The Effect of Recreational Marijuana on Crime: A
Synthetic Control Analysis
Ian Hoefer Marti
∗Supervisor: Dr. Erik Plug
July 13, 2018
AbstractUsing the synthetic control method, I assess the impact legalization of recreational mar-ijuana has had on crime in the states of Colorado and Washington, using a cohort of seven states with legal medical marijuana as controls. While the treatment states tend to have lower crime rates than their synthetic counterparts, this treatment impact seems to be insignificant compared to placebo synthetic treatments on each control state. These findings suggest there is little evidence to support the long-standing claim that legalization of marijuana will raise crime rates, or affect crime rates at all for that matter.
Keywords: legalization, marijuana, property crime, recreational, synthetic control, violent crime.
JEL Classification: I18, K42
∗Thesis submitted for the fulfillment of the requirements of MSc. Economics of student 11712678
Contents
1 Introduction 1
2 Literature Review 2
3 Data and Methodology 5
3.1 Data . . . 5 3.2 Methodology . . . 7
4 Results 9
4.1 Synthetic Control Analysis . . . 9 4.2 Placebo Testing . . . 14
5 Conclusions 18
1
Introduction
The debate around marijuana legislation and its social ramifications has been at the forefront of policymaking for many societies over the last few decades. Prohibition became the norm in the early 20th century across the globe as governments sought to reduce the use of what was considered a dangerous substance. In the United States more specifically, marijuana was prohibited at the Federal level in 1937 with the exception of restricted medical use, and again in 1970, this time also prohibiting medical use (Musto, 1991). Since 1973, however, many states have proceeded to decriminalize possession, with legalization for medical use being reintroduced first in California in 1996. Despite all this change in legislation, little research has been conducted concerning marijuana’s effects on both individuals and communities, in large part due to the prohibited status of the substance.
Concerns over the impact of drug usage on physical and mental health, and in turn on soci-ety, led to the prohibition of most recreational drugs, including marijuana. The principal aim of prohibition was to reduce consumption so as to protect potential users, ideally reducing usage, increasing their quality of life. Nonetheless in recent times evidence has shown that the consequential enforcement of prohibition, otherwise known as “The War on Drugs” has not had the desired effect on drug crime, usage and availability (Resignato, 2000; Baum, 1996). Furthermore, the causal relationship between marijuana usage and undesirable be-haviours such as crime has come into question in recent decades, as has the causality between long-term usage and neurological disorders. Beyond the persisting doubt over marijuana’s negative effects, the acceptance of medical marijuana has become more widespread as re-search has shown the positive results of using the drug in treatment for chronic pain or other psychiatric problems (Hill, 2015), casting further doubt on what should be the legal status of the drug.
One area of debate that is often of concern for policymakers is the impact of marijuana legalization on crime. This is certainly more so regarding recreational marijuana as opposed to medical marijuana, as the former will expectedly increase usage within communities more than the latter. As of July 2018, in the United States there are seven states that have legalized recreational marijuana since Colorado and Washington first began the legal sale of it in 2014.
This paper will attempt to assess the impact this effective change in law from medical to recreational marijuana in 2014 has had on crime rates in Colorado and Washington. Using a synthetic control analysis, I aim to deepen the understanding and add to the literature of the topic of marijuana legalization’s effect on crime.
The rest of this paper is structured as follows: Section 2 will present a literature review of the diverse effects of marijuana, so as to capture the essential arguments in favour and against the causal relation of marijuana on crime. Section 3 will discuss the data used to perform the analysis, as well as give a brief summary of the synthetic control model and how it can serve to determine the counterfactual trends of Colorado and Washington. Section 4 will present the results of this analysis, and Section 5 concludes with remarks on my findings, as well as on limitations to my study and potential further research possibilities.
2
Literature Review
In this Section I will present the principal channels through which marijuana legalization can affect crime rates both positively and negatively, providing a review of recent literature supporting these arguments.
