Crime and the Depenalization of Cannabis Possession: Evidence from a Policing Experiment

Crime and the Depenalization of Cannabis Possession: Evidence from a Policing ExperimentAuthor(s): Jérôme Adda, Brendon McConnell, and Imran RasulSource: Journal of Political Economy, Vol. 122, No. 5 (October 2014), pp. 1130-1202Published by: The University of Chicago PressStable URL: http://www.jstor.org/stable/10.1086/676932 .

Accessed: 08/11/2014 18:02

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

.JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

.

The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to Journalof Political Economy.

http://www.jstor.org

This content downloaded from 173.231.110.189 on Sat, 8 Nov 2014 18:02:50 PMAll use subject to JSTOR Terms and Conditions

http://www.jstor.org/action/showPublisher?publisherCode=ucpress

http://www.jstor.org/stable/10.1086/676932?origin=JSTOR-pdf

http://www.jstor.org/page/info/about/policies/terms.jsp


Crime and the Depenalization of CannabisPossession: Evidence from a PolicingExperiment

Jerome Adda

European University Institute

Brendon McConnell

University College London

Imran Rasul

University College London

We evaluate the impact on crime of a localized policing experimentthat depenalized the possession of small quantities of cannabis in theLondon borough of Lambeth. We find that depenalization policycaused the police to reallocate effort toward nondrug crime. Despitethe overall fall in crime attributable to the policy, we find that the totalwelfare of local residents likely fell, as measured by house prices. Weshed light on what would be the impacts on crime of a citywide de-penalization policy by developing and calibrating a structural modelof the market for cannabis and crime.

I. Introduction

In nearly every country the market for illicit drugs remains pervasive, de-spite long-running attempts to restrict such activities. Around the globe

We gratefully acknowledge financial support from the Economic and Social ResearchCouncil ðRES-000-22-2182Þ and ESRC Centre for Economic Learning and Social Evolution.

[ Journal of Political Economy, 2014, vol. 122, no. 5]© 2014 by The University of Chicago. All rights reserved. 0022-3808/2014/12205-0003$10.00

1130



various policy approaches have been tried, ranging from punitive ap-proaches as manifested in the US “war on drugs” to more liberal law en-forcement strategies, such as those inHolland or Portugal, that lead to thedecriminalization or depenalization of the possession of some forms ofillicit drug, most notably cannabis.1

Both approaches have been criticized on theoretical and empiricalgrounds ðGlaeser and Shleifer 2001; Becker, Murphy, and Grossman2006Þ: the historically tough US policy stance is estimated to cost tens ofbillions of dollars annually, and there remain an estimated 3.7 millionindividuals regularly using illicit drugs, the majority of whom consumecannabis ðDepartment of Health and Human Services 2008Þ. At the sametime, concerns over more liberal policy strategies relate to the inherentcharacteristics of the illicit drugs market: consumption might damage us-ers’ health ðArseneault et al. 2004; van Ours and Williams 2009Þ, the useof some drugs might provide a gateway to more addictive drugs ðvan Ours2003Þ, and there are potentially large spillover effects on crime and otherforms of antisocial behavior.We contribute to this policy debate by evaluating an increasingly com-

mon policy intervention in the illicit drug market: the depenalization ofcannabis possession, so that the possession of small quantities of cannabisis no longer a criminally prosecutable offense. We present evidence froma localized UK policing experiment that introduced such a policy andfocus attention on measuring its impact on crime, considered to be amajor social cost of illicit drug markets.Criminal activity and drug markets might be linked because ðiÞ the sub-

stance itself leads to more violent or criminal behavior by users, ðiiÞ userscommit property crimes to obtain money to buy drugs, and ðiiiÞ violenceoccurs betweendrug suppliers to control selling areas.Wepresent evidenceover a broad range of crime types to assess the impact of depenalization on

1 Donohue, Ewing, and Peloquin ð2012Þ categorize illicit drug policies into three types:ðiÞ legalization—a system in which possession and sale are lawful but subject to regulationand taxation; ðiiÞ criminalization—a system of proscriptions on possession and sale backedby criminal punishment, potentially including incarceration; and ðiiiÞ depenalization—ahybrid system, in which sale and possession are proscribed, but the prohibition on pos-session is backed only by sanctions such as fines or mandatory substance abuse treatment,not incarceration. Following Donohue et al., we prefer the use of depenalization over de-criminalization as best describing the policy experiment we evaluate and closely mapping intothe definition of depenalization used by criminologists.

Rasul gratefully acknowledges financial support from the Dr. Theo and Friedl SchoellerResearch Center for Business and Society. We thank the editor, David Blunkett, Mirko Draca,Jeffrey Grogger, Gavin Hales, Andrew Oswald, Steve Pischke, Andrea Prat, Peter Reuter, OlmoSilva, John Van Reenen, Frank Windmeijer, Ken Wolpin, and numerous seminar and confer-enceparticipants for valuable comments.Wealso thank theUKDataArchive and JennyOkwuluand Betsy Stanko at the Metropolitan Police Service for providing us with the data and MayRostom for research assistance. All errors remain our own.

depenalization of cannabis possession 1131



the size of illicit drug markets for cannabis and harder drugs as well as thepolicy impact on nondrug crime such as property and violent crime.2

The depenalization policy we evaluate was unilaterally introduced bythe local police force in one London borough, Lambeth, in July 2001, apolicy known as the Lambeth Cannabis Warning Scheme ðLCWSÞ. Wedescribe the motivation behind the policy and its implementation inmore detail later. It is, however, worth noting that many aspects of thepolicy reflect how other depenalization policies have been implementedaround the world: ðiÞ the possession of small quantities of cannabis forpersonal consumption was still a recordable offense but would no longerlead to the individual being arrested, ðiiÞ the primary motivation was tofree up police time and other resources to focus on crimes related toother drugs or other non-drug-related crimes, and ðiiiÞ the policy did notalter penalties for cannabis supply.The LCWS was first announced as a temporary policing experiment to

run for 6 months from July 2001. At the end of this trial period the policywas adjudged to have been a success with the support of local residents.The policy was then announced to have been extended for a further6 months. Following this announcement, media reports of the delete-rious effects of the policy on crime, drug tourism, and drug use by chil-dren began to steadily increase. As local support for the LCWSwaned, thepolicy came to an end by July 2002, having run for 13 months. We usethese various policy switches to assess the short- and long-run effects ofthe depenalization policy on the levels and composition of drug crimeand nondrug crime.When evaluating localized policy interventions in illicit drug markets,

it is important to recognize interlinkages between drug markets: theequilibriummarket size for cannabis in a given location is partly a functionof the endogenous choices of police and cannabis users in other locations.More precisely, a localized depenalization policy in Lambeth will likelyðiÞ affect the size of the market for cannabis in Lambeth as well as the restof London as drug users move there to purchase cannabis and ðiiÞ enablethe Lambeth police to reallocate effort toward other types of crime, con-

2 The size of drug markets has previously been linked to crime rates ðGrogger and Willis2000; Pacula and Kilmer 2003Þ, especially for property crime ðCorman and Mocan 2000Þ.On users, Fergusson andHorwood ð1997Þ report evidence of a link between the early onsetof cannabis use and subsequent crime using longitudinal data for a birth cohort of NewZealand children. Early-onset users had significantly higher rates of later substance use,juvenile offending, mental health problems, unemployment, and school dropout. On can-nabis and violence, there is no clear evidence between the two as cannabis is usually thoughtto inhibit aggressive behavior ðResignato 2000Þ. On crimes by drug suppliers, Kuziemko andLevitt ð2004Þ find that incarcerating drug offenders is almost as effective in reducing vio-lent and property crime as locking up other types of offenders. Levitt and Venkatesh ð2000Þshow that workers in the illicit drug market are not particularly well remunerated, and sopursuing property crime might provide additional income and the flexibility to continue work-ing in the drug trade.

1132 journal of political economy



sequently affecting the number of drug and non-drug-related crime in alllocations.3

We investigate whether such changing patterns of crime and policebehavior are observed during and after the depenalization policy is in-troduced in Lambeth. To do so, we use administrative records obtainedfrom the London Metropolitan Police Service to construct a panel dataset on crime for all 32 London boroughs, for each month from April 1998until January 2006. This contains information on the number of recordeddrugoffenses at two fine levels of detail: ðiÞ the number of criminal offensesrelated to any given drug type, for example, cannabis, heroin, cocaine, andso forth; and ðiiÞ for each drug type, the specific offense committed: pos-session, trafficking, intent to supply, and so forth. Such detailed measure-ment of drug crime allows us to assess the impact of the policy on the sizeof the cannabis market ðas proxied by the total number of cannabis of-fensesÞ and whether the change in market size is predominantly drivenby changes in demand-related offenses such as cannabis possession or bysupply-related offenses such as cannabis trafficking and others.A depenalization policy can free up police resources to tackle noncan-

nabis drug crime. The disaggregated drug crime data we exploit allow usto specifically measure such effects on other illicit drug markets, not justthe direct effects on the market for cannabis, as well as for seven types ofnondrug crime: violence against the person, sexual offenses, robbery, bur-glary, theft and handling, fraud and forgery, and criminal damage. Finally,we note that the administrative records also contain information on twomeasures more closely correlated to police behavior for each disaggre-gated crime type: the number of individuals arrested and the number ofcrimes cleared up. These margins help provide evidence on how police ef-fectiveness across crime types changes in response to the depenalizationpolicy.We present four classes of results. First, the depenalization of can-

nabis in Lambeth leads to a significant increase in cannabis-relatedcrime: offense rates for cannabis-related crime rise by 29.3 percent morein Lambeth relative to the rest of London between the prepolicy andpolicy periods; in a comparison of the prepolicy and postpolicy periods,they are 61.0 percent higher in Lambeth vis-a-vis the rest of London. Thislonger-term effect persists well after the policy experiment ends. At thesame time, we document significant falls in police effectiveness againstcannabis-related crime that also persist well after the policy officially ends.Second, we find some evidence that the policy causes the police to

reallocate their effort toward crimes relating to the supply of hard drugs,

3 This potential reallocation of police effort across crime types has been hinted at in pre-vious studies. For example, Single ð1989Þ notes that following depenalization in California,there is some evidence that the police targeted noncannabis crime to a greater extent.




such as heroin, crack, and cocaine ðwhich are known as “class A” drugs inthe UK drug classification systemÞ. However, the primary benefit of thepolicy is that it allows the Lambeth police to reallocate their effort towardnondrug crime: we observe significant reductions in five out of sevenother crime types in the long run and significant improvements in po-lice effectiveness against such crimes, as measured by arrest and clear-up rates.4 Overall, these channels cause total nondrug crime to fall by9.4 percent in the long term in Lambeth relative to the rest of London.This reduction occurs against a backdrop of unchanging offense ratesfor nondrug crime in the postpolicy period for the rest of London.Our third class of results document the welfare impacts of the depe-

nalization on local residents. The welfare effects of the policy are a prioriambiguous: although it caused total crime to fall, it also led to a dramaticchange in the composition of crime. There was an increase in cannabis-related offenses, but the rates of many other types of crime fell in thelonger term. To estimate the overall impact of the policy through thesechanging crime patterns, as well as through other noncrime channels,we estimate policy impacts on house prices in Lambeth relative to otherLondon boroughs. Intuitively, the total social cost of depenalization ðnotjust those costs arising from crimeÞ should be reflected in house pricesðRosen 1974; Thaler 1978Þ.Wefind thatdespite theoverall fall in crimeattributable to thepolicy, the

total welfare of local residents likely fell, as measured by house prices.These welfare losses are concentrated in Lambeth zip codes where theillicit drug market was most active. We provide a lower-bound estimate ofthe loss inproperty values inLambeth ðwhichhas around280,000 residentsand 119,000 property unitsÞ due to the policy to be around £200 million.Our final set of results use the lessons from the localized policing ex-

periment to shed light on the likely impacts on crime if the same policywere to be applied citywide. To do so we develop and calibrate a struc-tural model of the market demand for cannabis and nondrug crime, ac-counting for the behavior of police and cannabis users. The model makesprecise interlinkages across cannabis markets, where the number of in-dividuals purchasing cannabis from a given location depends on the polic-ing strategies in all locations. With citywide depenalization, an importantmechanism driving the impacts of the localized policing experiment—themovement of cannabis users toward Lambeth to purchase cannabis—is

4 Section II.A describes in far more detail the definitions of each monthly crime seriesdata related to offenses, arrests, and clear-ups. Here we note that we define the offense rate,for a given crime, as the number of offenses per 1,000 of the adult population ðaged 16 andaboveÞ. As individuals are not necessarily immediately arrested for offenses committed, wedefine the arrest rate as the number of arrests in period t divided by the number of offensescommitted between month t and the previous quarter within the borough. The clear-uprate is analogously defined.




shut down. Because of this, the counterfactual policy simulation highlightsthat many of the gains of the policy can be retained, and some of the del-eterious consequences ameliorated, if all jurisdictions simultaneously de-penalize cannabis possession.Our study builds on the evidence on the effects of depenalization or

decriminalization policies on crime. MacCoun and Reuter ð2001Þ reviewthese studies and find positive but modest impacts. One reason for thedifference with our findings stems from our research design exploitingwithin- and across-borough variation in crime rather than being basedon nationwide policy changes. US studies have exploited the fact that inthe 1970s some states depenalized cannabis and found weak impacts oncrime ðNational Research Council 2001Þ. However, Pacula, Chriqui, andKing ð2004Þ have questioned such studies because “½so-called� decrimi-nalized states are not uniquely identifiable based on statutory law as hasbeen presumed by researchers over the past twenty years” (26).We contribute to this literature by exploiting a localized policy change

and using detailed administrative records on crime and police behavior.Our evidence provides a nuanced picture of the impacts of an increas-ingly observed policy, the depenalization of cannabis ðiÞ across crimesrelated to cannabis, class A drugs, and seven nondrug crime types; ðiiÞ onmeasures of police behavior, by assessing its impact on arrest and clear-up rates; ðiiiÞ across time, by assessing the short- and long-run impacts ofthe LCWS; and ðivÞ on welfare, as measured by house prices, and howthis varies within Lambeth depending on the prevalence of the illicit drugmarket across different zip code sectors in Lambeth. Taken together withour structural model estimates, these results provide new evidence rele-vant to the policy debate on interventions in illicit drug markets.5

The paper is organized as follows. Section II describes the motivationbehind the LCWS and the reasons for its ending. Section III describesour administrative data and empirical method. Section IV presents theresults on the impact of depenalization on cannabis crime. Section Vinvestigates how the policy affects other drug crime and nondrug crime.Section VI uses house price information to provide a hedonic evaluationof the depenalization policy. This sheds light on how Lambeth residentsvalue the total social effects of depenalization in the long run, not just

5 We also contribute to the literature examining the impact of drug policies on drugusage. The earlier evidence is mixed: some studies find little evidence of increased drugusage either in the United Kingdom ðWarburton, May, and Hough 2005; May, Duffy, et al.2007; Pudney 2010Þ or in other countries ðSingle 1989; DiNardo and Lemieux 2001;MacCoun and Reuter 2005; Hughes and Stevens 2010Þ, and others find slight increasesðWilliams 2004; Damrongplasit, Hsiao, and Zhao 2010Þ. Our reduced-form results suggestthat there might have been a considerable increase in the equilibrium market size forcannabis in Lambeth. The structural model sheds light on how total usage might vary withcitywide depenalization.




those operating through changes in crime. In Section VII, we shed lighton what would be the impacts on crime if the same policy were to beapplied citywide, by developing and calibrating an equilibrium model ofcrime and the demand for cannabis. Section VIII presents conclusions.The Appendix contains further information related to the crime andhousing data and further robustness checks.

II. The Lambeth Cannabis Warning Scheme

A. Background

To understand why the LCWS policing experiment was introduced inLambeth in July 2001, we need to go back to the earlier UK policy debatestimulated by the publication of the Runciman Report in 2000. This wasa high-profile inquiry commissioned by the Police Foundation, whoseremit was to review and suggest amendments to the primary piece of UKlegislation governing the policing of illicit drugs: the Misuse of DrugsAct 1971. This laid out the three-tiered drug classification system used inthe United Kingdom, with assignment from class C to class A intended toindicate increasing potential harm to users: class A drugs are cocaine,crack, crystal meth, heroin, LSD, MDMA, and methadone; class B drugsare amphetamines and cannabis; and class C drugs are anabolic steroids,GHB, and ketamine. The Runciman Report called for the classificationsystem tomore closely follow the scientific evidence of relative harms and,consequently, that cannabis be reclassified from a class B to a class C drug.The report emphasized threebenefits of doing so: ðiÞ reducingnumbers ofindividuals being criminalized, ðiiÞ removing a source of friction betweenthe police and local communities, and ðiiiÞ freeing up police time.Subsequent to the Runciman Report, the Metropolitan Police Service

ðMPSÞ produced its own report on drugs policing, “Clearing the Decks.”This suggested the idea of a workable depenalization policy in May 2000.This report again emphasized that such a policy might enable the po-lice to divert resources toward areas of high priority if they were willingto explore alternatives to arrest for a number of minor crimes, includ-ing possession of cannabis. The notion that such a depenalization pol-icy might actually be implemented within London began to take hold ayear later in early 2001, when the police commander for the London bor-ough of Lambeth, Brian Paddick, conducted a staff consultation exerciseon drugs policing strategy. During the consultation, officers complainedthat they spent a considerable amount of time dealing with arrests for can-nabis possession, and this detracted from their ability to deal with high-priority crime such as street crime, to tackle class A drugs, and to respondto emergency calls.6

6 Police officers also reported concerns, following a recent disciplinary case, that theymight face formal sanctions if they continued to follow a long-standing unofficial practice




With the sanctioning of the Metropolitan Police Commissioner, SirJohn Stevens, the LCWS was introduced in Lambeth on July 4, 2001, as apilot project that was intended to run for 6 months. Under the scheme,those found in possession of small quantities of cannabis for their per-sonal use had the drugs confiscated, and an offense was still recorded,although individuals were given a warning rather than an arrest beingrecorded; prior to the policy, such individuals would have been arrestedðDark and Fuller 2002Þ. To be clear, the policy was designed to lead to nochange in how the police should record offenses related to cannabispossession, all else equal. Rather, it would reduce the penalties to offend-ing individuals such that they would not be arrested. As such, the LCWShad all the hallmarks of many policies tried around the world that havesought to depenalize rather than decriminalize the possession of smallquantities of cannabis ðDonohue et al. 2012Þ.There are various mechanisms through which such a depenalization

policy can affect drug crime, depending on whether and how such pol-icies alter the behavior of the police, cannabis users, and local residents.As emphasized throughout, it is likely that the policy induced changes inpolice behavior: under the policy the police can effectively reallocate re-sources from cannabis-related crime to other crimes. This has the obvi-ous benefit that it allows the police to better deal with non-drug-relatedcrime and should be evident in falling offense rates for other crimes andrising police effectiveness against such nondrug crime.7

Second, such changes in police behavior will induce endogenouschanges in behavior among cannabis users who perceive reduced penal-ties for being caught in possession of cannabis in Lambeth. As empha-sized in the structural model developed later, such users might origi-nate from Lambeth or other parts of London. If users assume there to belower penalties for being caught in possession of almost any quantity ofcannabis, then offense rates for cannabis possession should rise with theLCWS because the possession of such larger quantities of cannabis wouldstill be recorded as an offense and still lead to an arrest.8 Alternatively,the lower penalties might induce some individuals to start using canna-bis. If such new users then choose to possess sufficiently large quantities,this would again cause recorded cannabis offenses to increase with thepolicy, all else equal. Hence changes in police behavior can explain both a

7 Of course the behavior of illicit drug suppliers could also alter with depenalization.However, given the lack of information on the supply side and no reliable time series on drugprices by London borough, for the bulk of our analysis we do not focus on this channel. Wereturn to this issue in the conclusion.

8 Indeed, in a Metropolitan Police Authority ðMPAÞ review of the LCWS policy, Dark andFuller ð2002Þ note the ambiguity officers themselves faced in regard to establishing a clear

of dealing with people found in possession of cannabis by informally warning them anddestroying the drugs on the streets. Before the policy, such actions did not have officialsanction ðMay et al. 2002; Warburton et al. 2005; May, Duffy, et al. 2007Þ.




simultaneous increase in cannabis-related crime and a reduction in othertypes of nondrug crime.In an alternative scenario, any changes in police behavior induce no

change in the behavior of cannabis users, in terms of whether to pur-chase cannabis or where to purchase it. The LCWS should then lead tono change in recorded offenses in cannabis possession and mechani-cally reduce arrest and clear-up rates for cannabis possession: behaviorsthat previously would have been recorded as offenses would continue tobe classified as such, but the LCWS policy would lead to the number ofarrests and clear-ups for cannabis possession falling in this scenario.In the absence of any changes in behavior among cannabis users,

changes in offense rates for cannabis possession might also occur throughwhat criminologists refer to as a “net-widening effect” that operatesthrough changes in police reporting behavior ðChristie and Ali 2000;Warburton et al. 2005, May, Duffy, et al. 2007Þ. This states that depenali-zation policies allow the police to start formally dealing with cannabisoffenses where previously they might have issued informal warnings andno offense was recorded. Indeed, given the documented heterogeneityin behavior of individual police officers in relation to drugs policing ðMay,Duffy, et al. 2007Þ, we would certainly expect some element of net widen-ing to occur under the LCWS. In consequence, the LCWS would causerecorded offense rates for cannabis possession to increase. This channelalone does not suggest any impact on arrest and clear-up rates for canna-bis possession, nor does it imply any change in police effectiveness againstnondrug crime.Finally, the policy might also induce changes in reporting behavior

among local residents. If they view the policy as signaling that the policewere devoting less effort toward cannabis-related crimes, residents mightthen be less inclined to report incidents involving cannabis possession.All else equal, this would cause a reduction in recorded cannabis offenses,but this channel alone should have no impact on arrest and clear-up ratesfor cannabis possession, nor on the incidence of nondrug crime. As we se-quentially present evidence on the impacts of the LCWS policy on cannabisoffenses, on measures of police effectiveness related to cannabis crime,and on the incidence and police effectiveness against other types of non-drug crime, we will be able to narrow down the likely dominant channelsthrough which the policy operates. It is these first-order channels we thencapture in our structural model that allow us to take the key lessons fromthe localized LCWS policing experiment and predict the likely impacts ofa counterfactual citywide depenalization policy.

threshold for what constituted a small quantity of cannabis possessed. Christie and Alið2000Þ report that in the context of depenalization in South Australia, small quantitiescorresponded to less than 100 grams of cannabis or 20 grams of cannabis resin.




