
1

(a 30% drop), and thefts by about 100 per 100,000 inhabitants (a 20% drop ).
These results support Becker and Murphy’s conjectures, and are also in line with two
possible reasons that have been suggested for why illicit drugs may increase crime (Goldstein,
1985): stealing to buy expensive drugs, and drug wars within the system of drug distribution.
However, they stand in sharp contrast with the presumption that drugs cause crime, a major
argument in support of a prohibitionist approach to substance use. For instance, according
to the California Police Chiefs Association (2009), “public officials and criminal justice or-
ganizations who oppose medical marijuana laws often cite the prospect of increased crime”.
Case studies of crime reports found drugs to be, in fact, a contributing factor (Goldstein,
1985), and it has been observed that a higher percentage of persons arrested test positive
for illicit drugs compared with the general population (US Department of Justice). Yet,
research on the recent wave of legalization of cannabis for medical use (“medical marijuana
laws”, MML henceforth) in the US yields mixed results on the association between illicit
drug use and crime. Some researchers find no significant relationship between MML and
crime (Keppler and Freisthler, 2012; Braakman and Jones, 2014; Morris et al., 2014; Freisth-
ler et al., 2016; Shepard and Blackley, 2016), while others show that MML may reduce some
kind of non-drug crimes (Ingino, 2015) because of reduced activity by drug-trafficking orga-
nizations (Gavrilova et al., 2014). Using data from the UK, Adda et al. (2014) argue that
the decriminalizing marijuana allows the police to reallocate effort away from drug-related
crimes and towards other types of offenses. However, the estimation of a causal effect going
from legalizing cannabis to crime rates remains an elusive question because of the lack of
an experimental design (Miron, 2004). The present paper makes progress in this respect
by engineering a quasi-experiment that is able to provide first-pass causal evidence on the
relationship between recreational cannabis and crime rates.
At this level of analysis we cannot pin down the mechanisms operating behind the effects
we identify. Moving retail cannabis deals from degraded streets to safe, legal shops most
likely played a role. Anecdotal evidence is provided by this message posted on Twitter
by the Portland Police on June 10, 2016: “If you are looking to buy marijuana, go to a
legit business and avoid street dealers who might rob you”. Substitution away from drugs
which have remained illegal and from alcohol which makes consumers more aggressive than if
2

consuming cannabis is another possibility for which we provide evidence via a complementary
analysis that uses substance consumption as an outcome. We find that the legalization of
recreational marijuana in Washington induced an increase in the consumption of cannabis of
about 2.5 percentage points (off a base level of about 10%), a decrease in the consumption
of other drugs of about 0.5 points (off a base level of about 4%), and a decrease in the
consumption of both ordinary alcohol and binge alcohol of about 2 points (off base levels
of about 50% and 20%, respectively). Finally, the police reallocation channel suggested
by Adda et al. (2014) is certainly a plausible mechanism. We expand on mechanisms in
the concluding Section of the paper. In the next one, we summarize the legal details that
generate our quasi-experiment. The data and the results are presented in Section 3.
2 Legal framework
At the general election ballot of November 2012, voters in the state of WA approved with
about 56% of votes Initiative 502, which allows producing, processing, and selling cannabis,
subject to licensing and regulation by the Liquor Control Board, allows limited possession
by persons aged 21 and over (but not home cultivation), and taxes sales. Legal possession
began on December 9, 2012. Regulations for producers, processors and sellers were approved
in 2013 and retail sales of recreational cannabis began July, 8 2014 (Darnell, 2015). Shortly
after, the state of OR passed a similar reform. At the November 2014 general election
ballot, voters in OR approved with about 56% of votes Measure 91, a cannabis law reform
that is similar to the one passed in WA in terms of taxing sales and subjecting them to
regulation and licensing by the Liquor Control Commission, but is more permissive in terms
of possession and cultivation.3A previous legalization attempt in OR (Measure 80 of 2012),
quite permissive in terms of regulation and oversight, was marginally rejected with around
53% of votes in November 2012, thus enhancing the comparability with WA. Legalization of
possession, use and home cultivation started in OR in July 2015, recreational sales through
medical dispensaries in October 2015, and retail store licenses began in October 2016.
3Home cultivation of up to four plants per household is allowed. Adults over the age of 21 are allowed to
carry 1 ounce and keep 8 ounces at home, whereas WA establishes a possession limit of 1 ounce.
3