The long-standing assumption about drug use, and more specifically marijuana use, has been that abuse and crime have a causal relationship, an assumption strengthened through
mis-perceptions and misinformation concerning abusers (Inciardi, 1986). The assumed channel through which marijuana use could have an impact on crime is through its effect on the human psyche, particularly for young adults. Fergusson et al. (2002) found that among a representative sample of youth in New Zealand, marijuana use was significantly correlated to psychosocial outcomes such as depression, suicidal ideation and use of other illicit drugs as well as violent and property crime. More recently, Solowij et al. (2012) found in their study that marijuana use was highly related to risky and impulsive decision-making. Furthermore, that legalization of marijuana significantly increases use has been evidenced by Cerd´a et al. (2012), where they found that legalization of medical marijuana increased not only use, but abuse and dependance as well. These findings would seem to support the argument that legalization of recreational marijuana could result in a higher proportion of people vulnerable to using and abusing the substance and, in turn, lead to higher crime rates. Beyond the possible ramifications of marijuana use on one’s psychosocial wellbeing, there exist concerns that the increase in dispensaries for medical or recreational marijuana could influence an area’s criminal activity. Indeed, a micro-temporal geospatial analysis of medical marijuana dispensaries in Long Beach, California by Freisthler et al. (2016) found that while there was no relationship between the location of dispensaries and crime in that same location, areas spatially adjacent to dispensaries did experience a positively correlated shift in crime. Despite the findings supporting the sequential argument for legalized recreational marijuana positively affecting a region’s crime rates, the truth is that there is currently little evidence to back up this claim. In their paper on medical marijuana and crime between 1990 to 2006, Morris et al. (2014) use state panel data to find that there is almost no positive correlation between medical marijuana and crime between states; all but one of their models presented a negative or statistically insignificant impact. Focusing on the supply side of the marijuana market, there is evidence from Gavrilova et al. (2014) to show that the legalization and regulation of medical marijuana in states bordering Mexico has been linked to an ensuing
reduction in violent crime as local, legal marijuana farmers who are not dependant on com-mitting crime for business replace drug cartels in the United States. This negative impact on crime is strongest in counties closer to Mexico and for crimes most related to drug traffick-ing. Furthermore, inland states with medical marijuana laws appear to lead to a reduction in crime in bordering counties. While this effect of legalization makes sense as illicit economic activity with criminal spillovers is reduced, it is unclear whether there is also a reduction in crime on the demand side of the market; that is, amongst consumers. That drug use is correlated with social exclusion and with antisocial behaviours has been known for some time (March et al., 2006), however it is unclear whether the consumption of marijuana is what causes antisocial behaviours and thus social exclusion, or whether the act of illegally acquiring marijuana is what leads to social exclusion and thus antisocial behaviours such as crime. If the latter is the case, then perhaps legalization of recreational marijuana could serve to reduce the social exclusion of consumers, resulting in fewer antisocial behaviours and, consequently, lower crime rates.
Ultimately, while there is undoubtedly a negative effect of marijuana on individuals, possibly resulting in lower quality of life and a risk of criminal behaviour in said individuals, the evidence indicating the contrary following legalization for medical marijuana points towards a complex effect of recreational legalization on crime. Given recent findings, I expect to find no positive link at the state-level between legalization for recreational use and crime, nor do I expect to find a consistent and significant negative impact on crime, be it for violent or property crimes.
3
Data and Methodology
In this Section I will give a description of the data used for my study, along with a brief illustration of the synthetic control analysis I apply to the data so as to study the effect of the change in law.
3.1
Data
The data used are the National Incidence-Based Reporting System (NIBRS) crime data containing data for violent crime, including its four sub-classes murder, rape, robbery and aggravated assault, as well as property crime, also including its four sub-classes burglary, larceny-theft, motor vehicle theft and arson. The data I have acquired are given per city agency, per year, spans the periods 2005 to 2016 and also contains the total population within each city agency. Pre-treatment periods thus consist of nine years from 2005 to 2013, whilst the post-treatment periods are 2014, 2015 and 2016, as the first dispensaries available to the public in the treated states of Colorado and Washington opened in 2014.