B. Initial Public Reaction and the Evolution of the Policy

To gauge the initial local public reaction toward the LCWS, an Ipsos-MORI poll was commissioned during the 6-month policy experiment.This found broad support for the scheme among locals: 36 percent ofsurveyed residents approved outright of the policy; a further 47 percentapproved provided that the police actually reduced serious crime inLambeth. Following this groundswell of support, at the end of the trialperiod, the policy was then announced to have been extended for afurther 6 months. It is plausible that this extension might have beeninterpreted by cannabis users and the police as representing a perma-nent change in drug policing strategy.Anecdotal evidence then suggests that local support for the scheme

began to decline once the policy was announced to have been extendedbeyond the initial pilot. Media reports cited that local opposition arosebecause of concerns that children were at risk from the scheme and thatthe LCWS had led to an increase in drug tourism in Lambeth. TheLCWS formally ended on July 31, 2002. In part because of disagreementsbetween the police and local politicians over the policy’s true impacts,after the policy, Lambeth’s cannabis policing strategy did not returnidentically to what it had been before the policy. Rather, it adjusted tobe a firmer version of what had occurred during the pilot. More pre-cisely, the MPS announced that in Lambeth officers would continue torecord offenses for cannabis possession, and they would continue to is-sue warnings rather than necessarily arrest those in possession of can-nabis but would now also have the discretion to arrest where the offensewas aggravated. Aggravating factors included if ðiÞ the officer feareddisorder; ðiiÞ the person was openly smoking cannabis in a public place;ðiiiÞ those aged 17 or under were found in possession of cannabis; andðivÞ individuals found in possession of cannabis were in or near schools,youth clubs, or children’s play areas.

C. Other Police Operations

To place the LCWS into the wider context of other police operationsconducted in London, we have constructed a novel panel data set ofpolice operations by London borough-month for our sample period.This is described in Appendix table A1: As shown in panel A, for eachborough-specific police operation, we note the type of criminal offensetargeted and dates of operation. Some operations occur like the LCWS,within one borough; others are coordinated across boroughs. The lengthof police operations varies between a few months and 2 years. There isno evidence of a spike in police operations immediately after the LCWSis introduced, to perhaps reinforce or compensate for its effects. Panel Bshows borough-specific police operations for which we have incomplete




information on their dates of operation: many of these also operate withina single borough. Panel C shows police operations that are London-wide.Panel D records police operations that are referred to in MetropolitanPolice Authority ðMPAÞ reports, but for which we have insufficient detailin order to code in panels A–C. Overall, there is little evidence from ta-ble A1 suggesting that the impacts of the LCWS could be confounded withother police operations. In the Appendix we show the robustness of ourbaseline results when these other police operations are explicitly controlledfor.

III. Data, Descriptives, and Empirical Method

A. Data Sources

We exploit two sources of data to analyze how the LCWS affected crimein each London borough. First, we use administrative records obtainedfrom the London MPS to construct monthly panel data sets for variouscrime-related series. For any criminal act—such as the supply of can-nabis—the administrative records provide information on three crimeseries: the number of offenses, the number of arrests, and the number ofclear-ups. Each crime series panel covers all 32 London boroughs foreach month from April 1998. The crime series cover drug-related crimeas well as seven broad categories of nondrug crime: violence against theperson, sexual offenses, robbery, burglary, theft and handling, fraud andforgery, and criminal damage.Second, we use the Quarterly Labor Force Survey Local Area ðQLFS-

LAÞ data to obtain borough-level demographic and labor market char-acteristics. We interpolate this quarterly data set to the borough-monthlevel and use this to define our main outcome variable, offense rates forany given crime: the number of recorded offenses for that crime per1,000 of the adult population ðaged 16 and overÞ. We also use the QLFS-LA data to control for demographics and unemployment rates at theborough-month level in our empirical specifications, as described later.

1. Crime Data: Series Definitions

We describe the core definitional issues related to each crime series, fo-cusing on ðiÞ official HomeOffice guidelines for the recording of criminaloffenses, ðiiÞ the link between offenses and arrest data, ðiiiÞ the use ofwarnings by the police, and ðivÞ the definition of clear-ups and their linkto arrest data.9 The Appendix documents some of the important changestheHomeOffice has instigated in the way in which offenses and clear-ups

9 The HomeOffice is the UK government department that set the crime-recording rulesin our study period. It corresponds most closely to the Department of Homeland Securityand Department of Justice in the United States.




are defined over our study period. Such nationally determined defini-tional changes in crime series data apply equally in all London boroughsand so do not explain differences over time between Lambeth and otherLondon boroughs.Home Office guidelines state that as a result of a reported incident,

whether from victims, witnesses, or third parties, the incident will be re-corded as a crime by the police for offenses against an identified victimif, on thebalance of probability, ðaÞ the circumstances as reported amountto a crime defined by law ðthe police will determine this on the basis oftheir knowledge of the law and counting rulesÞ and ðbÞ there is no cred-ible evidence to the contrary. For offenses against the state, evidence ofan offense must be clearly presented before a crime is recorded.There are additional guidelines specifically related to how drug of-

fenses are counted. While these do not appear to provide any exceptionsto the above instructions for how drug-related offenses are recorded,these additional guidelines make clear that ðiÞ the general rule is onecrime per offender; so, for example, a stop and search of three indi-viduals all carrying cannabis will lead to three recordings of cannabispossession; ðiiÞ when an individual is found to be carryingmore than onedrug, the most serious class of drug possessed is that recorded; and ðiiiÞ ifan individual is found with several class B drugs including cannabis, thisis recorded as a cannabis offense.10

On the link between offenses and arrests, a recorded offense of canna-bis possession need not translate into an arrest if, for example, a memberof the public witnesses the offense but by the time the police show up tothe scene ðif at allÞ there are no individuals to arrest. Hence there can bea wedge between the number of offenses and the number of arrests, andthe size of this wedge differs across crime types because, for example,crimes vary in the extent to which they are reported by witnesses and theybring victims and perpetrators into direct contact, and so forth.On the issuance of warnings by police ðrather than arrestsÞ, we note that

for the bulk of our study period, warnings for cannabis possession werenot separately recorded for all boroughs. Our correspondence with thestatistical office of the MPS has also confirmed that during the period inwhich the LCWS was in operation, actual cannabis possession offenseswould continue to be recorded but no arrests made or clear-ups recorded.This is precisely as the policy was originally designed.11 Hence, if the be-

10 Home Office guidelines are available at http://www.gov.uk/government/uploads/system/uploads/attachment_data/file/177103/count-general-april-2013.pdf.

11 The Crime in England and Wales 2006/7 report states that, “From 1 April 2004 infor-mation on police formal warnings for cannabis possession started to be collected centrallyas part of the information held ðprior to this a pilot scheme was run in parts of LondonÞ.Those aged 18 and over who are caught in simple possession of cannabis can be eligible fora police formal warning which would not involve an arrest. An offence is deemed to be




havior of cannabis users remains unchanged, then the introduction ofthe LCWS policy should lead to no change in recorded offenses for can-nabis possession: the reason is that the policy was designed and practicedto lead to no change in how the police should record offenses relatedto cannabis possession, all else equal. However, under the policy, arrestand clear-up rates for cannabis possession should mechanically declinegiven that such incidents have been depenalized under the LCWS.Finally, for any crime to be counted as a clear-up, Home Office guide-

lines state that sufficient evidence must be available to claim a clear-up,and the following conditionsmust be met: ðiÞ a notifiable offense has beencommitted and recorded, ðiiÞ a suspect has been identified and has beenmade aware that he or she will be recorded as being responsible for com-mitting that crime and what the full implications of this are, and ðiiiÞ asanctioned clear-up or nonsanctioned clear-up method applies. In con-sequence, not every case in which the police know, or think they know,who committed a crime can be counted as a clear-up, and some crimesare counted as a clear-up even when the victimmight view the case as beingfar from solved. In short, a clear-up means that the case was closed, whetheror not anyone was actually sentenced.Hence, the primary reason why the series for arrests and clear-ups can

diverge is that an individual is arrested for an offense but is not charged.12

The relative frequency with which this occurs varies across crimes. Forsome offenses such as cannabis possession, arrest and clear-up time se-ries are nearly identical. For other crimes, such as violent crime or sexualoffenses, there is a greater divergence between the number of arrests andthe number of clear-ups. In studying the impacts of the LCWS on drugand nondrug crime, we exploit information on both arrests and clear-upseries: this information is crucial to measuring the police’s ability to ef-fectively reallocate resources toward nondrug crime as a result of the de-penalization of cannabis possession.

2. Drug Crime Data: Offense Types

For the crime series related to drug offenses, the administrative recordscontain information at two fine levels of detail. First, the records spec-ify the number of criminal offenses by drug type ðe.g., cannabis, heroin,cocaineÞ. We focus attention on cannabis and class A drug crimes as theseaccount for 95 percent of all drug crimes, as shown below. Second, for

12 Charging must occur within 24 hours of arrest, unless the crime is serious, in whichcase it may be extended by a police superintendent ð36 hoursÞ or a court ð96 hoursÞ.

cleared up if a formal warning for cannabis possession has been issued in accordance withguidance from the Association of Chief Police Officers” (Home Office 2007). Hence forthe bulk of our study period ðwhich runs from April 1998 until January 2006Þ, warnings forcannabis possession are not separately recorded for all boroughs.




each drug type, the data record the specific offense committed: possession,trafficking, intent to supply, and so forth. To shed light on whether anyobserved change in the number of cannabis offenses is driven predom-inantly by demand- or supply-side factors, we split cannabis offense typesinto two categories: we proxy changes in demand with the number of of-fenses related to cannabis possession, and we proxy changes in supply withthe number of offenses related to trafficking, intent to supply, and soforth.13 Both levels of disaggregation by drug and offense types are alsoavailable for the other two crime series: on arrests and clear-ups. We ex-ploit the full richness of this data set when studying the impacts of thedepenalization of cannabis on drug crime in Lambeth relative to the restof London.To make clear the levels and patterns of drug crime before the policy,

table 1 provides descriptive evidence on drug crime in Lambeth andother London boroughs before the LCWS was introduced.We define theoffense rate for cannabis-related crime as the number of offenses per1,000 of the adult population ðaged 16 and aboveÞ. Panel A highlightsthat Lambeth has historically higher rates of drug offenses than otherLondon boroughs: in the average month before the policy since April1998, there were 0.608 offenses per 1,000 of the adult population inLambeth, while the average for the rest of London was 0.400. To put thisinto perspective, we note that the prepolicy adult population in Lambethwas approximately 240,000, so around 146 drug-related offenses were be-ing recorded in Lambeth each month before the policy. Out of 32 bor-oughs, Lambeth would be ranked sixth highest in terms of drug-relatedoffense rates before the policy.Panel B highlights the composition of drug offenses by drug type. In

line with some of the motivations for depenalization, the majority ofdrug offenses relate to cannabis: 60 percent of all drug offenses relate tocannabis in Lambeth; for other London boroughs this figure is closer to74 percent. The incidence of offenses related to class B drugs ðexcludingcannabisÞ and class C drugs is relatively minor, corresponding to less than5 percent of all recorded drug offenses. In consequence, Lambeth hasrelatively more drug offenses related to class A drugs than other Londonboroughs.Panel C shows how cannabis offenses break down by crime types , which

can be roughly classified as demand- and supply-side offenses. In Lam-beth, 91 percent of cannabis offenses are for cannabis possession, with theremainder mostly related to intent to supply offenses. This breakdown bycannabis offense type is not significantly different between Lambeth and

13 These supply-side offenses include possession with intent, possession on a ship, pro-duction, supply, unlawful export, unlawful import, carrying on a ship, inciting others tosupply, manufacture, and money laundering. There are a very small number of other offensesthat cannot be classified as either demand or supply related.




other London boroughs. The levels of cannabis-related drug crime docu-mented in table 1 certainly make it plausible that a cannabis depenaliza-tion policy could save considerable amounts of police time and resourcesthat could potentially be reallocated toward class A drug crime or non-drug crime.

3. Descriptive Time-Series Evidence on Crime

To begin to establish whether and how the LCWS policy might have af-fected drug and nondrug crime in London, we present three pieces of

TABLE 1Detailed Drug Offenses, Prepolicy Period

Lambethð1Þ

Other LondonBoroughs

ð2ÞA. Total

Total drug offenses per 1,000 of adult population .608 .400ð.124Þ ð.298Þ

B. Drug Type

Share of drug offenses related to any cannabisoffenses .600 .735

ð.052Þ ð.108ÞShare of drug offenses related to class A drugs .344 .204

ð.054Þ ð.106ÞShare of drug offenses related to class B drugsðincluding cannabisÞ .628 .770

ð.057Þ ð.110ÞShare of drug offenses related to class C drugs .002 .004

ð.004Þ ð.010ÞC. Cannabis Offenses

Breakdown

Share of cannabis offenses related to having possessionof cannabis .907 .918

ð.044Þ ð.055ÞShare of cannabis offenses related to having possessionof cannabis with intent to supply .055 .049

ð.031Þ ð.043ÞShare of cannabis offenses related to production/being concerned in production of cannabis .015 .013

ð.016Þ ð.021ÞShare of cannabis offenses related to supply or offer tosupply cannabis .023 .019

ð.020Þ ð.027ÞNote.—Entries are means, and standard deviations are in parentheses. The prepolicy

period runs from April 1998 until June 2001. Other London boroughs are all London bor-oughs except Lambeth. Class A drugs are cocaine, crack, crystal meth, heroin, LSD,MDMA,and methadone; class B drugs are amphetamines and cannabis ðin the prepolicy periodÞ;class C drugs are anabolic steroids, GHB, and ketamine.




descriptive evidence. Figure 1A shows the monthly time series for thenumber of cannabis drug offenses per 1,000 of the adult population forLambeth and the average for all other London boroughs. The periodduring which the LCWS is in place is indicated by the dashed vertical lines.Four points are of note.First, prior to the introduction of the LCWS, there is a downward

trend in cannabis offense rates in Lambeth and London more generally.Second, there is a large increase in cannabis offense rates in Lambethduring the policy. On average within the prepolicy and policy periods,cannabis offenses in Lambeth rose by 61 percent in the policy periodrelative to the prepolicy period. For the rest of London, there was nosignificant change in cannabis offenses between these time periods. Third,the dramatic upturn in offenses occurs 6 months after the policy starts—precisely the time when the policy extension is announced—rather thanimmediately after the policy experiment is first introduced. This suggeststhat the impact of the announcement of the policy’s extension, ratherthan its mere introduction, is key for understanding changes in canna-bis crime. At face value this casts further doubt on whether all the changein cannabis offenses can be understood through merely a net-wideningeffect of changes in police reporting behavior or changes in reportingbehavior of local residents. Fourth, the rise in cannabis offenses is quan-titatively large and appears permanent. There is little evidence from fig-ure 1A that the time series for Lambeth begins to converge back to itsprepolicy level or those of the other boroughs in the postpolicy period.Indeed, before the policy, cannabis-related offenses continue to rise by afurther 46 percent in Lambeth.Figure 1B then focuses exclusively on offenses of cannabis possession.

This time series mimics the pattern for cannabis offenses as a whole sothat possession-related offenses, which constitute the bulk of cannabis-related crime as shown in table 1, do indeed drive the increase in can-nabis offenses in aggregate.It seems unlikely that these policy impacts simply reflect changes in the

likelihood that either police or local residents report the cannabis pos-session offenses that they witness. Before, during, and after the LCWSpolicy, the police were required to report all cannabis offenses they ob-served. Furthermore, there is no reason to expect local residents to be-come more likely to report cannabis offenses during the LCWS since theyhad reason to expect that the introduction of LCWS decreased the prob-ability that such reports would result in sanctions for offenders. Thus, ourevidence strongly suggests that, both in levels and relative to other bor-oughs, cannabis use in Lambeth increased substantially following the im-plementation of the LCWS. In the remainder of the paper, we focus onhow changes in the behavior of Lambeth police may have induced this in-crease in cannabis consumption.




FIG.1.—

A,A

ggregatecannab

isoffen

ses;B,cannab

ispossessionoffen

ses;C,agg

regatenondrugoffen

ses.Thesampleperiodrunsfrom

April199

8untilJan

uary20

06.

Thetwovertical

lines

representthestartan

den

doftheLam

bethpolicy

ðJuly20

01an

dJuly20

02,respectivelyÞ.I

neach

figu

re,theblack

timeseries

representsthe

relevanttimeseries

forLam

beth.Thegray

series

representsthemeanoffen

sesper

capitafortherestofLondon.Pan

elA

showsthetimeseries

forthenumber

of

cannab

is-related

offen

sesin

aggregateper

1,00

0ofthead

ultpopulation.P

anelBshowsthetimeseries

forthenumber

ofcan

nab

ispossessionoffen

sesper

1,00

0ofthe

adultpopulation.P

anel

Cshowsthetimeseries

ofthenumber

ofnondrugoffen

sesper

1,00

0ofthead

ultpopulation.N

ondrugoffen

sesincludethose

forviolence

againsttheperson,sexualoffen

ses,robbery,burglary,theftan

dhan

dling,frau

dorforgery,an

dcrim

inaldam

age.



A key dimension along which changes in police behavior could thenaffect crime is through nondrug crime. The final piece of descriptive ev-idence we therefore present is the time series for all nondrug offensesaggregated to a single series for Lambeth and the rest of London. Asfigure 1C shows, prior to the LCWS’s introduction, we observe upwardtrends in such crime rates in Lambeth and across London as a whole.However, a fewmonths into thepolicy period, rates of criminal offenses fornondrug crime begin declining in Lambeth, and this downward trendcontinues in the long run. In contrast to the rest of London, nondrug of-fenses remain relatively constant for the second half of the sample period.While far from definitive, this is the first piece of evidence that hints atthe importance of changes in police behavior and potential reallocationsof police resources from cannabis-related crime toward nondrug crime,which might then induce changes in behavior among cannabis users, tobest explain the full set of descriptive evidence.

B. Empirical Method

To establish whether there is a causal impact of the LCWS policy oncrime, we estimate the following panel data specification for borough bin month m in year y :

lnCbmy 5 b0Pmy 1 b1ðLb � PmyÞ1 b2PPmy 1 b3ðLb � PPmyÞ1 gXbmy 1 lb 1 lm 1 ubmy;

ð1Þ

where Cbmy is the offense rate for a given crime type. The offense rate isdefined as the number of criminal offenses per 1,000 of the adult pop-ulation ðaged 16 and overÞ. The terms Pmy and PPmy are dummies for thepolicy and postpolicy periods, respectively. The term Lb is a dummy forthe borough of Lambeth. The parameters of interest are estimated fromwithin a standard difference-in-difference research design: b1 and b3

capture differential changes in crime rates in Lambeth during and afterthe LCWS policy period, relative to other London boroughs; b0 and b2

capture London-wide trends in offense rates during the policy and post-policy periods.All other London boroughs are included as part of the sample when

estimating ð1Þ. Given the interlinkages across locations in cannabis mar-kets, it is likely that after the LCWS is introduced, some individuals willbe induced to start traveling to Lambeth to purchase cannabis there. Thisimpact is spread over all 31 other London boroughs ðand beyondÞ and sois unlikely to lead to a discernible upward bias in the coefficients of in-terest. However, to shed some light on this, in the Appendix we presenta robustness check that estimates ð1Þ when boroughs neighboring Lam-beth are excluded from the sample ðand find results very similar to thebaseline estimates presentedÞ.




While administrative data on offenses are available for each monthfrom April 1998 onward, the QLFS-LA data from which the denominatorfor offense rates is measured are available only until 2005:Q4. Hence ourstudy period for analyzing the impacts of the LCWS runs from April 1998until January 2006, covering 3 years before the policy, the 13 months ofthe policy, and 3½ years after the policy. In Xbmy, we control for the fol-lowing borough-specific time-varying variables: the share of the adult pop-ulation that is an ethnic minority; the share that is aged 20–24, 25–34,35–49, and above 50 ðthose aged 16–19 are the omitted categoryÞ; andthe male unemployment rate. The fixed effects capture remaining time-invariant differences in offense rates across boroughs ðlbÞ and monthlyvariation in crime ðlmÞ. We weight observations by borough population.Finally, defining time t as the number of months since January 1990—t 5 ½12 � ðy 2 1990Þ�1 m—in our baseline specification, we assume aPrais-Winsten borough-specific ARð1Þ error structure, ubmy 5 ubt 5 rbubt21 1 ebt ,where ebt is a classical error term and ubmy is borough specific, heteroske-dastic, and contemporaneously correlated across boroughs.

IV. Results

A. Cannabis Crime in Aggregate

Table 2 presents estimates of ð1Þ in which we focus on how the policyaffects the rate of cannabis offenses in aggregate. Column 1 estimatesequation ð1Þ conditioning only on borough andmonth fixed effects. Theresults replicate the descriptive evidence presented earlier: offense ratesfor cannabis-related crime rise by 32.5 percent more in Lambeth relativeto the rest of London between the prepolicy and policy periods. The co-efficient on the policy period dummy, b0, is close to zero, suggesting thatthere is no citywide time trend in cannabis crime rates during the policyperiod. When we compare the prepolicy and postpolicy periods, canna-bis offenses are 61.5 percent higher in Lambeth vis-a-vis the rest of Lon-don. The postpolicy period dummy, b2, is positive and significant, sug-gesting that the long-run rises in Lambeth occur against a backdrop ofsignificantly smaller, but rising, offense rates for the rest of London be-tween August 2002 and January 2006.Column 2 shows the results to be robust to including the full set of

covariates in ð1Þ. These baseline results suggest that the depenalizationof cannabis in Lambeth led to a significant increase in cannabis offensesboth during the policy period and well after the policy officially ended.The next two specifications additionally control for within-borough lin-ear and quadratic time trends, respectively. As expected, the policy effectsare less precisely estimated and of slightly smaller magnitude. As col-umns 3 and 4 show, once we also control for within-borough time trends,it is no longer possible to identify an effect of the policy during its period




TABLE2

TheEffectoftheDepenalizationonCannabisOffensesinAggregate

DependentVariable:LogðTotalRecorded

Can

nab

isOffen

ses,per

1,00

0ofAdultPopulationÞ

Fixed

Effects

ð1Þ

Baseline

ð2Þ

Borough

-Specific

LinearTim

eTrend

ð3Þ

Borough

-Specific

Quad

raticTim

eTrend

ð4Þ

Within-Policy

Dyn

amics

ð5Þ

Lam

beth�

policy

period

.325

***

.293

**.195

.182

ð.117

Þð.1

18Þ

ð.148

Þð.1

45Þ

Policy

period

.018

.034

.023

.182

***

.034

ð.056

Þð.0

56Þ

ð.065

Þð.0

51Þ

ð.056

ÞLam

beth�

postpolicy

period

.615

***

.610

***

.414

**.479

**.682

***

ð.092

Þð.0

96Þ

ð.201

Þð.1

86Þ

ð.076

ÞPostpolicy

period

.171

***

.181

***

.160

*.237

***

.180

***

ð.043

Þð.0

47Þ

ð.090

Þð.0

66Þ

ð.047

ÞLam

beth�

policy

periodð1–6

monthsÞ

2.026

ð.120

ÞLam

beth�

policy

periodð7–1

3monthsÞ

.647

***

ð.118

ÞBorough

andmonth

fixe

deffects

Yes

Yes

Yes

Yes

Yes

Sociodem

ograp

hic

controls

No

Yes

Yes

Yes

Yes

Observations

3,00

83,00

83,00

83,00

83,00

8

Note.—Allobservationsareat

theborough

-month-yearlevel.Thesample

periodrunsfrom

April19

98untilJanuary20

06.Controlborough

sareall

other

Londonborough

s.Pan

el-correctedstan

darderrors

arecalculatedusingaPrais-W

insten

regression,whereaborough

-specificARð1Þp

rocess

isassumed

.Thisalso

allowstheerrorterm

sto

beborough

specific,heterosked

astic,an

dco

ntemporaneo

uslyco

rrelated

acrossborough

s.Observationsare

weigh

tedbytheshareofthetotalLondonpopulationthat

month-yearin

theborough

.Thepolicy

perioddummyvariab

leiseq

ual

toonefrom

July20

01untilJuly20

02,andzero

otherwise.Thepostpolicy

perioddummyvariab

leiseq

ualto

onefrom

July20

02onward,andzero

otherwise.Column1ad

ditionally

controlsonlyforborough

andmonth

fixedeffects.In

cols.2–5,thefollowingsociodem

ograp

hiccontrolvariables,measuredin

logs,are

controlled

forat

the

borough

-month-yearlevel:theshareofthead

ult

population

that

isethnic

minority;that

isaged

20–24,

25–34,

35–4

9,an

dab

ove

50;an

dthemale

unem

ploym

entrate.Column3ð4Þa

dditionallycontrolsforaborough

-specificlinearðquad

raticÞ

timetren

d.