Therefore, the timing of the reforms was such that cannabis was legal on one side of the
border two years before the other side. Specifically, in 2013 and 2014 cannabis was legal in
WA but not in OR, a temporary 2-year window followed by a virtually identical legal status
across the border between two similar states where voters had a similar attitude towards
legalizing cannabis. This allows us to combine a difference-in-differences (DID) design (where
WA acts as the treatment group, OR as the control group, 2010-2012 is the pre-legalization
period and 2013-2014 is the post-legalization period) and a spatial regression discontinuity
(SRD) design (where the WA-OR border marks a discontinuity in the legal status of cannabis
in 2013-2014) to identify the causal impact of legal cannabis on violent and property crime.
Even after the legalization, there are counties in WA where cannabis business is pro-
hibited or where, according to the WA Liquor Control Board, Marijuana Sales Activity
by License Number, no recreational cannabis retailers are present. These are Columbia,
Franklin, Garfield, Wahkiakum, and Walla Walla County, all of them bordering Oregon ex-
cept Franklin County. We show later that our results are robust to excluding these counties
from the analysis.
A potential confounding factor in our analysis is that other relevant legal or institutional
changes affecting crime rates in WA may have taken place in 2013-2014. A search for such
changes reveals no relevant events that may have affected crime rates at the same time as the
legalization of cannabis possession and use. During this period, a reorganization of the 911
emergency call system took place in WA, and there were reforms related to health services,
regulation of wine and beer, and drug courts. There were also changes in the statute of
limitations for child molestation, incest (victim under age eighteen), and rape (victim under
age eighteen), as well as new norms concerning commercial sale of sex and commercial sexual
abuse, sexually violent predators, and sexual violence at school. However, all of these changes
were too marginal to exert a plausible first-order effect on crime.
3 Data and results
We employ data on criminal activity at the county level from the US Uniform Crime Re-
porting (UCR) statistics. The data base contains the number of offenses reported by the
4

sheriff’s office or county police department. For the reasons detailed below, these are not
necessarily the county totals, but they are the only publicly available information from the
UCR at the county level of disaggregation. We collected these crime data for years 2010
to 2014. For each county and each year, we have the total number of reported offenses for
murder, rape, assault, robbery, burglary, and theft. The final dataset is an unbalanced panel
(since not all counties report crime data every year) consisting of 335 observations for 75
counties, 36 in OR and 39 in WA. County-level population from the 2010 Census is used
to obtain crime rates per 100,000 inhabitants. The distance of each county’s centroid from
the WA-OR border is computed using a GIS software. Table 1 reports crime rates in WA
and OR counties between 2010 and 2014: all counties at the top of the table, counties at
the WA-OR border (where our comparison takes place) at the bottom. Because these rates
result from the aggregation of county-level reports in the UCR, they do not necessarily co-
incide with state-level counts. The reason of the discrepancy is twofold, as explained by the
FBI’s Criminal Justice Information Services Division at the UCR website. First, “only data
for city law enforcement agencies 10,000 and over in population and county law enforcement
agencies 25,000 and over in population are on this site”. That is, crimes occurring in smaller
cities are not counted for the published county-level totals. Second, “Because not all law
enforcement agencies provide data for complete reporting periods, it is necessary to estimate
for the missing data” when building statistics beyond the county level of aggregation. That
is, the FBI imputes crime counts to non-reporting agencies when building estimates at the
state and nation levels.
In addition, we employ data from the National Survey on Drug Use and Health (NSDUH)
to include in our analysis information on substance consumption. Such information may shed
some light on competing channels in the explanation of our results. Specifically, we pulled
from the NSDUH the rates of use over the previous month for marijuana, other Federal
illicit drugs, and alcohol. These statistics are publicly available only as averages over the
2010-2012 and 2012-2014 periods. Fortunately, these roughly correspond to the “pre” and
“post” periods in our DID-SRD analysis.4Table 2 reports these consumption rates for the
4For smaller counties the NSDUH data come as aggregates for larger units consisting of groups of
neighboring counties. In these cases, each county in the group is imputed the group-level average rate of
consumption.
5