Due to the historically gradual decriminalization and regulation of marijuana, no state has legalized the substance for recreational use without having legalized medical marijuana some-time prior. As such, the comparison I wish to study must be between the treatment states Colorado and Washington, both with medical marijuana from 2005 up until the legalization for recreational use in 2014, with control states that have had strictly legal medical mari-juana for the entire period from 2005 to 2016. These control states are Alaska, California, Maine, Montana, Nevada, Oregon and Vermont. Hawaii is another state that fits such cri-teria, however data for the state was not available for the years 2011 to 2014, a key period for my analysis. For this reason I chose to drop Hawaii from my dataset. Because I wish
to assess the impact of recreational legalization on crime rates for an entire state, I cluster the data by state, and transform the absolute numbers of crime into crime per one-thousand citizens.
As such, the dataset used is a balanced panel dataset containing entries for nine states across the periods 2005 to 2016, and rates for ten types of crime, including violent crime and property crime.
Violent Crime Murder Rape Robbery Aggravated Assault
Alaska
7.86
0.05
1.06
1.31
5.44
California
4.62
0.05
0.25
1.73
2.58
Colorado
4.00
0.04
0.55
0.91
2.50
Maine
1.47
0.02
0.34
0.38
0.73
Montana
3.17
0.02
0.45
0.35
2.35
Nevada
7.23
0.07
0.46
2.48
4.22
Oregon
3.14
0.02
0.37
0.86
1.89
Vermont
1.53
0.01
0.22
0.21
1.10
Washington
3.90
0.03
0.42
1.25
2.20
Table 1: average rate of violent crimes per state.
Property Crime Burglary Larceny-Theft Motor Vehicle Theft Arson
Alaska
37.39
4.91
29.15
3.34
0.30
California
28.70
6.00
17.67
5.04
0.23
Colorado
34.44
6.22
24.49
3.77
0.24
Maine
27.49
4.70
21.98
0.81
0.14
Montana
39.49
4.61
32.45
2.43
0.22
Nevada
33.04
8.90
17.78
6.36
0.17
Oregon
40.19
5.99
30.44
3.76
0.33
Vermont
25.75
4.48
20.60
0.67
0.10
Washington
48.22
8.94
33.30
6.03
0.25
Table 2: average rates of property crimes per state.
As we can see, the level of crime varies from state to state, and demonstrates that no two states are very similar. Some types of crime are explainable due to the characteristics of each state; for example, that Washington, California and Nevada, all sparse states generally requiring vehicles for daily life, have the highest rates of motor vehicle theft is no surprise (Copes, 1999). Neither is it unexpected that Nevada, with its dependence on tourism and gambling, presents such high rates of robbery, aggravated assault and burglary (Albanese, 1985).
3.2
Methodology
To assess the impact of a treatment on a cohort, one would first think of using a simple difference-in-difference analysis between states to estimate the post-treatment effect. In this particular case, in which the nine states vary greatly in demographic, economic and political conditions, pre-treatment crime trends differ too significantly for such a straightforward analysis. Another possibility would be to test the impact of legalization through its effect on consumption levels of marijuana and test to see if variance in consumption across regions correlates positively with crime rates; while there are possible proxies available for post-treatment marijuana consumption such as marijuana sales tax revenue data for Colorado, there are no similar and reliable proxies for consumption prior to legalization. As a result, this method is not feasible for estimating the impact of legalization, as there would be nothing prior with which to compare post-treatment trends in consumption or crime.
Instead, the approach I use is the synthetic control analysis first proposed by Abadie and Gardeazabal (2003) in their study on the impact of violent conflict on GDP in the Basque Country, and further discussed by Abadie et al. (2010) in their assessment of Proposition 99 on cigarette sales in California.