*Sign

ificantat

10percent.

**Sign

ificantat

5percent.

***Sign

ificantat

1percent.



of operation. This is hardly surprising given that the policy is in opera-tion for only 13 months. However, in both specifications the postpolicyeffect remains highly significant, suggesting that postpolicy offense ratesfor cannabis crime were at least 41.4 percent higher than for the rest ofLondon, all else equal.14

Following the time-series evidence in figure 1A , the specification incolumn 5 checks for differential policy responses during the first 6monthsof the policy, when the LCWS was announced to be a temporary policingexperiment, and the last 7 months, after it was announced to have beenextended. In line with the evidence in figure 1A , all the significant within-policy effect on cannabis offenses occurs after the second policy an-nouncement. We can only speculate on why this second announcement isthe trigger for cannabis offenses to rise. If, for example, it is interpreted asa signal of the policy’s permanence, then as there are fixed costs to re-structuring police resource allocations, the policemight have incentives todelay any large changes in their organization until the policy is presumedto be permanently in place.Clearly understanding such dynamic and announcement effects of

policy needs more research, but this finding does help to immediatelyaddress two issues. First, it suggests that the LCWS was not introduced inresponse to rising cannabis crime rates: as figure 1A shows, cannabis of-fenses were generally trending downward in Lambeth in the years priorto the introduction of the LCWS. Second, this casts doubt on whether allthe change in cannabis offenses can be understood through changes inreporting behavior of local residents or solely through a net-wideningeffect caused by changes in the way the police recorded cannabis of-fenses. If so, we would expect such effects to be picked up as soon as theLCWS comes into effect amid much media publicity, and we would ex-pect such effects to be affected by the policy officially ending.15

In the Appendix we detail robustness checks on the baseline specifi-cation estimated in column 2 of table 2. These address concerns relatedto ðiÞ excluding neighboring boroughs as valid controls, ðiiÞ accountingfor common citywide shocks to cannabis crime through the inclusion ofyear fixed effects, ðiiiÞ controlling for a series of dummies that capture

14 As a related robustness check, we estimated ð1Þ restricting the sample to a 12-monthwindow around the policy, i.e., from July 2000 until July 2003. Hence the policy and post-policy effects are not identified assuming any particular underlying long-run time trends.The previous results are robust to using this narrower time frame. Indeed, this specifica-tion shows that over this shorter time frame, when drug offenses are still found to have risenin Lambeth, drug offenses are declining elsewhere in London as suggested by App. fig. A1A .

15 We also estimated a specification breaking down the postpolicy response for each year.This confirmed the postpolicy effects on cannabis crime to be long-lasting: we cannot rejectthe null that the effect in Lambeth is the same in the first and fourth years after the policy.These helps address concerns that cannabis crime rates in Lambeth were naturally diverg-ing away from those in the rest of London.




each period when specific Home Office reporting guidelines are in place,ðivÞ controlling for other police operations in London, and ðvÞ estimatingstandard errors allowing for spatially correlated error structures. In allcases we find results qualitatively similar to the baseline estimates pre-sented: the magnitude of the long-run policy impact on cannabis offensesin aggregate varies between 41.4 percent and 68.2 percent across the ro-bustness checks and is significantly different from zero in each specifi-cation.

B. Cannabis Crime: Demand and Supply lmpacts

We now further unpack the mechanisms lying behind the main resultfrom table 2, that aggregate cannabis crime rises in Lambeth relative tothe rest of London, in both the short and long term, after the depenal-ization of cannabis possession in Lambeth. To do so we exploit the factthat the administrative crime records break down cannabis crime intospecific types of crime. We do so along two natural margins: ðiÞ offensesrelated to cannabis possession, which might be more attributable tochanges in the demand for cannabis; and ðiiÞ offenses related to canna-bis trafficking and supply, which might be more attributable to changesin cannabis supply.16

For both demand- and supply-side cannabis crimes, we also exploremeasures of police behavior such as ðthe log ofÞ arrest rates and clear-uprates. As individuals are not necessarily immediately arrested for cannabis-related offenses they commit, we define the arrest rate as the number ofarrests in the borough in period t divided by the number of offenses com-mitted between month t and the previous quarter within the borough.The clear-up rate is analogously defined: the number of clear-ups in theborough in period t divided by the number of offenses committed be-tween month t and the previous quarter within the borough.17

Table 3 presents the results. In each column, specifications analogousto ð1Þ are estimated, where the crime series now refer to subcategories ofcannabis crime. Columns 1–4 have as dependent variables Cbmy, crimeseries related to cannabis possession, proxying the demand for cannabis;columns 5–8 explore crime series related to cannabis supply ðthe sam-ple size drops slightly in these specifications because crimes related tocannabis supply do not necessarily occur in every borough-monthÞ. Fur-thermore, given the earlier finding in column 5 of table 2, we divide

16 Of course, this classification of offenses into demand and supply related is only ap-proximate. For example, itmight be substantiallymore difficult to prove an offense of intentto supply, so that in practice the police use their discretion so that some drug suppliers arecharged with a lesser offense of possession.

17 Ideally, the clear-up rate in time period t would be defined as the number of clear-upsin time t divided by the stock of unsolved offenses at the time, but such data are unavailable.




TABLE3

TheEffectoftheDepenalizationontheDemandforandSupplyofCannabis-RelatedCrime

CannabisPossessionðD

eman

dÞ

CannabisSupply

Offen

ses

ð1Þ

Arrests

ð2Þ

Clear-ups

ð3Þ

Clear-ups

per

Arrest

ð4Þ

Offen

ses

ð5Þ

Arrests

ð6Þ

Clear-ups

ð7Þ

Clear-ups

per

Arrest

ð8Þ

Lam

beth

�policy

periodð1–6

monthsÞ

2.036

2.436

**21.55

6***

21.19

9***

.236

2.250

2.287

*2.043

ð.127

Þð.1

92Þ

ð.349

Þð.2

12Þ

ð.167

Þð.1

76Þ

ð.173

Þð.0

87Þ

Lam

beth

�policy

periodð7–1

3monthsÞ

.675

***

2.946

***

21.55

8***

2.490

*.505

***

2.149

2.095

.039

ð.124

Þð.1

81Þ

ð.393

Þð.2

66Þ

ð.165

Þð.1

66Þ

ð.163

Þð.0

81Þ

Policy

period

.035

2.010

2.027

2.017

**2.016

2.024

2.023

.007

ð.055

Þð.0

63Þ

ð.065

Þð.0

08Þ

ð.064

Þð.0

43Þ

ð.043

Þð.0

15Þ

Lam

beth

�postpolicy

period

.686

***

2.094

21.04

7***

2.576

**.676

***

2.007

.077

.077

*ð.0

80Þ

ð.102

Þð.3

57Þ

ð.288

Þð.1

01Þ

ð.093

Þð.0

89Þ

ð.046

ÞPostpolicy

period

.192

***

2.049

2.028

.022

***

.034

2.069

**2.064

**.003

ð.046

Þð.0

47Þ

ð.048

Þð.0

07Þ

ð.043

Þð.0

32Þ

ð.031

Þð.0

12Þ

Borough

andmonth

fixe

deffects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Sociodem

ograp

hic

controls

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Observations

3,00

83,00

83,00

83,00

82,75

62,72

22,71

12,98

7


theborough


periodrunsfrom

April19

98untilJanuary20

06.Controlborough

sareall

other

Londonborough

s.Thedep

enden

tvariab

lein

cols.1an

d5is

thelogofthenumber

ofoffen

sesforeach

offen

setype,

per

1,00

0ofthead

ult

population.Thedep

enden

tvariab

lein

cols.2an

d6isthearrestrate

foreach

offen

setype,

defi

ned

asthelogofthenumber

ofarrestsdivided

bythe

number

ofoffen

sesin

theborough

inthesamemonth

andpreviousquarter.Thedep

enden

tvariab

lein

cols.3

and7istheclear-uprate

foreach

offen

setype,

defi

ned

asthelogofthenumber

ofclear-upsdivided

bythenumber

ofoffen

sesin

theborough

inthesamemonth

andpreviousquarter.The

dep

enden

tvariab

lein

cols.4

and8istheratioofclear-upsto

arrests,defi

ned

asthelogofthenumber

ofclear-upsdivided

bythenumber

ofarrestsin

the

samemonth.Inco

ls.1

–4theoffen

setyperelatesto

cannab

ispossession.Inco

ls.5

–8theoffen

setypeisthesum

ofalloffen

sesrelatedto

cannab

issupply

includingpossessionwithintent,possessiononaship,p

roduction,supply,u

nlawfule

xport,u

nlawfulimport,carryingonaship,incitingothersto

supply,

man

ufacture,an

dmoney

laundering.

Pan

el-correctedstan

darderrors


insten


-specificARð1Þ

process

isassumed

.Thisalso

allowstheerrorterm

sto

beborough

specific,

heterosked

astic,

andco

ntemporaneo

uslyco

rrelated

across

borough

s.Ob-

servationsareweigh


month-yearin

theborough

.Thepolicy

perioddummyvariab

leiseq

ual

toone

from

July20

01untilJuly20

02,andzero

otherwise.Thepostpolicy

perioddummyvariab

leiseq

ualto

onefrom

July20

02onward,andzero

otherwise.The

followingsociodem

ograp

hic

controlvariab

les,measuredin

logs,areco

ntrolled

forat

theborough


ultpopulation

that

isethnic

minority;that

isaged

20–2

4,25

–34,

35–49

,an

dab

ove

50;an

dthemaleunem

ploym

entrate.In

addition,thelogofthead

ultpopulation

isincluded

asaco

ntrolin

cols.2–

4an

dco

ls.6–

8.*Sign

ificantat

10percent.

**Sign

ificantat

5percent.

***Sign

ificantat

1percent.



the policy period into two halves tomore precisely understand the effectsof the LCWS on the market for cannabis when it is announced as atemporary policy experiment vis-a-vis a more permanent change in po-licing strategy.

1. Cannabis Demand

On the demand for cannabis, column 1 shows that offense rates for can-nabis possession rise only after the policy is announced to have been ex-tended: this increase of 67.5 percent in offense rates for cannabis posses-sion in the second half of the policy period closely matches the descriptiveevidence in figure 1B . We find no evidence that rates of cannabis posses-sion in other London boroughs change significantly during the policy pe-riod. In the longer term, postpolicy cannabis possession offense rates re-main 68.6 percent higher in Lambeth relative to the rest of London.To focus on changes in police behavior that the LCWS induced, we

next estimate equation ð1Þ but using as the dependent variable the arrestrate for cannabis possession. Column 2 shows that relative to the pre-policy period, arrest rates for cannabis possession in Lambeth signifi-cantly drop by 43.6 percent in the first half of the policy period and by94.6 percent in the second half of the policy period. However, after thepolicy, arrest rates return back to their prepolicy levels ðb3 5 0Þ.The next specification considers another dimension of police behav-

ior: clear-up rates for cannabis possession offenses. Column 3 shows asignificant fall in clear-up rates in Lambeth for cannabis possession assoon as the LCWS policy is introduced.18 In the longer term, police ef-fectiveness in Lambeth for crimes related to cannabis possession appearsweakened relative to the prepolicy period: clear-up rates remain signifi-cantly lower. This occurs at a time when there are no London-wide trendsin clear-up rates ðb2 is not significantly different from zero in col. 3Þ. Atthe same time, as previously noted in column 1, in the longer term, post-policy offense rates remain 68.6 percent higher in Lambeth than in theprepolicy period, suggesting that the demand for cannabis remains per-manently higher long after the LCWS policy officially ends.Perhaps the cleanest way to measure police effectiveness is to consider

the ðlog ofÞ clear-ups per arrest in any given period t as the dependentvariable in ð1Þ: this captures the rate of conversion of arrests into clear-ups as arrestees are charged for cannabis possession. The result in col-umn 4 shows a significant fall in clear-ups per arrest in Lambeth dur-ing the policy period and, more notably, a significant fall of 57.6 percent

18 The fact that the impacts on arrest and clear-up rates for cannabis possession are quali-tatively similar is not surprising: as described in Sec. III.A.1, the arrest and clear-up series divergeonly if individuals are arrested but not charged for cannabis possession. This occurs far morerarely for cannabis possession offenses than for some other nondrug crime we later analyze.




after the policy. This occurs against a backdrop of significantly risingclear-ups per arrest for cannabis possession in the rest of London in thepostpolicy period.In summary, the measures of police behavior used in columns 2–4 in-

dicate that once depenalization is in place, the police immediately de-vote less effort toward targeting cannabis users. On the one hand, thisis reassuring because it is precisely what the depenalization policy pre-scribes: cannabis possession no longer leads to arrests ðalthough offensesshould be recorded in the same way as before the policyÞ, and so we ex-pect to observe immediate falls in arrest and clear-up rates as soon as thepolicy is introduced. However, such a weakened deterrence effect of de-penalizationmight in turn affect the behavior of cannabis users, ultimatelyfeeding through to drive the significant rise in cannabis possession of-fenses 6 months into the policy, as shown in column 1.19

In the longer term, there remains evidence that police effectivenessagainst cannabis possession offenses is lower than in the prepolicy pe-riod, in line with the description of the policy evolution given in Sec-tion II.C: in the longer term, policing strategies in Lambeth did not re-vert to identically what was in place before the policy. This opens up thepossibility that in Lambeth, police resources are permanently reallocatedtoward class A drug crime and nondrug crime, as we explore in detail inSection V.

2. Cannabis Supply

The remaining columns of table 3 repeat the analysis for crime seriesrelated to the supply of cannabis. We find evidence that the LCWS sig-nificantly increased offenses related to cannabis supply during its offi-cial period of operation: by the second half of the policy period, offenserates for cannabis supply were 50.5 percent higher in Lambeth relativeto the prepolicy period, an impact significant at the 1 percent level. Wealso find that in the postpolicy period, cannabis supply offenses rose by67.6 percent more in Lambeth relative to the rest of London, and thereis no long-term citywide time trend in such crimes. On police effective-ness against crime related to supplying cannabis, columns 6–8 documentno changes during the policy period in terms of arrests and a fall in clear-up rates that is significant at the 10 percent level. For our preferred mea-sure of police effectiveness, clear-ups per arrest do not change signifi-cantly during the policy period and, in the longer term, rise slightly inLambeth relative to the rest of London ðan effect significant at the 10 per-cent levelÞ, at a time when citywide police effectiveness against cannabis

19 Durlauf and Nagin ð2012Þ provide a comprehensive overview of the literature on theevidence in favor of deterrence effects from a range of crime policies.




supply–related crime appears to be either falling ðcols. 6 and 7Þ or stableðcol. 8Þ.20Taken together the results suggest that any change in the underlying

size of the market for cannabis in Lambeth as a result of the policy wasdriven by demand- and supply-side factors. However, while police effec-tiveness against demand-side offenses remained permanently lower afterthe policy, police effectiveness against crimes related to cannabis supplymarginally improved in Lambeth in the longer term even after the LCWSwas officially ended.21This hints at thepossibility that thepolicewere able toreallocate their effort away from incidents related to cannabis possessiontoward other drug crime and nondrug crime.We now explore this inmoredetail.

V. The Reallocation of Police Effort

The results in tables 2 and 3 document changes in levels and composi-tion of cannabis-related crime following the depenalization of cannabispossession in Lambeth. These results suggest that the primary mechan-isms at play driving the policy impacts are changes in behavior of the po-lice and cannabis users. Focusing on these channels, we now investigatethe short- and long-term impacts the depenalization policy had on the in-cidence of, and police effectiveness against, crime related to class A drugsand nondrug crime in seven categories: violence against the person, sex-ual offenses, robbery, burglary, theft and handling, fraud and forgery, andcriminal damage.

A. Crime Related to Class A Drugs

As the administrative crime data record drug crime by drug type, we firstexamine whether the LCWS policy allowed police in Lambeth to real-locate their effort toward class A drugs, which constitute the bulk ofnoncannabis drug crime ðtable 1, panel BÞ. As described in Section II,that the policy might enable the retargeting of police resources towardcrime related to class A drugs was one motivation behind the introduc-tion of the LCWS, as is often the case for depenalization policies in othercontexts.

20 We note that all the results presented in cols. 2–4 and 6–8 are largely robust to de-fining arrest and clear-up rates as being per 1,000 of the adult population rather than perthe number of offenses in the previous quarter. The results are not therefore driven by theincrease in offenses previously noted.

21 For brevity, we have not shown the dynamic policy response along these margins whenwe split the postpolicy period year by year. Doing so, we find that the significant increase incannabis possession offenses remains in each of the 4 years after the policy, as does the in-crease in cannabis supply–related offenses.




We estimate specifications analogous to ð1Þ breaking the results downalong two margins: ðiÞ crime series related to the possession of class Adrugs, proxying the demand for such illicit substances; and ðiiÞ crimeseries related to the supply of class A drugs. As for cannabis crime, we doso for crime series on offense rates and measures of police effectivenesssuch as arrest and clear-up rates. Table 4 shows the results. To facilitatecomparison with the previously documented impacts on cannabis crime,we again divide the policy period into two halves.On the demand side, table 4 shows the following: ðiÞDuring the policy

period there is an impact of depenalizing cannabis possession on thedemand for class A drugs as proxied by possession offenses for such sub-stances ðcol. 1Þ. ðiiÞ In the longer term, offenses related to the posses-sion of class A drugs significantly rise by 12.0 percent in Lambeth relativeto the rest of London; this increase occurs against the backdrop of nochange in citywide offense rates for class A drug possession. ðiiiÞ There islittle robust evidence of a change in police effectiveness against crimerelated to the possession of class A drugs, as measured by arrest rates,clear-up rates, and clear-ups per arrest ðcols. 2–4Þ. Hence, the evidencedoes not suggest that the Lambeth police turned a blind eye toward classA drug possession in Lambeth during or after the LCWS policing ex-periment.The remaining columns of table 4 show crimes series related to the

supply of class A drugs. We find no evidence that the LCWS policy affectsoffense rates related to the supply of class A drugs during the policy periodbut a significant fall in such offenses after the policy. We also find some-what mixed evidence on any impact on the police effectiveness againstcrimes related to the supply of class A drugs: we observe no significantchanges in arrest or clear-up rates ðcols. 6 and 7Þ, but there is a significantincrease of 12.3 percent in clear-ups per arrest ðcol. 8Þ.Taken together, the results show that in the long term, the patterns

of demand-related class A drug crime in Lambeth along all three mar-gins of offenses, arrests, and clear-ups do not differ much from London-wide trends more generally. This is in sharp contrast to the previouslydocumented effects on cannabis demand offenses, arrests, and clear-upsshown in table 3. However, the evidence in the second half of table 4hints at the possibility that the police might have reallocated effort to-ward supply-related class A drug crime: offense rates for crimes relatedto the supply of class A drugs significantly fall in the longer term, and po-lice effectiveness against such crimes, at least as measured by clear-upsper arrest, significantly rise.

B. Nondrug Crime

Motivated by the earlier descriptive evidence from figure 1C on trends innondrug crime in Lambeth relative to other London boroughs, we now




TABLE4

TheEffectoftheDepenalizationontheDemandforandSupplyofClassADrugs–R

elatedCrime

ClassADrugsPossessionðD

eman

dÞ

ClassADrugsSupply

Offen

ses

ð1Þ

Arrests

ð2Þ

Clear-ups

ð3Þ

Clear-ups

per

Arrest

ð4Þ

Offen

ses

ð5Þ

Arrests

ð6Þ

Clear-ups

ð7Þ

Clear-ups

per

Arrest

ð8Þ

Lam

beth�

policy

periodð1–6

monthsÞ

2.236

**2.114

2.059

.034

2.343

2.380

2.335

.028

ð.115

Þð.1

55Þ

ð.149

Þð.0

24Þ

ð.340

Þð.3

47Þ

ð.389

Þð.1

10Þ

Lam

beth�

policy

periodð7–1

3monthsÞ

.081

2.070

2.098

2.026

2.330

.188

.210

.031

ð.109

Þð.1

44Þ

ð.138

Þð.0

23Þ

ð.303

Þð.3

20Þ

ð.362

Þð.1

02Þ

Policy

period

2.036

2.118

2.107

.007

.292

***

2.077

2.061

.013

ð.043

Þð.0

80Þ

ð.081

Þð.0

07Þ

ð.081

Þð.1

05Þ

ð.107

Þð.0

18Þ

Lam

beth

�postpolicy

period

.120

*2.032

2.028

2.001

2.316

**2.088

.019

.123

**ð.0

70Þ

ð.080

Þð.0

76Þ

ð.013

Þð.1

46Þ

ð.137

Þð.1

55Þ

ð.059

ÞPostpolicy

period

.005

2.040

2.015

.020

***

.241

***

2.096

2.078

2.003

ð.035

Þð.0

58Þ

ð.058

Þð.0

06Þ

ð.067

Þð.0

83Þ

ð.088

Þð.0

15Þ

Borough

andmonth

fixe

deffects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Sociodem

ograp

hic

controls

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Observations

2,95

02,94

42,94

33,00

52,55

82,54

32,51

72,97

8


theborough


periodrunsfrom

April19

98untilJanuary20

06.Controlborough

sareall

other

Londonborough

s.ClassAdrugs

areco

caine,

crack,

crystalmeth,h

eroin,L

SD,M

DMA,andmethad

one.

Thedep

enden

tvariab

lein

cols.1

and5is

thelogofthenumber

ofo

ffen

sesforeach

offen

setype,per

1,00

0ofthead

ultpopulation.T

hedep

enden

tvariab

lein

cols.2

and6isthearrestrate

foreach

offen

setype,

defi

ned

asthelogofthenumber

ofarrestsdivided

bythenumber

ofoffen

sesin

theborough

inthesamemonth

andpreviousquarter.The

dep

enden

tvariab

lein

cols.3an

d7istheclear-uprate

foreach

offen

setype,

defi

ned

asthelogofthenumber

ofclear-upsdivided

bythenumber

of

offen

sesin

theborough

inthesamemonth


enden

tvariab

lein

cols.4

and8istheratioofclear-upsto

arrests,defi

ned

asthe

logofthenumber

ofclear-upsdivided

bythenumber

ofarrestsin

thesamemonth.Inco

ls.1

–4theoffen

setyperelatesto

possessionofclassAdrugs.In

cols.5–

8theoffen

setypeisthesum

ofalloffen

sesrelatedto

classAdrugs

supply

includingpossessionwithintent,possessiononaship,production,

supply,unlawfulex

port,unlawfulim

port,carryingonaship,incitingothersto

supply,man

ufacture,an

dmoney

laundering.