Table 1: Crime rates at the county level
Year Murder Rape Assault Robbery Burglary Theft
All WA counties (N= 39)
2010 0.76 10.96 46.66 12.17 265.79 458.97
2011 0.85 9.65 40.84 10.30 265.08 440.87
2012 1.03 9.16 42.70 9.99 287.77 432.55
2013 0.80 9.07 41.23 9.21 258.73 419.59
2014 0.73 9.70 41.21 10.47 246.90 399.60
All OR counties (N= 36)
2010 0.80 7.22 34.31 6.82 132.96 393.71
2011 0.66 7.26 32.02 6.26 142.14 387.37
2012 0.84 7.51 29.31 6.75 150.93 412.93
2013 0.88 5.69 22.48 5.40 146.14 433.22
2014 0.66 7.22 30.21 4.72 115.17 335.12
Border WA counties (N= 11)
2010 0.35 15.37 33.69 8.51 224.00 529.80
2011 0.48 13.56 33.55 9.69 212.19 491.00
2012 0.75 12.80 42.00 7.58 223.30 445.11
2013 0.59 10.28 40.78 6.15 210.41 407.93
2014 0.71 10.52 39.48 6.97 184.76 357.10
Border OR counties (N= 10)
2010 0.34 1.58 13.40 3.04 41.88 163.57
2011 0.44 2.51 11.22 1.31 49.15 158.78
2012 0.31 2.59 10.76 1.14 56.88 176.11
2013 0.10 1.77 11.67 1.67 41.04 144.27
2014 0.11 0.91 14.89 2.39 40.91 128.08
Notes: Average crimes per 100,000 inhabitants in WA and OR counties, estimated from the county-level
counts reported in the Uniform Crime Reporting Statistics. The averages are weighted by county population.
6

Table 2: Substance Consumption rates at the county level
Year Marijuana Other drugs Alcohol Binge alcohol
All WA counties (N= 39)
2010-2012 0.102 0.044 0.560 0.222
2012-2014 0.127 0.039 0.542 0.206
All OR counties with consumption data (N= 34)
2010-2012 0.112 0.042 0.596 0.214
2012-2014 0.122 0.040 0.579 0.213
Border WA counties (N= 11)
2010-2012 0.093 0.042 0.535 0.223
2012-2014 0.101 0.034 0.486 0.199
Border OR counties (N= 10)
2010-2012 0.145 0.050 0.630 0.238
2012-2014 0.130 0.043 0.600 0.233
Notes: Average rates of substance use in WA and OR counties, estimated from the rates reported in the
National Survey on Drug Use and Health. The averages are weighted by county population.
same WA and OR counties used in Table 1.
Four features of our data are crucial for identification. First, WA and OR share similar
geographic, economic and institutional characteristics, including (quite crucially) a similar
attitude towards legal cannabis (see Section 2). Second, WA legalized the cannabis market
at the end of 2012, and OR (despite an attempt to legalize in that same year, marginally
failed) in 2014, which results in a 2-year period in which recreational cannabis is legal on one
side of the border and illegal on the other side. Third, the longitudinal dimension of the data
allows us to condition on county fixed effects and time effects, thus netting out unobserved
local characteristics that do not change over time, as well as those factors that vary over
time but are common to all counties. Fourth, the geographical features of the data allow us
to identify the effect of the policy at the WA-OR border, where treated and control counties
offer a better comparison: arguably, the similarity between two different states is maximized
when comparing bordering counties. Moreover, by conditioning on distance from the border
7