This method consists in recreating the pre-treatment crime rates for Colorado and Wash-ington by assigning weights to the control states’ crime rates such that the pre-treatment gap between the treatment state and its counterfactual, the synthetic treatment state, are minimized. Suppose Cit is a rate of crime C for the treatment state i at period t, where t0 is the period the treatment is implemented, then our perfectly estimated synthetic treatment state would be Citsynthetic where Citsynthetic+ αitDit = Cit. αit is the impact of the treatment and Dit is a dummy variable equal to 1 if t ≥ t0. Because we wish to create our synthetic treatment state using a weighted average of the control states’ pre-treatment values, we can drop αitDitand replace Citsynthetic with ˆC
synthetic it =
P
jωjCjt, our estimated synthetic control state, such that Cit = ˆCitsynthetic+ t. ωj is the weight assigned to control state j such that the error term t = Cit− ˆCitsynthetic is minimized for t < t0. Assuming this error term is i.i.d, has a mean equal to 0 and is uncorrelated with crime rates, we can interpret the gap between Cit and ˆCit
synthetic
for t ≥ t0 as being equivalent to αitDit, the impact of the treatment.
While this method often manages to produce a convincing replica of pre-treatment trends in the treatment state, the ensuing post-treatment trends cannot be immediately interpreted as the impact; as discussed by Abadie et al, there is a chance that the post-treatment gap produced from the synthetic control method is purely coincidental, and not indicative of the treatment effect. To test the significance of the findings of synthetic control, they propose running a set of placebo tests, consisting in treating each control state as if it had received the same treatment as a treatment state. They use all other states, including the true treatment states, in the donor pool for the synthetic placebo state. This would effectively give the counterfactual to a control state’s pre- and post-treatment crime trends had they not been affected by the placebo treatment. This placebo test will produce a variety of post-treatment outcomes, which we can then compare with the treatment impact on our true treatment state, to check if the treatment impact is significantly different to that of the placebo. In their study on the effect of Proposition 99, Abadie et al have an initial 39 control states, whilst
I have 7. The fact that they have this many control states helps their placebo test, as they can simply drop those states with the largest mean squared prediction error (MSPE) with respect to California. This presents a challenge for my statistical analysis of the significance of the treatment impact, as I will not be able to remove those control states with the largest differences from my placebo test without reducing the comparative group to almost nothing. Regardless, I run this placebo test to assess where the trends of Colorado and Washington lie with respect to the control states, as this will still be useful in demonstrating whether or not these two states differ significantly from the control group.
4
Results
This Section is dedicated to presenting my findings and discussing the possible interpretations of the results of my study.
4.1
Synthetic Control Analysis
The first step of the synthetic control method is to compute the weights assigned to each control state per type of crime. These weights are shown below for violent crime and property crime in Colorado in Tables 3 and 4, and in Washington in Tables 5 and 6.
For synthetic Colorado, all crime rates are a weighted combination of the seven control states, and no state is disregarded for any type of crime. The same cannot be said for synthetic Washington, where for property crime, burglary and larceny-theft synthetic Washington is composed simply of a single state, multiplied by 1. The reason for this is that for these three variables Washington is higher than any other state in the donor pool for almost every period, thus the synthetic control merely assigns a value of 1 to the control state closest to
Violent Crime Murder Rape Robbery Aggravated Assault
Alaska
0.136
0.151
0.284
0.133
0.134
California
0.141
0.164
0.105
0.12
0.142
Maine
0.149
0.122
0.122
0.16
0.15
Montana
0.145
0.126
0.129
0.163
0.144
Nevada
0.136
0.199
0.135
0.107
0.137
Oregon
0.144
0.126
0.124
0.142
0.145
Vermont
0.149
0.112
0.102
0.175
0.148
Table 3: assigned weights for synthetic Colorado, violent crimes.
Property Crime Burglary Larceny-Theft Motor Vehicle Theft Arson
Alaska
0.152
0.103
0.144
0.135
0.176
California
0.111
0.142
0.141
0.165
0.139
Maine
0.109
0.11
0.142
0.109
0.102
Montana
0.178
0.099
0.145
0.124
0.132
Nevada
0.139
0.294
0.141
0.215
0.12
Oregon
0.207
0.144
0.145
0.145
0.244
Vermont
0.104
0.108
0.142
0.108
0.087
Table 4: assigned weights for synthetic Colorado, property crimes.