Pan

el-correctedstan

dard

errors


insten

regression,w

hereaborough

-specificARð1Þp

rocessisassumed

.Thisalso

allowstheerrorterm

sto

beborough

specific,heterosked

astic,an

dco

ntemporaneo

uslyco

rrelated

acrossborough

s.Observationsareweigh

tedbytheshareofthetotalL

ondonpopulationthat

month-yearin

theborough

.Thepolicy

perioddummyvariab

leiseq

ual

toonefrom

July20

01untilJuly20

02,andzero

otherwise.

Thepostpolicy

period

dummyvariab

leiseq

ual

toonefrom

July

2002

onward,an

dzero

otherwise.

Thefollowingsociodem

ograp

hic

controlvariab

les,measuredin

logs,are

controlled

forat

theborough


ultpopulationthat

isethnicminority;that

isaged

20–2

4,25

–34,

35–49

,andab

ove

50;

andthemaleunem

ploym

entrate.In

addition,thelogofthead

ultpopulationisincluded

asaco

ntrolin

cols.2–4an

dco

ls.6–

8.*Sign

ificantat

10percent.

**Sign

ificantat

5percent.

***Sign

ificantat

1percent.



broaden the search for evidence of the reallocation of police effort byexamining seven types of nondrug crime. Table 5 reports the results. Incolumn 1 we first estimate equation ð1Þ using as the dependent variablethe ðlog ofÞ offense rate for total nondrug crime. During the policy pe-riod, offense rates for total nondrug crime were not significantly differ-ent in Lambeth from those in other London boroughs. Remarkably, inthe postpolicy period, the offense rate for total nondrug crime in Lam-beth significantly fell by 9.4 percent more than the London-wide aver-age. Quantitatively, this translates into a large reduction in total crimein Lambeth: before the policy, 97 percent of all offenses in Lambeth arenon–drug related. This long-term reduction in Lambeth occurred in aperiod when citywide offense rates for nondrug crimes are flat, as fig-ure 1C suggested.The remaining columns of table 5 show significant falls after the pol-

icy in recorded offense rates for five out of seven crime types. These cat-egories—robbery, burglary, theft and handling, fraud and forgery, andcriminal damage—account for 81 percent of all criminal offenses beforethe policy. The point estimates on the other three categories, violence,sexual offenses, and robbery, are all negative but not significantly differ-ent from zero. To aid exposition, figure 2A shows the eight coefficientsof interest ðb2Þ from table 5, along with their associated 95 percent con-fidence intervals.To pin down whether this long-run decline in nondrug crime is due to

a reallocation of police effort, Appendix table A3 estimates the short-and long-run policy effects on our measures of police effectiveness: ar-rest rates ðpanel AÞ, clear-up rates ðpanel BÞ, and clear-ups per arrestðpanel CÞ. Given the large number of coefficients to read in table A3,figures 2B–2D show the coefficients of interest of the long-run policy im-pacts from each specification, along with their associated 95 percent con-fidence interval.In terms of police effectiveness against nondrug crime, we find the

following: ðiÞ Arrest rates for total nondrug crime rose significantly ðta-ble A3, panel A, col. 1Þ: the long-run difference-in-difference estimate is28.4 percent for Lambeth relative to the rest of London. ðiiÞ Consideringspecific crime types, the remaining columns in panel A and figure 2Bhighlight how in the long run there are significant increases in arrestrates for nearly all crime types. ðiiiÞ Panel B of table A3 and figure 2Cshow that these higher arrest rates actually feed into significantly higherclear-up rates, again for nearly all crime types.22 ðivÞ Panel C of table A3

22 The one exception relates to crimes of theft and handling, where we see no long-rundifferential change between Lambeth and the rest of London in arrest or clear-up rates. Aswith some of the earlier evidence and existing literature, this might suggest that suchcrimes are especially colinear with themarket for cannabis, which is of course expanding inthe long run in Lambeth. In contrast to offenses related to cannabis possession, there isgenerally a divergence between arrest and clear-up numbers for these nondrug offenses.



TABLE5

TheEffectofDepenalizingCannabisonNon-Drug-RelatedCrime

DependentVariable:LogðRecorded

Offen

sesofaGiven

Type,

per

1,00


TotalCrime

ðwithoutDrugsÞ

ð1Þ

Violence

against

the

Person

ð2Þ

Sexu

alð3Þ

Robbery

ð4Þ

Burglary

ð5Þ

Theftan

dHan

dling

ð6Þ

Fraudor

Forgery

ð7Þ

Criminal

Dam

age

ð8Þ

Lam

beth�

policy

period

.023

.010

2.112

2.053

2.007

.064

*2.257

*2.046

ð.033

Þð.0

38Þ

ð.084

Þð.0

96Þ

ð.060

Þð.0

37Þ

ð.141

Þð.0

53Þ

Policy

period

.033

.077

***

.100

***

.223

***

2.012

.049

**2.031

2.012

ð.020

Þð.0

27Þ

ð.025

Þð.0

53Þ

ð.021

Þð.0

21Þ

ð.065

Þð.0

20Þ

Lam

beth

�postpolicy

period

2.094

***

2.046

2.096

2.321

***

2.250

***

2.083

**2.355

***

2.090

**ð.0

33Þ

ð.034

Þð.0

60Þ

ð.093

Þð.0

49Þ

ð.033

Þð.1

28Þ

ð.044

ÞPostpolicy

period

.024

.200

***

.110

***

.228

***

2.113

***

.039

**2.183

***

2.064

***

ð.018

Þð.0

24Þ

ð.020

Þð.0

46Þ

ð.017

Þð.0

18Þ

ð.055

Þð.0

18Þ

Shareofalloffen

sesprepolicy

.973

.155

.009

.034

.128

.401

.089

.159

Borough

andmonth

fixe

deffects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Sociodem

ograp

hic

controls

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Observations

3,00

83,00

83,00

83,00

83,00

83,00

83,00

83,00

8


theborough


periodrunsfrom

April19

98untilJanuary20

06.Controlborough

sareall

other

Londonborough

s.In

col.1thedep

enden

tvariab

leisthelogofthenumber

ofalln

on-drug-relatedcrim

esper

1,00

0ofthead

ultpopulation.P

anel-

correctedstan

darderrors


insten


-specificARð1Þp

rocess

isassumed

.Thisalso

allowstheerror

term

sto

beborough

specific,

heterosked

astic,

andco

ntemporaneo

uslyco

rrelated

across

borough


tedbytheshareofthetotal

Londonpopulationthat

month-yearin

theborough

.Thepolicy

perioddummyvariab

leiseq

ualto

onefrom

July20

01untilJuly20

02,andzero

otherwise.

Thepostpolicy

perioddummyvariab

leiseq

ual

toonefrom

July

2002

onward,an

dzero

otherwise.


ograp

hic

controlvariab

les,

measuredin

logs,are

controlled

forat

theborough


ultpopulationthat


isaged

20–2

4,25

–34,

35–49

,an

dab

ove

50;an

dthemaleunem

ploym

entrate.Wealso

showtheproportionofallcrim

inal

offen

sesðdrugan

dnon-drug-relatedÞt

hat

each

catego

rymakes

upin

theprepolicy

periodin

Lam

bethfrom

April19

98untilJune20

01.

*Sign

ificantat

10percent.

**Sign

ificantat

5percent.

***Sign

ificantat

1percent.




FIG.2.—

Impacts

ofthedep

enalizationpolicy

onnondrugcrim

es.Eachpointonthegrap

hrepresents

thepointestimateonthepostpolicy

�Lam

beth

interactionterm

from

aseparateregression.Thelines

represent95

percentco

nfiden

ceintervals.Thepointestimates

arefrom

regressionsas

described

asfollows.Allobservationsareat

theborough


periodrunsfrom

April19

98untilJanuary20

06.Controlborough

sareallother

Londonborough

s.Thedep

enden

tvariab

lein

theoffen

segrap

hðpan

elAÞisthelogofthenumber

ofoffen

sesforeach

offen

setypeper

1,00

0ofthead

ult

population.T

hedep

enden

tvariab

lein

thearrestgrap

hðpan

elBÞisthearrestrate

foreach

offen

setype,defi

ned

asthelogofthenumber

ofarrestsdivided

by

thenumber

ofoffen

sesin

theborough

inthesamemonth


enden

tvariab

lein

theclearance

grap

hðpan

elCÞistheclearance

rate

foreach

offen

setype,

defi

ned

asthelogofthenumber

ofclearancesdivided

bythenumber

ofoffen

sesin

theborough

inthesamemonth

andprevious

quarter.Thedep

enden

tvariab

lein

theclearance/arrestratiograp

hðpan

elDÞisdefi

ned

asthelogofthenumber

ofclear-upsdivided

bythenumber

ofarrests

inthesamemonth.P

anel-correctedstan

darderrors


insten

regression,w

hereaborough

-specificARð1Þp

rocessisassumed

.This

also

allowstheerrorterm

sto

beborough

specific,heterosked

astic,an

dco

ntemporaneo

uslyco

rrelated

acrossborough


tedbytheshare

ofthetotalLondonpopulationthat

month-yearin

theborough

.Thepolicy

perioddummyvariab

leiseq

ual

toonefrom

July20

01untilJuly20

02,an

dzero

otherwise.

Thepostpolicy

perioddummyvariab

leiseq

ual

toonefrom

July

2002

onward,an

dzero

otherwise.


ograp

hic

controlvar-

iables,measuredin

logs,are

controlled

forat

theborough


ultpopulationthat


isaged

20–2

6,25

–34,

35–49

,andab

ove

50;andthemaleunem

ploym

entrate.T

helogofthetotalb

orough

populationðbymonth-yearÞaged

16an

doverisalso

included

asaco

ntrol

inallex

cepttheoffen

seregressions.



and figure 2D show that clear-ups per arrest do not change for mostcrime types. Hence the likelihood that an arrestee is charged with theoffense is not driving the earlier result; rather any change in police effortleads to more arrests and clear-ups per se for these six broad crime typesand for nondrug crime overall.Taken together, the evidence suggests a significant reallocation of po-

licing intensity after the introduction of the LCWS away from cannabiscrimes and toward other nondrug crimes ðtable 5Þ, but not especiallytoward class A drug crime ðtable 4Þ. This reallocation appears to persistlong after the LCWS officially ends and is reflected in marked increasesin arrest and clear-up rates for a broad range of crime types ðtable A3,panels A and BÞ. These changes in police effectiveness of course feedback into lowering offense rates ðtable 5Þ.23

C. Police Resources

Given the central role the reallocation of policing effort plays in ex-plaining changing patterns of crime and police effectiveness as a resultof the depenalization policy, it is important to understand whether theresults could in part be confounded by a change in total police resourcesrather than a mere reallocation of existing resources. While detailedborough-month-level information on police manpower or task alloca-tions does not exist for our study period, there is evidence from MPAreports that police officer numbers in Lambeth rose in the postpolicy pe-riod ðall MPA reports are available at http://www.policeauthority.orgÞ.These suggest that in the summer of 2001 the Lambeth police were run-ning at 11 percent below their budgeted workforce target, equivalent to102 officers below strength. By January 2002 the situation had improved,with an additional 43 officers in Lambeth, reducing the deficit to 6.3 per-cent.To investigate whether this change in Lambeth can explain the dif-

ferential patterns of crime documented in table 5, we have collated theavailable data on annual police numbers for all 32 London boroughsfrom 1997 to 2010. This shows that police numbers certainly rose in Lam-beth during and after the policy: between 2001 and 2006, police num-bers increased by 20.5 percent in Lambeth. However, this pattern is byno means exceptional to Lambeth. Over the same period, the police num-bers for London as a whole rose by 22.7 percent, slightly more than inLambeth. This suggests that changing police strength in Lambeth vis-a-vis

23 These results are largely robust to defining arrest and clear-up rates as being per 1,000of the adult population rather than per offense in the previous quarter. Hence these patternsin arrest and clear-up rates likely reflect real changes in police behavior rather than beingdriven solely by declines in the number of offenses in each crime type.




other London boroughs is unlikely to explain the large reductions in non-drug crime documented.24

A second way to understand whether changing police numbers mightplausibly explain the documented impact on nondrugs crime is to useestimates from the literature on the elasticity of crime with respect topolice strength. In this setting, the estimates provided by Draca, Machin,and Witt ð2011Þ are perhaps most informative. They use the exogenousshift in police deployment following the July 2005 terror attacks in Lon-don to estimate an elasticity of crime with respect to police numbers tobe around20.3. For the LCWS, over the postpolicy period from January2002 to March 2006, police numbers in Lambeth increased by 13.2 per-cent. Ignoring the change in other London boroughs and so assumingthat the 13.2 percent increase in Lambeth represents the difference indifference with other boroughs, we can then combine the elasticity es-timate from Draca et al. and our regression coefficient; this should haveled to a 4 percent drop in nondrug crime. Hence, even under this mostconservative approach, where we ignore changing police numbers in otherboroughs, the drop innondrugs crime that canbe explained through thischannel is just less than half the actual long-run fall in nondrug crimewe find of 8.8 percent.In short, the evidence suggests that the documented reduction in

nondrug crime and increased police effectiveness against such crimeswere primarily due to a differential reallocation of police resources inLambeth relative to the rest of London rather than to increased num-bers of police officers per se. As such, the policy likely had small mon-etary costs of implementation. The next section moves on to establishingthe monetized welfare impacts of the policy on Lambeth residents.

VI. House Prices

Understanding the welfare consequences of any given drug policy isimportant given the large number of illicit drug users around the world.This is especially so for policies related to the market for cannabis, themost frequently used illicit drug in most countries. Miron ð2010Þ esti-mates, in the US context, the budgetary consequences of liberalizingdrug policy. We add to this nascent literature by evaluating the welfareeffects of the localized LCWS depenalization policy.From the documented impacts on crime, the welfare effects of the pol-

icy are ambiguous: although the policy caused total crime to fall, it alsocaused a dramatic change in the composition of crime. Depenalization ledto an increase in cannabis offenses; on the other hand, many other types

24 We have probed this time series on police numbers by borough-year to understandwhat drives changes in police strength. This suggests that police numbers track the boroughpopulation with some lag.




of crime were reduced in the longer term. To estimate the overall impact ofthe policy through these changing crime patterns, as well as through othernoncrime channels, we estimate the impact of the depenalization of can-nabis possession on house prices in Lambeth relative to other Londonboroughs. This approach uses the intuition that the total social costs ofdepenalization ðnot just those arising from crimeÞ should be reflected inhouse prices ðRosen 1974; Thaler 1978Þ.To do so, we exploit information at the zip code level on house prices

from the UK Land Registry to estimate a specification analogous to ð1Þ.The unit of observation is zip code sector s in quarter q in year y, wherezip code sectors are within a borough.25 This allows us to later explorewhether and how the effects of depenalization affect house prices withinLambeth. To begin with, we estimate a panel data specification of theform

lnhsqy 5 b0Pqy 1 b1ðLb � PqyÞ1 b2PPqy 1 b3ðLb � PPqyÞ1 gXbqy 1 ls 1 lq 1 usqy;

ð2Þ

where hsqy is the mean house price sale for terraced houses in zip codesector s in quarter q in year y, deflated to 1995:Q1 prices;26 Pqy and PPqy

are dummies for the policy and postpolicy periods, respectively; and Lb

is a dummy for whether the zip code sector is in Lambeth. To reflect thelag between house buying decisions and recorded house sales, all time-varying covariates are lagged one quarter. In Xbqy, we continue to controlfor sociodemographic controls, as in ð1Þ. We also allow for borough-specific time trends ðlb � qyÞ to capture common house price movementsand control for fixed effects for zip code and quarter. The sample runsfrom January 1995 until December 2005, standard errors are clusteredat the zip code sector level, and observations are weighted by the num-bers of terraced house sales in the zip code sector during the quarter.

25 A London zip code ðe.g., WC1E 6BTÞ is generally 10–12 neighboring addresses ðwhichwould include flats and maisonettes, as well as separate housesÞ. Our house price data wereobtained from the UK Land Registry at a lightly more aggregated level, that of a zip codesector ðe.g., WC1EÞ. In London there are an average of 215 zip codes per zip code sectorðso 2,000–2,500 addresses in each zip code sectorÞ. There are, on average, 20 zip code sectorsper borough. In Lambeth ða total of 10.36 square miles ½26.82 square kilometers�Þ, there are31 zip code sectors, so that each covers, on average, 0.33 square mile ð0.87 square kilometerÞ.

26 The house price data cover 25 of the 32 boroughs used for the crime analysis. Theboroughs not covered are Barking and Dagenham, Bexley, Harrow, Havering, Hillingdon,Kingston-upon-Thames, and Sutton. There are 509 distinct zip codes in the final sample, withan average of 25.3 zip codes per borough. House prices are deflated to 1995:Q1 prices usingthe Land Registry house price index for Greater London, which is based on repeat sales ðseehttp://www1.landregistry.gov.uk/houseprices/housepriceindex/Þ. We drop zip code sectorsthat have the lowest 10 percent of house sales as these are unlikely to correspond to residen-tial neighborhoods. The reported results are robust to dropping zip codes that straddle bor-ough boundaries.




House price information is available for terraced houses, detached,semidetached, and flats. When estimating ð1Þ, our baseline estimates fo-cus on terraced housing to strike a balance between using a housingtype that has both frequent sales and high values per sale. When docu-menting the total impact of the policy on house prices in Section VI.B, wedo so by aggregating the policy impacts across all four housing types.

A. Results

Table 6 reports the results. Column1presents the baseline finding: in thelong run after the LCWS is introduced, house prices fall by 5.0 percentmore in Lambeth relative to the London-wide average, an effect signifi-cant at the 1 percent level. Column 2 shows the impact to be even morenegative after we control for borough-specific linear time trends. To re-iterate, these negative effects on house prices in the long run occur de-spite the overall falls in total crime experienced in Lambeth after thepolicy: as table 5 showed, total nondrug crime fell by 9.4 percent. At thesame time, the results from table 2 showed that the incidence of cannabis-related crime rose by at least 40 percent in the longer term. To reconcilethese policy impacts on crime andhouse prices, either Lambeth residentsmight place disproportionate weight on cannabis-related crime relativeto all other crimes or there might exist other social costs beyond crimeassociated with a rapidly expanding market for cannabis.27

As house price data are available by zip code, the remaining specifi-cations in table 6 examine whether there are heterogeneous effects ofdepenalization on house prices within Lambeth and other boroughs.Theheterogeneity we focus on relates to the location of drug crimewithineach borough and leads us to designate each zip code sector as a drugcrime “hot spot” or not. The Appendix describes in detail how we usedisaggregated drug crime data to determine whether a zip code sectoris a hot spot. We then explore whether house prices vary differentiallywithin a borough between hot spots and non–hot spots, using a triple-differenced estimation strategy, across boroughs, time, and hot spot/non–hot spot areas.The disaggregated data from which hot spots are defined are “ward”-

level crime statistics published by the MPS. Wards are small administra-tive districts nested within boroughs. There are, for example, 21 wards in-Lambeth, which closely matches the London borough average. However,

27 Other studies have found a negative association between certain crime types and houseprices: Gibbons ð2004Þ documents how a one standard deviation increase in property crime isassociated with a 10 percent reduction in house prices in the United Kingdom; Linden andRockoff ð2008Þ present evidence from the United States that the revelation of information ofa sex offender residing next door leads to a 12 percent reduction in house prices. Our resultslikely differ because the policy we evaluate affects both the level and composition of crime.