and by allowing for different effects of the spatial gap before and after the legalization, the
SRD design controls for the effect of distance from the border on crime rates, including
possible spillovers due to cross-border activity in response to the different legal status of
cannabis.
Preliminary graphical evidence about the causal effect of interest is offered in Figure 1.
The figure plots nonparametric estimates of the difference between county-level crime rates
before (2010-2012) and after (2013-2014) the WA legalization, as a function of the distance
(measured in hundreds of kilometers) of the county centroid from the WA-OR border. In
each panel of Figure 1, the difference between the variations in crime rates at the border (i.e.,
the jump at zero distance) is therefore a nonparametric estimate of the effect of legalizing
cannabis. Except for murders (for which the variation is essentially zero on both sides of
the border) and assaults, the drop in crime on the WA side of the border is much larger
than the corresponding drop on the OR side. Figure 2 illustrates the analogous evidence for
consumption.
Figure 1: Variation in crime between before and after the WA legalization
-1 0 1 2
-4 -3 -2 -1 0 1 2 3 4
murders
-6 -4 -2 0 2
-4 -3 -2 -1 0 1 2 3 4
rapes
-20 -10 0 10 20
-4 -3 -2 -1 0 1 2 3 4
assaults
-3 -2 -1 0 1 2
-4 -3 -2 -1 0 1 2 3 4
robberies
-40 -20 0 20 40 60
-4 -3 -2 -1 0 1 2 3 4
burglaries
-200 -100 0 100 200
-4 -3 -2 -1 0 1 2 3 4
thefts
Notes: Variation in county-level crimes per 100k inhabitants (vertical axis) as a function of the distance of the county centroid
from the OR-WA border measured in hundreds Km (horizontal axis). A positive distance means that the county is located in
WA, and a negative distance means that the county is located in OR. The jump at zero distance is a non-parametric DID-SRD
estimate of the effect of the legalization policy on crime. The lines are smoothed county-level differences in crime rates obtained
from local linear regressions, weighted by county population, employing a triangular kernel and a bandwidth of 100 Km.
8

Figure 2: Variation in consumption between before and after the WA legalization
-.02 0 .02 .04 .06
-4 -3 -2 -1 0 1 2 3 4
marijuana consumption
-.01 -.008-.006-.004-.002 0
-4 -3 -2 -1 0 1 2 3 4
other drugs consumption
-.06 -.04 -.02 0
-4 -3 -2 -1 0 1 2 3 4
alcohol consumption
-.03 -.02 -.01 0 .01
-4 -3 -2 -1 0 1 2 3 4
binge alchool
Notes: Variation in county-level rates of use of substances (vertical axis) as a function of the distance of the county centroid
from the OR-WA border measured in hundreds Km (horizontal axis). A positive distance means that the county is located in
WA, a negative distance means that it is located in OR. The jump at zero distance is a non-parametric DID-SRD estimate of
the effect of the legalization policy on consumption. The lines are smoothed county-level differences in crime rates obtained
from local linear regressions, weighted by county population, employing a triangular kernel and a bandwidth of 100 Km.
To provide a more formal statistical analysis, we employ a parametric model that allows us
to condition on unobserved county and time effects. Let cit be the crime rate in county iand
year t, and define the following binary variables: first, wi= 1 if county iis located in WA
(treatment), and wi= 0 if county iis located in OR (control); second, pt= 1 if year t > 2012
(post), and pt= 0 if year t≤2012 (pre). The DID-SRD design, sometimes referred to as
the Difference-in-Spatial-Discontinuity design (Dickert-Conlin and Elder, 2010; Gagliarducci
and Nannicini , 2013) can be represented by the following model:
cit =k+αpt+βwipt+f(di)pt+g(di)wipt+θi+ξit ,(1)
where kis a constant, f(.) and g(.) are polynomials of the same order (but possibly different
coefficients) in distance difrom the WA-OR border, θiare county fixed effects, and ξit are
residual determinants of crime. Coefficient βis the difference in the SRD estimates between
the pre and post periods, i.e., by how much liberalizing recreational cannabis in WA changed
the difference in crime rates right across the WA-OR border. We estimated Eq. (1) by OLS,
employing quadratic polynomials in distance as is appropriate in a parametric framework
(Gelman and Imbens, 2014). The resulting estimates of βare reported in Table 3.
9