Violent Crime Murder Rape Robbery Aggravated Assault
Alaska
0.136
0.121
0.162
0.142
0.096
California
0.141
0.111
0.137
0.172
0.134
Maine
0.148
0.16
0.14
0.108
0.201
Montana
0.145
0.153
0.142
0.107
0.142
Nevada
0.137
0.096
0.143
0.241
0.11
Oregon
0.144
0.152
0.141
0.128
0.148
Vermont
0.148
0.206
0.136
0.102
0.169
Table 5: assigned weights for synthetic Washington, violent crime.
Washington. In the case of motor vehicle theft, Nevada accumulates most of the explanatory power of the control states, with a weight of 0.775. This makes interpreting post-treatment
Property Crime Burglary Larceny-Theft Motor Vehicle Theft Arson
Alaska
0
0
0
0.037
0.185
California
0
0
0
0.092
0.133
Maine
0
0
0
0.01
0.085
Montana
0
0
1
0.024
0.123
Nevada
0
1
0
0.775
0.107
Oregon
1
0
0
0.052
0.302
Vermont
0
0
0
0.009
0.066
Table 6: assigned weights for synthetic Washington, property crime.
gaps between Washington and synthetic Washington a challenge for property crimes, as the pre-treatment trends do not match as well as they do for violent crimes or for property crimes in Colorado.
Figure 1 plots the synthetic and treatment paths for each type of crime for Colorado, while Figure 2 plots the same for Washington. Where violent crimes are concerned, we can observe that post-treatment Colorado is generally below synthetic Colorado, however these gaps are small and follow the same trends as those of treated Colorado. Washington presents a more compelling case for the argument that legalization might negatively affect crime rates; not only are crime rates for treated Washington notably lower than synthetic Washington for the post-treatment periods, but the trends also differ significantly for murder, rape and robbery, as well as for the total number of violent crimes. More specifically, for the four aforementioned crime rates, the break in trends would appear to happen around the treatment period, between 2013 and 2014 in the graphs.
As for property crimes, we see that treated Colorado is even more similar to synthetic Colorado than is the case for violent crimes. Post-treatment, the two follow very similar paths, with the exception of motor vehicle thefts, for which treated Colorado is visibly above its synthetic counterpart. As mentioned previously, there is little to be interpreted from the synthetic control analysis of property crimes in Washington. As is to be expected, treated
Figure 1: synthetic control analysis of Colorado, per crime type (synthetic in grey, treated in dark blue)
Figure 2: synthetic control analysis of Washington, per crime type (synthetic in grey, treated in dark blue)
Washington lies above synthetic Washington for post-treatment, as well as significant parts of pre-treatment trends for crimes where the weights were skewed heavily in favour of a single control state. The only interpretable case is arson, which is consistent with the results for arson and other property crimes in Colorado.
4.2
Placebo Testing
As proposed by Abadie et al, I proceed to run the same synthetic control analysis to every control state, placing the remaining six control states and the two treatment states in the donor pool. I then measure the distance between each synthetic state and the corresponding true state. Figure 3 shows how the treatment effect for Colorado and Washington compares to the placebo treatment effect for each control state per type of crime.
Ideally, one would have enough control states to remove those with high pre-treatment MSPEs, resulting in a clear range of placebo treatment effects. Unfortunately, the fact that I can only rely on seven control states means that eliminating such states would result in a very limited comparative analysis between a few states for most crimes. Fortunately it is still possible to see how the post-treatment trends for Colorado and, where relevant, Washington compare to the seven control states’ placebo trends.
The true impact of the legalization of recreational marijuana on violent crimes would seem to be insignificant. While there are a few cases, such as rape for Alaska or robbery for Nevada, where the pre-treatment MSPE is far too great for the state to be of any use, the placebo tests generally manage to illustrate a range of possible post-trend placebo paths. The only case in which either of the treated states diverges notably from the placebo group is rape in Washington, however for every other case both Colorado and Washington seem to show no significant differences with the placebo treatment effects. Thus where violent crimes
Figure 3: placebo synthetic control analysis, per crime type (placebos in grey, Colorado in dark blue, Washington in red)
are concerned, there appears to be no significant effect of legal recreational marijuana on crime.