TABLE 6The Effect of Depenalizing Cannabis on House Prices

Dependent Variable: LogðZip Code Quarter MeanHouse Price, Deflated to 1995:Q1 PricesÞ

Baselineð1Þ

TimeTrendsð2Þ

Ex PostHot Spot

ð3Þ

Ex AnteHot Spot

ð4Þ

Higher-Level

Clusteringð5Þ

Lambeth � policy period .026** 2.028 .022 2.021 .022ð.013Þ ð.019Þ ð.037Þ ð.021Þ ð.016Þ

Policy period .004 2.025*** 2.054*** 2.036** 2.054***ð.006Þ ð.006Þ ð.011Þ ð.014Þ ð.013Þ

Lambeth � postpolicy period 2.050*** 2.126*** 2.016 2.011 2.016ð.016Þ ð.034Þ ð.030Þ ð.031Þ ð.029Þ

Postpolicy period .033*** 2.046*** 2.111*** 2.108*** 2.111***ð.010Þ ð.011Þ ð.015Þ ð.017Þ ð.028Þ

Lambeth � hot spot 2.087** 2.084* 2.087**ð.044Þ ð.046Þ ð.039Þ

Hot spot .039 2.211*** .039ð.024Þ ð.019Þ ð.026Þ

Lambeth � policy period �hot spot 2.062* 2.009 2.062***

ð.036Þ ð.021Þ ð.012ÞPolicy period � hot spot .033*** .012 .033**

ð.011Þ ð.015Þ ð.012ÞLambeth � postpolicy period

� hot spot 2.134*** 2.135*** 2.134***ð.022Þ ð.020Þ ð.021Þ

Postpolicy period � hot spot .073*** .066*** .073***ð.014Þ ð.016Þ ð.021Þ

Zip code and quarter fixedeffects Yes Yes Yes Yes Yes

Borough-specific linear timetrend No Yes Yes Yes Yes

Sociodemographic controls Yes Yes Yes Yes YesObservations 17,331 17,331 17,331 17,331 17,331

Note.—All observations are at the zip code sector-quarter-year level. House prices aredeflated to 1995:Q1 prices, using the Land Registry house price index for Greater London,which is based on repeat sales. More information on the index can be found at http://www1.landregistry.gov.uk/houseprices/housepriceindex/. For all specifications, the sam-ple runs from January 1995 until December 2005, and observations are weighted by thenumbers of sales for terraced housing in that quarter-year in the specific zip code sector.Standard errors are clustered by zip code sector in cols. 1–4 and by borough in col. 5. Toreflect the lag between the house buying decision and the recorded sale of the house, alltime-varying explanatory variables are lagged by one quarter. The ðone-quarter-laggedÞ pol-icy period dummy variable is equal to one from the fourth quarter ðstarts October 1Þ of 2001until the third quarter of 2002 ðends September 30Þ, and zero otherwise. The ðone-quarter-laggedÞ postpolicy period dummy variable is equal to one from the fourth quarter of 2002onward, and zero otherwise. The following sociodemographic control variables, measuredin logs, are controlled for at the borough-month-year level: the share of the adult populationthat is ethnic minority; that is aged 20–24, 25–34, 35–49, and above 50; and the male unem-ployment rate. All these socioeconomic variables are lagged one quarter. We also control forfixed effects for zip code and quarter throughout. In cols. 2–5, we also control for a borough-specific linear time trend. In cols. 3 and5, zip code sectors aredefined tobehotpots on thebasisof ex post ward-level crime data. In col. 4, we use ex ante ward-level crime data.* Significant at 10 percent.** Significant at 5 percent.*** Significant at 1 percent.



such ward-level crime data exist for each month only from April 2001onward. Hence for our baseline results, we classify zip code sectors intohot spots on the basis of crime rates measured ex post in 2008/9, long af-ter the LCWS is initially implemented. Given obvious concerns over usingsuch ex post data to define hot spots, we also use the available crimeward data for the few months before the policy to reestimate our mainspecification, classifying zip code sectors into hot spots on the basis of exante crime rates. To provide evidence of the geographic stability of hotspot locations in Lambeth over time, Appendix figure A1 shows the clas-sification of each Lambeth zip code sector into hot spots based on bothdefinitions: reassuringly, there is considerable stability in these classifica-tions over time. The Appendix presents further robustness checks basedon alternative hot spot definitions.Column 3 of table 6 then presents estimates of this triple-differenced

specification in which we allow the policy impacts to vary across hotspots within each borough. We find that all the previously documentedlong-run negative effect of depenalization on house prices within Lam-beth occurs in drug crime hot spots. There is no significant effect of de-penalization on house prices on non–hot spot zip codes in Lambeth. Asa result, the magnitude of the house price fall in Lambeth hot spots,213.4percent, is significantly larger than in the earlier all-Lambeth estimates.28

In the postpolicy period, hot spot areas in other boroughs appear to havepositive and significant house price rises, consistent with there being con-vergence in house prices across neighborhoods.Column 4 then shows the main results to be very stable using ex ante

ward-level crime data to classify zip code sectors as hot spots: the relativehouse price decline in Lambeth hot spots is very similar at 213.5 per-cent, and we still observe rising prices in hot spots in other London bor-oughs in the postpolicy period ð6.6 percentÞ. The similarity of findingsusing ex ante and ex post hot spots is unsurprising given the geographicstability over time where drug crime is concentrated in Lambeth, as fig-ure A1 shows.The remaining column demonstrates the robustness of the results to

alternative methods by which to calculate standard errors. In column 5we cluster at a higher level of aggregation: given the baseline estimatescluster by zip code sector, the natural next level of aggregation is to clus-ter by borough. Comparing this specification in column 5 to the base-line definition using ex post hot spots in column 3, we see the standard

28 May, Cossalter, et al. ð2007Þ provide detailed descriptive evidence on drug dealing in Brix-ton: a hot spot area in our definition covering more than one zip code, and the most impor-tant commercial center in Lambeth. They describe the geography of drug crime in Brixton,how it affects other crimes.




errors to be considerably smaller when clustering by borough, support-ing the view that the baseline approach is conservative.29

The Appendix presents robustness checks that probe these results intwo directions: ðiÞ the policy impacts on other housing types and ðiiÞ theuse of alternative definitions of crime hot spots. In each case we find re-sults very much in line with these baseline findings. For all variant spec-ifications we see that after the policy, house prices are significantly lowerin Lambeth hot spots than in other boroughs, where the magnitude ofthe impact varies between 7.7 percent and 13.9 percent.The results from table 6 suggest that for local residents, the total wel-

fare impacts of depenalizing the possession of small quantities of can-nabis likely went far beyond the impacts on crime. For example, theremight have been other deleterious impacts on behaviors associated withthe market for illicit drugs, such as alcohol use and other forms of visi-ble antisocial behavior. These are important channels through whichthe effects of depenalization might operate in the long run ðMiron andZweibel 1995Þ and that we are investigating in ongoing research.30 Suchwider changes appear to reduce the willingness to pay to reside in theseneighborhoods and increase within-borough inequality in house pricesbetween high– and low–drug crime zip codes.Themagnitudeof thesehouseprice impacts canbecompared relative to

other studies, although in some cases, we have to extrapolate out of sam-ple to have changes in local characteristics that would correspond to anequivalent reduction in house prices of 213.4 percent. Notwithstandingthis caveat, comparing our estimates to those linking house prices with

29 Cameron, Gelbach, and Miller ð2008Þ note that cluster-robust standard errors may bedownward biased when the number of clusters is small, leading to an overrejection of thenull of no effect. The authors propose various asymptotic refinements using bootstrap tech-niques, finding that the wild cluster bootstrap t technique performs particularly well in theirMonte Carlo simulations. We have implemented this method on our preferred specificationin col. 3, with 1,000 bootstrap iterations and using Rademacher weights for the procedure.The resulting estimated standard errors are very similar to those reported, and all the reportedcoefficients remain with the same significance.

30 For example, Kelly and Rasul ð2014Þ evaluate the impact of the LCWS on hospitaladmissions related to illicit drug use. They exploit administrative records on individualhospital admissions classified by International Classification of Diseases ðrev. 10Þ diagnosiscodes. They find that the depenalization of cannabis had significant longer-term impacts onhospital admissions related to the use of hard drugs, raising hospital admission rates formen. Among Lambeth residents, the impacts are concentrated amongmen in younger agecohorts. Model ð1993Þ explores the effect decriminalizing cannabis in 12 US states between1973 and 1978 had on hospital emergency room drug episodes. He finds evidence thatdecriminalization was accompanied by a significant reduction in episodes involving drugsother than marijuana and an increase in marijuana episodes, suggesting that consumerssubstitute toward the less severely penalized drug. There is mixed evidence on whetheralcohol and cannabis are substitutes for young individuals: DiNardo and Lemieux ð2001Þand Conlin, Dickert-Conlin, and Pepper ð2005Þ find that they are substitutes; Pacula ð1998Þfinds them to be complements.




school quality implies that an equivalent reduction in house prices couldbe generated by ðiÞ a 19 percent reduction in pupils achieving UK gov-ernment targets at the end of primary school ðGibbons and Machin 2003Þ,ðiiÞ a four standard deviation decrease in value-added scores of primaryschools in the United Kingdom ðGibbons, Machin, and Silva 2013Þ, andðiiiÞ test scores that are 32 percent below the mean, on the basis of USdata estimates ðBlack 1999Þ. Comparing our estimates to those linkinghouse prices with crime, we find that an equivalent reduction in houseprices could be generated by a greater than one standard deviation in-crease in property crime, based on UK data ðGibbons 2004Þ; for the UnitedStates, Linden and Rockoff ð2008Þ show that the revelation of a sex of-fender residing next door reduces house prices by 12 percent. Finally,we can also benchmark our findings against the documented impacts ofenvironmental quality on house prices: for the United States, Davis ð2004Þshows that a severe increase in the risk of pediatric leukemia is associatedwith a 14 percent reduction in house prices.

B. Interpretation

The documented impacts of the LCWS on house prices can reflectchanging amenity values of residing in Lambeth, changes in the qualityof the existing housing stock, or changes in value of newly constructedhomes in Lambeth. To tease apart these explanations would require farmore detailed information on housing characteristics that is not easilyavailable. Although we use data on house prices and sales from the mainUK data source, the Land Registry, even its most disaggregated admin-istrative records on individual sales provide little information on housecharacteristics: they relate only to whether the house is a new build andinformation on its freehold/leasehold status.31

We now focus attention on estimating the total implied loss in prop-erty values in Lambeth as a result of the policy, proceeding as follows.First, we run our preferred house price specification ð2Þ for each of thefour housing categories in the Land Registry data: terraced houses, flats,semidetached houses, and detached houses. Table 7 shows the estimatedb coefficients from each specification, where each column refers to a dif-ferent housing type. The relevant parameter of interest is the long-runpostpolicy impact on house prices: b3. This is negative and significant forthree of the four house types: semidetached, terraced, and flats.These parameter values are then multiplied by the base level of house

prices in Lambeth before the policy, for each property type, and then

31 However, we note that there is very limited scope for new builds in Lambeth ðas for allinner London boroughsÞ. With more that 280,000 residents, Lambeth is one of the mostdensely populated boroughs in the country, with more than 100 residents per hectare. Assuch, our prior is that the documented house price effects reflect changing amenity valuesand changing quality of the existing housing stock.




multiplied by the number of property types actually sold over the post-policy sample period. Rows A and B show themean andmedian prepolicysales for each housing type. There is little divergence between the two,and so for the remainder of the analysis we focus attention on using themean price in row A. Row C shows the number of house price sales in thepostpolicy period until December 2005 by housing type.Combining this information then provides an implied total loss in

value for a given property type. We first provide a lower-bound estimateon this implied loss by assuming that only those houses that are actuallysold experience any loss in value. Then row D shows, for each housingtype, the implied loss in value over the postpolicy period. Summing acrossthe four housing types in columns 1–4, column 5 of table 7 gives the totalimplied loss: this amounts to £233 million.32

This corresponds to a lower bound on welfare losses because it ignoresany reductions in property price values that are experienced by thoseresidents who chose not to sell. To better capture any such impacts, weconduct another thought experiment assuming that all properties inLambeth of a given type experience the same implied loss in value, ir-respective of whether or not they are actually sold before the policy. Thisapproach requires additional information on the total housing stock.This is shown in column 5 of row E: there are 119,000 properties inLambeth. However, as the information to break this down by propertytype does not exist, we assume that the share of all properties sold of agiven type for the postpolicy period ðbased on row CÞ is the same as theshare of all households that exist of a given type in Lambeth. This shareis then given in row F.Using this information, we are then able to derive something more akin

to an upper-bound estimate of the implied total loss in property value:row G gives the implied loss for the entire postpolicy period: £1.1 billion,almost five times the lower-bound estimate derived in row D. In short,whichever way the implied loss inLambethproperty values is calculated, itdwarfs any direct costs of the LCWS, a policing change that largelyamounted to a change in how existing police resources were allocatedrather than any change in the level of resources per se.

VII. Citywide Depenalization

The reduced-form analysis emphasized how a localized depenalizationof cannabis possession affects the levels and composition of crime. Wenow build on the key lessons from this policing experiment to shed light

32 This aggregate loss in property value is almost unchanged if we ignore any impacts ondetached houses, as shown in col. 1. The likely reason for a nonsignificant impact for suchhouse types is that there are only 52 recorded sales of such homes in Lambeth after thepolicy.




TABLE7

ImpliedLossinHousePricesDuetotheDepenalizationPolicy

DependentVariable:LogðZip

CodeQuarterMeanHouse

Price,

Defl

ated

to19

95:Q

1PricesÞ

Detached

Housing

ð1Þ

Semidetached

Housing

ð2Þ

Terraced

Housing

ð3Þ

Flats

ð4Þ

Row

Total

ð5Þ

Lam

beth�

policy

period

2.244

***

2.028

2.028

2.018

ð.087

Þð.0

31Þ

ð.019

Þð.0

18Þ

Policy

period

2.017

2.030

***

2.025

***

2.024

***

ð.026

Þð.0

08Þ

ð.006

Þð.0

06Þ

Lam

beth�

postpolicy

periodðb

3Þ

2.070

2.118

***

2.126

***

2.099

***

ð.121

Þð.0

41Þ

ð.034

Þð.0

31Þ

Postpolicy

period

2.087

***

2.050

***

2.046

***

2.089

***

ð.033

Þð.0

11Þ

ð.011

Þð.0

09Þ

A.Meanprepolicy

house

price

ðdefl

ated

to19

95:Q

1pricesÞ

£201

,653

£140

,697

£122

,691

£70,20

8B.Med

ianprepolicy

house

price

ðdefl

ated

to19

95:Q

1pricesÞ

£185

,792

£118

,086

£110

,311

£62,48

7

Lower-BoundEstim

ate:

AssumeUnsold

HousesExp

erience

NoLoss

inValue

C.Postpolicy

salestotal

511,20

05,79

617

,707

24,754

D.Meanloss

based

onpostpolicy

salestotal5

b3�

A�

C2£7

19,903

2£1

9,92

2,65

32£8

9,60

0,52

72£1

23,073

,484

2£2

33,316

,567

1170





Price,

Defl

ated

to19

95:Q

1PricesÞ

Upper-BoundEstim

ate:

AssumeAllHouseholdsExp

erience

SameLoss

inValue

E.Number

ofhouseholdsin

Lam

bethin

2001

Unkn

own

Unkn

own

Unkn

own

Unkn

own

119,00

0F.

Housingtypeshareofpostpolicy

salestotal

.002

.048

.234

.715

G:Meanloss

based

onpostpolicy

total5

b3�

A�

E�

F2£3

,460

,791

2£9

5,77

4,24

62£4

30,736

,962

2£5

91,651

,636

2£1

,121

,623

,634


thezipco

desector-quarter-year

level.House

pricesaredefl

ated

to19

95:Q

1prices,usingtheLan

dReg

istryhouse

price

index

forGreater

London,w

hichisbased

onrepeatsales.More

inform

ationontheindex

canbefoundat

http://www1.landregistry.gov.uk/

houseprices

/housepriceindex

/.Forallspecifications,

thesample

runsfrom

January19

95untilDecem

ber

2005

,stan

darderrors

areclustered

byzipco

de,

and

observationsareweigh

tedbythenumbersofsalesforthehousingtypein

that

quarter-year

inthespecificzipco

desector.Toreflectthelagbetweenthe

house

buyingdecisionan

dthereco

rded

sale

ofthehouse,alltime-varyingex

planatory

variab

lesarelagg

edbyonequarter.Theðone-quarter-lagg

edÞ

policy

perioddummyvariab

leiseq

ualto

onefrom

thefourthquarterðst

artsOctober

1Þof20

01untilthethirdquarterof20

02ðendsSe

ptember

30Þ,an

dzero

otherwise.

Theðone-quarter-lagg

edÞp

ostpolicy

perioddummyvariab

leiseq

ual

toonefrom

thefourthquarterof20

02onward,andzero

otherwise.


ograp

hicco

ntrolvariables,measuredin

logs,are

controlled

forat

theborough


ultpopulation

that


isaged

20–2

4,25

–34,35

–49

,andab

ove

50;andthemaleunem

ploym

entrate.A

llthesesocioeconomicvariab

lesarelagg

edone

quarter.When

calculatingthehigher

house

price

estimates

ðrowEdownÞ,wedonotkn

owthenumber

ofhouseholdsin

Lam

bethforeach

propertytype.

In20

01,therewere11

9,00

0householdsðso

urce:

https://www.gov.uk/

governmen

t/statistical-d

ata-sets/live-tab

les-on-household-projectionsÞ.

Wethen

estimatethenumber

ofeach

typeofh

ouses,usingthesalesshares

from

thepostpolicy

periodmultiplied

bythetotaln

umber

ofowned

housesin

Lam

beth.

*Sign

ificantat

10percent.

**Sign

ificantat

5percent.

***Sign

ificantat

1percent.

1171



on what would be the impacts on crime if the same policy were to beapplied citywide, as is relevant for many current policy debates aroundthe world. To do so, we develop a structural model of the market de-mand for cannabis, accounting for the endogenous choices of the policeand cannabis users. We first calibrate the model to the localized policingexperiment in Lambeth and then consider a counterfactual policy ex-periment of citywide depenalization.33

A. A Model of Cannabis Use, Nondrug Crime, and Policing

1. Cannabis Users

There are two locations, indexed by b : the borough of Lambeth ðb 5 1Þand the rest of London ðb 5 0Þ, with a population Nbt in location b attime t . Individuals make two choices: whether to buy ðand thus con-sumeÞ cannabis and, if they buy, which location b to buy cannabis from.Individuals are heterogeneous in two dimensions: the propensity to con-sume cannabis and the cost of moving from one location to another.We assume that individuals can be caught for cannabis crime only in thelocation of purchase.The utility of consuming cannabis comprises three components: an

individual-specific utility d, the moving cost incurred if the individualtravels to the other location to purchase cannabis, lD , and a cost of beingapprehended with cannabis by the police if purchasing in location b , de-noted abtp

Dbt . The term pD

bt is the ðendogenousÞ likelihood that an in-dividual is caught in possession of cannabis, and we refer to this as the“detection rate.” The term abt is the location-specific cost when appre-hended. This is indexed by time t as the LCWS experiment in Lambethcan be seen as partly operating though a reduction in a1t relative to a0t ,as those caught in possession of cannabis are no longer arrested unlessthere are additional aggravating factors, as described in Section II.Assume that individual i resides in location b . Her utility from con-

suming in her own borough is denoted uDibt , her utility from consuming

in the other borough is uDi;2bt , and her utility from not consuming is uND

it ,

33 The structural model does not emphasize how the behavior of cannabis suppliers mightalter with depenalization, and as such, the model is not used to make price predictions oncannabis across locations. We make this modeling choice because ðiÞ information about thecriminal supply side is lacking and ðiiÞ information on drug prices at the borough-month levelis also unavailable; it is unclear how reliable such price information would be given that it isoften based on selective samples of drug busts, and there is considerable dispersion in price-quality ratios for illicit drugs ðGalenianos, Pacula, and Persico 2012Þ. Our approach is relatedto those used by Imrohoroglu, Merlo, and Rupert ð2004Þ, Conley and Wang ð2006Þ, and Fuand Wolpin ð2013Þ, who develop equilibriummodels of crime and policing. Our approachdiffers as we allow for endogenous mobility across locations and specialization in differenttypes of crime. Moreover, identification of the parameters of the model is achieved usingquasi-experimental variation through the introduction of the LCWS policy.




and we normalize this last term to zero. Hence the utility of consumingcannabis is given by uD

it :uDit 5max½uD

ibt ; uDi;2bt �; ð3Þ

uDibt 5 d2 �db 2 abtp

Dbt if consuming in b;

uDi;2bt 5 d2 �db 2 lD 2 a2btp

D2bt if consuming in 2b:

ð4Þ

An individual purchases cannabis from some location if uDit> 0. We as-

sume that d is uniformly distributed over ½0, 1�. The parameter �db de-termines the share of the population that consumes cannabis in theabsence of policing ðif pD

bt 5 0Þ. We allow this parameter to vary acrosslocations to capture different preferences between Lambeth residentsand the rest of London. We assume that the moving cost lD is uniformlydistributed over ½0; �l� and that d and lD are uncorrelated.34

The term Dbt denotes the market demand for cannabis in location band period t , namely, the number of cannabis users in b . This is the sumof the number of users who reside in location b and prefer to consumethere and users from location 2b who prefer to move and buy cannabisfrom b :

DbtðpDbt ; p

D2btÞ5 NbtPrðuD

ibt> uD

i;2bt ; uDibt

> 0Þ1 N2btPrðuD

i;2bt> uD

i;2bt ; uDi;2bt

> 0Þ: ð5Þ

The model makes precise the interlinkages in cannabis markets acrosslocations. The equilibrium market size for cannabis in each borough is afunction of ðiÞ the detection rates in both boroughs ðpD

bt , pD2btÞ that are

endogenously determined as described below, ðiiÞ the punishment forcannabis-related criminal activities in both locations ðabt , a2btÞ, andðiiiÞ the populations of both boroughs ðNbt , N2btÞ. As cannabis marketsacross locations are interlinked, depenalization policies in one bor-ough will change the behavior of cannabis users in all boroughs andpotentially induce drug tourism across boroughs.As the population in the rest of London ðN 0tÞ is orders of magnitude

larger than that in Lambeth ðN 1tÞ, there can be very large impacts on thesize of the cannabis market in Lambeth as the result of a localized de-penalized policy. As made precise below, this channel of consumers mov-ing location to buy cannabis would be considerably weakened in thepresence of a citywide depenalization policy that ensures that the pun-

34 The assumption that d and lD are uncorrelated is driven by the available data: we donot have individual crime data to identify the provenance of offenders, so any correlationbetween these parameters cannot be identified.




ishment for cannabis-related criminal activities remained homogeneousacross locations ðabt 5 a2btÞ.

2. Policing and Arrests for Cannabis Offenses

Each borough has its own police force, and we assume that each acts in-dependently of the other.35 The size of the police force, or total policeresources, in location b is denoted Pbt . A fraction, fbt , of these resourcesare devoted to cannabis-related crime. The number of individuals ar-rested for cannabis crime is a function of police resources allocated to-ward such crime and the market demand for cannabis in location b , Dbt .

36

We postulate a Cobb-Douglas specification for this relation:

ArrestsDbt 5 gDðfbtPbtÞqDD12qDbt ; qD ∈ ½0; 1�: ð6Þ

3. Nondrug Crime

Individuals from both locations choose whether to commit nondrugcrime and where to commit it. Following a formulation similar to theone above, we assume that individuals are heterogeneous in two dimen-sions: the propensity to commit crime and the cost of moving from oneborough to another. The utility of committing crime depends on ðiÞ anindividual-specific utility component, x; ðiiÞ the moving cost if there is achange of location, lC ; and ðiiiÞ the cost of being apprehended by thepolice, bpC

bt : pCbt is the ðendogenousÞ detection rate for nondrug crime in

location b at time t , where we assume that individuals are caught for non-drug crime in the location of the crime. The term b is the cost of commit-ting nondrug crime when apprehended and is the same across locations.Normalizing the utility from not committing crime to zero, the utility ofcommitting crime in one of the two locations is then given by uC

it , where

uCit 5max½uC

ibt ; uCi;2bt �; ð7Þ

uCibt 5 x2 �xb 2 bpC

bt if committing crime in b;

uCi;2bt 5 x2 �xb 2 lC 2 bpC

2bt if committing crime in 2b:ð8Þ

Individual i commits crime if uCit> 0. We assume that x is uniformly dis-

tributed over ½0, 1�; �xb determines the share of individuals who commit

35 This matches the evidence in table A1 on police operations in London boroughs inour study period: there is little evidence of a spike in police operations in other Londonboroughs around the time of the LCWS to potentially offset any of its impacts.