Table 3: Effect of recreational cannabis on crime
Murder Rape Assault Robbery Burglary Theft
Estimated β0.23 –4.21** –1.30 –1.26 –36.32 –105.62*
(0.45) (1.26) (8.79) (1.92) (22.20) (40.21)
Observations 335 335 335 335 335 335
Notes: The table reports estimates of βfrom OLS on Equation 1, a coefficient that represents the difference in
the spatial regression discontinuity estimates between the pre and post periods, i.e., by how much liberalizing
recreational cannabis in WA changed the difference in crime rates right across the WA-OR border. Ordinary
standard error are reported in parentheses (robust standard errors clustered at the county level are smaller
than the ordinary ones displayed here). Each county is weighted in the regression based on the size of its
population in the 2010 Census. Significance level: * 5%; ** 1% or better.
There is evidence in this table that the legalization of recreational cannabis enacted in
WA caused a decrease in crime rates. The point estimates for rape, assault, robbery, burglary
and theft are all negative. This conclusion is reinforced by the statistical significance of the
drop in rapes (p-value = 0.001) and thefts (p-value = 0.01). For rapes, the reduction is 4.2
offenses per 100,000 inhabitants, which is about 30% of the 2010-2012 rate. For thefts, the
reduction is 105.6 offenses per 100,000 inhabitants, which is about 20% of the 2010-2012
rate.5Note that the parametric estimates of βin Table 3are in the same ballpark of the
jump at zero-distance in Figure 1(except for burglaries). This indicates that our parametric
choices are not driving the results.
As a robustness check, we re-estimate the DID-SRD model after excluding 5 WA counties
where cannabis business is prohibited and where, according to the Liquor Control Board,
Marijuana Sales Activity by License Number, no non-medical cannabis retailers are present.
These are Columbia, Franklin, Garfield, Wahkiakum, and Walla Walla County, all of them
bordering Oregon except Franklin County. Results are reported in Table 4 . These confirm
negative point estimates for all of the categories considered, and significant drops in rapes
and thefts.
The analogous estimates using consumption as an outcome are reported in Table 5. Our
DID-SRD estimates reveal that the legalization increased consumption of cannabis by about
2.5 percentage points (off a base level of about 10%), decreased in the consumption of other
5Although the point estimate for murders is positive, it is imprecise and not statistically significant.
10

drugs by about 0.5 points (off a base level of about 4%), and decreased consumption of both
ordinary alcohol (in a marginally significant way) and binge alcohol of about 2 points (off
base levels of about 50% and 20%, respectively). These effects on consumption suggest that
one of the mechanisms underlying the reduction in crime may be a substitution away from
other drugs which have remained illegal substances, such as alcohol, which makes consumers
more aggressive than if consuming cannabis. We expand on this point in the next section.
Table 4: Effect of recreational cannabis on crime: robustness check
Murder Rape Assault Robbery Burglary Theft
Estimated β0.20 –3.77** –0.36 –1.19 –41.84 –117.51**
(0.49) (1.49) (9.14) (2.04) (25.40) (39.67)
Observations 310 310 310 310 310 310
Notes: The table reports estimates of βfrom OLS on Equation 1, a coefficient that represents the difference in
the spatial regression discontinuity estimates between the pre and post periods, i.e., by how much liberalizing
recreational cannabis in WA changed the difference in crime rates right across the WA-OR border. WA
counties are excluded were cannabis business is prohibited and where, according to the Liquor Control
Board, Marijuana Sales Activity by License Number, no non-medical cannabis retailers are present. These
are Columbia, Franklin, Garfield, Wahkiakum, and Walla Walla County, all of them bordering Oregon except
Franklin County. Ordinary standard error are reported in parentheses (robust standard errors clustered at
the county level are smaller than the ordinary ones displayed here). Each county is weighted in the regression
based on the size of its population in the 2010 Census. Significance level: + 10%; * 5%; ** 1% or better.
Table 5: Effect of recreational cannabis on consumption
Marijuana Other drugs Alcohol Binge alcohol
Estimated β0.025** –0.005** –0.023+–0.020**
(0.009) (0.001) (0.014) (0.007)
[0.016] [0.002] [0.016] [0.010]
Observations 135 135 135 135
Notes: The table reports estimates of βfrom OLS on Equation 1when measures of consumption are used
as an outcome, a coefficient that represents the difference in the spatial regression discontinuity estimates
between the pre and post periods, i.e., by how much liberalizing recreational cannabis in WA changed
the difference in consumption right across the WA-OR border. Ordinary standard error are reported in
parentheses, and robust standard errors clustered at the county level are reported in brackets. Each county
is weighted in the regression based on the size of its population in the 2010 Census. Significance level: * 5%;
** 1% or better.
11