While the placebo tests for violent crimes produced somewhat interpretable results due to the generally good pre-treatment replication of the synthetic placebo states, the same cannot be said for property crimes. As can clearly be seen, the MSPEs of the placebo synthetic con-trols for pre-treatment periods are quite elevated, and the post-treatment ranges of possible placebo effects are mostly about as large as the pre-treatment variation. Despite this, and despite synthetic Washington being a very poor imitation of pre-treatment Washington, we can see that, as is the case for violent crimes, both Colorado and Washington again present post-treatment trends that are not significantly different from the control states’ placebo trends.
One further step proposed by Abadie et al to assess the relative impact of the treatment is to see how the ratio of post-treatment/pre-treatment MSPEs compares between states. A low ratio would indicate that both MSPEs are similar, either because they are small and there is no significant treatment effect, or because they are both large, and the explanatory power of that state is weak. Regardless, a higher ratio would indicate a large impact of the treatment on the respective state with respect to its pre-treatment trends.
Tables 7 and 8 show each state’s post-treatment/pre-treatment ratio for violent crimes and property crimes respectively. As is expected, Colorado and Washington are generally not the states with the largest ratios, although Washington’s ratios for violent crimes are far larger than Colorado’s, as the treatment impact is visibly larger for Washington. This would indicate that the treatment states are not the states with the largest post-treatment change, relative to their pre-treatment fit. The only case for which one of the treatment states seems to be significantly impacted is rape, where both states experience a negative treatment effect, and Washington presents a very high MSPE ratio.
Violent Crime Murder
Rape
Robbery Aggravated Assault
Alaska
8.127
7.335
2.916
8.379
2.479
California
5.711
11.096
2.410
9.112
2.016
Maine
0.208
1.434
12.985
2.759
0.208
Montana
13.677
2.251
5.758
1.394
9.845
Nevada
6.261
2.262
1.085
1.995
3.084
Oregon
10.554
0.313
11.593
21.255
5.968
Vermont
0.566
0.232
3.435
1.283
1.511
Colorado
0.767
1.008
3.800
0.351
1.226
Washington
17.446
3.748
16.695
4.861
9.570
Table 7: post-/pre-treatment MSPE ratios, violent crime.
Property Crime Burglary Larceny-Theft Motor Vehicle Theft Arson
Alaska
13.763
0.724
4.473
24.040
1.279
California
5.960
0.706
1.581
0.228
2.019
Maine
1.697
4.712
2.546
14.526
0.007
Montana
4.600
0.518
0.706
10.921
0.550
Nevada
0.629
0.513
3.106
0.278
1.456
Oregon
0.404
0.644
1.154
1.907
0.670
Vermont
2.828
6.541
3.356
13.565
0.364
Colorado
0.130
0.393
0.174
2.009
0.870
Washington
0.739
0.518
0.612
1.099
0.109
Table 8: post-/pre-treatment MSPE ratios, property crimes.
In summary, these results demonstrate how Colorado and Washington present lower levels of violent crime than their respective synthetic states constructed from the seven controls. In the case of property crime the results indicate that the treatment states follow almost the same path as their synthetic counterparts. While this may be indicative of a possible negative effect of legalization for recreational use on violent crimes, the placebo tests show that effects of this magnitude using the synthetic control method are common amongst the control states as well, reducing the significance of the treatment impact.
5
Conclusions
My analysis of the effect of state-wide legalization of recreational marijuana on crime rates using the synthetic control method shows that there is little difference between a treated state’s trends in crime and its synthetic counterpart. Despite the limited number of control states with which to test the significance for Colorado and Washington, the placebo test manages to illustrate how the impact of legalization compares to the trends in crime of those states unaffected by recreational marijuana legalization. This placebo test suggests that any post-treatment differences between a treatment state and the synthetic state are insignificant, with the possible exception of rape rates in Washington, which may indicate a negative impact of legalization on this type of crime.