36 We are implicitly assuming that all ðor a fixed fraction of Þ cannabis crimes are re-ported to the police, so the number of cannabis offenses equals DbtðpC

bt ; pC2btÞ ðor some

fraction of Dbtð�ÞÞ. As discussed earlier, the depenalization policy should have no impacts onpolice behavior in terms of their searching for cannabis offenses. Hence we focus on howthese offenses convert to arrests, which is a margin directly affected by the policy.




crime in the absence of policing ðif pCbt 5 0Þ. Again, we allow for different

propensities to commit crime between Lambeth and the rest of Londonby allowing �xb to vary across locations. We assume that x is uncorrelatedwith the moving cost lC . In consequence, x is also then uncorrelated withd so that an individual’s underlying propensity to use cannabis is unre-lated to his underlying propensity to commit nondrug crime.37 The num-ber of crimes committed in location b is then given by

CbtðpCbt ; p

C2btÞ5 NbtPrðuC

ibt> uC

i;2bt ; uCibt

> 0Þ1 N2btPrðuC

i;2bt> uC

i;2bt ; uCi;2bt

> 0Þ; ð9Þ

and we assume that all crimes are reported to the police, so the numberof nondrug criminal offenses equals CbtðpC

bt ; pC2btÞ. As with the market de-

mand for cannabis, the number of crimes committed in location b de-pends on characteristics and police behavior across both locations.Finally, the number of arrests for nondrug crime in location b will then

depend on the fraction ð12 fbtÞ of police resources Pbt that are de-voted to nondrug crime in location b and the actual number of nondrugcrimes committed. We again assume a Cobb-Douglas relationship so that

ArrestsCbt 5 gC ½ð12 fbtÞPbt �qC C 12qCbt ; qC ∈ ½0; 1�: ð10Þ

4. Equilibrium Detection Rates

The key endogenous outcomes in the model are detection rates forcannabis and nondrug crime in each location, ðpD

bt , pCbtÞ. Detection rates

are the ratio of the number of offenders caught by the police to the totalnumber of offenders. Hence they are determined through an interac-tion of the police and cannabis users and are the solution to the fol-lowing system of equations:

pDbt 5

gDðfbtPbtÞqDDbtðpDbt ; p

D2btÞqD

DbtðpDbt ; p

D2btÞ

;

pCbt 5

gC ½ð12 fbtÞPbt �qC CbtðpCbt ; p

C2btÞqC

CbtðpCbt ; p

C2btÞ

:

ð11Þ

Given the nonlinearity of this system, there are no closed-form solutionsfor ð11Þ. We therefore solve the model numerically by searching for thedetection rates that bring the left- and right-hand sides in ð11Þ as close aspossible, where a solution consists of four detection rates: fpD

0t ;pD1t ;p

C0t ;

37 Of course this assumption could be relaxed to capture the fact that cannabis marketsmight correlate with some nondrug crimes, such as property crime ðFergusson and Hor-wood 1997; Corman and Mocan 2000Þ. However, we would need to find more detailedindividual crime data ðe.g., which recorded multiple offenses where relevantÞ to incorporatethis feature into the model.




pC1tg. By looking at the whole support of the detection rates, ½0, 1�, we find

all the sets of detection rates that solve the system of equations ð11Þ for agiven value of the parameters. For any set of equilibrium detection rates,we then compute the market demand for cannabis, the number of of-fenses for cannabis and nondrug crime, and the number of arrests fornondrug crimes in all locations. This is done by using equations ð5Þ, ð6Þ,ð9Þ, and ð10Þ. There can be multiple equilibria generated, and how wemake the choice between these equilibria is explained below when wedetail the calibration procedure.

5. Modeling the Localized Policing Experiment

We define two time periods and denote by t 5 t B the time period beforethe policy is implemented ðcorresponding to the period from April 1998to June 2001Þ and denote by t 5 t A the time period after the policy isintroduced ðfrom January 2002 to March 2004Þ. We discard the first6 months of the policy to allow for transitional dynamics. We model thelocalized policing experiment in Lambeth as operating through two chan-nels. The first is a reduction in the penalty of being caught in possession ofcannabis, closelymatching the policy description in Section II: the penaltyis a1tB in Lambeth before the policy and decreases to a1tA

< a1tB in thepostpolicy period.38 We assume that in the rest of London, the penalty forcannabis arrest is the same during the two periods and that before thepolicy it is similar to the penalty in Lambeth ða0tB 5 a0tA 5 a1tB Þ.Second, we allow the police to reallocate their resources between can-

nabis and nondrug crime. In our model, this is captured by the fact that,in Lambeth, f1tA

< f1tB . We assume that in the rest of London, there is nochange in the fraction of the police force dealing with cannabis crimeðf0tB 5 f0tAÞ. This channel creates a linkage between cannabis crime andnondrug crime, which the reduced-form evidence suggested was an im-portant policy impact to consider.Such a localized policy change operating only in Lambeth ðb 5 1Þ will

then have two impacts on the market demand for cannabis in LambethðD1tAðpD

1tA; pD

0tAÞÞ: ðiÞ Lambeth residents will be more prone to consuming

cannabis and ðiiÞ residents in the rest of London will be more inclined totravel to Lambeth to purchase cannabis. These changes will affect theequilibrium detection rates for cannabis crime ðpD

btAÞ, which will in turn

determine the equilibrium proportion of the population consuming can-nabis and the number of cannabis users caught by the police. If thepolicy allows a reallocation of the police force toward nondrug crime

38 As described in Sec. II.B, Lambeth’s cannabis policing strategy did not return iden-tically to what it had been before the policy. Rather, it adjusted to be a firmer version ofwhat had occurred during the pilot. As evidenced in cols. 3 and 4 of table 3, there was apermanent reduction in police effectiveness against cannabis possession crime in Lam-beth.




ðso that 12 fbt increasesÞ, the policy impacts will then spill over to othercrimes, changing the equilibrium detection rates for nondrug crime ðpC

btAÞ

and thus the proportion of the population that chooses to commit non-drug crime.

B. Calibrating the Model to the Localized Policing Experiment

1. Calibration Method

The model has 16 parameters: ðiÞ five parameters describe preferencestoward cannabis consumption, moving across boroughs, and penaltiesassociated with arrests: �d0, �d1, �l, a1tB , and a1tA ; ðiiÞ three parameters de-scribe nondrug crime preferences and penalties: �x0, �x1, and b; andðiiiÞ eight parameters describe the arrest production functions: gD0, gD1,qD, gC , qC , f0tB , f1tB , and f1tA . We allow the arrest technology parameterfor cannabis crime gD to vary between boroughs as the two locationshave different arrest rates conditional on offenses. For nondrug crime,a good model fit is achieved with a common parameter gC.We calibrate all but two of the model parameters on the basis of the

localized LCWS policy to reproduce key features in the data. It is difficultto identify the parameter fbt, which is the fraction of the police forcedevoted to cannabis crime, using only observed crime in the prepolicyperiod. We therefore identify this parameter from other sources of dataas detailed below. The variation introduced by the LCWS, and in par-ticular the differential change in nondrug crimes across boroughs andtime, then allows us to identify the change in the fraction of police timedevoted to cannabis crime in Lambeth ði.e., f1tA=f1tB Þ.We rely on data moments computed for Lambeth and the rest of

London and for two periods: before the LCWS policy is in place and thepostpolicy period. We have a total of 17 moments that describe ðiÞ theprevalence of cannabis consumption, ðiiÞ the number of recorded of-fenses for cannabis, ðiiiÞ the number of offenses for other crimes, ðivÞ thenumber of arrests for other crimes, and ðvÞ the share of cannabis users inLambeth from other London boroughs before the policy. These mo-ments are chosen because they are direct outputs of the model and be-cause they best capture all the key policy impacts documented in theearlier reduced-form evidence. We now describe how each of these em-pirical moments is measured.For moment i, data on cannabis consumption for the rest of London

are derived from the British Crime Survey ðBCSÞ, which asks about can-nabis usage. We use the 2000/2001 and 2006 survey waves to measurecannabis consumption before and after the policy in the rest of London.As the BCS has only a few respondents in Lambeth, we estimate theprevalence of cannabis consumption in Lambeth by scaling the BCS-derived figure for the rest of London by the ratio of cannabis offenses




in Lambeth to those in the rest of London. We do so for the pre- andpostpolicy periods. Implicit in this scaling is the assumption that the re-lationship between cannabis use and offenses for cannabis possession isthe same in all locations. As highlighted throughout, LCWS policy wouldnot alter how the police would track or record offenses, all else equal.For moments ii–iv, data on offenses and arrests are taken from the

same administrative crime records from the MPS as used in the reduced-form analysis. For the calibration exercise, offense and arrest rates areexpressed per 1,000 inhabitants. Finally, for moment v, the share of can-nabis consumers in Lambeth from outside the borough in the preperiodis recovered from an MPA document.39

Our model requires three additional inputs: population size, the num-ber of police officers in each location, and the fraction of police timedealing with cannabis crime. The former is obtained from the QLFS-LAdata described earlier. For the second, we use data from MPA reportsdescribed in Section V.C, which report the number of police officers inboth Lambeth and the rest of London, during the prepolicy and the post-policy periods. As described earlier, during this time span, the numbers ofofficers have increased in both locations, at approximately an equal rate.To compute the fraction of police devoted to cannabis crime before

the policy, fbtB , we rely on additional data that characterize the numberof hours taken up by arrests linked to cannabis possession and total ef-fective police time. We denote by Hours ProcD the hours taken to pro-cess a cannabis arrest, which include the transfer of the offender to thepolice station, file processing, and time spent in prosecution.We use datafrom police reports that evaluate the time required to process each ar-rest linked to cannabis to be about 7 hours ðWood 2004Þ.40We obtain an estimate of total effective police time by multiplying the

size of the police force in a given borough, as recorded by the MPA anddiscussed in Section V.C, by an estimate of the time spent by the averagepolice officer on effective policing in London each year ðnamely, net oftime on holiday, sick days, training attendance, and other administrativeworkÞ. Herbert et al. ð2007Þ provide an estimate of this effective police

39 The share of cannabis consumers in Lambeth fromoutside the borough in the prepolicyperiod ismentioned inapp.6of theminutesof the followingMPAcommitteemeeting:http://policeauthority.org/Metropolitan/committees/mpa/2002/020926/17/index.html.

40 The Political Risk Services consultancy group, which evaluated the pilot scheme at the6-month point, estimated that for every individual apprehended with cannabis in which acaution rather than an arrest was issued, 3 police hours were saved by avoiding custodyprocedures and interviewing time. However, the MPA noted that the 3 hours per offensefigure was conservative, as it “was based on the premise of an officer working alone. It tookno account of the time spent transporting the arrested person to a police station and thetime waiting to book them in on arrival.” A later MPA report following the nationwidedeclassification stated that the time saving was 5 hours dealing with a cannabis arrest and2 more hours of operational time at police stations ðWood 2004Þ. We use this stated 7-hourreduction in processing time to calibrate the model.




time. The fraction of police time devoted to cannabis arrests is then ob-tained by

fbtB 5Arrests

D

btB�Hours ProcD

Total effective police timebtB

; ð12Þ

where ArrestsD

btBis the average number of arrests for cannabis offenses in

borough b in the prepolicy period.41

Given these inputs to the model from other data sources, the cali-bration of the remaining parameters is obtained using a minimum dis-tancemethod, where weminimize the quadratic distance between the ob-served and predicted moments, equally weighting each moment. For agiven value of the parameters, we may have several predictions due tomultiple equilibria. We compute the distance for all possible equilibriaand select the one that brings the predicted and observed moments theclosest. Themodel was solved numerically using 20,000 simulation draws,a number large enough so that increases in simulations did not changethe objective function. The search was done using a gradient-free opti-mizer built on the Simplex method. Finally, we note that the estimationwas started with many different initial parameter values to ensure that itconverged to a global minimum.

2. Results

Panel A of table 8 presents the observed and predicted moments de-scribed above: the model does a good job in matching the moments.For eight ð15Þ of the moments, the difference between the observed andpredicted moments is less than 5 percent ð10 percentÞ. A x2 goodness offit does not reject the hypothesis that the predicted moments are jointlythe same as the observed ones. Column 5 of table 8 displays a transfor-mation of the key moments related to crime: the difference in differencefor recorded log offenses of cannabis and nondrug crimes. These arecalculated across locations and time and transformed into percentages,and they are therefore comparable to the reduced-form results discussedearlier. Along this dimension, our model is able to reproduce two of thekey impacts of the policy quite well: ðiÞ the model predicts a 66.4 percentincrease between the pre- and postpolicy periods in recorded canna-bis offenses in Lambeth relative to the rest of London ðcompared to an

41 Hencewe focus onmodeling the timedevoted toprocessing arrests rather than the timedevoted to recorded offenses or warnings. On the time devoted to offenses, there should beno change in how offenses are recorded because of the policy, as discussed earlier. On timedevoted to warnings, we make the simplifying assumption that the time involved issuing awarning is negligible compared to the time involved in arresting and processing an offender.This seems reasonable as a warning can be issued verbally with no formal paperwork beingrequired.




TABLE8

FitoftheStructuralModel

PrepolicyPeriodðA

pril19

98to

June20

01Þ

PostpolicyPeriodðJa

nuary

2002

toMarch

2004

ÞLam

beth

ð1Þ

RestofLondon

ð2Þ

Lam

beth

ð3Þ

RestofLondon

ð4Þ

Differencein

Difference:

Lam

bethvs.Rest

ofLondonð%

Þð5Þ

A.Match

edMomen

ts

Can

nab

isco

nsumption:

Observed

.184

.123

.187

.123

Predicted

.176

.135

.197

.117

Can

nab

iscrim

eoffen

serate:

Observed

.366

.284

.825

.335

64.8

Predicted

.366

.288

.820

.332

66.4

Nondrugcrim

eoffen

serate:

Observed

18.9

14.1

18.2

14.6

27.26

Predicted

18.2

14.9

18.6

16.0

24.95

Nondrugcrim

earrest

rate:

Observed

2.30

2.40

2.04

1.96

Predicted

2.02

1.88

2.00

1.96

B.DrugTourism

Shareofcannab

isoffen

dersin

Lam

beth

from

therest

ofLondon:

Observed

.39

Predicted

.39

.60

C.DetectionProbab

ilities

Can

nab

iscrim

e:Predicted

.001

27.002

2.001

7.003

125.13

%NondrugCrime:

Predicted

.111

.126

.107

.122

2.444

%

Note.—Offen

sesan

darrestsareex

pressed

per

1,00

0inhab

itan

ts.T

hedifference

indifference

inpercentage

sreported

inco

l.5iscalculated

asðco

l.32

col.1Þ

2ðco

l.42

col.2Þ

foreach

observed

andpredictedmomen

t,whereeach

valueisfirstlogg

ed.Thedataonoffen

sesan

darrestsaretake

nfrom

thead

ministrativecrim

ereco

rdsfrom

theMPS.

Dataoncannab

isco

nsumptionarederived

from

theBritish

Crime

Survey

ðwhichhas

borough

iden

tifiersÞ

andfrom

dataonreco

rded

offen

ses.



observed difference-in-difference increase of 64.8 percentÞ, and ðiiÞ themodel predicts a 4.95 percent reduction in nondrug crime comparedto an observed decrease of 7.26 percent.Moreover, the model highlights an important mechanism that was not

captured in the reduced-form results, shown in panel B: there is a re-location of cannabis consumers from the rest of London toward Lam-beth after the policy. The share of cannabis consumers in Lambeth whoare from the rest of London matches the observed one ð39 percentÞbefore the policy was in place. The model predicts that this share risesfrom 39 percent before the policy to 60 percent under the localized de-penalization policy. This near doubling of drug tourists shows how theinterlinkage in cannabis markets across locations is a key reason whyoffense rates for cannabis-related crime in Lambeth rise so much withthe localized depenalization policy.Appendix table A5 shows the calibrated parameter values from this

exercise. Panel A focuses on the two parameters describing the initial ðex-ogenousÞ channels through which the policy operates as discussed above:a1tA=a1tB and f1tA=f1tB . As shown in the first row, the data are matched witha reduction in the penalty of getting caught with cannabis in Lambeth byabout 82 percent. This captures the fact that all recorded offenses leadto arrests before the policy, while most offenders were left with only acaution afterward ðthe exception being those offenses that occurred afterthe policy that had aggravating factorsÞ. The policy is also associated witha reallocation of about 53 percent of police time in Lambeth devoted tocannabis before the policy to nondrug crime afterward. To be clear, thischange in Lambeth should be interpreted as the combined effect fromany reallocation of police resources, changes in processing times for ar-rests after the policy, or the differential hiring of police for cannabis andother crimes after the policy: all these channels are captured in a reductionin f1tB relative to f1tA .

42

Panel C of table 8 displays the equilibrium detection rates for cannabisand nondrug crime, for each location and period. The detection ratesfor cannabis consumption are very small, reflecting the fact that a sizablefraction of the population uses cannabis and very few of them are ac-tually arrested each year. For nondrug crimes, offenses are rarer and

42 On the other calibrated parameters, panel B of table A5 shows that the preferenceparameters are such that a higher share of the Lambeth population would consume can-nabis in the absence of policing ð�d1 < �d0Þ but that the disutility from committing crime isnearly identical across locations ð�x1 5 �x0Þ. Panel C shows the calibrated policing technol-ogy parameters and suggests that the total factor productivity ðTFPÞ like parameter on theapprehension technology for cannabis crime is higher in London than inLambeth ðgD0

> gD1Þ.The corresponding TFP-like parameter for nondrug crime is fixed to be the same acrosslocations, but we note that its value is orders of magnitude higher ðgC

> gD0, gD1Þ so that in-dividuals are far more likely to be arrested for nondrug crime than for cannabis-related crime.




arrests relatively more frequent: Panel C shows that around 12 percentof nondrug crimes lead to an arrest ðin contrast, only 0.2 percent of can-nabis users are arrestedÞ. In column 5 of table 8, we also report the dif-ference in difference for the detection probabilities, again normalizedby their prepolicy levels in Lambeth. Detection rates for cannabis crimedeclined in Lambeth relative to the rest of London by around 5.13 per-cent, while the detection rate for nondrug crimes remains almost un-changed.To assess the plausibility of our calibrated model, we compute the elas-

ticity of total recorded criminal offenses with respect to the size of thepolice force, namely, the elasticity of CbtðpC

bt ;pC2bt ; �Þ1 DbtðpC

bt ;pC2bt ; �Þ with

respect to Pbt , the total number of police officers in location b . Earlierstudies have estimated this elasticity, exploiting very different research de-signs. Our structural model predicts an elasticity of 20.3 in Lambeth andabout 20.9 in the rest of London. The estimates of this elasticity in theliterature range from 0 ðMcCrary 2002Þ to 20.9 ðLin 2009Þ, and manystudies find an elasticity on the order of 20.3 to20.5 ðLevitt 1997, 2002;Corman andMocan 2000; Draca et al. 2011Þ. Hence, although ourmodelwas not calibrated to match these elasticities, they appear to be consistentwith previous results and provide external validity to our method.

C. A Counterfactual Policy Experiment: Citywide Depenalization

We now use the calibrated model to perform a counterfactual policyanalysis, which decreases the penalty of cannabis consumption citywide.Hence in both locations we allow the penalty to fall by the same ex-tent, as captured by the ratio a1tA=a1tB . We also adjust the police time de-voted to cannabis crime in each borough to match the change we observeðf1tA=f1tB Þ. Table 9 shows the change over time in a number of key statis-tics, expressed as a percentage change from the baseline level of the sta-tistic in the prepolicy period as a result of a citywide depenalization.This exercise shows the following. First, panel A highlights that the city-

wide depenalization of cannabis possession leads to a modest increase inthe prevalence of cannabis consumption, of about 1 percent in Lambethand 2 percent in London ðwhere the baseline prevalence is lowerÞ.43 Sec-ond, other crimes in the rest of London would actually fall in the city-wide policy ðby around 0.3 percentÞ as all police forces reallocate efforttoward nondrug crime.

43 This result contributes to the literature on the impact of drug policies on drug usage,on which the evidence remains mixed ðDiNardo and Lemieux 2001; Damrongplasit et al.2010; Pudney 2010Þ. Braakmann and Jones ð2012Þ evaluate the impact of the declassifi-cation of cannabis in the United Kingdom in 2004 on cannabis consumption: they find thepolicy to increase cannabis consumption, predominantly because of individuals starting toconsume cannabis.




Third, panel B highlights that in Lambeth, the share of cannabis usersoriginating from outside the borough decreases by 4 percent comparedto the baseline ðand by more than 60 percent compared to the actuallocalized policy periodÞ. In short, a citywide policy would much elimi-nate drug tourism, which is a key driving force in the localized experi-ment. Fourth, panel C highlights how a citywide policy would affectequilibrium detection probabilities across crime types: in both locationsthe structural model predicts a fall in the detection rate for cannabis con-sumption by around 7 percent and an increase in equilibrium detectionrates for nondrug crimes of around 0.2 percent.44

Linking these findings back to the documented welfare impacts inSection VI, we see that because citywide depenalization eliminates incen-tives for drug tourism, the cannabis market in Lambeth increases in sizeless dramatically than under a localized depenalization policy. As such, anyantisocial behaviors that are correlated to the size of the cannabis mar-ket but arenot captured in crime ratesmight thenbe reduced.Hence citywidedepenalization might then have far smaller negative impacts on propertyprices in Lambeth compared to the documented impacts of a localized po-licing experiment.

44 We can validate some of the model’s predictions using the actual nationwide depenal-ization of cannabis possession that took place from January 2004 until January 2009. This wasimplemented in a way rather similar to the Lambeth policy. We estimate the reduced-formimpacts of this policy on crime using a simple before-after comparison, which is obviouslysubject to far more caveats than the difference-in-difference design we used to evaluate theLCWS. In addition, the demographic controls from theQLFS-LA data are available only until2006:Q1, so these have to be extrapolated until 2010 to estimate the impacts of the nation-wide policy. Doing so, we find that crimes related to cannabis possession significantly risewhen the nationwide policy is in place and that offense rates for other nondrug crimessignificantly fall during this period ðand police effectiveness against them risesÞ.

TABLE 9Predicted Impacts of Citywide Depenalization ð%Þ

Lambethð1Þ

Rest of Londonð2Þ

A. Cannabis and Crime

Cannabis consumption 1 2Cannabis crime offense rate 27.4 24.0Nondrug crime offense rate 2.3 2.3Nondrug crime arrest rate 0 2.1

B. Drug Tourism

Share of cannabis offenders in Lambethfrom the rest of London

24

C. Detection Probabilities

Cannabis crime 26.9 27.4Nondrug crime .20 .22

Note.—Offenses and arrests are expressed per 1,000 inhabitants.




VIII. Conclusion

Cannabis users account for 80 percent of the 200 million illicit drugusers in the world ðUN Office on Drugs and Crime 2010Þ. Understand-ing the impacts of government intervention in the market for cannabisis of huge importance. In this paper we study the impacts of a commonintervention: thedepenalizationof cannabis, where thepossessionof smallquantities of cannabis no longer leads to individuals being arrested ðal-though such incidents are still recorded as offensesÞ. More precisely, weevaluate the impacts on the level and composition of crime, and socialwelfare as measured by house prices, of a localized depenalization policythat was implemented in the London borough of Lambeth.We have documented how the policy changed crime patterns during

and after the depenalization policy using administrative records on crim-inal offenses by drug type, by specific drug offenses that proxy demand-and supply-side criminal activities, and for seven types of nondrug crime.We find that depenalization in Lambeth led to an increase in cannabispossession offenses that persisted well after the policy experiment ended.We find evidence that the policy enables the police in Lambeth to beable to reallocate their effort toward nondrug crime: there are signif-icant long-run reductions in five nondrug crime types and significant im-provements in police effectiveness against such crimes as measured by ar-rest and clear-up rates.The totality of evidence is best interpreted through the depenalization

policy causing a behavioral response of the police among two dimensions:to reduce the penalties of being caught in possession of cannabis and toreallocate resources toward nondrug crime. Both channels then cause anendogenous response among potential users of cannabis in terms of thechoices over whether and where to buy and consume cannabis.We use thekey lessons from this localized policing experiment to shed light on whatwould be the impacts on crime if the same policy were to be applied city-wide by developing and calibrating amodel of themarket for cannabis andcrime, accounting for the behavior of police and cannabis users.While our model highlights some novel and important channels

through which a depenalization of cannabis affects the level and compo-sition of crime, it still leaves open areas for future research on how illicitdrug policy affects the behavior of drug suppliers and the police. In par-ticular, on drug suppliers, research on how drug policies change the or-ganization of criminal activity remains scarce; and on police behavior,much remains to be understood regarding the extent to which policeacross jurisdictions should coordinate strategies.We have provided a comprehensive review of the impact of depenal-

ization policies along four margins: drug and nondrug crimes, the lo-cation of crimes, short- and long-run policy responses, and impacts onwelfare as measured by house price changes. Our detailed and nuanced




reduced-form and structural form results are relevant for other settingsgiven that the depenalization policy we study reflects how liberal drugpolicies have been implemented by many other countries ðDonohueet al. 2012Þ, and the issue of whether and how governments should in-tervene in illicit drug markets remains at the top of the political agendaacross the world.45

Appendix

A. Crime Data: Definitions

Home Office counting rules for criminal offenses are periodically revised, includ-ing in 1998, so coinciding with the start of our sample period. Importantly, changesin Home Office guidelines/definitions are uniformly applied across all London bor-oughs and hence will not drive the difference-in-difference estimates on crime. Therewas another revision in the recording of crime in April 2002, with the introductionof the National Crime Recording Standard ðNCRSÞ. The Crime in England and Wales2004/2005 Report states the NCRS “aimed to introduce greater consistency to theprocess of recording crime and to establish a more victim-oriented approach torecording.The impactof theNCRS . . .was to increase thenumbersof crimes recordedand less serious violent offences were particularly affected” (Home Office 2005). Ina robustness check in table A2, we reestimate our baseline results on the impact of theLCWSpolicy on drug crime by additionally adding in a series of dummies equal toone for when each data regime is in place, and zero otherwise.