4 Concluding remarks
Our analysis of the causal effects on crime of the legalization of cannabis for recreational use
reaches conclusions in line with what Becker and Murphy (2013) expected when advocating
the full decriminalization of the drugs market, namely a crime drop. What are the possible
possible channels through which legalizing the production and sales of cannabis affects crim-
inal behavior? The effects may work through a change in market price and market structure,
as well as through institutional changes.
First, the policy leads to the emergence of a legal market, which offers more safety and
more reliable product quality. It thus reduces the risk of being victimized while buying,
the risk of being sanctioned, search costs (especially for first-time buyers), as well as the
psychological unease possibly related to purchasing an illegal product. From the consumer’s
point of view, this amounts to a reduction in quality-adjusted relative prices. Moreover,
retail prices should be expected, on average, to drop when the market is legalized due to a
corresponding lower risk on the supply side. Provided that cannabis is a normal good, a price
reduction should lead to an increase in its consumption, which is what we find analyzing
consumption data. Such increase may take place both at the extensive and intensive margin:
the number of consumers may increase and existing ones may consume more. Since cannabis
use determines a variety of psychoactive effects, which include a state of relaxation and
euphoria (Hall et al., 2001; Green et al, 2003), an increase in consumption may reduce the
likelihood of engaging in violent activities. This would hold, in particular, if cannabis is a
substitute for violence-inducing substances such as alcohol, cocaine and amphetamines.
Interestingly, the evidence is mixed in this respect. Some studies find that marijuana and
alcohol are substitutes (Anderson, Hansen, and Rees 2014; Crost and Guerrero 2012; Kelly
and Rasul, 2014; DiNardo and Lemieux, 2001), while others find that they are complements
(Williams et al., 2004; Wen et al., 2014). As observed in Sabia et al. (2016), who study the
effects of MML on body weight and health, the substitutability/complementarity between
alcohol and marijuana seems to be heterogeneous, depending on age.
Our results are in line with Gavrilova et al. (2016), who find that in US states bordering
Mexico the introduction of MML leads to a decrease in violent crimes such as homicides,
12

aggravated assaults and robberies, and that this reduction in crime rates is mainly due to
a drop in drug-law and juvenile-gang related homicides. The introduction of MML is found
to reduce the violent crime rate in Mexican-border states by 15-25 percent. This is a large
effect, but it is fully compatible with our estimates on the impact of recreational marijuana.
Besides directly affecting cannabis price and consumption, legalizing cannabis also changes
market structure. Entry of new legal sellers, who provide better quality than illegal com-
petitors, may drive the latter out of the market. Some illegal dealers might survive if legal
consumption is severely taxed, and they will surely survive during the time it takes to open
legal dispensaries. Yet, one may expect their profitability to fall – certainly their expected
future profits do. One reason is the increase in competitive pressure. Another one is that
product quality is not only likely to be higher in the legal part of the market, but it is
presumably also easier to identify, so that legalization might in principle introduce price
divergence: prices might increase in the legal relative to the illegal part of the market. The
likely result is an increase in average product quality and market exit by illegal suppliers.
This change in market structure is likely to reduce the presence of drug-trafficking criminal
organizations, together with drug-related conflicts and associated crimes. Yet, we do not
really know what previous dealers do after legalization, so this argument remains necessarily
incomplete. Moreover, one might be concerned that even legal dispensaries attract criminals,
e.g., to steal cash or marijuana. Yet, this concern is mitigated by the fact that dispensaries
may take measures to reduce crime and increase guardianship, such as doormen or video
cameras (Kepple and Freisthler, 2012). What seems more obvious is that the legalization
may not just affect the behavior of potential offenders, but also of potential victims. The
availability of cannabis through legal channels arguably makes consumers substantially less
willing to take risks in the illegal market. This might also contribute to explain the drop in
assaults, robberies and thefts that we document.
On top of altering behavior through changes in the cannabis market, legalization may
also generate a reallocation of police efforts. A lower rate of drug-related crimes opens the
possibility for the police to divert resources toward preventing non-cannabis related crimes,
as shown by Adda et al. (2014) for the decriminalization of possession of small quantities
of cannabis in London, UK. Interestingly, such reallocation may be driven by expectations,
13