These results run counter to the claim that through legalization of recreational marijuana, crime will rise. Rather, crime rates generally appear to continue on the same path as their predicted non-treatment counterparts, and present mostly negative differences in crime. This is more consistent with the findings of the 2014 paper by Morris et al, who found no positive relationship between medical marijuana legalization and crime rates.
Whilst my findings may support existing evidence against a positive relationship, they do not contribute significant evidence to support the argument that legalization can lower crime rates. The most likely scenario is that legalization of recreational marijuana has an impact on both communities and black markets in ways that cannot be observed through state-level crime rates. It has been shown that the presence of dispensaries is positively linked to crime in adjacent areas, as has it been shown that drug-related crime is affected negatively by more relaxed marijuana laws. To date, however, not much is known about the driving causes of shifts in crime due to legalization.
different areas of Colorado or Washington are correlated to marijuana consumption, using proxies such as tax revenue data. Another possibility would be to attempt to estimate how quickly legal marijuana sales have phased out the black market, if at all, and test to see if this change in the nature of the marijuana market is correlated to crime rates, however there would expectedly be difficulties in estimating the volume of the black market, given the very limited data on illicit drug sales. Lastly, it will be interesting to observe how crime rates evolve in the future as more and more states and countries (most recently Canada) opt to legalize marijuana for recreational use.
References
Abadie, A., Diamond, A., and Hainmueller, J. (2010). Synthetic control methods for compar-ative case studies: Estimating the effect of california’s tobacco control program. Journal of the American statistical Association, 105(490):493–505.
Abadie, A. and Gardeazabal, J. (2003). The economic costs of conflict: A case study of the basque country. American economic review, 93(1):113–132.
Albanese, J. S. (1985). Effect of casino gambling on crime, the. Fed. Probation, 49:39. Baum, D. (1996). Smoke and mirrors: The war on drugs and the politics of failure. Little,
Brown Boston, MA.
Cerd´a, M., Wall, M., Keyes, K. M., Galea, S., and Hasin, D. (2012). Medical marijuana laws in 50 states: investigating the relationship between state legalization of medical marijuana and marijuana use, abuse and dependence. Drug and alcohol dependence, 120(1):22–27. Copes, H. (1999). Routine activities and motor vehicle theft: A crime specific approach.
Fergusson, D. M., Horwood, L. J., and Swain-Campbell, N. (2002). Cannabis use and psychosocial adjustment in adolescence and young adulthood. Addiction, 97(9):1123–1135. Freisthler, B., Ponicki, W. R., Gaidus, A., and Gruenewald, P. J. (2016). A micro-temporal geospatial analysis of medical marijuana dispensaries and crime in long beach, california. Addiction, 111(6):1027–1035.
Gavrilova, E., Kamada, T., and Zoutman, F. (2014). Is legal pot crippling mexican drug trafficking organisations? the effect of medical marijuana laws on us crime. The Economic Journal.
Hill, K. P. (2015). Medical marijuana for treatment of chronic pain and other medical and psychiatric problems: a clinical review. Jama, 313(24):2474–2483.
Inciardi, J. A. (1986). The war on drugs: Heroin, cocaine, crime, and public policy, volume 1. Mayfield Publishing Company Palo Alto, CA.
March, J. C., Oviedo-Joekes, E., and Romero, M. (2006). Drugs and social exclusion in ten european cities. European addiction research, 12(1):33–41.
Morris, R. G., TenEyck, M., Barnes, J. C., and Kovandzic, T. V. (2014). The effect of medical marijuana laws on crime: evidence from state panel data, 1990-2006. PloS one, 9(3):e92816.
Musto, D. F. (1991). Opium, cocaine and marijuana in american history. Scientific American, 265(1):40–47.
Resignato, A. J. (2000). Violent crime: a function of drug use or drug enforcement? Applied Economics, 32(6):681–688.
Solowij, N., Jones, K. A., Rozman, M. E., Davis, S. M., Ciarrochi, J., Heaven, P. C., Pesa, N., Lubman, D. I., and Y¨ucel, M. (2012). Reflection impulsivity in adolescent cannabis users: a comparison with alcohol-using and non-substance-using adolescents. Psychopharmacology,