There have been a number of changes to recording practices and the sanc-tions available that have affected the recorded clear-up ðdetectionÞ rates. TheHome Office counting rules for recorded crime changed in April 1998. Thesebrought new offenses into the series with varying clear-up rates. It is estimatedthat the effect of the changes was to increase the overall clear-up rate from 28 per-cent to 29 percent. Additional changes were implemented with effect from April1999. Any recorded clear-up required “sufficient evidence to charge” and an in-terview with the offender and notification to the victim. In addition, clear-ups ob-tained by the interview of a convicted prisoner ceased to count. The overall effectof the April 1999 change is estimated as a single percentage point decrease in clear-up rates ðalthough the effect varied between crime typesÞ. Finally, the implemen-tation of the NCRS in April 2002 is thought to have had an inflationary effect onrecorded crime, and the assumption is that it has depressed clear-up rates sinceadditional recorded crimes are generally less serious andpossibly harder to clear up.

B. Cannabis Crime: Robustness Checks

Table A2 presents a series of robustness checks on the baseline result docu-mented in table 2, that the LCWS policy led to a significant increase in offenserates for cannabis-related crime in Lambeth relative to the rest of London be-

45 For example, the states of Colorado and Washington legalized possession of 1 ounce orless of marijuana for recreational use by adults ðthose 21 years or olderÞ in November 2012.At least 12 other states are considering similar policies. In Europe, Croatia decriminalized thepossession of small amounts of cannabis in 2013. In Latin America, Uruguayan President JoseMujica has proposed to put into place a legal state-controlled market for cannabis.




tween the prepolicy and policy periods; this effect persisted in the long run afterthe policy.

The first robustness check excludes geographic neighbors to Lambeth whenestimating ð1Þ. We define the geographic neighbors of Lambeth to be the bor-oughs that have contiguous land borders with Lambeth: Croydon,Merton, South-wark, and Wandsworth. Given the interlinkages between cannabis markets andthe dense network of public transport across boroughs in Lambeth, we expectcannabis users to travel to Lambeth to purchase cannabis in response to the policyðthe lower costs of apprehension and the endogenous reduction in detectionratesÞ. If such users originate only from neighboring boroughs, then by excludingsuch neighbors from ð1Þ, we will estimate the true impact of the policy on cannabiscrime in Lambeth. The result in column 2 shows the impacts to be almost un-changed if the neighbors to Lambeth are excluded as controls. This suggests thatcannabis users who switch to purchasing cannabis from Lambeth because of thepolicy are likely to originate from all over London.

The next robustness check in column 2 accounts for common citywide shocksto cannabis crime through the inclusion of year fixed effects into ð1Þ. The dif-ferential impacts of the LCWS policy in Lambeth during and after the policyperiod are identified because these periods cut across years. We see that the co-efficients of interest are very similar to the baseline specification. Column 3 thenshows the results to be robust to including a series of dummy variables for whendifferent data regimes are in place ðas described in the subsection aboveÞ; col-umn 4 shows the baseline results to be robust to additionally including the fullset of police operations operating in single or groups of boroughs ðpanel A intable A1Þ where start and end dates of the operation are known.

Finally, column 5 allows for spatially correlated error structures. Given the in-terlinkages in cannabis markets across locations, as well as the possibility of po-lice across boroughs coordinating strategies, there might be correlation in theerror structure in ð1Þ. To account for this possibility, we model the error term asfollows:

ubmy 5 ubt 5 rWubt 1 ebt ; ðA1Þ

where r is the coefficient on the spatially correlated errors, and W is the spatialweighting matrix of dimension 32 � 32 as there are 32 London boroughs. Wespecify W to be a contiguity spatial weighting matrix, where wij 51 if borough jneighbors borough i , and 0 otherwise. The result in column 5 is similar to thebaseline estimate: the parameters of interest remain with the same sign and sig-nificance, and both point estimates are marginally larger. For this model, r5 :346with a standard error of .0224, indicating the presence of spatial correlation in theerror terms. We have also experimented with several other W specifications, in-cluding inverse distance and inverse distance squared matrices ðdistance is cal-culated as the Euclidian distance from the centroid of each borough to all oth-ersÞ, and found results to be robust to these different weightingmatrices.

C. Defining Crime Hot Spots

In analyzing the impact of the depenalization policy on house prices in Lambethrelative to other London boroughs, we exploit the fact that data on house pricesand crime are available, for some years, at a more disaggregated level within each




borough. House price data are available at the zip code sector level from theUK Land Registry. Ward-level crime data are available monthly from April 2001from the MPS. We use the ward-level crime data to first define each ward as acrime hot spot, and we first describe how this is done. We then describe how wematch ward-level crime data to zip code sectors for which house price data areavailable ðas wards and zip codes do not correspond to the same geographicareasÞ to ultimately define zip codes as being crime hot spots.

Given our policy focus, our primary hot spot measure is based on the inci-dence of drug crime in each ward. A ward is defined as a hot spot if drug offensesare above the median for all wards in the same borough. One of the robustnesschecks described below experiments with using an alternative threshold for de-fining ward hot spots.

The ward-level crime data are available monthly from April 2001. We use themto create hot spots on the basis of two definitions: ðiÞ ex ante levels of drug crime,using the 3 months of data prior to the start of the LCWS, and ðiiÞ ex post levels ofdrug crime, based on ward-level drug crime rates in the period October 2007 toSeptember 2009.

Once the ward-level hot spots are defined, these must be mapped onto zipcode sectors to be able to create zip code sector hot spot markers to include asthree-way interactions in the house price regression ð2Þ. In general, zip codesectors are smaller than wards, but more importantly, the two do not perfectlyoverlap. The average number of wards in a zip code sector is 4.1 ðeven though zipcode sectors are the smaller unit of the twoÞ. For our baseline specifications, wethen define a zip code sector ðe.g., WC1EÞ to be a hot spot if any ward within azip code sector is defined as a drug crime hot spot. A second set of robustnesschecks described below experiment with using alternative methods for defininga zip code as being a hot spot. Each zip code sector is then ascribed either to bea hot spot or not. Figure A1A shows for Lambeth the classification of zip codesectors into hot spots and non–hot spots based on the ex post definition. Giventhe concerns described of using an ex post definition, figure A1B shows theclassification of zip code sectors into hot spots if we use the 3 months of ex anteward-level crime data to define hot spots. Reassuringly, there is considerable sta-bility in the definition of hot spots over this time period and using this method;as a result, the empirical house price results are very similar when using eitherdefinition ðcols. 3 and 4, table 6Þ.

D. House Price Impacts: Robustness Checks

Table A4 presents robustness checks on the main house price regression in table 6.Column 1 repeats the baseline specification using zip code sector hot spots, butfor another housing type, flats, which actually correspond to the most frequenthouse type sale in our study period ðalthough the lowest price per sale for anyhouse typeÞ. The basic pattern of results holds for this housing type also: afterthe policy, house prices for flats are significantly lower in Lambeth than in otherboroughs, and there is enormous variation within Lambeth between zip codesclassified as hot spots ðwhere house prices are 20.2 percent lower than in otherLondon boroughs after the policyÞ and zip codes in Lambeth that are not clas-sified as hot spots ðwhere house prices are actually 5.3 percent higher in Lambeththan comparable areas in other London boroughs after the policyÞ.




The remaining robustness checks examine the robustness of the findingsto alternative definitions of hot spots. The first check redefines how a ward is firstdefined to be a hot spot. More precisely, we define a ward as a hot spot if drugoffenses are above the 75th percentile median for all wards in the borough. Wethen define a zip code to be a hot spot if it contains any hot spot wards so de-fined. Column 3 examines the robustness of the baseline result to changing howwe translate ward hot spots into defining a zip code sector as being a hot spot.While the baseline specification denotes the zip code sector to be a hot spot ifany ward is defined to be a hot spot, in column 3 the zip code is defined to be ahot spot if the modal ward is itself defined to be a hot spot. Column 4 then usesan alternative method to define zip code sectors as hot spots that uses infor-mation on all wards in the zip code sector. In this case, the hot spot variable is nolonger binary, but rather a weighted average of all wards’ hot spot classificationswithin the zip code sector. These weights are based on the percentage of the zipcode that overlaps with the ward. Finally, column 5 uses information on totalcrimes ðnot drug crimeÞ to redefine wards and then zip codes as hot spots usingotherwise the same method as the baseline specification.

The results in columns 2–5 of table A4 are all very much in line with thebaseline findings in table 6. In particular, for all variant specifications we see thatafter the policy, house prices are significantly lower in Lambeth hot spots thanin other boroughs, where the magnitude of the impact varies between 7.7 per-cent and 13.9 percent.

FIG. A1.—A , Ex post drug hot spots in Lambeth; B , ex ante drug hot spots in Lambeth.Hot spots are set to one if total drug offenses in the ward are equal to or above the medianwithin the borough,. The ex post period runs from October 2007 to September 2009. Theex ante period runs from April to June 2001. The darker-shaded wards are those that aredefined to be a hot spot using the ex post and ex ante data. The lighter-shaded wards arethose defined to be non–hot spot wards under each definition.




E.

Tables

TABLEA1

CodingPoliceOperations

Inform

ationSo

urcean

dOperationNam

eBorough

Start

End

Focu

s

A.Borough

-SpecificPolice

Operations;Complete

Inform

ationonStartan

dEndDates

MPA:

Recover

Green

wich,Lew

isham

,So

uthwark,

Bromley,Croydon

10/20

0512

/17

/20

07Recovery

ofab

andoned

stolenvehicles

Blunt

Lam

beth,So

uthwark,

Hackn

ey,New

ham

,Haringe

y,Tower

Ham

lets,Brent,Croydon,Waltham

Forest,

Lew

isham

,Enfield,Ham

mersm

ithan

dFulham

11/20

0411

/20

05Knifecrim

eSaferStreets

Lam

beth,Westm

inster,So

uthwark,

Hackn

ey,Haringe

y,Cam

den

,Tower

Ham

lets,Brent,Islington

2/4/

2002

3/31

/20

02Street

crim

eSaferStreetsPhase2

Lam

beth,Westm

inster,So

uthwark,

Hackn

ey,Haringe

y,Cam

den

,Tower

Ham

lets,Brent,Islington,New

ham

,Ealing,

Waltham

Forest,Lew

isham

,Wan

dsworth,

Croydon

4/15

/20

023/

31/20

03

Street

crim

e

Strongb

ox-Windmill

Lam

beth

5/8/

1999

7/2/

1999

Volumecrim

e:burglary,

robbery,vehicle

crim

e,drugs

Strongb

ox-Empire

Hackn

ey7/

17/19

999/

10/19

99Strongb

ox-Reg

isCam

den

,Islington

10/2/

1999

12/3/

1999

Strongb

ox-Victory

Westm

inster

1/22

/20

013/

18/20

01Strongb

ox-Castille

Haringe

y4/

17/20

016/

10/20

01Strongb

ox-Claym

oor

Brent

7/16

/20

019/

9/20

01Strongb

ox-Sabre

Tower

Ham

lets

9/17

/20

0112

/9/

2001



TABLEA1(C

ontinued

)

Inform

ationSo

urcean

dOperationNam

eBorough

Start

End

Focu

s

Planning,

Perform

ance

and

ReviewCommittee

reportsarch

ive:

SaferHomes

Barnet,Bromley,Croydon,Enfield,Green

wich,Harrow,

Hillingd

on,Hounslow,Lew

isham

,Red

bridge

,So

uth-

wark,

Waltham

Forest,Wan

dsworth

10/28

/20

026/

2004

Burglary

MPAan

nual

reports:

Solstice

Brent,Hackn

ey,Westm

inster,Ham

mersm

ithan

dFulham

,Lew

isham

,Cam

den

12/1/

2003

12/8/

2003

Transport

crim

e

Alnwick

Haringe

y9/

16/20

0210

/13

/20

02Street

crim

eDraca

etal.ð200

8Þ:

Theseu

sWestm

inster,Cam

den

,Islington,Tower

Ham

lets,

Ken

singtonan

dChelsea

7/7/

2005

8/17

/20

057/

7bombings

B.Borough

-SpecificPolice

Operations;Inco

mplete

Inform

ationonStartan

dEndDates

MPAan

nual

reports:

Ban

tam

Hackn

ey11

/20

01Unkn

own

Triden

trelated

Footbrake

Red

bridge

4/20

033/

2004

Veh

icle

crim

eAnuric

Ken

nington

Drugtrafficking

Dobbi

Enfield

Unlicensedminicab

sMichaelm

asEnfield

Street

crim

e,burglary

Garm

Tower

Ham

lets

Robbery

Lew

ark

Lew

isham

,So

uthwark

Robbery

Challenge

rLam

beth,So

uthwark,

Hackn

ey,Brent,Lew

isham

,Tower

Ham

lets

Robbery

Orion

Hackn

eyDrugs

Foist

Hackn

ey,Haringe

y,New

ham

Uninsuredcars

Other

sources:

Alliance

5borough

sSo

uth

London

11/20

07Unkn

own

Gan

gcrim

eKartel

11borough

s2/

25/20

08Coalmont

Southward,Lam

beth,Lew

isham

Guncrim

e

1190



Inform

ationSo

urcean

dOperationNam

eBorough

Start

End

C.London-W

idePolice

Operations

MPA:

Blunt2

AllLondon

5/14

/20

08Present

Youth

knifecrim

eBlunt

AllLondon

12/20

05Unkn

own

Knifecrim

ePlanning,

Perform

ance

and

ReviewCommittee

reportsarch

ive:

Maxim

AllLondon

3/24

/20

03Unkn

own

Immigration,peo

ple

trafficking

SaferHomes

AllLondon

10/25

/20

0210

/27

/20

02Burglary

MPAan

nual

reports:

Payback

AllLondon

9/20

03Hatecrim

eRainbow

AllLondon

Terrorism

Copernicos

AllLondon

High-valued

property

theft

Halifax

IVAllLondon

1/17

/20

052/

28/20

05Failto

appearwarrants

Bluesky

AllLondon

Immigration

Jigsaw

AllLondon

Sexoffen

ders

Anch

orage

2AllLondon

Violentcrim

eErica

AllLondon

Antisocial

beh

avior

Argon

AllLondon

9/20

071/

2008

Guncrim

ein

nightclubs

Other

sources:

Curb

AllLondon

6/20

073/

2008

Youth

violence

Kontiki

AllLondon

Human

trafficking

Sterling

AllLondon

Fraud

Evader

AllLondon

1191



TABLEA1(C

ontinued

)

Inform

ationSo

urcean

dOperationNam

eBorough

Start

End

Focu

s

D.Police

Operations;Inco

mplete

Inform

ation

MPAan

nual

reports:

Enver

12/19

/20

03Tam

ilcrim

inals

Tullibardine

Grafton

4/20

03CrimearoundHeathrow

Brigh

tStar

Antiterror

Amethyst

Childsexab

use

Nem

oDrugs

Van

adium

Drugs

Chicago

Buscrim

eBusTag

Busvandalism

Overt

Antiterror

Overamp

Antiterror

Other

sources:

Suki

Lateen

Violentcrim

e

Note.—AllMPAreportscanbeaccessed

athttp://www.policeau

thority.org.

1192



TABLEA2

TheEffectoftheDepenalizationonCannabisOffensesinAggregate:RobustnessChecks


Can

nab

isOffen

ses,per

1,00


Neigh

bors

Excluded

asControlBorough

sð1Þ

Year

Fixed

Effects

ð2Þ

DataReg

ime

Fixed

Effects

ð3Þ

Police

Operation

Controls

ð4Þ

SpatiallyCorrelated

Errors

ð5Þ

Lam

beth�

policy

period

.298

**.335

***

.349

***

.259

**.151

***

ð.117

Þð.1

05Þ

ð.103

Þð.1

12Þ

ð.028

ÞPolicy

period

.038

2.008

.066

.019

.001

ð.056

Þð.0

65Þ

ð.055

Þð.0

53Þ

ð.008

ÞLam

beth�

postpolicy

period

.606

***

.623

***

.636

***

.555

***

.253

***

ð.095

Þð.0

82Þ

ð.080

Þð.0

91Þ

ð.020

ÞPostpolicy

period

.185

***

.034

.072

.179

***

.052

***

ð.047

Þð.0

94Þ

ð.081

Þð.0

46Þ

ð.006

Þ

1193



TABLEA2(C

ontinued

)


Can

nab

isOffen

ses,per

1,00


Neigh

bors

Excluded

asControlBorough

sð1Þ

Year

Fixed

Effects

ð2Þ

DataReg

ime

Fixed

Effects

ð3Þ

Police

Operation

Controls

ð4Þ

SpatiallyCorrelated

Errors

ð5Þ

Borough

,month

fixe

deffects

Yes

Yes

Yes

Yes

Yes

Sociodem

ograp

hic

controls

Yes

Yes

Yes

Yes

Yes

Observations

2,63

23,00

83,00

83,00

83,00

8


theborough

-month-yearlevel.Thesampleperiodrunsfrom

April1

998untilJan

uary20

06.F

orallcolumnsex

ceptco

l.1,

controlborough

sareallother

Londonborough

s.In

col.1,

Lam

beth’s

neigh

bors

ðCroydon,Merton,So

uthwark,

andWan

dsworthÞa

reex

cluded

asco

ntrols.Pan

el-correctedstan

darderrors


insten


-specificARð1Þp

rocess

isassumed

.Thisalso

allowstheerrorterm

sto

beborough

specific,heterosked

astic,an

dco

ntemporaneo

uslyco

rrelated

acrossborough

s.Theex

ceptionisco

l.5,whereaspatial

errormodel

isestimated

.Thespatialweigh

tingmatrixusedhereis

aco

ntigu

itymatrix;

allneigh

bors

areallocatedones,an

dallnonneigh

bors

are

allocatedzeroes.W

ealso

experim

entedwithseveralo

ther

spatialw

eigh

tingmatrices,includinginversedistance

ðbetweenborough

centroidsÞan

dinverse

distance

squared

weigh

tingmatrices.Theresultsarerobustto

thesedifferentspatialerrorspecifications.Observationsareweigh

tedbytheshareofthe

totalLondonpopulationthat

month-yearin

theborough

.Theex

ceptionsagainareco

ls.1

and5.

Inco

l.1,

observationsareweigh

tedbytheshareofthe

ðnonneigh

boringborough

ÞtotalL

ondonpopulationthat

month-yearin

theborough

.Inco

l.5,

observationsarenotweigh

ted.T

hepolicy

perioddummy

variab

leiseq

ual

toonefrom

July20

01untilJuly20

02,andzero

otherwise.

Thepostpolicy

perioddummyvariab

leiseq

ual

toonefrom

July20

02onward,

andzero

otherwise.Thefollowingsociodem

ograp

hicco


logs,are

controlled

forat

theborough

-month-yearlevel:theshareof

thead

ultpopulationthat


isaged

20–2

4,25

–34,

35–4

9,an

dab

ove

50;andthemaleunem

ploym

entrate.D

ataregimefixe

deffects

allowforan

ych

ange

sin

thereco

rdingofthedatain

each

oftheseseparatetimeperiods,as

wellasadummyforthech

ange

incrim

ereco

rdingrulesfrom

April20

02onward.T

hepolice

operationco

ntrolvariab

lesareindicators

forwhether

theborough

was

partofarecentpolice

operation.O

perationsthat

targeted

agroupofspecificborough

sincludetheSaferStreetsInitiative

Phase1ð2/4/

2002

–3/31

/20

02Þa

ndPhase2ð4/15

/20

02–3

/31

/20

03Þ,O

peration

Recoverð10/

2005

–12/

17/20

07Þ,O

perationBlunt1ð11/

2004

–11/

2005

Þ,OperationSaferHomes

ð10/

28/20

02–6

/20

04Þ,a

ndOperationSo

lstice

ð12/

01/

2003

–12/

8/20

03Þ.Lam

bethwas

partofSaferStreetsPhase1an

d2an

dBlunt1.

Further

operationsðpartofalarger

operationnam

edStrongb

oxÞ

that

targeted

singleborough

sincludeOperationWindmillðL

ambeth:5/

8/19

99–7

/2/

1999

Þ,OperationEmpireðH

ackn

ey:7/

17/19

99–9

/10

/19

99Þ,Opera-

tionReg

isðC

amden

,Islington:10

/2/

1999

–12/

3/19

99Þ,OperationVictory

ðWestm

inster:1/

22/20

01–3

/18

/20

01Þ,OperationCastilleðH

aringe

y:4/

17/

2001

–6/10

/20

01Þ,OperationClaym

oorðB

rent:7/

16/20

01–9

/9/

2001

Þ,an

dOperationSabre

ðTower

Ham

lets:9/

17/20

01–1

2/9/

2001

Þ.*Sign

ificantat

10percent.

**Sign

ificantat

5percent.

***Sign

ificantat

1percent.