and therefore need not wait for the actual opening of new dispensaries.
Summing up, the WA-OR quasi-experiment provides first-pass evidence that legalizing
cannabis may well cause a drop in crime. What we estimate is the short-run response. As
new data become available over time, for these states as well as for the other ones that
legalized in 2016, it will be possible to appropriately distinguish between short and long-run
effects.
References
Adda, J´erˆome, McConnell, Brendon, and Rasul, Imran (2014). Crime and the depenalization
of cannabis possession: Evidence from a policing experiment. Journal of Political Economy
122(5), 1130–1202.
Becker, Gary, and Kevin Murphy (2013). Have We Lost the War on Drugs? The Wall Street
Journal, January 4, 2013.
Becker, Gary (1968). Crime and Punishment: An Economic Approach. Journal of Political
Economy, 76(2), 169-217.
Becker, Gary (2014). Why Marijuana Should be Decriminalized. The Becker-Posner Blog,
February 23, 2014.
Braakman, Nils, and Jones, Simon (2014). Cannabis depenalisation, drug consumption and
crime: Evidence from the 2004 cannabis declassification in the UK. Social Science & Medicine
115, 29–37.
Darnell, Adam J (2015). I-502 evaluation plan and preliminary report on implementation
(Document Number 1509-3201). Olympia: Washington State Institute for Public Policy.
Retrieved from
http ://www .wsipp.wa.gov/ReportF ile/1616/W sippI−502 −Evaluation −P lan −and −
P reliminar y −Report −on −ImplementationReport.pdf (accessed 27 August 2016).
Dickert-Conlin, S., and T. Elder (2010). Suburban Legend: School Cutoff Dates and the
Timing of Births. Economics of Education Review, 29, 826-841.
DiNardo, John, and Lemieux, Thomas (2001). Alcohol, marijuana, and American youth:
the unintended consequences of government regulation. Journal of Health Economics 20(6),
991–1010.
Freisthler, Bridget, Ponicki, William R., Gaidus, Andrew, and Gruenewald, Paul J. (2016).
A micro-temporal geospatial analysis of medical marijuana dispensaries and crime in Long
Beach, California. Addiction 111(6), 1027-1035.
Gagliarducci, S., and T. Nannicini (2013). Do Better Paid Politicians Perform Better?
14

Disentangling Incentives from Selection. Journal of the European Economic Association, 11,
369-398.
Gavrilova, Evelina, Kamada, Takuma, and Zoutman, Floris T (2014). Is legal pot crippling
Mexican drug trafficking organizations? The effect of medical marijuana laws on US crime.
Discussion Paper. Institutt for Foretaksokinimi, Department of Business and Management
Science.
Gelman, A, and G. Imbens (2014). Why High-order Polynomials Should not be Used in Re-
gression Discontinuity Designs. NBER Working Papers 20405, National Bureau of Economic
Research.
Goldstein, Paul J (1985). The drugs/violence nexus: A tripartite conceptual framework.
Journal of Drug Issues 15(4), 493-506.
Green, Bob, Kavanagh, David, and Young, Ross (2003). Being stoned: a review of self-
reported cannabis effects. Drug and Alcohol Review 22(4), 453–460.
Hall, Wayne, Degenhardt, Louisa, and Lynskey, Michael (2001). The health and psycholog-
ical effects of cannabis use. National Drug and Alcohol Research Centre, Monograph Series
44.
Ingino, Francesco (2016). The heterogeneous effect of medical marijuana laws, Mimeo.
Keppler, Nancy J, and Freisthler, Bridget (2012). Exploring the ecological association be-
tween crime and medical marijuana dispensaries. Journal of Studies on Alcohol and Drugs
73(4), 523–530.
Miron, Jeffrey A (2004). Drug war crimes: The consequences of prohibition. Oakland, CA:
The Independent Institute.
Morris, Robert G, TenEyck, Michael, Barnes, James C, and Kovandzic, Tomislav V (2014).
The effect of medical marijuana laws on crime: evidence from state panel data 1990-2006.
PloS One 9(3), e92816.
Shepard, Edward M, and Blackley, Paul R (2016). Medical marijuana and crime: Further
evidence from the western states. Journal of Drug Issues 46(2), 122–134.
Sabia, Joseph J., Swigert, Jeffrey, and Young, Timothy (2017). The effect of medical mari-
juana laws on body weight. Health Economics 26(1), 6–34.
Williams, Jenny, Pacula, Rosalie Liccardo, Chaloupka, Frank J. and Wechsler, Henry (2004).
Alcohol and marijuana use among college students: economic complements or substitutes?.
Health Economics 13(9), 825–843.
15