1194



TABLEA3

TheEffectoftheDepenalizationonPoliceEffortonNondrugCrime

CrimeType

Totalðw

ithout

DrugsÞ

ð1Þ

Violence

against

the

Person

ð2Þ

Sexu

alð3Þ

Robbery

ð4Þ

Burglary

ð5Þ

Theftan

dHan

dling

ð6Þ

Fraudor

Forgery

ð7Þ

Criminal

Dam

age

ð8Þ

A.Dep

enden

tVariable:LogðArrestRateforaGiven

CrimeCateg

oryÞ

Lam

beth�

policy

period

.065

.096

.158

.383

***

2.197

2.152

*.058

.024

ð.108

Þð.1

28Þ

ð.182

Þð.1

42Þ

ð.142

Þð.0

90Þ

ð.160

Þð.1

58Þ

Policy

period

2.101

*2.178

**2.164

***

2.242

***

.128

***

2.173

***

2.154

*2.168

***

ð.058

Þð.0

87Þ

ð.054

Þð.0

54Þ

ð.049

Þð.0

44Þ

ð.080

Þð.0

62Þ

Lam

beth�

postpolicy

period

.284

***

.344

***

.454

***

.417

***

.325

***

2.062

.567

***

.299

**ð.1

05Þ

ð.124

Þð.1

32Þ

ð.106

Þð.1

05Þ

ð.072

Þð.1

21Þ

ð.130

ÞPostpolicy

period

2.015

2.076

2.114

***

2.112

***

.185

***

2.209

***

2.056

2.033

ð.048

Þð.0

72Þ

ð.043

Þð.0

43Þ

ð.039

Þð.0

35Þ

ð.062

Þð.0

48Þ

Shareofallarrestsprepolicy

.861

.281

.016

.034

.086

.297

.049

.098

Borough

,month

fixe

deffects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Sociodem

ograp

hic

controls

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Observations

3,00

83,00

82,93

62,98

63,00

83,00

83,00

63,00

8

B.Dep

enden

tVariable:LogðClear-upRateforaGiven

CrimeCateg

oryÞ

Lam

beth�

policy

period

.028

.066

.161

.317

**2.192

2.119

.063

.131

ð.112

Þð.1

29Þ

ð.179

Þð.1

45Þ

ð.146

Þð.0

90Þ

ð.274

Þð.1

59Þ

Policy

period

2.073

2.159

*2.169

***

2.176

***

.154

***

2.154

***

.001

2.157

**ð.0

62Þ

ð.088

Þð.0

53Þ

ð.054

Þð.0

48Þ

ð.045

Þð.0

46Þ

ð.063

ÞLam

beth�

postpolicy

period

.270

**.319

**.484

***

.436

***

.314

***

2.077

.554

***

.305

**ð.1

15Þ

ð.128

Þð.1

31Þ

ð.109

Þð.1

09Þ

ð.072

Þð.1

94Þ

ð.134

ÞPostpolicy

period

.067

2.022

2.094

**2.039

.242

***

2.145

***

.396

***

.026

ð.052

Þð.0

73Þ

ð.042

Þð.0

42Þ

ð.038

Þð.0

36Þ

ð.041

Þð.0

49Þ

Shareofallclear-upsprepolicy

.846

.311

.019

.029

.084

.293

.007

.104

Borough

,month

fixe

deffects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Sociodem

ograp

hic

controls

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Observations

3,00

83,00

82,93

42,98

03,00

73,00

82,63

03,00

8

1195



TABLEA3(C

ontinued

) CrimeType

Totalðw

ithout

DrugsÞ

ð1Þ

Violence

against

the

Person

ð2Þ

Sexu

alð3Þ

Robbery

ð4Þ

Burglary

ð5Þ

Theftan

dHan

dling

ð6Þ

Fraudor

Forgery

ð7Þ

Criminal

Dam

age

ð8Þ

C.Dep

enden

tVariable:LogðClear-upper

ArrestÞ

Lam

beth�

policy

period

.018

.021

*2.006

2.037

.012

.030

**2.027

.057

*ð.0

14Þ

ð.012

Þð.0

38Þ

ð.070

Þð.0

39Þ

ð.015

Þð.1

45Þ

ð.031

ÞPolicy

period

.023

***

.022

***

.002

.072

***

.029

**.018

***

.194

***

.027

***

ð.008

Þð.0

06Þ

ð.012

Þð.0

21Þ

ð.015

Þð.0

07Þ

ð.054

Þð.0

08Þ

Lam

beth�

postpolicy

period

.010

.006

.015

.014

2.014

2.020

*2.019

.025

ð.010

Þð.0

09Þ

ð.028

Þð.0

51Þ

ð.028

Þð.0

11Þ

ð.099

Þð.0

22Þ

Postpolicy

period

.081

***

.066

***

.030

***

.088

***

.056

***

.064

***

.465

***

.061

***

ð.006

Þð.0

05Þ

ð.010

Þð.0

17Þ

ð.012

Þð.0

05Þ

ð.039

Þð.0

07Þ

Shareofallclear-upsprepolicy

.846

.311

.019

.029

.084

.293

.007

.104

Borough

,month

fixe

deffects

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Sociodem

ograp

hic

controls

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Observations

3,00

83,00

83,00

23,00

23,00

73,00

82,63

23,00

8


theborough


periodrunsfrom

April19

98untilJanuary20

06.Controlborough

sareall

other

Londonborough

s.In

pan

elAthedep

enden

tvariab


ofarrestsdivided

bythenumber

ofoffen

sesin

theborough

inthe

samemonth

andpreviousquarter,foreach

crim

etype.

Inpan

elBthedep

enden

tvariab


ofclear-upsdivided

bythenumber

of

offen

sesin

theborough

inthesamemonth

andpreviousquarter,foreach

crim

etype.In

pan

elCthedep

enden

tvariab


ofclear-

upsdivided

bythenumber

ofarrestsin

theborough

,inthegivenmonth.P

anel-correctedstan

darderrorsarecalculatedusingaPrais-W

insten

regression,

whereaborough

-specificARð1Þp

rocess

isassumed

.Thisalso

allowstheerrorterm

sto

beborough

specific,

heterosked

astic,

andco

ntemporaneo

usly

correlated

across

borough



month-yearin

theborough

.Thepolicy

period

dummyvariab

leiseq

ual

toonefrom

July20

01untilJuly20

02,andzero

otherwise.

Thepostpolicy

perioddummyvariab

leiseq

ual

toonefrom

July20

02onward,andzero

otherwise.Thefollowingsociodem

ograp

hicco


logs,are

controlled

forat

theborough

-month-yearlevel:the

shareofthead

ultpopulationthat


isaged

20–2

4,25

–34,

35–49

,andab

ove

50;andthemaleunem

ploym

entrate.E

achpan

elshows

theproportionofallarrestsan

dclear-upsðdrugan

dnon-drug-relatedÞthat

each

catego

rymakes

upin

theprepolicy

periodin

Lam

bethfrom

April19

98untilJune20

01.

*Sign

ificantat

10percent.

**Sign

ificantat

5percent.

***Sign

ificantat

1percent.

1196



TABLEA4

RobustnessChecksontheEffectofDepenalizingCannabisonHousePrices



Price,Defl

ated

to19

95:Q

1PricesÞ

Zip

CodeSe

ctorHotSp

otDefi

nition

Flats

ð1Þ

WardHotSp

ot

Defi

nition:

75th

Percentile

ð2Þ

Modal

Ward

ð3Þ

Weigh

tedAverage

ofWards

ð4Þ

HotSp

ots

Based

on

TotalCrime

ð5Þ

Lam

beth�

policy

period

.011

2.013

2.032

2.027

.010

ð.022

Þð.0

23Þ

ð.024

Þð.0

27Þ

ð.020

ÞPolicy

period

2.050

***

2.044

***

2.033

***

2.041

***

2.057

***

ð.015

Þð.0

07Þ

ð.008

Þð.0

08Þ

ð.011

ÞLam

beth�

postpolicy

period

.070

**2.083

**2.099

***

2.064

2.010

ð.027

Þð.0

41Þ

ð.038

Þð.0

39Þ

ð.028

ÞPostpolicy

period

2.155

***

2.079

***

2.068

***

2.091

***

2.107

***

ð.016

Þð.0

12Þ

ð.012

Þð2

.013

Þð.0

12Þ

Lam

beth�

hotspot

.006

2.056

*2.126

***

2.296

***

2.086

*ð.0

39Þ

ð.029

Þð.0

27Þ

ð.081

Þð.0

44Þ

Hotspot

2.058

**2.091

*2.045

**2.001

2.001

ð.027

Þð.0

54Þ

ð.018

Þð.2

12Þ

ð.014

ÞLam

beth�

policy

period�

hotspot

2.033

2.017

.011

2.002

2.044

**ð.0

30Þ

ð.025

Þð.0

23Þ

ð.028

Þð.0

18Þ

Policy

period�

hotspot

.031

**.031

***

.016

**.031

***

.035

***

ð.015

Þð.0

08Þ

ð.008

Þð.0

10Þ

ð.012

ÞLam

beth�

postpolicy

period�

hotspot

2.199

***

2.070

***

2.077

***

2.139

***

2.133

***

ð.030

Þð.0

25Þ

ð.022

Þð.0

26Þ

ð.018

Þ

1197



TABLEA4(C

ontinued

)



Price,Defl

ated

to19

95:Q

1PricesÞ

Zip

CodeSe

ctorHotSp

otDefi

nition

Flats

ð1Þ

WardHotSp

ot

Defi

nition:

75th

Percentile

ð2Þ

Modal

Ward

ð3Þ

Weigh

tedAverage

ofWards

ð4Þ

HotSp

ots

Based

on

TotalCrime

ð5Þ

Postpolicy

period�

hotspot

.080

***

.051

***

.043

***

.085

***

.066

***

ð.016

Þð.0

10Þ

ð.011

Þð.0

13Þ

ð.011

ÞZip

codean

dquarterfixe

deffects

Yes

Yes

Yes

Yes

Yes

Borough

-specificlineartimetren

dYes

Yes

Yes

Yes

Yes

Sociodem

ograp

hic

controls

Yes

Yes

Yes

Yes

Yes

Observations

20,706

17,331

17,331

17,331

17,331


thezipcodesector-quarter-year

level.House

pricesaredefl

ated

tothefirstquarterof19

95prices,usingtheLan

dRegistry

house

price

index

forGreater

London,whichisbased

onrepeatsales.More

inform

ationontheindex

canbefoundat

http://www1.landregistry.gov.uk

/houseprices/housepriceindex/.Fo

rallspecifications,thesample

runsfrom

January19

95untilDecem

ber

2005

.In

col.1ðco

ls.2–5Þ,observationsare

weigh

tedbythenumbersofsales

forflatsðte

rraced

housingÞ

inthatquarter-year

inthespecificzipcodesector.Stan

darderrorsareclustered

byzipcodesector

through

out.Toreflectthelagbetweenthehouse

buyingdecisionan

dtherecorded

saleofthehouse,alltime-varyingexplanatory

variab

lesarelagg

edbyone

quarter.Theðone-quarter-lagg

edÞp

olicy

perioddummyvariab

leiseq

ualto

onefrom

thefourthquarterðst

artsOctober

1Þof20

01untilthethirdquarterof

2002

ðendsSeptember

30Þ,an

dzero

otherwise.

Theðone-quarter-lagg

edÞp

ostpolicy

perioddummyvariab

leiseq

ualto

onefrom

thefourthquarterof20

02onward,an

dzero

otherwise.


ograp

hic

controlvariab

les,measuredin

logs,arecontrolled

forat

theborough

-month-yearlevel:the

shareofthead

ultpopulationthat

isethnic

minority;that

isaged

20–24,

25–34,

35–49

,an

dab

ove

50;an

dthemaleunem

ploym

entrate.Allofthese

socioeconomicvariab

lesarelagg

edonequarter.Wealso

controlforfixedeffectsforzipcodean

dquarterthrough

outandaborough

-specificlineartimetren

d.

Incol.2,wedefi

neawardas

ahotspotifd

rugoffen

sesareab

ove

the75

thpercentilemed

ianforallw

ardsin

theborough

.Wethen

defi

neazipcodeto

beahot

spotifitcontainsanyhotspotw

ardssodefi

ned

.Incol.3,thezipcodesectorisdefi

ned

tobeahotspotifthemodalwardisitselfdefi

ned

tobeahotspot.In

col.4,

thehotspotvariab

leisnolongerbinary,butrather

aweigh

tedaverageofallw

ards’hotspotclassificationswithin

thezipcodesector.Theseweigh

tsarebased

onthepercentage

ofthezipcodethat

overlap

swiththeward.Finally,col.5usesinform

ationontotalcrim

esðnotdrugcrim

eÞto

redefi

newardsan

dthen

zipcodes

ashotspotsusingotherwisethesamemethodas

thebaselinespecification.

*Sign

ificantat

10percent.

**Sign

ificantat

5percent.

***Sign

ificantat

1percent.



References

Arseneault, Louise, Mary Cannon, JohnWitten, and RobinMurray. 2004. “CausalAssociation between Cannabis and Psychosis: Examination of the Evidence.”British J. Psychiatry 184:110–17.

Becker, Gary S., Kevin M. Murphy, andMichael Grossman. 2006. “TheMarket forIllegal Goods: The Case of Drugs.” J.P.E. 114:38–60.

Black, Sandra E. 1999. “Do Better Schools Matter? Parental Valuation of Ele-mentary Education.” Q.J.E. 114:577–99.

Braakmann, Nils, and Simon Jones. 2012. “Cannabis Consumption, Crime andVictimization—Evidence from the 2004 Cannabis Declassification in the UK.”Manuscript, Newcastle Univ.

Cameron, A. Colin, Jonah B. Gelbach, and Douglas L. Miller. 2008. “Bootstrap-Based Improvements for Inference with Clustered Errors.”Rev. Econ. and Statis.90:414–27.

Christie, Paul, and Robert Ali. 2000. “Offences under the Cannabis ExpiationNotice Scheme in South Australia.” Drug and Alcohol Rev. 19:251–56.

Conley, John P., and Ping Wang. 2006. “Crime and Ethics.” J. Urban Econ. 60:107–23.

TABLE A5Calibrated Parameters

Notation LocationCalibratedParameter

A. Direct Depenalization Policy Channels

Penalty reduction during policy apolicy Lambeth .178Reduction in police hours duringpolicy for cannabis crime r1 Lambeth .530

B. Preference Parameters

Disutility of cannabis consumption �d1 Lambeth .799�d0 Rest of London .823

Disutility of committing nondrugcrime �x1 Lambeth .956

�x0 Rest of London .955Maximum mobility cost �l All .753

C. Policing Technology

Apprehension technology for cannabiscrime gD;1 Lambeth .0127

gD;0 Rest of London .0191Apprehension technology for nondrugcrime gC All .218

Cobb-Douglas parameter, cannabiscrime arrests qD All .270

Cobb-Douglas parameter, nondrugcrime arrests

qC All .356

D. Other

Penalty of arrest, cannabis crime a1 All 21Penalty of arrest, nondrug crime b All .229




Conlin, Michael, Stacey Dickert-Conlin, and John Pepper. 2005. “The Effects ofAlcohol Prohibitions on Illicit Drug Related Crimes.” J. Law and Econ. 48:215–34.

Corman, Hope, and H. Naci Mocan. 2000. “A Time-Series Analysis of Crime,Deterrence, and Drug Abuse in New York City.” A.E.R. 90:584–604.

Damrongplasit, Kannika, Cheng Hsiao, and Xueyan Zhao. 2010. “Decriminal-ization and Marijuana Smoking Prevalence: Evidence from Australia.” J. Bus.and Econ. Statis. 28:344–59.

Dark, Stuart, and Michael Fuller. 2002. “The Lambeth Cannabis Warning PilotScheme.” Report 17, Metropolitan Police Authority, London.

Davis, Lucas W. 2004. “The Effect of Health Risk on Housing Values: Evidencefrom a Cancer Cluster.” A.E.R. 94:1693–1704.

Department of Health and Human Services. 2008. Substance Abuse and MentalHealth Services Administration: Results from the 2007 National Survey on Drug Useand Health: National Findings . National Survey on Drug Use and Health, SeriesH-34. Publication no. SMA 08-4343. Rockville, MD: Office Appl. Studies, Dept.Health and Human Services.

DiNardo, John, and Thomas Lemieux. 2001. “Alcohol, Marijuana, and AmericanYouth: The Unintended Consequences of Government Regulation.” J. HealthEcon. 20:991–1010.

Donohue, John J., Benjamin Ewing, and David Peloquin. 2012. “RethinkingAmerica’s Illegal Drug Policy.” In Controlling Crime: Strategies and Tradeoffs , editedby Philip J. Cook, Jens Ludwig, and Justin McCrary. NBER Conference Report.Chicago: Univ. Chicago Press.

Draca, Mirko, Stephen Machin, and Robert Witt. 2011. “Panic on the Streets ofLondon: Police, Crime, and the July 2005 Terror Attacks.” A.E.R. 101:2157–81.

Durlauf, Steven N., and Daniel S. Nagin. 2012. “The Deterrent Effect of Impris-onment.” In Conrolling Crime: Strategies and Tradeoffs , edited by Philip J. Cook,Jens Ludwig, and Justin McCrary. NBER Conference Report. Chicago: Univ.Chicago Press.

Fergusson, David M., and L. John Horwood. 1997. “Early Onset Cannabis Useand Psychosocial Adjustment in Young Adults.” Addiction 92:279–96.

Fu, Chao, and Kenneth I. Wolpin. 2013. “Structural Estimation of a Becker-Ehrlich EquilibriumModel of Crime: Allocating Police across Cities to ReduceCrime.” Manuscript, Univ. Pennsylvania.

Galenianos, Manolis, Rosalie L. Pacula, and Nicola Persico. 2012. “A Search-Theoretic Model of the Retail Market for Illicit Drugs.” Rev. Econ. Studies 79:1239–69.

Gibbons, Steve. 2004. “The Costs of Urban Property Crime.” Econ. J. 114:F441–F463.

Gibbons, Steve, and Stephen Machin. 2003. “Valuing English Primary Schools.”J. Urban Econ. 53:197–219.

Gibbons, Steve, Stephen Machin, and Olmo Silva. 2013. “Valuing School QualityUsing Boundary Discontinuities.” J. Urban Econ. 75:15–28.

Glaeser, Edward L., and Andrei Shleifer. 2001. “A Reason for Quantity Regula-tion.” A.E.R. Papers and Proc. 91:431–35.

Grogger, Jeff, and Mike Willis. 2000. “The Emergence of Crack Cocaine and theRise in Urban Crime Rates.” Rev. Econ. and Statis. 82:519–29.

Herbert, Nick, Oscar Keeble, Aidan Burley, and Blair Gibbs. 2007. Policing for thePeople . Interim Report. London: Police Reform Taskforce.

HomeOffice. 2005.Crime in England andWales 2004/2005 . London:HomeOffice.———. 2007. Crime in England and Wales 2006/7 . London: Home Office.




Hughes, Caitlin E., and Alex Stevens. 2010. “What Can We Learn from thePortuguese Decriminalization of Illicit Drugs?” British J. Criminology 50:999–1022.

Imrohoroglu, Ayse, Antonio Merlo, and Peter Rupert. 2004. “What Accounts forthe Decline in Crime?” Internat. Econ. Rev. 45:811–43.

Kelly, Elaine, and Imran Rasul. 2014. “Policing Cannabis and Drug Related Hos-pital Admissions: Evidence from Administrative Records.” J. Public Econ. 112:89–114.

Kuziemko, Ilyana, and Steven D. Levitt. 2004. “An Empirical Analysis of Im-prisoning Drug Offenders.” J. Public Econ. 88:2043–66.

Levitt, Steven D. 1997. “Using Electoral Cycles in Police Hiring to Estimate theEffect of Police on Crime.” A.E.R. 87:270–90.

———. 2002. “Using Electoral Cycles in Police Hiring to Estimate the Effect ofPolice on Crime: Reply.” A.E.R. 92:1244–50.

Levitt, Steven D., and Sudhir A. Venkatesh. 2000. “An Economic Analysis of aDrug-Selling Gang’s Finances.” Q.J.E. 115:755–89.

Lin, Ming-Jen. 2009. “More Police, Less Crime: Evidence from US State Data.”Internat. Rev. Law and Econ. 29:73–80.

Linden, Leigh L., and Jonah E. Rockoff. 2008. “Estimates of the Impact of CrimeRisk on Property Values from Megan’s Laws.” A.E.R. 98:1103–27.

MacCoun, Robert J., and Peter Reuter. 2001. Drug War Heresies: Learning fromOther Vices, Times and Places . Cambridge: Cambridge Univ. Press.

———. 2005. “Does Europe Do It Better?” In Drug War Deadlock: The Policy BattleContinues , edited by L. E. Huggins. Stanford, CA: Stanford Univ. Press.

May, Tiggey, Stefano Cossalter, Isabella Boyce, and Ian Earnden. 2007. Drug Deal-ing in Brixton Town Centre . London: Inst. Criminal Policy Res., King’s College.

May, Tiggey, Martin Duffy, Hamish Warburton, and Mike Hough. 2007. PolicingCannabis as a Class C Drug: An Arresting Change? Report. York: Joseph RowntreeFound.

May, Tiggey, Hamish Warburton, Paul J. Turnbull, and Mike Hough. 2002. TimesThey Are A-Changing: Policing of Cannabis . Report. York: Joseph Rowntree Found.

McCrary, Justin. 2002. “Using Electoral Cycles in Police Hiring to Estimate theEffect of Police on Crime: Comment.” A.E.R. 92:1236–43.

Miron, Jeffrey A. 2010. “The Budgetary Implications of Drug Prohibition.” Man-uscript, Harvard Univ.

Miron, Jeffrey A., and Jeffrey Zweibel. 1995. “The Economic Case against DrugProhibition.” J. Econ. Perspectives 9 (Autumn): 175–92.

Model, Karyn E. 1993. “The Effect of Marijuana Decriminalization on HospitalEmergency Room Drug Episodes: 1975–1978.” J. American Statis. Assoc. 88:737–47.

National Research Council. 2001. Informing America’s Policy on Illegal Drugs: WhatWe Don’t Know Keeps Hurting Us . Washington, DC: Nat. Acad. Press.

Pacula, Rosalie L. 1998. “Does Increasing the Beer Tax Reduce Marijuana Con-sumption?” J. Health Econ. 17:557–85.

Pacula, Rosalie L., Jamie F. Chriqui, and Joanna King. 2003. “Marijuana De-criminalization: What Does It Mean in the United States?” Working Paper no. 9690,NBER, Cambridge, MA.

Pacula, Rosalie L., and Beau Kilmer. 2003. “Marijuana and Crime: Is Therea Connection beyond Prohibition?” Working Paper no. 10046, NBER, Cam-bridge, MA.

Pudney, Stephen E. 2010. “Drugs Policy: What Should We Do about Cannabis?”Econ. Policy 25:165–211.




Resignato, Andrew J. 2000. “Violent Crime: A Function of Drug Use or DrugEnforcement?” Appl. Econ. 32:681–88.

Rosen, Sherwin. 1974. “Hedonic Prices and Implicit Markets: Product Differ-entiation in Pure Competition.” J.P.E. 82:34–55.

Single, Eric W. 1989. “The Impact of Marijuana Decriminalization: An Update.”Public Health Policy 10:456–66.

Thaler, Richard. 1978. “A Note on the Value of Crime Control: Evidence fromthe Property Market.” J. Urban Econ. 5:137–45.

UNOffice onDrugs and Crime. 2010.World Drug Report . Vienna: United Nations.Van Ours, Jan. 2003. “Is Cannabis a Stepping-Stone for Cocaine?” J. Health Econ.22:539–54.

Van Ours, Jan, and Jenny Williams. 2009. “Why Parents Worry: Initiation intoCannabis Use by Youth and Their Educational Attainment.” J. Health Econ.28:132–42.

Warburton, Hamish, Tiggey May, and Mike Hough. 2005. “Looking the OtherWay.” British J. Criminology 45:113–28.

Williams, Jenny. 2004. “The Effects of Price and Policy on Marijuana Use: WhatCan Be Learned from the Australian Experience?” Health Econ. 13:123–37.

Wood, K. 2004. “Evaluation Report Following the Reclassification of Cannabis.”Report 14, Metropolitan Police Authority, London.




Documents

Crime and the Depenalization of Cannabis Possession: Evidence from a Policing Experiment